JP2022087585A - Mechanical learning device, prediction device, set number control parameter output device, and control device - Google Patents

Abstract

To create a prediction model for predicting a maximum vapor content and an average vapor use amount.SOLUTION: A mechanical learning device 7 learns a prediction model 75 for predicting a maximum vapor use amount and an average vapor use amount of a boiler group 2 at a second-time ahead unit time from time-series data at each unit time of an arbitrary first time on the basis of first-time band time-series data being time-series data of a vapor content including a maximum vapor use amount at each unit time which is included in a first time band and an average vapor use amount, which are calculated on the basis of time-series data on a sampling order including measurement data including an entire output vapor content of a boiler group 2 of a boiler system 1 having the boiler group 2 which is constituted of a plurality of boilers 20, and a vapor content including a maximum vapor use amount at a second-time ahead unit time immediately after the first time.SELECTED DRAWING: Figure 1

Description

The present invention relates to a machine learning device, a prediction device, a number control parameter output device, and a control device related to a boiler system including a boiler group having a plurality of boilers.

Conventionally, the combustion state of the boiler group is controlled based on the boiler group having a plurality of boilers, the steam header that collects the steam generated in the boiler group, and the header pressure value which is the steam pressure inside the steam header. A boiler system has been proposed.
Conventionally, as a boiler, a step value control boiler capable of increasing or decreasing the combustion amount stepwise according to a selected combustion position and a continuous control boiler capable of continuously increasing or decreasing the combustion amount are known.
Here, the stage value control boiler is a multi-position control boiler that burns at multiple (3 or more) stages of combustion positions by selectively turning on / off combustion, adjusting the size of the flame, and the like. Say. In such a step value control boiler, the combustion rate is changed step by step. The multi-position control means that the combustion amount of the step value control boiler can be controlled stepwise to a plurality of positions including the combustion stop position.
Then, as described in Patent Document 1, in the combustion control of a boiler system including a boiler group having a plurality of stage value control boilers, a high efficiency operation point indicating a combustion position where the boiler efficiency becomes high efficiency (hereinafter, "" It is known to control each boiler to burn at the "eco-driving point" as much as possible, aiming at the "eco-driving point").

For example, the combustion control described in Patent Document 1 includes a maximum evaporation amount as the maximum steam amount used, a first evaporation amount as an average steam usage amount, a second evaporation amount as a variable steam amount, and the number of boilers to be operated. And it is premised that the combustion priority is set in advance. On top of that, for example, if you want to secure a long burning time at the "eco-driving point" throughout the day, consider the daily operating conditions and set the first steam amount as the average steam usage to the high-efficiency combustion position. It is disclosed that the number of boilers to be operated is set in advance and the combustion priority is set so as to burn at the medium combustion position.

On the other hand, in a boiler system composed of a boiler group consisting of one or more continuous control boilers, when controlling the combustion state of each continuous control boiler, a target steam pressure value is set in advance and the header pressure value is set as the target steam. It is known to control the number of boilers to be controlled by calculating the control amount according to the deviation between the header pressure value and the target steam pressure value so as to keep the pressure value.
As described in Patent Document 2, the continuous control boiler can continuously change the combustion rate in the range from the minimum combustion state S1 to the maximum combustion state S2, but the boiler efficiency (continuous control) depends on the combustion rate. Boiler thermal efficiency) is different. Therefore, for example, the combustion rate at which the boiler efficiency is the highest (for example, 98%) is set as the eco-driving point, and the range of the combustion rate at which the boiler efficiency is higher than the predetermined value (for example, 97%) is set as the eco-driving point. Set as a zone. For example, if the boiler efficiency is highest when the combustion rate of the continuous control boiler is 50%, the eco-driving point is the combustion rate of 50%, and when the combustion rate of the continuous control boiler is in the range of 30% to 70%, the boiler When the efficiency becomes a combustion rate higher than a predetermined value, the eco-driving zone is in the range of the combustion rate of 30% to 70%.
Even in the combustion control of a boiler system including a boiler group having a plurality of continuously controlled boilers, even if the combustion rate of the continuous control boiler decreases, the combustion rate of each continuous control boiler is within the range of the eco-driving zone. It is desired to control each continuous control boiler so as to burn within the range of the eco-driving zone as much as possible by setting the predetermined amount of steam so as to be.

Japanese Unexamined Patent Publication No. 2012-017940 Japanese Unexamined Patent Publication No. 2016-176679

However, it is difficult to actually grasp the average daily steam amount, and when the steam load pattern changes depending on the day of the week, holiday, etc. such as month, season, or quarter, the number of boilers to be operated is changed each time. It is difficult to adjust the setting and the setting of the combustion priority. For many users, first set the average daily steam amount, variable steam amount, and maximum steam amount used, and then the subsequent operations are operated so as to control the number of units based on these parameters. Is.
In order to improve the stability of the output steam amount while keeping the boiler efficiency high, for example, even if the steam load pattern differs depending on the day, holiday, etc. such as month, season, or quarter, for example, the immediately preceding time zone. Based on the history of the total output steam amount by the boiler group in (multiple hours), the average steam usage amount, the variable steam amount, and the maximum steam usage amount from immediately after the relevant time zone to a predetermined time ahead (for example, one hour ahead) are calculated. A predictable boiler system is desired.

In the present invention, in a boiler system including a boiler group having a plurality of boilers, even if the steam load pattern differs depending on the day, holiday, etc. of the month, season, quarter, etc., for example, the immediately preceding time zone (plural hours). To provide a machine learning device capable of creating a prediction model for predicting the maximum steam usage and the average steam usage in a predetermined time based on the history of the total output steam amount by the boiler group in the boiler group. The purpose.

The present invention is a sampling time and a sampling time output from a boiler system having a boiler group having a plurality of boilers and a control device for controlling the combustion state of the boiler group for each preset sampling cycle. At least the maximum steam usage of the boiler group for each predetermined unit time and calculated based on the measurement data including at least the total output steam amount of the boiler group in the above and the time series data of the sampling order for each working day including the above. From the set of time series data of unit time steam amount including average steam usage amount, the first time zone time series data which is a subset of the time series data of the unit time steam amount included in the preset first time zone. And by extracting the steam amount of the unit time of the second time ahead preset from immediately after the first time zone from the set of time series data of the unit time steam amount, the first time zone. Based on the teacher data creation unit that creates teacher data including time-series data and the amount of steam in the unit time of the second hour ahead, and the teacher data created by the teacher data creation unit, the first on any working day. From the time series data of the unit time steam amount included in one time zone, the unit time including at least the maximum steam usage amount and the average steam usage amount of the boiler group in the unit time immediately after the first time zone and the second hour ahead. The present invention relates to a machine learning device including a learning unit that generates a prediction model for predicting the amount of steam.

Further, the working day may be associated with a preset steam load pattern, and the learning unit may generate the prediction model corresponding to the steam load pattern.

Further, the learning unit uses the first time zone time-series data as input data, and based on the teacher data in which the steam amount in the unit time from immediately after the first time zone to the second time ahead is used as label data. Supervised learning may be performed to generate a trained model as the predictive model.

Further, the machine learning device may be composed of one or more servers communicably connected to the control device via a network.

Further, the present invention acquires the prediction model generated by the machine learning device and the first time zone time series data immediately before an arbitrary time on the operating day of the boiler system for each unit time, and obtains the prediction model. The predicted value of the steam amount in the unit time of the second hour ahead from the time is acquired for each unit time, and the predicted value of the steam amount of the unit time of the second hour ahead is acquired for each unit time. A prediction including a predictor for predicting at least the maximum steam usage and the average steam usage of the boiler group in a time zone from a predetermined time on the working day of the boiler system to a second hour ahead based on the set. Regarding the device.

Further, the boiler group included in the boiler system is a boiler group having a plurality of stage value control boilers capable of controlling combustion at a plurality of staged combustion positions, and the first from the predetermined time predicted by the prediction device. Two or more of the maximum steam usage, average steam usage, and variable steam usage of the boiler system in the time zone based on the maximum steam usage and the average steam usage in the time zone up to two hours ahead. The present invention relates to a number control parameter output device that calculates a value and outputs the value to the control device of the boiler system.

Further, the present invention is based on any two or more values of the maximum steam amount used, the average steam amount, and the variable steam amount acquired from the number control parameter output device, from the predetermined time to the second hour ahead. The present invention relates to a control device for a boiler system, which sets an appropriate number control parameter in the time zone of the above and burns and controls the boiler group based on the set number control parameter.

Further, in the present invention, the boiler group included in the boiler system is a boiler group having a plurality of stage value control boilers capable of controlling combustion at a plurality of stage combustion positions, and the predetermined boiler group predicted by the prediction device. The most suitable number control parameter group ID in the time zone is selected based on the maximum steam usage amount and the average steam usage amount in the time zone from the time of The present invention relates to a number control parameter output device including a number control parameter output unit for output.

Further, in the present invention, an appropriate number control parameter in the time zone from the predetermined time to the second hour ahead is set and set based on the number control parameter group ID acquired from the number control parameter output device. The present invention relates to a control device for a boiler system that controls combustion of the boiler group based on the number control parameter.

According to the present invention, the machine learning device is, for example, in a boiler system including a boiler group having a plurality of boilers, even if the steam load pattern differs depending on the day, holiday, etc. such as month, season, or quarter. Based on the history of the total output steam amount by the boiler group in the immediately preceding time zone (multiple hours), it is possible to create a prediction model for predicting the maximum steam usage amount and the average steam usage amount at a predetermined time ahead. ..

It is a figure which shows the outline of the boiler system with a prediction function which concerns on one Embodiment of this invention. It is a functional block diagram which shows the structure of the number control apparatus of the said embodiment. It is a functional block diagram which shows the structure of the time series data storage apparatus of the said embodiment. It is a figure which shows the outline of the 1st time zone time series data extracted by the time series data storage apparatus of the said embodiment. It is a functional block diagram which shows the structure of the machine learning apparatus of the said embodiment. It is a figure which shows the outline of the teacher data in the machine learning apparatus of the said embodiment. It is a figure which shows an example of the prediction model (learned model) provided to the prediction device. It is a figure which illustrates the input / output processing situation in the prediction apparatus of the said embodiment. It is a functional block diagram which shows the structure of the prediction apparatus of the said embodiment. It is a functional block diagram which shows the structure of the number control parameter output device of the said embodiment. It is a figure which showed an example of the operation of the prediction device and the number control parameter output device in the operation phase of the said embodiment.

Hereinafter, a preferred embodiment of the boiler system with a predictive function of the present invention will be described with reference to the drawings. FIG. 1 is a diagram showing an outline of a boiler system 100 with a prediction function according to the present embodiment. As shown in FIG. 1, the boiler system 100 with a prediction function includes a boiler system 1, a time series data storage device 6, a machine learning device 7, a prediction device 8, and a number control parameter output device 9.
The time-series data storage device 6, the machine learning device 7, the prediction device 8, and the number control parameter output device 9 are arithmetic processing units such as a CPU (Central Processing Unit) in order to realize the operation of each predetermined function. In addition to being equipped with, auxiliary storage devices (not shown) such as ROM (Read Only Memory) and HDD that store various control programs, and data temporarily required for the arithmetic processing unit to execute the programs are stored. It is provided with a main storage device (not shown) such as a RAM for using the data. Then, the arithmetic processing device reads the OS and application software from the auxiliary storage device, and while deploying the read OS and application software to the main storage device, performs arithmetic processing based on these OS and application software. Based on this calculation result, each device controls each hardware. As a result, the functions of each device can be realized by the cooperation of hardware and software.
In the following description, a step value control boiler will be illustrated as a boiler. However, the number control device 3 is replaced with the number control device 3A that controls the continuous control boiler, and the time series data storage device 6, the machine learning device 7, and the prediction device 8 are directly applied to the continuous control boiler. By doing so, the boiler system 100 with a prediction function can also be applied to the boiler system 1 including a boiler group having a plurality of continuous control boilers.
Next, the boiler system 1 will be described.

As shown in FIG. 1, the boiler system 1 includes a boiler group 2 including a plurality of stage value control boilers 20, a steam header 4 for collecting steam generated in the plurality of stage value control boilers 20, and a steam header 4. It is provided with a steam pressure sensor 5 for measuring the internal pressure (hereinafter, also referred to as “header pressure”), and a number control device 3 having a control unit 30 for controlling the combustion state of the boiler group 2.
The boiler group 2 generates steam to be supplied to the steam-using equipment 18 as a load device.

The number control device 3 may be directly connected to, for example, the time series data storage device 6 and the number control parameter output device 9 via a connection interface (not shown). Further, the number control device 3 may be connected to the time series data storage device 6 and the number control parameter output device 9 via a network (not shown) such as a LAN (Local Area Network) or the Internet. In this case, the number control device 3, the time series data storage device 6, and the number control parameter output device 9 include a communication unit (not shown) for communicating with each other by such a connection. The machine learning device 7 is also communicably connected to the time-series data storage device 6 and the number control parameter output device 9. As will be described later, the number control device 3 may include any one of the time series data storage device 6, the machine learning device 7, and the number control parameter output device 9.
Next, the boiler system 1 will be briefly described.

Each of the plurality of stage value control boilers 20 includes a boiler main body 21 in which combustion is performed, and a local control unit 22 for controlling the combustion position of the stage value control boiler 20.
In the present embodiment, as the step value control boiler 20, 1) combustion stop position (first combustion position: 0%),
2) Low combustion position L (second combustion position: set at, for example, 5 to 35% of the maximum combustion amount, 20% in this embodiment),
3) Medium combustion position M (third combustion position: set at, for example, 40 to 60% of the maximum combustion amount, 40% in this embodiment),
4) An example is a step value control boiler (hereinafter, also referred to as "four position control boiler") having a stepwise combustion position of a high combustion position H (fourth combustion position: 100% (maximum combustion amount)).
The stage value control boiler 20 of the present embodiment is a boiler having a characteristic that the medium combustion position has high operating efficiency. That is, in each stage value control boiler 20, the medium combustion position is set as the high efficiency combustion position.

Priority is set for each of the plurality of stage value control boilers 20. The priority order is used to select the stage value control boiler 20 that gives a combustion instruction or a combustion stop instruction. The priority can be set, for example, by using an integer value so that the smaller the numerical value, the higher the priority. Note that this priority may be changed at predetermined time intervals (for example, 24 hour intervals) under the control of the control unit 30 described later.
The step value control boiler 20 is not limited to the 4-position boiler. It may be any multi-position boiler.

The boiler main body 21 is provided with a water pipe and a burner, and heats canned water supplied from a water source (water supply tank) (not shown) in the water pipe to generate steam.

The local control unit 22 changes the combustion position of the step value control boiler 20 according to the amount of steam consumed. Specifically, the local control unit 22 controls the combustion position of the stage value control boiler 20 based on the number control signal transmitted from the number control device 3 via the signal line 16. Further, the local control unit 22 transmits the signal used by the number control device 3 to the number control device 3 via the signal line 16. Examples of the signal used in the number control device 3 include the actual combustion position of the stage value control boiler 20, the output steam amount, and other data.

The steam header 4 is connected to a plurality of stage value control boilers 20 constituting the boiler group 2 via a steam pipe 11. The downstream side of the steam header 4 is connected to the steam use facility 18 via the steam pipe 12.

The steam header 4 collects and stores the steam generated in the boiler group 2. The steam header 4 adjusts the mutual pressure difference and pressure fluctuation of one or a plurality of step value control boilers 20 to be burned, and supplies steam having a constant steam pressure value to the steam use facility 18.

The vapor pressure sensor 5 is electrically connected to the number control device 3 via the signal line 13. The steam pressure sensor 5 measures the steam pressure value of the steam header 4 (hereinafter, also referred to as “header pressure value”), and the steam pressure signal corresponding to the steam pressure value is transmitted to the number control device 3 via the signal line 13. Send to.

Next, the number control device 3 will be described.
As shown in FIG. 1, the number control device 3 is electrically connected to a plurality of stage value control boilers 20 via a signal line 16. The number control device 3 calculates the required steam amount of the boiler group 2 according to the required load based on the steam pressure value of the steam header 4 measured by the steam pressure sensor 5, and based on the calculated required steam amount. , The combustion position (combustion amount) of the stage value control boiler 20 (hereinafter, also referred to as “control target boiler”) to be controlled in the boiler group 2 is controlled.
In the boiler system 1, a stage value control boiler 20 (hereinafter, also referred to as “preliminary boiler”), which is not a control target, can be provided in the boiler group 2. Then, the number control device 3 (“control unit 30” described later) adds a spare boiler 20 to the control target and changes it to the control target boiler 20 or changes to the control target boiler 20 according to the combustion state of the control target boiler 20. Can be removed from the control target and changed to the spare boiler 20.

FIG. 2 is a functional block diagram showing the configuration of the number control device 3. As shown in FIG. 2, a control unit 30 and a storage unit 32 as control means are provided. Before explaining the configuration of the control unit 30, the storage unit 32 will be described first.

The storage unit 32 is, for example, a ROM, a RAM, an HDD, or the like, and together with various control programs, as shown in FIG. 2, the maximum used steam amount storage unit 321, the average steam amount storage unit 322, and the variable steam amount storage unit 323. , The number control parameter group ID storage unit 324, and the number control parameter group storage unit 325.
The maximum steam consumption storage unit 321 is the maximum steam usage amount which is the maximum value of the total output steam amount supplied by the boiler group 2 (that is, the maximum steam consumption of the steam consumption in the steam usage equipment 18 (required load)). Amount) is memorized.
The average steam amount storage unit 322 stores the average steam usage amount (also referred to as “average steam amount”) which is an average value of the total output steam amount supplied by the boiler group 2.
The variable steam amount storage unit 323 stores the variable steam amount corresponding to the fluctuation of the steam supply amount in the boiler group 2. Usually, the amount of fluctuating steam = the maximum amount of steam used-the average amount of steam.
The number control parameter group ID storage unit 324 stores the number control parameter group ID selected in advance from the plurality of parameter groups. Here, the number control parameter group ID may be any number, characters, or a combination of numbers and characters that can identify the number control parameter group.
The number control parameter group storage unit 325 stores a plurality of parameter groups preset by the user. Here, each parameter group is a parameter such as each parameter value for setting the number control pattern (for example, proportional distribution set pressure, proportional distribution control width, maximum steam amount used, average steam amount, variable steam amount, combustion priority control mode, etc.). Value) is a set containing.
The user can select the most appropriate parameter group for controlling the number of boiler systems 1 from the plurality of parameter groups, which can be reflected in the control of the number of boiler systems 1.

Further, as described above, the storage unit 32 includes the content of the instruction given to each stage value control boiler 20 under the control of the number control device 3 (control unit 30) and the combustion received from each stage value control boiler 20. Stores information such as position, priority setting information, and setting information related to priority change (rotation).

Next, the control unit 30 will be described. As shown in FIG. 2, the control unit 30 includes a number control function unit 301, a time series data output unit 302, and a number control parameter setting unit 303.
including.

The number control function unit 301 is a function unit known to those skilled in the art, and calculates the required steam amount of the boiler group 2 according to the required load based on the steam pressure value of the steam header 4 measured by the steam pressure sensor 5. Then, based on the calculated required steam amount, a functional unit that controls the combustion position (combustion amount) of the stage value control boiler 20 (hereinafter, also referred to as “control target boiler”) to be controlled in the boiler group 2. Is.
For example, the number control function unit 301 calculates the required steam amount of the boiler group 2 according to the required load based on the steam pressure value of the steam header 4 measured by the steam pressure sensor 5, and the calculated required steam amount. The combustion position (combustion amount) of the stage value control boiler 20 (hereinafter, also referred to as “control target boiler”) to be controlled in the boiler group 2 is controlled based on the above. For example, the control unit 30 may control the number of units by the proportional distribution control method.
Specifically, the control unit 30 sets the control pressure band determined by the two preset values of the proportional distribution control pressure and the proportional distribution control width into a plurality of pressure bands based on the number of boilers to be controlled and the type of boiler. It is divided, and the number of boilers burned and the combustion state are assigned in advance so that the amount of boiler combustion increases as the header pressure decreases for each divided pressure band. By doing so, the control unit 30 can control the number of boilers burned and the combustion state so as to change according to the divided pressure band including the header pressure value according to the fluctuation of the header pressure.

The number control function unit 301 can control the number of units from a plurality of number control parameter groups based on a preset parameter group.
For example, when the efficiency priority mode is set as the combustion priority control mode value included in the selected unit control parameter group, the unit control function unit 301 is in the most efficient combustion state (for example, medium combustion state). The boilers to be controlled and the number of boilers are determined so that the output steam amount corresponding to the average steam amount can be secured, and the combustion boilers are gradually increased so as to secure the surplus capacity of steam corresponding to the variable steam amount. Is controlled to be the most efficient combustion state. That is, a loss occurs when the boiler that has stopped functioning is started for a load exceeding the average steam amount. Therefore, the unit control function unit 301 temporarily stops the boiler that is efficiently burning without starting a new boiler. It is controlled so that it is in a higher combustion state.
Further, when the response priority mode is set as the combustion priority control mode value included in the selected unit control parameter group, the unit control function unit 301 outputs the total amount of the average steam amount and the variable steam amount. In order to secure the amount of steam, the boiler to be controlled and the number of boilers are determined, and the boiler is controlled so as to improve the responsiveness by not putting the boiler in the standby state as much as possible. That is, the number control function unit 301 controls the boiler so as not to be in a standby state as much as possible by preferentially securing a low combustion boiler when the load increases and the boiler is sequentially set to medium combustion when the load decreases. Improves responsiveness.
As the combustion priority control mode, there is a basic priority mode suitable for a three-position control type boiler. In this case, the unit control function unit 301 determines the boiler to be controlled and the number of units so as to secure the maximum amount of steam used, and sequentially increases the combustion boilers so as to secure the surplus capacity of steam corresponding to the variable amount of steam. Control the boiler so that it is in a high combustion state.
The function of the number control function unit 301 has been described above.

The time-series data output unit 302 sets each combustion state of the stage value control boiler 20 to be controlled transmitted from the local control unit 22 at each preset sampling cycle (or at each preset time interval). Based on this, the output steam amount output by the boiler group 2 is calculated for each cycle. Then, the time-series data output unit 302 generates time-series data including a time stamp indicating the time in the cycle (hereinafter, also referred to as “sampling time”) and at least the total output steam amount of the boiler group 2 at the time. , Is output to the time-series data storage device 6 described later. Here, the sampling cycle (preset time interval) may be, for example, 1 second. Further, the sampling cycle may be used as the control cycle. Hereinafter, unless otherwise specified, the sampling period will be illustrated and described as 1 second. The sampling cycle is not limited to 1 second. Any value may be used.

As will be described later, the unit control parameter setting unit 303 is output from the unit control parameter output device 9 every predetermined time (for example, 1 hour) from immediately after the time to a predetermined time ahead (for example, 1 hour ahead). The maximum used steam amount, the average steam amount, the variable steam amount, and the parameter group ID calculated based on the predicted values of the average steam used amount, the maximum steam used amount, and the variable steam amount of the above are acquired, and the maximum used steam is obtained, respectively. It is stored in the quantity storage unit 321, the average steam amount storage unit 322, the variable steam amount storage unit 323, and the number control parameter group ID storage unit 324. By doing so, the number control device 3 can automatically control the number of units appropriate for the user based on the various steam amounts predicted in real time and the parameter group selected from the plurality of parameter groups.
For example, when the user selects a mode (efficiency priority mode) in which the user wants to secure a long combustion time at the "eco-driving point", the number control device 3 is predicted while the boiler system 1 is in operation, for example, every hour. The number of boilers to be operated can be set so that the average amount of steam used up to one hour ahead is burned at the medium combustion position, which is the high-efficiency combustion position, and the parameters can be adjusted to set the combustion priority.
Further, when the user selects the response priority mode, the number control device 3 is the minimum number of units that can secure the maximum amount of steam used up to one hour, which is predicted every hour, for example, while the boiler system 1 is in operation. The number of boilers is controlled, the boiler is set to medium combustion in order when the load increases, the low combustion boiler is preferentially secured when the load decreases, and the parameters are adjusted so that the boiler is not in the standby state as much as possible, and the boiler system is installed. Responsiveness can be improved by controlling.
The unit control device 3 has been described above.

Next, the time-series data storage device 6 will be described. FIG. 3 shows a functional block diagram showing the configuration of the time-series data storage device 6.
As described above, the time-series data storage device 6 is directly connected to each other via the connection interface with the number control device 3, the machine learning device 7, and the number control parameter output device 9, respectively, or is connected to each other via a LAN (Local Area Network), the Internet, or the like. Devices that are interconnected via a network (not shown).
The time-series data storage device 6 may be, for example, an edge device installed in the field. Further, it may be a data server connected so as to be communicable, a virtual storage device located on the cloud, or a virtual data server.
As shown in FIG. 3, the time-series data storage device 6 includes a control unit 60 and a storage unit 62. Before explaining the configuration of the control unit 60, first, the storage unit 62 will be described.

The storage unit 62 is, for example, a ROM, RAM, HDD, etc., and together with various control programs, as shown in FIG. 3, a total output steam amount time-series information storage unit 621 and a unit-time steam amount time-series information storage unit. 622 and a steam load pattern information storage unit 623 are included.

As will be described later, the total output steam amount time-series information storage unit 621 provides time-series information (“total output steam amount time series”) of the total output steam amount of the boiler group 2 for each sampling cycle for each operating day of the boiler system 1. Information ") is memorized. In this embodiment, 1 second is exemplified as the sampling cycle. Then, in the total output steam amount time series information for one day, a set of the value of the total output steam amount of the boiler group 2 and the time information (sampling time) indicating the time when the total output steam amount was measured is set. It includes information of a maximum of 86400 sets (= 24hh × 60mm × 60ss) as information and date information in which the total output steam amount is measured. The time information may be, for example, hhmmss (hours, minutes, seconds). Hereinafter, unless otherwise specified, the sampling cycle is exemplified and described as 1 second, but the sampling cycle is not limited to 1 second. Any value may be set.

The unit time steam amount time-series information storage unit 622 has the maximum value and the average value (also referred to as "unit time steam amount") of the total output steam amount per unit time for each operating day of the boiler system 1 as described later. ) Is associated with the time information, and the time-series information (also referred to as "unit-time steam amount time-series information") is stored. In this embodiment, 1 minute is exemplified as the unit time. Further, the start of the unit time (1 minute) is set to mm minutes 00 seconds, and it is set to be between mm minutes 00 seconds and 1 minute. Then, in the unit time steam amount time series information for one day, the value of the unit time steam amount of the boiler group 2 for each unit time and the time information indicating the time of the unit time are set as one set of information. Information of a maximum of 1440 (= 24hh × 60mm) and date information in which the total output steam amount is measured are included. The time information may be, for example, hhmm (hours and minutes). That is, if the date is yymmdd (year / month / day) and the unit time steam amount including (maximum steam usage amount, average steam usage amount) at hhmm of the date is _xymmdd (hhmm), the unit time steam amount time. The series information can be described as a set of unit-time steam amounts { _xymmdd (hhmm)} of the date information on which the steam amount was measured. Hereinafter, unless otherwise specified, the unit time is exemplified and described as 1 minute, but the unit time is not limited to 1 minute. Any value may be set.

The steam load pattern information storage unit 623 stores a table (also referred to as a "steam load pattern table") in which a steam load pattern preset by the user is associated with the calendar date.
For example, the steam load pattern is a "high load pattern" corresponding to, for example, an average operating rate of "large" manufacturing equipment in a factory, and a "medium" average operating rate of manufacturing equipment in a factory and a large amount of equipment consumed in a short time. "Medium load and large fluctuation pattern" corresponding to "high" number of operations, "medium load and large fluctuation pattern" corresponding to "low" number of operations of equipment that consumes a large amount in a short time with "medium" average operating rate of manufacturing equipment in the factory Examples include a "small fluctuation pattern", a "low load pattern" in which the average operating rate of manufacturing equipment in a factory is "small", a "holiday pattern" in which all manufacturing equipment in a factory is stopped, and only air conditioning is used. For example, when the steam load pattern is set for each day of the week, the day of the week = steam usage pattern may be set, and an arbitrary steam load pattern may be set at irregular times such as holidays. Here, the user may preset a threshold value for discriminating between large, medium, and small average operating rates. Similarly, regarding the number of operations of equipment that consumes a large amount of time in a short time, the threshold value for identifying the equipment that consumes a large amount of time and a short time, and the threshold value for identifying the number of times of operation and the number of operations are set in advance by the user. You may. Further, for each month, a table for associating the date belonging to the month with the steam load pattern may be set in advance.

Next, the control unit 60 will be described. As shown in FIG. 3, the control unit 60 includes a time series information recording unit 601, a unit time data creation unit 602, a first time zone time series data extraction unit 603, and a steam load pattern acquisition unit 604. ..

The time-series information recording unit 601 performs time-series information (total output steam amount time-series information) of the total output steam amount of the boiler group 2 for each sampling cycle (for example, 1 second) of the boiler system 1 for each working day. Is taken in and recorded in the storage unit 62 (total output steam amount time series information storage unit 621).

The unit time data creation unit 602 outputs all the output for each working day for each working day based on the total output steam amount time series information recorded in the total output steam amount time series information storage unit 621. The maximum value and the average value (unit time steam amount) of the steam amount are calculated and recorded in the storage unit 62 (unit time steam amount time series information storage unit 622) as the unit time steam amount time series information. When calculating the average value, the average value may be calculated based on the moving average.

The first time zone time series data extraction unit 603 is based on the unit time steam amount time series information, and is included in the operating time zone of the boiler system 1 for each unit time in a preset first time (for example, 2 as the first time). A subset (referred to as "first time zone time series data") of unit time steam amount time series information (with time = 120 minutes) is extracted. Hereinafter, unless otherwise specified, the first hour will be illustrated and described as 2 hours. In this case, the start of each first hour (2 hours) is every 1 minute at an arbitrary time, specifically, between hh hours mm minutes and 2 hours. Here, h is the time included in the operating time zone, and mm is an integer satisfying 0 ≦ mm ≦ 59. In this way, the first time zone time series data extraction unit 603 can extract the first time zone time series data every minute from the unit time steam amount time series information.
FIG. 4 is a diagram showing an outline of the first time zone time series data extracted by the time series data storage device 6. As shown in FIG. 4, the first time zone time series data extraction unit 603 is, for example, the first time zone time series two hours after the time corresponding to the first unit time steam amount of the working day of the boiler system 1. The data (1) is extracted, and then the first time zone time series data (i) (2≤i) is extracted every minute interval, and then, for example, the operation end time of the operation day (the first unit time of the operation day). The final first time zone time series data (N) from 1 hour before (x minutes after the time corresponding to the steam amount) to 1 hour before (after (x-60 minutes) from the time corresponding to the first unit time steam amount on the working day). ) Is extracted. Here, N indicates the number of time-series data in the first time zone extracted every minute on the working day. As will be described later, the first time-series time-series data extraction unit 603 uses the machine learning device 7 and / or the number of first-time time-series data (i) (1 ≦ i ≦ N) extracted at 1-minute intervals. It may be provided to the control parameter output device 9.

As described above, the steam load pattern acquisition unit 604 sets the steam load pattern corresponding to the date on which the steam amount is measured based on the steam load pattern preset by the user corresponding to the calendar date, for each working day. Corresponds to the total output steam amount time series information and the unit time steam amount time series information, respectively.
By doing so, the steam load pattern acquisition unit 604 can incorporate the operation pattern of each working day into the prediction model. For example, it becomes possible to learn the prediction model for each steam load pattern, which leads to improvement in prediction accuracy. Can be expected.

As described above, the time-series data storage device 6 has a total output steam amount time-series information and a unit-time steam amount in which the machine learning device 7 and / or the number control parameter output device 9 is associated with the steam load pattern. It can function as a time-series database that provides time-series information and first time-series time-series data.
The time-series data storage device 6 has been described above.

Next, the machine learning device 7 will be described. FIG. 5 shows a functional block diagram showing the configuration of the machine learning device 7.
As shown in FIG. 5, the machine learning device 7 includes a control unit 70 and a storage unit 72. Before explaining the configuration of the control unit 70, first, the storage unit 72 will be described.

The storage unit 72 is, for example, a ROM, a RAM, an HDD, or the like, and together with various control programs, is a teacher data (1 ≦ k ≦ M) generated for each steam load pattern k (1 ≦ k ≦ M) generated by the teacher data creation unit 701. The training data) and the prediction model 75 (k) (learned model 75 (k)) being constructed or constructed for each steam load pattern k (1 ≦ k ≦ M) by the learning unit 702 are stored.
As shown in FIG. 5, the storage unit 72 includes a teacher data storage unit 721 and a predictive model storage unit 722.

The teacher data storage unit 721 uses each first time zone time series data (i) for each working day as input data, and the second hour (for example) immediately after the time zone corresponding to the first time zone time series data (i). Teacher data (training data) using the unit time steam amount after (1 hour = 60 minutes) as label data is stored in association with the steam load pattern k (1 ≦ k ≦ M) on the working day.
FIG. 6 is a diagram showing an outline of teacher data (training data) in the machine learning device 7. The first time zone time series data shown in FIG. 6 corresponds to the first time zone time series data shown in FIG. As shown in FIG. 6, the teacher data (training data) uses 120 consecutive unit-time steam amount time-series data corresponding to each first time-zone time-series data (i) as input data, and the first time. The unit time steam amount data after the second hour (1 hour = 60 minutes) immediately after the time zone corresponding to is used as the label data. By doing so, the learning unit 702, which will be described later, will use the 120 consecutive unit-time steam amount time-series data corresponding to the first time-zone time-series data (i) stored in the teacher data storage unit 721. A prediction model (trained model) that predicts the unit time steam amount immediately after the time zone corresponding to the first time and after the second time (1 hour = 60 minutes) is learned.

The predictive model storage unit 722 stores the predictive model (learned model) being trained by the learning unit 702. As will be described later, a predictive model (learned model) being trained is generated for each steam load pattern.

Next, the control unit 70 will be described. As shown in FIG. 5, the control unit 70 includes a teacher data creation unit 701, a learning unit 702, and a prediction model providing unit 703.

The teacher data creation unit 701 is associated with the steam load pattern k (1 ≦ k ≦ M) from the time-series data storage device 6 based on the working days, and each first time zone time-series data (i) for each working day. ), The unit time steam amount data immediately after the first hour and after the second hour (1 hour = 60 minutes) is acquired, and the first time zone time series data (i) is input data, the first. Teacher data (also referred to as "training data") is created using unit-time steam amount data immediately after one hour and after the second hour (1 hour = 60 minutes) as label data. By doing so, the teacher data creation unit 701 can create N teacher data (training data) for each working day. The teacher data creation unit 701 may create a set of teacher data (training data) for each steam load pattern k (1 ≦ k ≦ K) from the teacher data (training data) created for each working day.
It is desirable to prepare a large number of teacher data (training data) for supervised learning. For example, training data (training data) is prepared for each steam load pattern k (1 ≦ k ≦ M) based on the time series information of the boiler system 1 stored in the time series data storage device 6 for the past several years. Is desirable.
By doing so, the learning unit 702, which will be described later, performs machine learning based on a large number of teacher data (training data) generated for each steam load pattern k (1 ≦ k ≦ M), thereby performing the steam load pattern k. A prediction model can be created for each (1 ≦ k ≦ M).

The learning unit 702 performs supervised learning based on the teacher data (training data) generated for each steam load pattern k (1 ≦ k ≦ M) by the teacher data creation unit 701, thereby performing the steam load pattern k. Corresponding to (1 ≦ k ≦ M), the prediction model 75 (k) that predicts the unit time steam amount from immediately after the first hour to after the second hour (60 minutes later) from the time series data in the first time zone. (Also referred to as "trained model 75 (k)") can be constructed.

As an example of machine learning in this embodiment, deep machine learning using LSTM (Long Short Term Memory), which is a recurrent neural network, will be illustrated. LSTM is a special kind of recurrent neural network RNN (RNN) known to traders, which can learn long-term dependencies of past data.
The learning unit 702 optimizes parameters such as weight vectors in the prediction model 75 (k) (learned model 75 (k)) configured by the LSTM network for each steam load pattern k (1 ≦ k ≦ M). Therefore, each first time zone time series data created for each steam load pattern k (1 ≦ k ≦ M) is input from the input layer, and the output information of the LSTM network is taken in. The learning unit 702 performs processing by a well-known error back propagation method or the like by using, for example, a preset error function. By doing so, the learning unit 702 will use the steam load pattern k (1 ≦ k ≦ M) based on the teacher data (training data) created from the time-series information of the predetermined period or the predetermined number of teacher data (training data). ), The parameters such as the weight vector in the LSTM network are optimized, and the prediction model 75 (k) (trained model 75 (k)) is generated. Here, the learning unit 702 may end the learning by satisfying a predetermined condition. Predetermined conditions include, but are not limited to, learning a predetermined number of teacher data, learning teacher data corresponding to a predetermined operating period, and the like. The user may set arbitrary conditions.
Although LSTM is exemplified as an example of machine learning, it is not limited to this. A known RNN may be applied. Further, a known feedforward neural network may be applied.

FIG. 7 is a diagram showing an example of a prediction model 75 (k) (learned model 75 (k)) provided to the prediction device 8 described later. Here, as shown in FIG. 7, the prediction model 75 (k) (trained model 75 (k)) is a time series of the first time zone of the boiler system 1 operating on the working day corresponding to the steam load pattern k. A multi-layer neural network that uses data (120 consecutive unit-time steam amount time-series data) as an input layer and unit-time steam amount in a unit time (for example, 1 minute) immediately after the first hour to after the second hour as an output layer. Is illustrated.

Further, after the learning unit 702 builds the prediction model 75 (k) (trained model 75 (k)), the learning unit 702 periodically, for example, every month, every three months, every six months, or every year. The time-series data (i) of each first time zone of the working day corresponding to the steam load pattern k of the boiler system 1 newly captured in the series data storage device 6 (storage unit 62) is acquired, and the first time series data (i) is acquired. Prediction model 75 (k) (learned model 75 (k) by acquiring unit-time steam amount data immediately after the time and after the second hour (1 hour = 60 minutes) and creating teacher data (training data). )) May be updated.
Further, even when a deviation exceeding a predetermined threshold value between the predicted value and the measured value occurs more than a predetermined frequency, the boiler system 1 newly incorporated into the time series data storage device 6 (storage unit 62) In addition to acquiring the time-series data (i) of each first time zone of the working day corresponding to the steam load pattern k, the unit time steam amount data immediately after the first hour and after the second hour (1 hour = 60 minutes). May be updated and the prediction model 75 (k) (trained model 75 (k)) may be updated by acquiring the teacher data (training data). At that time, in addition to the newly created teacher data (training data), a new prediction model is obtained from the time-series information of the amount of steam captured in the recent period such as one year in the past from that time. The 75 (k) (trained model 75 (k)) may be recreated. By doing so, the influence of the time-series information of the amount of steam captured in the old period is eliminated, and the prediction model 75 (k) (trained model 75 (k)) reflecting the time-series information of the recent amount of steam is reflected. ) Can be updated.

The above-mentioned supervised learning may be performed by online learning, batch learning, or mini-batch learning.
Online learning is a learning method in which one of the teacher data (training data) is taken out and the weight and bias are updated. Specifically, during the operation of the boiler system 1, the time-series data storage device 6 gives the time-series data (i) in the first time zone, and immediately after the first hour and after the second hour (1 hour = 60 minutes). It refers to a form in which unit-time steam amount data is acquired each time, teacher data (training data) is created each time, and machine learning is performed each time.
Moreover, batch learning is a learning method in which all teacher data (training data) are used for updating weights and biases. In the batch learning, after the time-series data storage device 6 stores the time-series data for a predetermined period, for example, monthly time-series data, the time-series data in the first time zone of the month and the second hour immediately after the first time ( Unit-time steam amount data after 1 hour = 60 minutes) is acquired, a plurality of teacher data (training data) are generated, and all the generated teacher data (training data) are used for supervised learning. Further, mini-batch learning is a learning method in which teacher data (training data) is divided into small chunks (batch) and weights and biases are updated for each chunk. The learning interval is an intermediate period between online learning and batch learning.

The prediction model providing unit 703 uses the learning unit 702 to predict the prediction model 75 (k) (learned model 75 (k)) constructed for each pattern k (1 ≦ k ≦ M), which will be described later according to the user's instruction. Provided to the device 8.
The machine learning device 7 has been described above.

Next, the prediction device 8 will be described. The prediction device 8 is a device used in the prediction model operation phase in which the prediction model 75 (k) (learned model 75 (k)) constructed by the machine learning device 7 is applied to the boiler system 1.
In the prediction model operation phase, the prediction device 8 uses the prediction model 75 (k) (learned model 75 (k)) constructed by the machine learning device 7, for example, in the first hour from the start of operation of the boiler system 1. Immediately after (for example, 2 hours) has elapsed, every unit time (for example, 1 minute), the unit time steam amount predicted value for the second hour ahead (for example, 1 hour ahead) is calculated, and the calculated unit time steam amount predicted value is calculated. Is stored in the storage unit 82 as time-series data. FIG. 8 is a diagram illustrating an input / output processing status in which unit-time steam amount predicted values acquired by the prediction unit 802 are sequentially stored as time-series data in the second time-series time-series data storage unit 823. By doing so, as shown in FIG. 8, when the time series data of 1 hour (60 minutes) (hereinafter referred to as "second time zone time series data") is stored for the unit time steam amount predicted value, the number of units is stored. The control parameter output device 9 statistically determines the maximum steam usage predicted value and the average steam usage predicted value to be supplied by the boiler system 1 in the period (for example, 1 hour) based on the second time zone time series data. Can be calculated in.
FIG. 9 shows a functional block diagram showing the configuration of the prediction device 8. As shown in FIG. 9, the number control parameter output device 9 includes a control unit 80 and a storage unit 82.

First, the storage unit 82 will be described. The storage unit 82 is, for example, a ROM, a RAM, an HDD, or the like, and together with various control programs, the storage unit 82 includes a prediction model storage unit 821, a first time-series data storage unit 822, and a second storage unit 82, as shown in FIG. Includes a time zone time series data storage unit 823.

The prediction model storage unit 821 stores the prediction model 75 (k) (learned model 75 (k)) constructed by the machine learning device 7 for each steam load pattern k (1 ≦ k ≦ M).

The first time zone time-series data storage unit 822 performs time-series data every minute from the start of the first hour (2 hours) of the working day of the boiler system 1 to one hour before the end time of the working day. The first time zone time series data created by the storage device 6 is stored. The first time-series time-series data storage unit 822 stores, for example, 120 unit-time vapor amount data corresponding to at least 120 minutes in, for example, a round-robin format so that it can be acquired every minute. You may.

As will be described later, the second time zone time series data storage unit 823 is predicted based on each first time zone time series data, from immediately after the first time (2 hours = 120 minutes) to the second hour ( The unit time steam amount data (predicted value) after 1 hour = 60 minutes) is stored as the second time zone time series data. The second time-series data storage unit 823 stores, for example, 60 unit-time vapor amount data corresponding to at least 60 minutes in, for example, a round-robin format so that it can be acquired every 60 minutes. You may.

Next, the control unit 80 will be described. As shown in FIG. 9, the control unit 80 includes an input unit 801 and a prediction unit 802.

The input unit 801 acquires the latest version of the prediction model 75 (k) (learned model 75 (k)) constructed from the machine learning device 7, for example, according to the instruction of the user, and stores the prediction model in the prediction model storage unit 821. It is stored together with the version information of.
The input unit 801 acquires the operation start information of the acquired prediction model 75 (k) (learned model 75 (k)) by, for example, an input instruction from the user. The operation start information may include, for example, the operation start date of the prediction model 75 (k) (learned model 75 (k)). By doing so, after the operation start date of the prediction model 75 (k) (learned model 75 (k)), the functional unit related to the prediction of the number control parameter output device 9 may be automatically started.
Further, the input unit 801 acquires, for example, the steam load pattern k on the operation day, and as described above, the time is from the start of the first hour (2 hours) to one hour before the end time of the working day. The first time zone time series data (i) is acquired from the series data storage device 6 every minute.

Next, the function of the prediction unit 802 will be described.
As shown in FIG. 7, the prediction unit 802 uses the first time zone time series data (i) input by the input unit 801 every minute as the prediction model 75 (k) (trained model 75 (k)). By inputting to, the unit time steam amount predicted value in the unit time (1 minute) from immediately after the first hour to after the second hour (1 hour) is acquired. As shown in FIG. 8, the prediction unit 802 sequentially stores the acquired unit-time steam amount prediction value in the second time zone time-series data storage unit 823 as time-series data.
By doing so, as shown in FIG. 8, when the time series data (second time zone time series data) for one hour (60 minutes) is stored, the prediction unit 802 uses the second time zone time series data. Based on the above, the maximum steam usage predicted value and the average steam usage predicted value to be supplied by the boiler system 1 in the second time are statistically calculated. By doing so, the prediction unit 802 predicts the maximum steam usage amount to be steamed by the boiler system 1 and the average in the time zone up to the second hour (1 hour) every hour (60 minutes). The predicted value of steam usage can be calculated statistically.
Here, the maximum amount of steam used determines the upper limit of the number of boilers to be burned, and even if a load exceeding the maximum amount of steam used is generated, it is controlled so that no more boilers are burned. In the calculation of the steam amount, for example, the prediction unit 802 may set a correction value α in advance and calculate the maximum steam usage amount as the maximum steam usage amount (predicted value) + α. In addition, the average steam amount is for determining the number of boilers to be burned in the most efficient combustion state (eco-driving point) as described above, and if a small value is set, it is easy to deviate from the eco-driving point. Therefore, in calculating the average steam amount, for example, the prediction unit 802 may set a correction value β in advance and calculate the average steam amount as the average steam usage amount (predicted value) + β. The variable steam amount is calculated by subtracting the average steam amount from the maximum steam amount used.
The prediction device 8 has been described above.

Next, the number control parameter output device 9 will be described. FIG. 10 shows a functional block diagram showing the configuration of the number control parameter output device 9. As shown in FIG. 10, the number control parameter output device 9 includes a control unit 90 and a storage unit 92.

The storage unit 92 is, for example, a ROM, RAM, HDD, or the like, and includes various control programs and a number control parameter group storage unit 921 as shown in FIG. The number control parameter group storage unit 921 stores a plurality of parameter groups stored in the storage unit 32 (unit control parameter group storage unit 325) of the number control device 3.

As shown in FIG. 10, the control unit 90 includes a unit control parameter output unit 901.
The unit control parameter output unit 901 uses the maximum steam amount, the average steam amount, and the variable steam amount based on the maximum steam usage predicted value and the average steam usage predicted value in the second time predicted by the prediction device 8. Any two or more of these values may be calculated, and the calculated values may be set in the number control parameter.
The unit control parameter output unit 901 outputs any two or more values of the maximum steam amount, the average steam amount, and the variable steam amount to the unit control device 3 (unit control parameter setting unit 303), and these values. The number control parameter group may be selected based on the above, and the selected number control parameter group ID may be output to the number control device 3 (unit control parameter setting unit 303). The unit control parameter output unit 901 may select, for example, a unit control parameter group having the calculated maximum steam amount and average steam amount values (or the closest values). By doing so, the number control device 3 automatically controls the number of units by the parameter group appropriate for the user, which is selected from the predicted various steam amounts and a plurality of parameter groups preset by the user. Can be done.
Specifically, for example, appropriate values of the maximum steam usage amount and the average steam usage amount are defined in advance for each unit control parameter group ID.
For example, the maximum steam usage amount is defined as 12,000 to 15,000 kg / h and the average steam usage amount is defined as 6,000 to 8,000 kg / h as appropriate values of pattern number 1 in advance.
As appropriate values for pattern number 2, the maximum steam usage is defined as 12,000 to 15,000 kg / h, and the average steam usage is defined as 4,000 to 6,000 kg / h.
As appropriate values for pattern number 3, the maximum steam usage is defined as 12,000 to 15,000 kg / h, and the average steam usage is defined as 2,000 to 4,000 kg / h.
As appropriate values for pattern number 4, the maximum steam usage is defined as 6,000 to 9,000 kg / h, and the average steam usage is defined as 2,000 to 3,000 kg / h.
It is assumed that the maximum steam usage amount is defined as 3,000 to 6,000 kg / h and the average steam usage amount is defined as 1,000 to 2,000 kg / h as appropriate values of pattern number 5.
Then, the unit control parameter output unit 901 compares the calculated maximum steam amount and average steam amount value with the appropriate value of the maximum steam usage amount and the appropriate value of the average steam usage amount defined by each pattern number, and calculates. The most suitable (for example, the closest) pattern number among the pattern numbers in which the values of the maximum steam amount and the average steam amount are within the range of the maximum steam usage amount and the average steam usage amount defined by each pattern number. May be selected.
The unit control parameter output device 9 has been described above.

Next, every 1 hour (60 minutes) of the prediction device 8 according to the present embodiment, the maximum steam usage amount and the average steam usage amount prediction processing and the number of units in the time zone up to the second hour (1 hour) ahead. The operation related to the setting process of the number control parameter to the number control device 3 by the control parameter output device 9 will be described.
FIG. 11 is a flowchart illustrating the processing of the prediction device 8 and the number control parameter output device 9 in the operation phase. It is assumed that the prediction device 8 has acquired the prediction model 75 (k) (learned model 75 (k)) in advance from the machine learning device 7. It is assumed that the steam load pattern k on the day of operation has been acquired. Further, the number N of the first time zone time series data extracted every minute on the working day is set in advance.

In step S11, the prediction device 8 (input unit 801) initializes i and j (i = 1, j = 1).

In step S12, the prediction device 8 (input unit 801) acquires the first time-series time-series data (i) from the time-series data storage device 6 while the boiler system 1 is in operation.

In step S13, the prediction device 8 (prediction unit 802) inputs the first time zone time series data (i) collected every minute into the prediction model 75, and immediately after the first hour to the second hour (1). The unit time steam amount predicted value (j) in the unit time (1 minute) of (after time) is acquired and stored in the second time zone time series data storage unit 823 in the time series order (in order of j).

In step S14, the prediction device 8 (prediction unit 802) sets i = i + 1 and j = j + 1.

In step S15, it is determined whether or not j = 60. If j <60, the process proceeds to step S12. When j = 60, the process proceeds to step S16.

In step S16, the prediction device 8 (prediction unit 802) has a second time (1 hour = 60 minutes) from the present time based on the second time zone time series data stored in the second time zone time series data storage unit 823. ) Statistical calculation of the maximum steam usage predicted value and the average steam usage predicted value to be supplied by the boiler system 1 in the previous time zone.

In step S17, the unit control parameter output device 9 (unit control parameter output unit 901) uses the maximum steam usage amount, the average steam amount, and the variable steam based on the maximum steam usage predicted value and the average steam usage predicted value. The quantity is calculated, or the optimum number control parameter group ID is selected from the number control parameter groups described above.

In step S18, the number control parameter output device 9 (number control parameter output unit 901) sets the maximum used steam amount, the average steam amount, the variable steam amount, and the number control parameter group ID calculated in step S17 to the number control device 3 (number control device 3 (unit control parameter output unit 901). Output to the unit control parameter setting unit 303).
As a result, the number control device 3 stores the number control parameters input from the number control parameter output device 9 in the storage unit 32, and automatically sets the boiler 20 and the number of units to be controlled based on the set number control parameters. The number of units can be controlled by the combustion priority control mode.

In step S19, the number control device 3 (number control parameter setting unit 303) determines whether or not i = N. If i <N, the process proceeds to step S1A. If i = N, the process ends.
In step S1A, the number control device 3 (number control parameter setting unit 303) moves to step S12 with j = 1.

As described above, as described above, the number control device 3 is replaced with the number control device 3A that controls the continuous control boiler, so that the time series data storage device 6, the machine learning device 7, and the prediction device 8 are continuous. It can be applied as it is to the control boiler. Therefore, the prediction device 8 according to the embodiment is set in the boiler system 1 and the boiler system 1A including the boiler group 2 having a plurality of boilers 20 in accordance with the calendar date (for example, season or month, day, holiday, etc.). Based on the steam load pattern and the history of the total output steam amount by the boiler group 2 in the first hour, the maximum steam usage amount and the average steam usage amount in the second hour from immediately after the first hour to the predetermined destination, It is possible to predict the amount of fluctuating steam used.
Further, in the boiler system 1 including the boiler group 2 having a plurality of step value control boilers 20 capable of burning at a plurality of stepwise combustion positions, the steam system 1 predicted as described above is immediately after the first hour to the second hour ahead. Based on the maximum steam usage predicted value and the average steam usage predicted value in the time zone up to, the number control parameter output device 9 is the maximum steam usage amount and the average in the time zone from immediately after the first hour to the second hour ahead. By calculating the steam amount, the variable steam amount, and the number control parameter and outputting them to the number control device 3, the number control device 3 automatically performs appropriate number control for the user based on various steam amounts predicted in real time. Can be done.

Although one embodiment has been described above, the time-series data storage device 6, the machine learning device 7, and the prediction device 8 are not limited to the above-described embodiment, and are modified and improved within a range in which the object can be achieved. Etc. are included.

<Modification 1>
In the above-described embodiment, the machine learning device 7 acquires the first time data (i) from the time-series data storage device 6 at unit time (1 minute) intervals for each working day based on the history information, and also obtains the first time data (i). It was configured to acquire the unit time steam amount data immediately after the time zone corresponding to the first hour and after the second hour (1 hour = 60 minutes). However, for example, the machine learning device 7 (teacher data creation unit 701) is stored in the storage unit 62 (total output steam amount time series information storage unit 621) of the time series data storage device 6 in the machine learning phase. Based on the data (history information), the maximum value and average value (unit time steam amount) of the total output steam amount are directly calculated for each unit time of each working day, and the calculated unit time steam amount time series information is used. It may be recorded in the storage unit 72. By doing so, the machine learning device 7 (teacher data creation unit 701) has the first time zone time series data and the second time based on the unit time steam amount time series information of each working day stored in the storage unit 72. Teacher data (training data) may be created by extracting time-series data.

<Modification 2>
In the above-described embodiment, on the working day of the prediction model operation phase, the prediction device 8 acquires the first time data (i) from the time series data storage device 6 every unit time (1 minute), and also obtains the first time data (i). Predict the unit time steam amount data immediately after the second hour (for example, 1 hour = 60 minutes) from immediately after the time zone corresponding to 1 hour, and the unit time immediately after the 1st hour to the 2nd hour (1 hour later). The unit time steam amount prediction value in (1 minute) was sequentially configured to be the second time zone time series data. However, for example, instead of the prediction device 8, the time-series data storage device 6 stores the first time data (i) based on the unit-time steam amount time-series information stored in the unit-time steam amount time-series information storage unit 622. In addition to extracting, the unit time steam amount data from immediately after the time zone corresponding to the first hour to the second hour (1 hour = 60 minutes) is predicted, and from immediately after the first hour to after the second hour (1 hour). The unit time steam amount predicted value in the unit time (1 minute) of the latter) may be sequentially stored in the storage unit 62. In that case, the time-series data storage device 6 may store the prediction model in the storage unit 62. Alternatively, the prediction model stored in the machine learning device 7 may be requested for prediction. By doing so, the prediction device 8 acquires the first time data (i) and the second time zone time series data from the time series data storage device 6, thereby supplying steam to the boiler system 1 in the second time. The maximum steam usage predicted value and the average steam usage predicted value may be calculated statistically.

<Modification 3>
In the above embodiments, the predictive model is constructed according to the steam load pattern set corresponding to the calendar date (eg, season or month, day of the week, holiday, etc.), but is not limited to this. For example, the steam load pattern may be set corresponding to a predetermined operating time zone of the calendar date. Thereby, for example, even if the steam load pattern differs depending on the time zone, it is determined based on the history of the total output steam amount by the boiler group in the immediately preceding time zone (for example, 2 hours) in the time zone. It is possible to provide a machine learning device for creating a prediction model for predicting the maximum steam usage and the average steam usage up to a time (for example, one hour) ahead.

<Modification example 4>
In the above-described embodiment, the machine learning device 7 (learning unit 702) uses the first time zone time series data as input data and the unit time steam amount data after the second time (1 hour = 60 minutes) as label data. Although supervised learning by a multi-layer neural network is illustrated based on the supervised learning data (training data), the learning method is not limited to this.
For example, the learning unit 702 executes a time-series analysis based on the time-series data in the first time zone and the unit-time steam amount data after the second time (1 hour = 60 minutes), and generates a prediction model 75. You may try to do it.

<Modification 5>
In the above-described embodiment, the machine learning device 7 is exemplified as a device different from the number control device 3, the time series data storage device 6, the prediction device 8, and the number control parameter output device 9, but is a part of the machine learning device 7. Alternatively, all the functions may be provided in the number control device 3, the edge device as the time series data storage device 6, the prediction device 8, or the number control parameter output device 9. Alternatively, a part or all of the functions of the machine learning device 7 may be realized by using the virtual server function or the like on the cloud.
Similarly, a part or all of the functions of the prediction device 8 may be provided in the edge device as the number control device 3 or the time series data storage device 6. Further, the function of the number control parameter output device 9 may be provided in the edge device as the number control device 3 or the time series data storage device 6.

<Modification 6>
In the present embodiment, the number control device 3 executes combustion control of each stage value control boiler 20 of the boiler group 2 so that the vapor pressure inside the steam header 4 falls within a preset set pressure range. However, it is not limited to this proportional distribution control method.
For example, in order to keep the steam pressure inside the steam header at a preset target steam pressure value, the number control device 3 controls the combustion state of the stage value control boiler 20 to be burned in the boiler group 2. You may.

<Modification 7>
Further, as described above, the boiler is not limited to the step value control boiler. It can also be applied to a group of boilers consisting of continuously controlled boilers.
Specifically, for example, as described above, the number control device 3 is replaced with the number control device 3A for controlling the number of continuous control boilers known to those skilled in the art, and is replaced with the number control parameter output device 9, for example. When outputting the average steam amount, control parameters may be set, for example, set so that the combustion rate of the combustion boiler is within the range of the eco-driving zone.

According to the machine learning device 7 or the number control parameter output device 9 included in the boiler system 100 with a prediction function of the present embodiment described above, the following effects are obtained.

(1) The machine learning device 7 of the present embodiment is sampled preset from a boiler system 1 having a boiler group 2 having a plurality of boilers 20 and a number control device 3 for controlling the combustion state of the boiler group 2. A predetermined value calculated based on the sampling time output for each cycle, the measurement data including at least the total output steam amount of the boiler group 2 at the sampling time, and the time-series data in the sampling order for each working day including the sampling time. Time series of unit time steam amount included in the preset first time zone from the set of time series data of unit time steam amount including at least maximum steam usage amount and average steam usage amount of boiler group 2 for each unit time. In addition to extracting the time-series data for the first time zone, which is a subset of the data, the amount of steam for the second hour ahead, which is preset immediately after the first time zone from the set of time-series data for the unit-time steam amount. By extracting Based on, from the time series data of the unit time steam amount included in the first time zone on any working day, at least the maximum steam usage amount of the boiler group 2 in the unit time immediately after the first time zone to the second hour ahead and A learning unit 702 that generates a prediction model 75 that predicts a unit time steam amount including an average steam usage amount is provided.
By doing so, in the boiler system 1 including the boiler group 2 having a plurality of boilers 20, even if the steam load pattern differs depending on the day, holiday, etc. of the month, season, quarter, etc., for example, the immediately preceding time zone. Based on the history of the total output steam amount by the boiler group 2 in (plural hours), a prediction model 75 for predicting the maximum steam usage amount and the average steam usage amount in a predetermined time ahead can be created.

(2) In the present embodiment, the working day may be associated with a preset steam load pattern, and the learning unit 702 may generate a prediction model 75 corresponding to the steam load pattern.
By doing so, the machine learning device 7 of the present embodiment can be used, for example, in the immediately preceding time zone (plural hours) even if the vapor load pattern differs depending on the day, holiday, etc. of the month, season, quarter, or the like. Based on the history of the total output steam amount by the boiler group 2, a prediction model 75 for predicting the maximum steam usage amount and the average steam usage amount after a predetermined time can be created.

(3) The learning unit 702 of the machine learning device 7 of the present embodiment uses the first time zone time series data as input data, and labels the amount of steam in the unit time immediately after the first time zone and the second hour ahead. Based on the teacher data, supervised learning may be executed to generate a trained model as a predictive model 75.
By doing so, the learning unit 702 of the machine learning device 7 of the present embodiment can exert the same effect as (1) or (2).

(4) The learning unit 702 of the machine learning device 7 of the present embodiment includes time-series data for each unit time in the first time zone, the amount of steam in the unit time immediately after the first time zone and the second time ahead. A time series analysis may be performed based on the above to generate a prediction model 75.
By doing so, the learning unit 702 of the machine learning device 7 of the present embodiment can exert the same effect as (1) or (2).

(5) The machine learning device 7 of the present embodiment may be composed of one or more servers that are communicably connected to the number control device 3 via a network.
By doing so, the machine learning device 7 of the present embodiment can be a device independent of the boiler system 1.

(6) The prediction device 8 of the present embodiment is immediately before an arbitrary time on the working day of the prediction model 75 generated by the machine learning device 7 according to any one of (1) to (5) and the boiler system 1. The first time zone time series data of At least the maximum amount of steam used by the boiler group 2 in the time zone from the predetermined time on the working day of the boiler system 1 to the second hour ahead based on the set of the acquired predicted values of the steam amount in the unit time of the second hour ahead. , And a prediction unit 802 for predicting the average amount of steam used.
By doing so, the prediction device 8 of the present embodiment is a boiler system 1 including a boiler group 2 having a plurality of stage value control boilers 20 capable of burning at a plurality of staged combustion positions, such as month, season, or quarter. Even if the steam load pattern differs depending on the day, holiday, etc., for example, from immediately after the first hour to the second hour ahead based on the history of the total output steam amount by the boiler group 2 in the first time zone immediately before. The maximum amount of steam used and the average amount of steam used in the second time zone can be predicted.

(7) The boiler group 2 included in the boiler system 1 of the present embodiment is the boiler group 2 having a plurality of stage value control boilers 20 capable of controlling combustion at a plurality of staged combustion positions, and is a unit control parameter output device 9. Is the maximum steam usage of the boiler system in the time zone, based on the maximum steam usage in the time zone from a predetermined time to the second hour ahead and the average steam usage predicted by the prediction device 8. A unit control parameter output unit 901 that calculates any two or more values of the average steam amount and the variable steam amount and outputs the values to the number control device 3 of the boiler system 1 is provided.
By doing so, the boiler system 1 of the present embodiment has an appropriate maximum steam amount and average used in the time zone from a predetermined time to the second hour ahead based on the predicted values of various steam amounts predicted in real time. The amount of steam and the amount of fluctuating steam can be set automatically.

(8) The number control device 3 of the present embodiment has a predetermined time based on any two or more values of the maximum used steam amount, the average steam amount, and the variable steam amount acquired from the number control parameter output device 9. An appropriate number control parameter may be set in the time zone from to the second hour ahead, and the boiler group 2 may be controlled for combustion based on the set number control parameter.
By doing so, the boiler system 1 of the present embodiment can perform optimum number control in the time zone from a predetermined time to the second hour ahead based on the predicted values of various steam amounts predicted in real time. can.

(9) The boiler group 2 included in the boiler system 1 of the present embodiment is the boiler group 2 having a plurality of stage value control boilers 20 capable of controlling combustion at a plurality of stage combustion positions, and is a unit control parameter output device 9. Is the most suitable number control parameter group ID in the time zone based on the maximum steam usage amount and the average steam usage amount in the time zone from the predetermined time to the second hour ahead predicted by the prediction device 8. A unit control parameter output unit 901 to be selected and output to the control device of the boiler system is provided.
By doing so, the boiler system 1 of the present embodiment has an appropriate maximum steam amount and average used in the time zone from a predetermined time to the second hour ahead based on the predicted values of various steam amounts predicted in real time. The amount of steam and the amount of fluctuating steam can be set automatically.

(10) The number control device 3 of the present embodiment has an appropriate number control parameter in the time zone from the predetermined time to the second hour ahead based on the number control parameter group ID acquired from the number control parameter output device 9. May be set and the boiler group may be controlled for combustion based on the set number control parameter.
By doing so, the boiler system 1 of the present embodiment can perform optimum number control in the time zone from a predetermined time to the second hour ahead based on the predicted values of various steam amounts predicted in real time. can.

100 Boiler system with prediction function 1 Boiler system 2 Boiler group 20-step value control boiler 3 Unit control device 30 Control unit 301 Unit control function unit 302 Time series data output unit 303 Unit control parameter setting unit 32 Storage unit 321 Maximum steam consumption storage Unit 322 Average steam amount storage unit 323 Variable steam amount storage unit 324 Unit control parameter group ID storage unit 325 Number control parameter group storage unit 4 Steam header 5 Steam pressure sensor 6 Time-series data storage device 60 Control unit 601 Time-series information recording unit 602 Unit-time data creation unit 603 1st time zone time-series data extraction unit 604 Steam load pattern acquisition unit 62 Storage unit 621 Total output steam amount time-series information storage unit 622 Unit-time steam amount time-series information storage unit 623 Steam load pattern information Storage unit 7 Machine learning device 70 Control unit 701 Teacher data creation unit 702 Learning unit 703 Prediction model provision unit 72 Storage unit 721 Teacher data storage unit 722 Prediction model storage unit 75 Prediction model (learned model)
8 Prediction device 80 Control unit 801 Input unit 802 Prediction unit 82 Storage unit 821 Prediction model storage unit 822 1st time zone time series data storage unit 823 2nd time zone time series data storage unit 9 Unit control parameter output device 90 Control unit 901 Unit control parameter output unit 92 Storage unit 921 Unit control parameter group storage unit

Claims (10)

A boiler group having a plurality of boilers, a control device for controlling the combustion state of the boiler group, and a sampling time output from a boiler system having each preset sampling cycle, and at least the boiler at the sampling time. At least the maximum steam usage and average steam usage of the boiler group for each predetermined unit time, calculated based on measurement data including the total output steam amount of the group and time series data of the sampling order for each working day including. From the set of time series data of unit time steam amount including, the first time zone time series data which is a subset of the time series data of the unit time steam amount included in the preset first time zone is extracted. ,
By extracting the steam amount of the unit time of the second time ahead preset from immediately after the first time zone from the set of the time series data of the unit time steam amount, the first time zone time series data and the said The teacher data creation department that creates teacher data including the amount of steam in the unit time of the second hour ahead,
Based on the teacher data created by the teacher data creation unit, from the time-series data of the unit time steam amount included in the first time zone on an arbitrary working day, immediately after the first time zone and the second hour ahead. A learning unit that generates a prediction model that predicts the unit time steam amount including at least the maximum steam usage amount and the average steam usage amount of the boiler group in the unit time of the above.
A machine learning device equipped with. The working days are associated with a preset steam load pattern.
The machine learning device according to claim 1, wherein the learning unit generates the prediction model in response to the steam load pattern. The learning unit is supervised based on teacher data in which the time-series data of the first time zone is used as input data and the amount of steam in the unit time immediately after the first time zone and the second time ahead is used as label data. The machine learning device according to claim 1 or 2, wherein the learning is executed and the trained model as the prediction model is generated. The learning unit executes a time-series analysis based on the time-series data for each unit time in the first time zone and the amount of steam in the unit time immediately after the first time zone and the second time ahead. The machine learning device according to claim 1 or 2, wherein the prediction model is generated. The machine learning device according to any one of claims 1 to 4, wherein the machine learning device is composed of one or more servers communicably connected to the control device via a network. The prediction model generated by the machine learning apparatus according to any one of claims 1 to 5.
The first time zone time series data immediately before an arbitrary time on the working day of the boiler system is acquired for each unit time, and using the prediction model, the unit time two hours ahead of the time for each unit time. Get the predicted value of steam amount of
At least in the time zone from a predetermined time on the operating day of the boiler system to the second hour ahead, based on the set of the predicted steam amount values of the unit time in the second hour ahead acquired for each unit time. A predictor that predicts the maximum steam usage and average steam usage of the boiler group,
A predictor equipped with. The boiler group included in the boiler system is a boiler group having a plurality of stage value control boilers capable of combustion control at a plurality of stage combustion positions.
The maximum amount of the boiler system in the time zone based on the maximum steam usage amount and the average steam usage amount in the time zone from the predetermined time to the second hour ahead predicted by the prediction device according to claim 6. A unit control parameter output device including a unit control parameter output unit that calculates any two or more values of the steam amount used, the average steam amount, and the variable steam amount and outputs the values to the control device of the boiler system. From the predetermined time to the second hour ahead based on any two or more values of the maximum steam amount used, the average steam amount, and the variable steam amount acquired from the number control parameter output device according to claim 7. Set the appropriate number control parameters in the time zone and
The control device according to claim 1, wherein the boiler group is burned and controlled based on the set number control parameter. The boiler group included in the boiler system is a boiler group having a plurality of stage value control boilers capable of combustion control at a plurality of stage combustion positions.
The most suitable unit control in the time zone based on the maximum steam usage amount and the average steam usage amount in the time zone from the predetermined time to the second hour ahead predicted by the prediction device according to claim 6. A unit control parameter output device including a unit control parameter output unit that selects a parameter group ID and outputs it to the control device of the boiler system. Based on the number control parameter group ID acquired from the number control parameter output device according to claim 9, an appropriate number control parameter in the time zone from the predetermined time to the second hour ahead is set.
The control device according to claim 1, wherein the boiler group is burned and controlled based on the set number control parameter.

Images