[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

WO2018016776A1 - Procédé et dispositif de commande optimale de réfrigérateur au moyen d'un système de gestion énergétique du bâtiment - Google Patents

Procédé et dispositif de commande optimale de réfrigérateur au moyen d'un système de gestion énergétique du bâtiment Download PDF

Info

Publication number
WO2018016776A1
WO2018016776A1 PCT/KR2017/007133 KR2017007133W WO2018016776A1 WO 2018016776 A1 WO2018016776 A1 WO 2018016776A1 KR 2017007133 W KR2017007133 W KR 2017007133W WO 2018016776 A1 WO2018016776 A1 WO 2018016776A1
Authority
WO
WIPO (PCT)
Prior art keywords
refrigerator
cold water
variable
power consumption
random forest
Prior art date
Application number
PCT/KR2017/007133
Other languages
English (en)
Korean (ko)
Inventor
서원준
신한솔
추한경
박철수
Original Assignee
성균관대학교 산학협력단
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from KR1020160092772A external-priority patent/KR101754536B1/ko
Priority claimed from KR1020160098519A external-priority patent/KR101727434B1/ko
Application filed by 성균관대학교 산학협력단 filed Critical 성균관대학교 산학협력단
Publication of WO2018016776A1 publication Critical patent/WO2018016776A1/fr

Links

Images

Classifications

    • F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
    • F24—HEATING; RANGES; VENTILATING
    • F24F—AIR-CONDITIONING; AIR-HUMIDIFICATION; VENTILATION; USE OF AIR CURRENTS FOR SCREENING
    • F24F11/00—Control or safety arrangements
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
    • G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
    • G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators

Definitions

  • the present invention relates to a method and apparatus for optimal control of a refrigerator through a building energy management system, and more particularly, collects BEMS data necessary for determining the performance of a refrigerator through a building energy management system (BEMS).
  • BEMS building energy management system
  • the present invention relates to an optimal control method and apparatus of a refrigerator through a building energy management system using a random forest algorithm by applying the machine learning method and an optimization algorithm.
  • In-building refrigerators are installed to handle cooling loads in the building, and in the process, the refrigerator consumes energy such as gas / power.
  • the ratio of the amount of heat removed by the freezer to the energy consumed by the freezer is referred to as a COP (Coefficient of Performance).
  • the efficiency of the refrigerator is not a fixed value, and varies freely depending on various physical quantities (cold water temperature, cooling water temperature, cold water flow rate, freezer capacity, etc.) related to the operation of the freezer and the ambient environmental conditions in which the freezer is installed. . That is, the efficiency of the refrigerator can be changed in a very wide range according to the setting of the control value of the refrigerator such as the set temperature of the cold water and the temperature of the incoming water of the cooling water. Therefore, the use efficiency of the refrigerator varies depending on how to optimize and set the control value of the refrigerator, which may be a very important factor for reducing the cooling energy consumption of building energy.
  • a dynamic simulation tool eg, EnergyPlus
  • EnergyPlus a dynamic simulation tool
  • the actual refrigerator must be accurately simulated, and in order to do so, all relevant input variables required for modeling the refrigerator must be accurately input. Therefore, building a freezer model based on a dynamic simulation tool requires much time, effort, and expertise, and thus it is difficult to obtain satisfactory results.
  • the state of the refrigerator also changes, and thus there is a problem in that modeling is performed by continuously correcting and updating the simulation model according to the changing state.
  • an object of the present invention is to solve such a conventional problem, building energy management system that can set the optimal refrigerator control value by using the machine learning model and optimization algorithm using the collected BEMS data as input variables It is to provide an optimal control method and apparatus of the refrigerator.
  • the object according to the present invention, the step of collecting building energy management system (BEMS) data required for the performance control of the refrigerator; Selecting sets of input variables for operating the refrigerator at optimum efficiency using a machine learning model and an optimization algorithm constructed to predict power consumption of the refrigerator using the BEMS data as an input variable; Inputting each of the set of input variables into the machine learning model to predict power usage of the refrigerator for each of the set of input variables; And setting the control value of the refrigerator by comparing the predicted power consumption with the current power consumption of the refrigerator.
  • BEMS building energy management system
  • the BEMS data is at least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the refrigerator, the flow rate of the cold water, the inlet temperature of the cooling water, the outlet temperature of the cooling water, the capacity of the freezer and the power consumption of the refrigerator. It can include.
  • the BEMS data may include at least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the freezer, the inlet temperature of the coolant, the outlet temperature of the coolant, the capacity of the freezer and the power consumption of the freezer. .
  • At least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the refrigerator, the inlet temperature of the cooling water, the outlet temperature of the cooling water, the capacity of the refrigerator and the power consumption of the refrigerator After collecting the BEMS data, at least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the refrigerator, the inlet temperature of the cooling water, the outlet temperature of the cooling water, the capacity of the refrigerator and the power consumption of the refrigerator.
  • the variable may further include the predicted flow rate of cold water.
  • the machine learning method of the machine learning model may be any one of artificial neural networks, support vector machines, Gaussian process modeling, random forest, and genetic programming.
  • the random forest method may include: (a) constructing a random forest model by setting the BEMS data as an input variable and setting an output variable of the power consumption of the refrigerator; (b) determining the importance of the input variable in the process of constructing the random forest model; (c) selecting a part of the input variables in the order of high importance among the input variables as a new input variable; And (d) rebuilding a random forest model with the selected new input variable, and using the random forest model constructed in step (d), determine the power consumption of the refrigerator.
  • the variable constructed data in the step (a) can be added as an input variable .
  • variable construction may be performed based on a relational expression correlated with the output variable.
  • step (b) the importance may be determined in consideration of the effect on the value of the output variable when the value of any one of the input variables is changed.
  • the optimization algorithm may be any one of genetic algorithm, simulated annealing, tabu search, particle swarm optimization, or ant colony optimization. Can be.
  • the object is, according to the present invention, a data receiving unit for receiving the BEMS data necessary for determining the performance of the refrigerator from a building energy management system (BEMS) server;
  • a power usage prediction module for predicting power usage of the refrigerator with a machine learning model constructed to predict power consumption of the refrigerator, and a set of the input variables for operating the refrigerator at optimum efficiency using the machine learning model and an optimization algorithm.
  • an input variable candidate selection module for selecting (sets) and inputting each of the sets of input variables as input variables to predict power usage of the refrigerator for each of the set of input variables through the power usage prediction module, Power consumption of the refrigerator by comparing It can be achieved by the optimum control device of the refrigerator through the building energy management system including the optimum control value setting unit for setting the control value.
  • the BEMS data is at least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the refrigerator, the flow rate of the cold water, the inlet temperature of the cooling water, the outlet temperature of the cooling water, the capacity of the freezer and the power consumption of the refrigerator. It can include.
  • the BEMS data may include at least one of the cold water inlet temperature of the refrigerator, the cold water outlet temperature of the freezer, the inlet temperature of the coolant, the outlet temperature of the coolant, the capacity of the freezer and the power consumption of the freezer. .
  • the at least one of the cold water inlet temperature of the freezer, the cold water outlet temperature of the freezer, the inlet temperature of the coolant, the outlet temperature of the coolant, the capacity of the freezer and the power consumption of the freezer may be used as input variables.
  • the apparatus further includes a cold water flow rate prediction module configured to predict the flow rate of the cold water using a machine learning model constructed to predict a flow rate, wherein the input variable of the machine learning model constructed to predict power consumption of the refrigerator is determined by the predicted cold water. It may further comprise a flow rate.
  • the machine learning method of the machine learning model may be any one of artificial neural networks, support vector machines, Gaussian process modeling, random forest, and genetic programming.
  • the BEMS data is used as an input variable
  • the power consumption of the refrigerator is set as an output variable to construct a first random forest model, and in the process of constructing the first random forest model. Determining an importance of the input variable, selecting a part of the input variables in the order of high importance among the input variables as a new input variable, rebuilding a second random forest model with the selected new input variable, and The power consumption of the refrigerator may be determined using a second random forest model.
  • variable construction may be performed to increase the number of input variables from the BEMS data, and the variable randomized data may be added as an input variable to construct the first random forest model. have.
  • variable construction may be performed based on a relational expression correlated with the output variable.
  • the importance may be determined in consideration of the influence on the value of the output variable when the value of any one of the input variables is changed when the first random forest model is constructed.
  • the optimization algorithm may be any one of genetic algorithm, simulated annealing, tabu search, particle swarm optimization, or ant colony optimization. Can be.
  • a machine learning model using BEMS data as an input variable is a control value for allowing a refrigerator installed in a building to operate at an optimum efficiency;
  • control value set in real time operating the refrigerator at the optimum efficiency has the advantage that it is possible to reduce the energy used to operate the refrigerator in the building.
  • a random forest model can be constructed to accurately predict the performance of the refrigerator.
  • variable construction when the variable construction is performed, it is possible to further improve the accuracy of the predicted result by performing the variable construction based on a relation that correlates with the output variable.
  • FIG. 1 is a structural diagram schematically showing a structure of a cooling device according to an embodiment of the present invention.
  • FIG. 2 is a flow chart of a method for optimal control of a refrigerator through a building energy management system according to an embodiment of the present invention.
  • FIG. 3 is a flow chart of a method for optimal control of a refrigerator through a building energy management system according to another embodiment of the present invention.
  • 4 is a three-dimensional graph showing the relationship between the cooling water efficiency and the cooling water efficiency according to the cooling water extraction temperature.
  • FIG. 5 is a diagram illustrating a process of selecting an input variable of a machine learning model constructed to predict power consumption of a refrigerator by an optimization algorithm according to an exemplary embodiment of the present invention.
  • 6 and 7 are graphs showing the relationship between the power consumption predicted from the machine learning model according to the present invention and the actually measured power consumption.
  • 8 to 10 are graphs showing the relationship between the predicted cold water flow rate and the actual measured cold water flow rate from the machine learning model according to the present invention, with different input variables.
  • FIG. 11 is a block diagram of an optimum control device of the refrigerator via a building energy management system according to an embodiment of the present invention.
  • FIG. 12 is a block diagram of an optimum control device for a refrigerator via a building energy management system according to another embodiment of the present invention.
  • FIG. 13 is a flowchart illustrating a method of determining a performance of a refrigerator using a random forest model according to an embodiment of the present invention.
  • FIG. 14 is a diagram illustrating a modeling process according to a random forest algorithm.
  • 15 is a comparison table for explaining a method of determining the importance of an input variable according to an embodiment of the present invention.
  • FIG. 16 is a graph illustrating a comparison of a power consumption of a refrigerator and actual measured power consumption by constructing a random forest model from BEMS data without data preprocessing.
  • FIG. 17 is a graph showing the importance determined according to the present invention for BEMS data and input variables that are randomly generated data.
  • FIG. 18 is a graph illustrating the distribution of RMSE and CVRMSE when the input variables are removed in the order of low importance with respect to the input variable of FIG. 17.
  • FIG. 19 is a graph showing importance determined according to the present invention for input variables, which are data generated based on BEMS data and relational expressions correlated with output variables therefrom.
  • 20 is a graph showing the distribution of RMSE and CVRMSE when the input variables are removed in the order of low importance for the input variable of FIG. 19.
  • ' ⁇ module' means a hardware component such as software, FPGA or ASIC, and the module performs certain functions.
  • modules are not meant to be limited to software or hardware.
  • the module may be configured to be in an addressable storage medium and may be configured to play one or more processors.
  • a module may include components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, procedures, subroutines. , Segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables.
  • the functionality provided within the components and modules may be combined into a smaller number of components and modules or further separated into additional components and modules.
  • the components and modules may be implemented to reproduce one or more CPUs in a device.
  • FIG. 1 is a structural diagram schematically showing the structure of a refrigeration apparatus according to an embodiment of the present invention.
  • Cooling devices used in buildings can be divided into central and individual, depending on the size of the building and the ventilation system. Among them, in order to supply cooled air to a large space such as a large building or a subway station, a central cooling device for supplying cool air to each room of the building through the control of the large refrigerator 10 is frequently used.
  • Indirect cooling method to be supplied to the room by the use is used, in this embodiment based on the cooling device of the indirect cooling method shown in Figure 1 Optimal control of the refrigerator through the building energy management system according to an embodiment of the present invention The method will be described.
  • a reciprocating freezer a rotary freezer, a scroll freezer, a screw freezer, a turbo freezer, an absorption freezer and an absorption cold / hot water freezer, and the like can be used, but are not limited thereto.
  • FIG. 2 is a flow chart of a method for optimal control of a refrigerator via a building energy management system according to an embodiment of the present invention
  • Figure 3 is a method of optimal control of a refrigerator via a building energy management system according to another embodiment of the present invention.
  • Figure 4 is a three-dimensional graph showing the relationship between the efficiency of the freezer according to the exit temperature of the cold water and the exit temperature of the cooling water
  • Figure 5 is an optimization algorithm according to an embodiment of the present invention predicts the power consumption of the refrigerator
  • FIG. 6 and FIG. 7 are graphs showing the relationship between the power consumption predicted from the machine learning model according to the present invention and the actually measured power consumption.
  • 8 to 10 illustrate the measured cold water flow rate and the actual measured values from the machine learning model according to the present invention, with different input parameters. It is a graph which shows the relationship between cold water flow volume.
  • a method of optimally controlling a refrigerator through a building energy management system includes collecting building energy management system (BEMS) data (S110), and inputting all or part of BEMS data to input variables.
  • BEMS building energy management system
  • BEMS data is collected from the building energy management system server 300 (S110).
  • the building energy management system is a system that collects and analyzes various information of energy management facilities in buildings such as buildings in real time to improve energy use efficiency.
  • the present invention is for optimal efficiency operation of the refrigerator 10 as described above with reference to Figure 1 can collect the BEMS data required for the performance control of the refrigerator 10 from the BEMS data.
  • BEMS data required for the performance control of the refrigerator 10 the cold water inlet temperature of the refrigerator 10, the cold water outlet temperature of the refrigerator 10, the flow rate of cold water, the inlet temperature of the cooling water, the outlet temperature of the cooling water, the freezer ( It may include the capacity of 10) and the power consumption of the refrigerator 10, but is not limited thereto.
  • a machine learning model for estimating the power consumption of the refrigerator 10 using all or a part of the collected BEMS data as an input variable can be constructed. That is, using the BEMS data collected and stored at regular time intervals to build a machine learning model to predict the power consumption and / or the efficiency of the refrigerator (10).
  • an algorithm such as an artificial neural network, a support vector machine, a Gaussian process modeling, a random forest technique, or a genetic programming may be used as the machine learning method of the machine learning model according to the present invention.
  • a random forest technique is used, and a conventionally known random forest technique is modified. This will be described later with reference to FIGS. 13 to 20.
  • a set of input variables for operating the refrigerator 10 at an optimum efficiency is selected by using a machine learning model and an optimization algorithm constructed to predict power consumption (S120). .
  • Figure 4 shows the relationship between the cold water outflow temperature and the cooling water outlet temperature (COP) when the input temperature as the input variable in a three-dimensional contour (contour), the actual refrigerator 10 is the cold water outflow
  • the actual refrigerator 10 is the cold water outflow
  • it may be influenced by the inlet temperature of the cold water, the flow rate of the cold water, the inlet temperature of the coolant, the capacity of the freezer, etc., which is not the three-dimensional contour as shown in FIG. Can be defined.
  • the performance of the refrigerator in such a multi-dimensional space changes more rapidly with the change of the input variable, and the change range or the width thereof is wider. Therefore, it is necessary to consider together the performance of the refrigerator which changes according to the various input variables as described above.
  • the present invention selects a set of input variables for operating the refrigerator at the optimum efficiency by using the above-described machine learning model and optimization algorithm.
  • the input variables are the inlet temperature of the cold water, the outlet temperature of the cold water, the flow rate of the cold water, the inlet temperature of the coolant, the outlet temperature of the coolant, and the capacity of the freezer as shown in FIG.
  • Performance of the refrigerator 10 by applying the above-described machine learning model for predicting the power consumption of the refrigerator 10 and / or the performance of the refrigerator 10 from the collected BEMS data and an optimization algorithm for obtaining an optimal set of input variables therefrom.
  • the candidate group of the input variables is selected to enable the refrigerator 10 to operate at optimum efficiency under the conditions of the currently collected BEMS data.
  • the optimization algorithm used may be genetic algorithm, simulated annealing, tabu search, particle swarm optimization, or ant colony optimization. It is not limited to this.
  • the power consumption of the refrigerator 10 for each of the three sets of input variables may be predicted through the machine learning model that predicts the above-described power usage by using each of the candidate groups of the input variables selected as the input variables (S130). .
  • the power consumption of the refrigerator 10 predicted for each candidate group is compared with the power consumption of the refrigerator 10 that is actually consumed in the current state to set a control value of the refrigerator 10 (S140).
  • a control value of the refrigerator 10 S140.
  • the power consumption of the refrigerator 10 predicted with the input variable set 1 with the power usage of the refrigerator 10 collected by the current BEMS data if the predicted power consumption is large, for optimal control Since the energy consumption may be large, the condition of the input variable of set 1 is not preferable, and the power consumption of the refrigerator 10 predicted using the input variable set 2 and the power consumption of the refrigerator 10 collected by the current BEMS data are compared.
  • the optimal control is performed and the energy consumption can be reduced. Therefore, the control value of the refrigerator 10 can be set based on the input variable of set 2.
  • control value of the refrigerator 10 may be a condition related to the operation of the refrigerator 10 that can be arbitrarily changed, such as the temperature of the cooling water and the flow rate of the cold water.
  • the latest BEMS data may be collected and the control value may be set again to control the refrigerator at the optimum efficiency.
  • the machine learning model built to predict the state of the refrigerator 10 can continuously reflect the state of the refrigerator 10 in the building, and the state of the refrigerator can be accurately predicted.
  • 6 and 7 illustrate the chiller predicted by the machine learning module using the inlet temperature of the cold water, the outlet temperature of the cold water, the flow rate of the cold water, the inlet temperature of the coolant, the outlet temperature of the coolant, the inlet temperature of the coolant, and the capacity of the freezer as input variables.
  • the graph shows a comparison of the measured power consumption and the actual measured value.
  • the CVRMSE (Coefficient of Variation of the Root Mean Square Error) value is 2.4%. Therefore, it could be verified that the machine learning model can be used to accurately predict the power consumption of the refrigerator 10.
  • the optimal control method of the refrigerator using the building energy management system is a step of collecting building energy management system (BEMS) data (S210), and the flow rate of cold water using the collected BEMS data as an input variable.
  • BEMS building energy management system
  • Predicting the power consumption of the refrigerator 10 for each of the variable set (S240), and the power consumption of the refrigerator 10 and the current power consumption of the refrigerator 10 Comparing the capacity may include the step (S250) of setting the control value of the refrigerator (10).
  • step S220 is added in comparison with the above-described embodiment with reference to FIG. 2, the following description focuses on the difference from the method described with reference to FIG. 2. Let's do it.
  • the collected BEMS data includes the flow rate of cold water. Although it is possible to measure the flow of cold water by installing a flow meter in the pipe through which cold water flows, most of the buildings do not have a flow meter installed in the cold water system of the refrigerator. There is a problem that the reliability of the measured value is inferior.
  • the flow rate of the cold water is measured using a flow meter, not collected as BEMS data, and the collected cold water flow rate is predicted by using a machine learning method using a machine learning model.
  • a machine learning method of a machine learning model well-known algorithms, such as an artificial neural network, a support vector machine, Gaussian process modeling, a random forest technique, or genetic programming, can be used.
  • FIG. 8 illustrates the difference between the inlet temperature of the cold water and the outlet temperature of the cold water as the input variable of the machine learning model constructed to predict the cold water flow rate, and the power consumption of the refrigerator.
  • FIG. 10 shows the cold water outlet temperature, the cold water inlet temperature, the cold water outlet temperature, the coolant inlet temperature, and the freezer power consumption as input parameters of the machine learning model. It is a graph which compares and shows the flow rate of the cold water predicted by the machine learning model in case of setting, and the flow of the cold water actually measured using the actual flow meter.
  • Model 1 ( Figure 8) Model (2) ( Figure 9) Model 3 (FIG. 10) RMSE (kg / min) 0.47 0.46 0.43 CVRMSE (%) 5.5% 5.3% 5.0% MBE (%) 0.3% 0.24% 0.07%
  • the flow rate of cold water as one of the input variables of the machine learning model for predicting the power consumption of the refrigerator 10 is not collected as BEMS data using a flow meter in this embodiment, but instead of collecting the cold water flow rate from the machine learning model using other collected BEMS data. There is a difference in the prediction, and the rest is almost the same as the content described with reference to FIG.
  • FIG. 13 is a flowchart illustrating a method of determining a performance of a refrigerator using a random forest model according to an embodiment of the present invention
  • FIG. 14 is a diagram illustrating a modeling process according to a random forest algorithm
  • FIG. 15 is an embodiment of the present invention.
  • the method of determining the performance of the refrigerator 10 using the random forest model collecting BEMS data (S310) and performing variable construction to increase the number of input variables from BEMS data (S320). ), Building a random forest model for determining the performance of the refrigerator 10 based on the input variable (S330), determining the importance of the input variable in the process of building the random forest model (S340), of the input variable Reselecting an input variable having a high importance as an input variable according to the importance (S350), constructing a random forest model for determining the performance of the refrigerator 10 with the selected input variable (S360), and finally constructing the random forest. It may include the step (S370) of determining the performance of the refrigerator 10 by using the model.
  • BEMS data is collected from the building energy management system server (S310).
  • the cold water inlet temperature of the refrigerator 10 the cold water outlet temperature of the refrigerator 10
  • the cold water flow rate the cold water inlet temperature, the coolant outlet temperature, and the coolant flow rate
  • the capacity of the refrigerator 10 and the amount of power used by the refrigerator 10 may be included, but are not limited thereto.
  • a random forest which is an example of a machine learning model
  • a machine learning model is a data processing algorithm that learns the relationship between input and output variables to predict output variables according to input variables.
  • an input variable is required for learning, and the BEMS data becomes an input variable, and the output variable indicates the performance of the refrigerator 10.
  • the collected BEMS data values may be incomplete and there may be noise, and some of the collected BEMS data may not be important for determining the performance of the refrigerator 10, and thus may lower the performance of the random forest model.
  • the performance degradation may be a modeling time for constructing the random forest model, a computation time for performance determination, or the accuracy of an output variable determined by the random forest model.
  • data preprocessing may be performed on BEMS data as an input variable to improve the performance of the random forest model.
  • variable construction may be performed to increase the number of input variables from the collected BEMS data by the method of performing data preprocessing (S320). Constructing variables creates new variables from existing input variables that can improve the performance of the model. If you build a random forest model by adding input variables that correlate with the output variables you want to predict, you can increase the accuracy of the predicted results.
  • variable Arbitrary relation variable Arbitrary relation x13 x5 ⁇ x6-x4 x22 x8 ⁇ x9-x11 x14 x6 ⁇ x3-x3 x23 x3 ⁇ x3 ⁇ x8 x15 x4-x2 + x6 x24 x2-x4 ⁇ x3 x16 x8-x1 ⁇ x2 x25 x10-x1 + x3 x17 x6 ⁇ x13-x5 x26 x10 + x4 ⁇ x3 x18 x11 + x9 ⁇ x6 x27 x8 + x11 ⁇ x1 x19 x12 ⁇ x8 + x8 x28 x6-x12-x2 x20 x8 ⁇ x7 ⁇ x11 x29 x12 + x5 ⁇ x2 x21 x11 + x8 ⁇ x10 x30
  • x1 to x12 are collected BEMS data
  • x13 to x30 are input variables newly generated by constructing variables in arbitrary relations from BEMS data, respectively.
  • variable construction may be performed in the same manner as described above, it is preferable to perform variable construction based on a relation that correlates with an output variable.
  • Q is calorie
  • C is the specific heat of cold water
  • Breiman proposed a random forest algorithm that combines random input selection, first proposed by Amit et al, with bagging (Bootstrap Aggregating).
  • the Ensemble Method a machine learning method that combines multiple decision trees to predict the state of the system, averages the prediction results of each decision tree as the average for regression and votes for classification. Calculate Therefore, it is known that generalization performance is superior to single decision trees.
  • Random Forest features include bagging, random input variable selection, and out-of-bags error rate.
  • the bagging method proposed by Breiman is one of the ensemble methods, and complements the instability of decision trees by using bootstrap, a random reconstruction method.
  • the entire training data used for the learning of bagging is extracted by the bootstrap method, divided into n bootstrap samples, and aggregating n models trained on different training data. Since a typical decision tree has small deviations and large variances, very deeply grown decision trees are overfitted with training data.
  • Bootstrap improves the performance of the overall model because it reduces variance while maintaining the variation of each tree. That is, one decision tree is very sensitive to the noise of the training data, but if the correlation between trees is small, the average of several trees becomes strong against noise. Therefore, bagging is a way to uncorrelate each tree by training each decision tree with a different data set.
  • -Random input variable selection This method selects m input variables among all input variables and finds the optimal node segmentation criteria among the selected input variables when the node of the decision tree is split. The result is a set of decision trees that have different structures but perform well. This technique is important to determine the number of randomly selected variables (m). The closer m is to 1, the greater the variance and variance of each decision tree, making it more effective for the average or voting method that is characteristic of the ensemble method. The closer m is to the total number of input variables M, the smaller the variation and variance of each tree, and the less effective the ensemble method is.
  • OOB error rate Approximately 37% of the total data is not included in the bootstrap sample. Observations not included in the bootstrap sample are used as out-of-bags (OOB) data to evaluate the generalization performance of the model. The OOB error rate is used as a basic indicator to verify the performance of the random forest model. In addition, random forest can evaluate the influence of input variables on the prediction of output variables with variable importance, which helps to improve model interpretation ability.
  • bootstrap randomly extracts n bootstrap samples from BEMS data and newly constructed variable data therefrom. Each sample is used as training data for a decision tree. In addition, 37% of the data not included in the training data is extracted as OOB samples.
  • OOB sample The OOB error rate, which is a generalized performance index of the random forest model, is calculated. The user can also verify that an appropriate number (n) of decision trees have been generated based on the error rate. In addition, it is also possible to calculate the importance of the variable to be described later using the OOB sample.
  • the importance level is determined in consideration of the effect on the output variable value when any one of the input variable values is changed. That is, if the value of the output variable is large when the value of one input variable is changed, the input variable may be judged to have high importance. On the contrary, if the value of the output variable is small when the value of the input variable is changed, It can be determined that the importance of the input variable is low.
  • Fig. 15B when predicting the output variable Y using a random forest model constructed using the input variables X1 and X2, three OOB samples not used in the model are cycled and inputted. You can determine the importance of a variable.
  • Fig. 15B the X1 value is circulated compared to Fig. 15A, which is the reference data, and the error rate of the model predicted by this data is 1/3.
  • FIG. 15C the X2 value is cycled compared to FIG. 4A, and the error rate of the model predicted by this data is 2/3. Accordingly, X1 may be determined as an input variable that does not significantly affect the prediction of the output variable compared to X2, and X2 may be determined as an input variable that has a great influence on the prediction of the output variable compared to X1.
  • a portion of the input variables in the order of high importance may be selected as the new input variable (S350). For example, if there are 30 input variables, the importance is judged in the same manner as above, and when they are arranged in the order of importance, the top 5 variables can be selected as the input variables in order of importance. . In this case, a method of determining the number of input variables to be selected will be described later.
  • a random forest model is again constructed using input variables selected in a predetermined number in order of high importance (S360). Therefore, since the number of input variables required for constructing the random forest model is reduced, the time required for constructing the random forest model can be reduced. In addition, since the input variables of high importance are considered, prediction accuracy can be improved.
  • the input variable data used to construct the random forest model may be collected again to determine the performance of the refrigerator 10 (S370). According to the determined performance of the refrigerator 10, the operation of the refrigerator 10 may be controlled to operate the refrigerator 10 at an optimum efficiency.
  • FIG. 16 is a graph illustrating a comparison between the power consumption of the refrigerator 10 predicted therefrom and the actual measured power consumption by constructing a random forest model from BEMS data without data preprocessing
  • FIG. 17 is a random generation from the BEMS data. It is a graph showing the importance determined according to the present invention for the input variable, which is the data
  • FIG. 18 shows the distribution of RMSE and CVRMSE when the input variables are removed in the order of low importance for the input variable of FIG. 17.
  • 19 is a graph showing importance determined according to the present invention with respect to an input variable which is data generated based on BEMS data and a relational expression correlated with an output variable therefrom
  • FIG. 9 is an input of FIG. 8.
  • the graph shows the distribution of RMSE and CVRMSE when the input variables are removed in order of low importance for the variables.
  • FIG. 16 illustrates the power consumption of the refrigerator 10 and the actual measured power consumption predicted by constructing a random forest model from BEMS data without preprocessing of the above-described data.
  • the root mean square error (RMSE) and the coefficient of variation of root mean square error (CRMRMSE) were 6.21 kW and 8.56%, respectively, indicating that the accuracy of the model constructed by the random forest is excellent.
  • RMSE and CVRMSE are 3.74 kW and 5.16%, respectively. Investigations show that the accuracy is improved compared to the case of building a random forest model using only BEMS data.
  • FIG. 17 is a graph illustrating the importance of input variables according to the present invention for a total of 30 input variables as shown in Table 2.
  • FIG. Accordingly, in consideration of the importance of the input variable, a variable having a high importance among the input variables is selected as a new input variable.
  • the model was constructed while removing the input variables having the least importance one by one, and the result is shown in FIG. 18.
  • the CVRMSE and RMSE values of the model begin to change and grow rapidly as soon as 25 input variables are removed (i.e., when the 5 critical input variables are used).
  • RMSE and CVRMSE are 3.95 kW and 5.44%, respectively.
  • the relationship between the BEMS data of X1 to X12 and the output variable therefrom (for example, In the case of constructing random forest model with 18 input variables by newly constructing 6 input variables of X31 to X36, RMSE and CVRMSE were investigated at 2.45kW and 3.37%, respectively. In comparison, the accuracy is further improved.
  • FIG. 19 is a graph illustrating the importance of input variables according to the present invention with respect to a total of 18 input variables, which are newly constructed data based on BEMS data and relational expressions correlated with output variables.
  • a variable having a high importance among the input variables is selected as a new input variable.
  • the model was constructed while removing the input variables having the least importance one by one, and the result is shown in FIG. 9.
  • RMSE and CVRMSE were 3.11 kW and 4.28%, respectively.
  • FIG 11 is a block diagram of an optimum control device of the refrigerator via a building energy management system according to an embodiment of the present invention
  • Figure 12 is an optimum control device of the refrigerator via a building energy management system according to another embodiment of the present invention.
  • the configuration diagram is a block diagram of an optimum control device of the refrigerator via a building energy management system according to an embodiment of the present invention.
  • Optimal control device 400 of the refrigerator via the building energy management system is a data receiving unit 410, power usage prediction module 420, input variable candidate selection module as shown in FIG. 430 and the optimum control value setting unit 440 may be configured.
  • the data receiver 410 receives BEMS data necessary for determining the performance of the refrigerator 10 from the building energy management system server 300.
  • the received BEMS data is the cold water inlet temperature of the freezer 10, the cold water outlet temperature of the freezer 10 , the flow rate of cold water, the inlet temperature of the coolant, the outlet temperature of the coolant, the capacity of the freezer 10 and the freezer 10
  • At least one of the power consumption may include, but is not limited thereto.
  • the power usage prediction module 420 is a machine learning model constructed to predict the power usage of the refrigerator 10 by using all or a part of BEMS data received from the data receiver 410 as an input variable, and the power usage of the refrigerator 10. To predict.
  • the neural network may be used as the machine learning method of the machine learning model.
  • the random forest algorithm is described as described with reference to FIGS. 13 to 20. It is preferable to use.
  • BEMS data is used as an input variable and power consumption of the refrigerator is set as an output variable to construct a first random forest model, and a first random forest model is constructed.
  • the importance of the input variable is determined, a portion of the input variables having the highest priority among the input variables is selected as a new input variable, the second random forest model is rebuilt with the selected new input variable, and the second random forest is selected.
  • the model can be used to determine the power consumption of the refrigerator.
  • the first random forest model when constructing the first random forest model, it is preferable to construct the first random forest model by performing variable construction to increase the number of input variables from the BEMS data and adding the variable constructed data as an input variable. Do. In addition, it is preferable to perform variable construction based on a relational expression correlated with an output variable as described above. Furthermore, the importance may be determined in consideration of the influence on the value of the output variable when the value of any one of the input variables is changed when the first random forest model is constructed.
  • the input variable candidate selection module 430 optimizes the refrigerator 10 as described with reference to FIG. 5 using a machine learning model constructed to predict power consumption of the refrigerator 10 and an optimization algorithm based thereon.
  • the candidate group of the set of input variables to operate the controller is selected.
  • an optimization algorithm known algorithms of genetic algorithm, simulated annealing, tabu search, particle swarm optimization, or ant colony optimization can be used. There is a number.
  • the optimum control value setting unit 440 inputs each of the input variable sets selected by the input variable candidate selection module 430 as an input variable in the machine learning model of the power usage prediction module 420 to provide a refrigerator for each input variable set.
  • the power consumption of 10 may be predicted, and the control value of the refrigerator 10 may be set by comparing the predicted power consumption with the current power consumption of the refrigerator 10.
  • the apparatus 500 includes a data receiver 510 and a cold water flow rate prediction module. 520, the power usage prediction module 530, the input variable candidate selection module 540, and the optimum control value setting unit 550 may be configured.
  • the apparatus 500 of FIG. 12 is an apparatus 500 corresponding to the method described above with reference to FIG. 3, and further includes a cold water flow rate prediction module 520 as compared to the apparatus 400 of FIG. 11. As described above, the cold water flow rate is different in that the cold water flow rate prediction module 520 predicts the cold water flow rate using the machine learning model without directly measuring the flow rate and receiving the data by the data receiving unit 510.
  • the cold water flow rate prediction module 520 may determine at least one of cold water inlet temperature, cold water outlet temperature, cooling water inlet temperature, cooling water outlet temperature, freezer capacity, and power consumption of the refrigerator, which are BEMS data received from the data receiver 510. Predict the cold water flow rate using a machine learning model constructed to predict the cold water flow rate as an input variable.
  • the machine learning method of the machine learning model used herein may use the above-described artificial neural network, support vector machine, Gaussian process modeling, random forest technique or genetic programming.

Landscapes

  • Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Chemical & Material Sciences (AREA)
  • Combustion & Propulsion (AREA)
  • Mechanical Engineering (AREA)
  • General Engineering & Computer Science (AREA)
  • Feedback Control In General (AREA)
  • Air Conditioning Control Device (AREA)

Abstract

La présente invention concerne un procédé et un dispositif de commande optimale d'un réfrigérateur au moyen d'un système de gestion énergétique du bâtiment (SGEB). Selon la présente invention, le procédé et le dispositif de commande optimale du réfrigérateur au moyen du SGEB consistent : à collecter des données de SGEB nécessaires pour une régulation du rendement du réfrigérateur ; à sélectionner des ensembles de variables d'entrée pour faire fonctionner le réfrigérateur à une efficacité optimale au moyen de l'utilisation d'un modèle d'apprentissage automatique et d'un algorithme d'optimisation construit de façon à prédire la consommation d'énergie du réfrigérateur au moyen de l'utilisation des données de SGEB en tant que variables d'entrée ; à introduire chacun des ensembles de variables d'entrée dans le modèle d'apprentissage automatique et à prédire la consommation d'énergie du réfrigérateur pour chacun des ensembles de variables d'entrée ; et à comparer la consommation d'énergie prédite avec la consommation d'énergie actuelle du réfrigérateur et à régler une valeur de régulation du réfrigérateur.
PCT/KR2017/007133 2016-07-21 2017-07-05 Procédé et dispositif de commande optimale de réfrigérateur au moyen d'un système de gestion énergétique du bâtiment WO2018016776A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
KR10-2016-0092772 2016-07-21
KR1020160092772A KR101754536B1 (ko) 2016-07-21 2016-07-21 건물 에너지 관리 시스템을 통한 냉동기의 최적 제어 방법 및 장치
KR10-2016-0098519 2016-08-02
KR1020160098519A KR101727434B1 (ko) 2016-08-02 2016-08-02 랜덤 포레스트 모델을 이용한 냉동기의 성능 판단 방법

Publications (1)

Publication Number Publication Date
WO2018016776A1 true WO2018016776A1 (fr) 2018-01-25

Family

ID=60993137

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/KR2017/007133 WO2018016776A1 (fr) 2016-07-21 2017-07-05 Procédé et dispositif de commande optimale de réfrigérateur au moyen d'un système de gestion énergétique du bâtiment

Country Status (1)

Country Link
WO (1) WO2018016776A1 (fr)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113240070A (zh) * 2021-05-19 2021-08-10 上海应用技术大学 基于pso-aco-bp神经网络的自动评分方法
KR102548894B1 (ko) * 2022-10-20 2023-06-29 한국전자기술연구원 건물 거주자의 개인성향 및 센싱 데이터 기반의 거주만족도 평가 시스템 및 방법
US12051082B2 (en) 2022-10-20 2024-07-30 Korea Electronics Technology Institute System and method for evaluating residential satisfaction based on resident's preferred section

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110137464A (ko) * 2010-06-17 2011-12-23 삼성물산 주식회사 건물 에너지 관리 시스템을 통한 냉동기 최적 운전 시스템 및 방법
KR101168153B1 (ko) * 2011-12-29 2012-08-07 지에스네오텍 주식회사 건물의 에너지 수요 예측 방법 및 그 시스템
KR20130122193A (ko) * 2012-04-30 2013-11-07 에스케이텔레콤 주식회사 냉수 왕복온도차 제어를 통한 에너지 관리 방법 및 이를 위한 에너지관리장치
KR20150041526A (ko) * 2013-10-08 2015-04-16 성균관대학교산학협력단 가우시안 프로세스 에뮬레이터를 이용한 건물 운영의 확률적 제어 방법
KR20160042673A (ko) * 2014-10-10 2016-04-20 엘지전자 주식회사 설비를 관제하는 중앙 제어 장치, 이를 포함하는 설비 제어 시스템 및 설비 제어 방법

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20110137464A (ko) * 2010-06-17 2011-12-23 삼성물산 주식회사 건물 에너지 관리 시스템을 통한 냉동기 최적 운전 시스템 및 방법
KR101168153B1 (ko) * 2011-12-29 2012-08-07 지에스네오텍 주식회사 건물의 에너지 수요 예측 방법 및 그 시스템
KR20130122193A (ko) * 2012-04-30 2013-11-07 에스케이텔레콤 주식회사 냉수 왕복온도차 제어를 통한 에너지 관리 방법 및 이를 위한 에너지관리장치
KR20150041526A (ko) * 2013-10-08 2015-04-16 성균관대학교산학협력단 가우시안 프로세스 에뮬레이터를 이용한 건물 운영의 확률적 제어 방법
KR20160042673A (ko) * 2014-10-10 2016-04-20 엘지전자 주식회사 설비를 관제하는 중앙 제어 장치, 이를 포함하는 설비 제어 시스템 및 설비 제어 방법

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113240070A (zh) * 2021-05-19 2021-08-10 上海应用技术大学 基于pso-aco-bp神经网络的自动评分方法
CN113240070B (zh) * 2021-05-19 2024-05-21 上海应用技术大学 基于pso-aco-bp神经网络的自动评分方法
KR102548894B1 (ko) * 2022-10-20 2023-06-29 한국전자기술연구원 건물 거주자의 개인성향 및 센싱 데이터 기반의 거주만족도 평가 시스템 및 방법
US11983729B1 (en) 2022-10-20 2024-05-14 Korea Electronics Technology Institute System and method for evaluating residential satisfaction based on resident's personal tendency and sensing data
US12051082B2 (en) 2022-10-20 2024-07-30 Korea Electronics Technology Institute System and method for evaluating residential satisfaction based on resident's preferred section

Similar Documents

Publication Publication Date Title
CN112577161B (zh) 空调能耗模型训练方法与空调系统控制方法
WO2018016776A1 (fr) Procédé et dispositif de commande optimale de réfrigérateur au moyen d'un système de gestion énergétique du bâtiment
KR101754536B1 (ko) 건물 에너지 관리 시스템을 통한 냉동기의 최적 제어 방법 및 장치
CN111795488B (zh) 一种分布式机房温度的智能调控系统及方法
CN117411189B (zh) 一种微电网协调控制器的监测数据增强方法
KR101727434B1 (ko) 랜덤 포레스트 모델을 이용한 냉동기의 성능 판단 방법
CN114901057B (zh) 一种数据中心机房内部多点能耗检测与动态调节系统
CN107105028A (zh) 一种基于云计算的机房环境智能调节系统
WO2017095101A1 (fr) Système pour mesurer et évaluer les performances énergétiques d'un bâtiment et son procédé de commande
WO2021145577A1 (fr) Procédé et appareil de prédiction de données de série temporelle
CN115542824B (zh) 一种基于能耗管控的中央空调机组控制方法及系统
CN113048807A (zh) 一种空冷机组背压异常检测方法
CN117553404A (zh) 一种提高大型水冷中央空调系统能效的方法及系统
CN113268913B (zh) 一种基于pso-elm算法的智能建筑空调冷机系统运行优化方法
MirhoseiniNejad et al. ALTM: Adaptive learning-based thermal model for temperature predictions in data centers
CN111814388A (zh) 一种基于神经网络的下送风数据中心cfd仿真校验方法
CN118265275B (zh) 风冷线控冷优化控制方法
CN117634217B (zh) 基于数字孪生的建筑物智能化管理系统及方法
CN118350600A (zh) 基于混合模型和多变量分析的供热负荷预测方法及系统
CN116757354A (zh) 一种基于多层感知机的烟叶复烤段关键参数筛选方法
CN116954329A (zh) 制冷系统的状态调节方法、装置、设备、介质及程序产品
CN113205178B (zh) 人工智能红外图像传感系统及方法
CN111210060B (zh) 一种工作日期间机房温度预测方法
CN115540202A (zh) 一种基于多维泰勒网的空调机组故障预测方法
Chumnanvanichkul et al. Three-level Classification of Air Conditioning Energy Consumption for Building Energy Management System Using Data Mining Techniques

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17831249

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17831249

Country of ref document: EP

Kind code of ref document: A1