CN116205375B - Pump station unit running state prediction method and system - Google Patents

Abstract

The application provides a method and a system for predicting the running state of a pump station unit. The method comprises the following steps: collecting self-variable data and running state actual measurement data of unit running state prediction, deleting invalid data in the self-variable data and carrying out data preprocessing to obtain the self-variable data after data preprocessing; based on a bidirectional LSTM neural network, a unit running state prediction model is established; training and testing a unit running state prediction model by applying the self-variable data and running state actual measurement data which are subjected to data preprocessing; and predicting the change trend of the real-time running state of the pump station unit by applying the trained unit running state prediction model. According to the scheme provided by the application, the performance influence caused by the introduction of the abnormal value is fully considered, a data processing method of missing labels and multi-point fusion is designed, and the training calculation parameters of a network are effectively reduced; and secondly, multi-step long multi-element data preprocessing is realized, so that the difficulty in use of the multi-element data preprocessing in different scenes is reduced.

Description

A method and system for predicting the operating state of a pumping station unit

technical field

The invention belongs to the field of predicting the operating state of a pumping station unit, in particular to a method and system for predicting the operating state of a pumping station unit.

Background technique

The unit status change trend prediction technology is an important means to realize unit maintenance in advance. According to the history and current operation status of the unit, the future development trend of the unit’s working status can be predicted. The pumping station unit is a large-scale engineering unit, which is composed of multiple sets of equipment such as the main motor unit and the water unit. Its operating conditions are complex, and the main indicators for judging its operation safety are temperature, vibration and swing. Mainly rely on the unit electrical data, temperature parameters and vibration parameters. Electrical and temperature sensors are relatively mature and easy to obtain in practical applications. However, the vibration and swing parameters of the unit are easily affected by noise and mechanical factors in the operating environment when they are acquired, resulting in invalid values. Therefore, the first step to evaluate the operating status of the pumping station unit It is necessary to carry out the prediction of vibration and swing data, and then realize the prediction of the state change trend of the pumping station unit.

The pumping station unit has accumulated a certain amount of historical operation data, including the real-time operation data of the unit, the start-up time data of the unit and the data of the power supply module, etc., to find the characteristic parameters that characterize the operation status of the main equipment of the unit, and realize the monitoring and monitoring of the operation status of the pumping station unit. Prediction: Evaluate the health status of the main equipment of the unit through the forecast of the unit’s operating status trend, which can effectively judge the current operating status of the pumping station, and then predict the possibility and time of the abnormal operation of the unit, and analyze the working status of various monitoring equipment. Detect equipment abnormalities and fault symptoms, and generate alarm records in the system according to the type and level of abnormalities, providing technical support for the intelligent operation and maintenance of auxiliary units of the pumping station and long-term guarantee of production safety of the units.

At present, the research on the trend prediction of the operating state of the unit is mainly through the collection of state data such as vibration and swing of the pumping station unit, and displaying the unit status monitoring diagram, bar graph, table, etc., showing the value, peak-peak value and some parameters of each parameter. Time-frequency characteristics, showing the status of equipment in a statistical way. In the face of such massive monitoring data, which often has time-varying and multivariate coupling, and there are complex correlations and change mechanisms among variables, it is difficult for traditional data statistical methods to monitor and control the operating status of units in the absence of prior knowledge. Trend forecasting.

Contents of the invention

In order to solve the above-mentioned technical problems, the present invention proposes a technical proposal of a method for predicting the operating state of pumping station units to solve the above-mentioned technical problems.

The first aspect of the present invention discloses a method for predicting the operating state of a pumping station unit, the method comprising:

Step S1, based on the time-series data of unit operation, collect the independent variable data of unit operating state prediction and the actual measurement data of operating state, and delete invalid data in the independent variable data;

Step S2, performing data preprocessing on the independent variable data from which invalid data is deleted, to obtain the independent variable data after data preprocessing;

Step S3, based on the bidirectional LSTM neural network, establish a unit operation state prediction model; apply the independent variable data after data preprocessing and the actual measurement data of the operation state to train and test the unit operation state prediction model;

Step S4, applying the trained unit operation state prediction model to predict the change trend of the real-time operation state of the pumping station unit.

According to the method of the first aspect of the present invention, in the step S1, the independent variable data include: thrust bearing bush temperature, upper guide shoe temperature, stator winding temperature, lower guide shoe temperature, current, voltage, active power, reactive power Work power, power factor, and unit vibration swing; the measured operating state data includes: unit vibration and swing state.

According to the method of the first aspect of the present invention, in the step S1, the method for deleting invalid data in the argument data includes:

Sampling and diluting the independent variable data, and then deleting null or out-of-range abnormal values in the sample-diluted independent variable data, and recording the time stamp of the abnormal value;

In the independent variable data with outliers removed, a missing matrix is generated for the data points and time stamps with outliers removed, and the data points and time stamps are marked as data anomalies.

According to the method of the first aspect of the present invention, in the step S2, the method for performing data preprocessing on the independent variable data for deleting invalid data includes:

The weighted average fusion is performed on the values of the independent variable data of multiple collection points of the same equipment.

According to the method of the first aspect of the present invention, in the step S2, the method of performing data preprocessing on the independent variable data for deleting invalid data further includes:

Sample instance normalization is performed on the independent variable data that contains data points that are anomalous in the labeled data.

According to the method of the first aspect of the present invention, in the step S2, the method of performing data preprocessing on the independent variable data for deleting invalid data further includes:

Stratified normalization of the independent variable data for different forecast durations.

According to the method of the first aspect of the present invention, in the step S3, the method of establishing a unit operating state prediction model based on the bidirectional LSTM neural network includes:

Use a multi-layer stacked bidirectional LSTM neural network to establish a unit operating status prediction model;

The number of neurons in the input layer of the unit operating state prediction model is equal to the number of dimensions of the independent variable data.

The second aspect of the present invention discloses a system for predicting the operating state of pumping station units, the system comprising:

The first processing module is configured to, based on the time-series data of unit operation, collect independent variable data of unit operating state prediction and operating state measured data, and delete invalid data in the independent variable data;

The second processing module is configured to perform data preprocessing on the independent variable data from which invalid data is deleted, to obtain the independent variable data after data preprocessing;

The third processing module is configured to, based on the bidirectional LSTM neural network, establish a unit operation state prediction model; apply the independent variable data after data preprocessing and the operation state measured data to train and test the unit operation state prediction model;

The fourth processing module is configured to use the trained unit operation state prediction model to predict the change trend of the real-time operation state of the pumping station unit.

According to the system of the second aspect of the present invention, the first processing module is configured such that the independent variable data includes: thrust bearing pad temperature, upper guide pad temperature, stator winding temperature, lower guide pad temperature, current, voltage, Active power, reactive power, power factor, and vibration swing of the unit; the measured data of the operating state includes: the vibration and swing state of the unit.

According to the system of the second aspect of the present invention, the first processing module is configured such that the deleting invalid data in the argument data includes:

Sampling and diluting the independent variable data, and then deleting null or out-of-range abnormal values in the sample-diluted independent variable data, and recording the time stamp of the abnormal value;

In the independent variable data with outliers removed, a missing matrix is generated for the data points and time stamps with outliers removed, and the data points and time stamps are marked as data anomalies.

According to the system of the second aspect of the present invention, the second processing module is configured such that the data preprocessing of the independent variable data for deleting invalid data includes:

The weighted average fusion is performed on the values of the independent variable data of multiple collection points of the same equipment.

According to the system according to the second aspect of the present invention, the second processing module is configured such that the data preprocessing of the independent variable data for deleting invalid data further includes:

Sample instance normalization is performed on the independent variable data that contains data points that are anomalous in the labeled data.

According to the system according to the second aspect of the present invention, the second processing module is configured such that the data preprocessing of the independent variable data for deleting invalid data further includes:

Stratified normalization of the independent variable data for different forecast durations.

According to the system according to the second aspect of the present invention, the third processing module is configured such that, based on the bidirectional LSTM neural network, establishing a unit operation state prediction model includes:

Use a multi-layer stacked bidirectional LSTM neural network to establish a unit operating status prediction model;

The number of neurons in the input layer of the unit operating state prediction model is equal to the number of dimensions of the independent variable data.

The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the method for predicting the operating state of the pumping station unit in any one of the first aspects of the present disclosure are realized.

A fourth aspect of the present invention discloses a computer readable storage medium. A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps in the method for predicting the operating state of the pumping station unit in any one of the first aspects of the present disclosure are realized.

The solution proposed by the present invention fully considers the performance impact of the introduction of outliers without losing the semantic information of the original data, and designs a data processing method for missing labels and multi-point fusion, which effectively reduces the network training calculation parameters , which greatly reduces the computational complexity; secondly, multi-step multivariate data preprocessing is realized, which reduces the difficulty of its use in different scenarios and increases its applicability.

Description of drawings

In order to more clearly illustrate the specific implementation of the present invention or the technical solutions in the prior art, the following will briefly introduce the accompanying drawings that need to be used in the specific implementation or description of the prior art. Obviously, the accompanying drawings in the following description The drawings show some implementations of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

1 is a flow chart of a method for predicting the operating state of a pumping station unit according to an embodiment of the present invention;

Fig. 2 is a schematic diagram of independent variable data according to an embodiment of the present invention;

Fig. 3 is a schematic diagram of the structure of an LSTM neural network according to an embodiment of the present invention;

Fig. 4a is a single-step duration rolling prediction unit operating state change trend diagram according to an embodiment of the present invention;

Fig. 4b is a multi-step duration rolling prediction trend diagram of unit operating state change according to an embodiment of the present invention;

5 is a structural diagram of a system for predicting the operating state of a pumping station unit according to an embodiment of the present invention;

Fig. 6 is a structural diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is only some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The first aspect of the present invention discloses a method for predicting the operating state of a pumping station unit. Fig. 1 is a flow chart of a method for predicting the operating state of a pumping station unit according to an embodiment of the present invention. As shown in Fig. 1 , the method includes:

Step S1, based on the time-series data of unit operation, collect the independent variable data of unit operating state prediction and the actual measurement data of operating state, and delete invalid data in the independent variable data;

Step S2, performing data preprocessing on the independent variable data from which invalid data is deleted, to obtain the independent variable data after data preprocessing;

Step S3, based on the bidirectional LSTM neural network, establish a unit operation state prediction model; apply the independent variable data after data preprocessing and the actual measurement data of the operation state to train and test the unit operation state prediction model;

Step S4, applying the trained unit operation state prediction model to predict the change trend of the real-time operation state of the pumping station unit.

In step S1, based on the time-series data of unit operation, the independent variable data of unit operating state prediction and the actual measurement data of operating state are collected, and invalid data in the independent variable data (invalid data due to sensor failure and other reasons) are deleted.

In some embodiments, in the step S1, as shown in FIG. 2 , the independent variable data includes: thrust bearing pad temperature, upper guide pad temperature, stator winding temperature, lower guide pad temperature, current, voltage, active power power, reactive power, power factor, and unit vibration swing; the measured operating state data includes: unit vibration and swing state.

The unit thrust bearing bush temperature, upper guide bush temperature, stator winding temperature and lower guide bush temperature are part of the temperature parameters of the unit;

Current, voltage, active power, reactive power and power factor are the electrical parameters of the unit;

The method for deleting invalid data in the independent variable data includes:

Sampling and diluting the independent variable data, and then deleting null or out-of-range abnormal values in the sample-diluted independent variable data, and recording the time stamp of the abnormal value;

In the independent variable data with outliers removed, a missing matrix is generated for the data points and time stamps with outliers removed, and the data points and time stamps are marked as data anomalies.

Specifically, 17-dimensional data in total of 6-dimensional electrical parameters, 9-dimensional electrical parameters, and 2-dimensional water machine parameters of the industrial control network of the pumping station unit are used as independent variable data, and the vibration of the unit is used as the actual measurement data of the operating state. Because the data in the industrial control database of the pumping station has a large time span and a high sampling frequency, such data is not conducive to the monitoring of the operating status of the unit and the prediction of the trend in a short period of time. In order to ensure the diversity of the data, two of them are selected during training The one-year operating status data of a pumping station unit in relatively supplementary operation is used as training data, and the data is sampled and diluted at a ratio of 4:1. After deleting invalid data due to sensor failure and other reasons, mark the deletion point and deletion time. stamp.

In step S2, data preprocessing is performed on the independent variable data from which invalid data has been deleted, and the independent variable data after data preprocessing is obtained.

In some embodiments, in the step S2, the method for performing data preprocessing on the independent variable data for deleting invalid data includes:

The weighted average fusion is carried out on the values of the independent variable data of multiple collection points of the same device. The weighted value range is: 0.95-1.05, which initially reduces the data dimension and eliminates the influence of errors.

The sample instance normalization is performed on the independent variable data containing the abnormal data points of the labeled data, and it is stored in the time series.

Stratified normalization of the independent variable data for different forecast durations.

Specifically, the independent variable data whose invalid data is deleted is divided into training set and test set according to 8:2. The training process is combined with the actual unit operation, and a single-duration rolling forecast experiment and a multi-duration (for example, 12 time intervals) rolling forecast are set up. Experiment 2 prediction experiments, according to the different forecasting time, first judge the deletion point, perform sample instance normalization on the independent variable data containing the deletion point, and then perform layer normalization on the batch samples.

In step S3, based on the bidirectional LSTM neural network, a unit operation state prediction model is established; the unit operation state prediction model is trained and tested using the independent variable data after data preprocessing and the operation state measurement data.

In some embodiments, in the step S3, the method of establishing a unit operating state prediction model based on a bidirectional LSTM neural network includes:

Use a multi-layer stacked bidirectional LSTM neural network to establish a unit operating status prediction model;

The number of neurons in the input layer of the unit operating state prediction model is equal to the dimension of the independent variable data.

Specifically, based on the bidirectional LSTM neural network, a unit operating state prediction model is established to realize single-step rolling prediction and multi-step rolling prediction of multivariate time-series data of unit operation: in order to better fit multivariate long-distance unit operation time-series data, multi-layer Stacked bidirectional LSTM neural network, the number of neurons in the input layer is equal to the data dimension, the output layer is the predicted value of the unit's operating status, and the results are passed to the output layer in a linear full connection manner.

The core of the LSTM neural network is to add, delete or update the data of the unit state through three gates, and finally achieve the control of the information flow. Specifically, the Sigmoid layer and the dot product operation form the so-called "gate". The LSTM neural network has three gates, namely the input gate, the output gate, and the forgetting gate. These gates can selectively pass information. All LSTM neural networks have a chain form of repeated neural network modules, and the repeated modules interact in a layer and stacked manner. The network structure of the LSTM neural network is shown in Figure 3.

Among them, f _t is the forget gate, which determines which information in the unit state c _(t-1) at the previous moment is saved in the unit state c _t at the current moment; _it is the input gate, which determines which useful information needs to be is recorded in the unit state; o _t is the output gate, how much information is output in the control unit state to the current output value h _t . The calculation formula of each door is as follows:

Among them, σ represents the sigmoid layer, h _t-1 represents the output value of the hidden layer at the previous moment, x _t represents the input data at the current moment, [h _t-1 ,x _t ] represents the splicing of two vectors together, w _f , w _i , and w _o represent the weight matrices of the forget gate, input gate, and output gate, respectively, and b _f , _bi , and b _o represent the bias items of the forget gate, input gate, and output gate, respectively. The final output h _t of the LSTM neural network is jointly determined by o _t and the unit state c _t . Calculated as follows:

Among them, _ct represents the cell state, ct _-1 represents the cell state at the previous moment, and represents the temporary cell state at the current moment, _bc represents the bias of the cell state, and tanh() represents the activation function.

In the initialization of network parameters, random numbers from 0 to 1 are used to set the input gate state in _t , the forgetting gate state ^{f n t} _, the output gate state ^on _t , the cell state C ⁿ _t , and the hidden layer state h ⁿ _t of each layer. Initialized; use random numbers from 0 to 0.5 to weight matrix W ⁿ _fh of the backpropagation forgetting gate of each layer, weight matrix W ⁿ _ih of the input gate of backpropagation, and weight matrix of the state of the backpropagation calculation unit W ⁿ _ch , the weight matrix W ⁿ _oh of the backpropagation output gate, the weight matrix W ⁿ _fx of the backpropagation forgetting gate, the weight matrix W ⁿ _ix of the backpropagation input gate, the state of the backpropagation calculation unit The weight matrix W ⁿ _cx and the weight matrix W ⁿ _ox of the output gate of the backpropagation along the network line are initialized, and n represents the layer of the bidirectional LSTM neural network stacked by multiple layers. In this embodiment, five single-layer LSTM neural networks with 100 hidden nodes are connected in series in a multi-layer stacking manner, and a two-way operation mechanism of the model is established so that the model can better obtain the semantic features of the input data.

Considering actual usage requirements, set rolling forecasting methods with different time intervals, such as one day or several hours in advance, and load the training set in batches into the multi-layer stacked bidirectional LSTM neural network for forecasting training;

The MAE and MSE errors of the training process will decrease as the iterative process proceeds;

In the formula, n is the batch size, which is the predicted value, and y _i is the real value.

In order to speed up the model training process, consider using the loss of MAE plus N times MSE as the loss penalty return. Whether the model can achieve the ideal performance depends on the model evaluation index. R ² , MAPE and single batch time consumption are selected as the evaluation index system of the model. The formula for calculating ^R2 is as follows:

The calculation formula of MAPE is as follows:

The experimental platform is the deep learning environment of the PyTorch framework under the Ubuntu 18.04 operating system, and the model is iteratively trained in combination with the evaluation indicators. The training process uses a high number of iterations. The number of iterations is 2000. The rate is 0.0001, the model optimizer uses the Adam algorithm, and uses the MSE error and N times the MAE error to calculate the error between the predicted value and the true value, and feedback to optimize the training of the model; algorithm selection R ² , MAPE, and single-batch time consumption are used as the evaluation index system of the model. The example prediction results are shown in Figure 4a, which is the single-step time-length rolling prediction of the vibration change trend in the operating state of the unit. From the upper left to the lower right, it represents the whole batch and part The error after numerical normalization between the predicted value and the true value in the batch, when the model converges, the model MSE loss is 0.00065, the MAE is 0.0101, and the R ² is 0.9409. The model test MSE loss is 0.00244, the MAE is 0.0224, and the R ² is 0.7487; Figure 4b is a multi-step time-length rolling forecast, from upper left to lower right, respectively represents the error after numerical normalization between the predicted value and the real value in all batches and some batches, when the model converges, the model The MSE loss is 0.0014, the MAE is 0.017, and the R ² is 0.943. The test results show that the model MSE loss is 0.0091, the MAE is 0.0461, and the R ² is 0.7819. The comparison between the two experiments shows that the operating state of the pumping station unit has a more rapid development in time. Many correlations.

In step S4, the trained unit operating state prediction model is used to predict the change trend of the real-time operating state of the pumping station unit.

Specifically, in combination with the network characteristics and equipment characteristics of the industrial control network of the pumping station unit, the Python language-related development environment and related applications are deployed. The functions refer to the actual needs. At the same time, the unit operation status change trend prediction model is deployed, and the prediction time and GPU device usage are set.

Combined with the data acquisition method of the industrial control network of the pumping station unit, use Python programming to obtain the real-time data of the unit operation, dynamically add it to the target database, store it in time series, filter and label the data through step 1 and reduce the dimension and sum of step 2 Normalized processing.

Predict the status change trend of the unit according to the forecast time requirement, and visually display the forecast results, and combine the relevant regulations of the pumping station management to evaluate the risk level of the predicted value and give an alarm.

To sum up, the solution proposed by the present invention can be applied to unit operation status monitoring and development prediction, without excessive data preprocessing, low computational complexity, low prediction error, multi-step time series prediction for time series data, and wide application range. Experiments show that this method can effectively predict the changing trend of unit operating status.

The second aspect of the present invention discloses a system for predicting the operating state of pumping station units. Fig. 5 is a structural diagram of a system for predicting the operating state of pumping station units according to an embodiment of the present invention; as shown in Fig. 5, the system 100 includes:

The first processing module 101 is configured to, based on the time-series data of unit operation, collect independent variable data of unit operating state prediction and operating state measured data, and delete invalid data in the independent variable data;

The second processing module 102 is configured to perform data preprocessing on the independent variable data from which invalid data is deleted, to obtain the independent variable data after data preprocessing;

The third processing module 103 is configured to, based on the bidirectional LSTM neural network, establish a unit operation state prediction model; apply the independent variable data after data preprocessing and the operation state actual measurement data to train and test the unit operation state prediction model;

The fourth processing module 104 is configured to use the trained unit operation state prediction model to predict the change trend of the real-time operation state of the pumping station unit.

According to the system of the second aspect of the present invention, the first processing module 101 is configured such that the independent variable data includes: thrust bearing pad temperature, upper guide pad temperature, stator winding temperature, lower guide pad temperature, current, voltage , active power, reactive power, power factor and unit vibration swing; the measured operating state data includes: unit vibration and swing state.

According to the system of the second aspect of the present invention, the first processing module 101 is configured such that the deletion of invalid data in the argument data includes:

Sampling and diluting the independent variable data, and then deleting null or out-of-range abnormal values in the sample-diluted independent variable data, and recording the time stamp of the abnormal value;

In the independent variable data with outliers removed, a missing matrix is generated for the data points and time stamps with outliers removed, and the data points and time stamps are marked as data anomalies.

According to the system of the second aspect of the present invention, the second processing module 102 is configured such that the data preprocessing of the independent variable data for deleting invalid data includes:

The weighted average fusion is performed on the values of the independent variable data of multiple collection points of the same equipment.

According to the system of the second aspect of the present invention, the second processing module 102 is configured such that the data preprocessing of the independent variable data for which invalid data is deleted further includes:

Sample instance normalization is performed on the independent variable data that contains data points that are anomalous in the labeled data.

According to the system of the second aspect of the present invention, the second processing module 102 is configured such that the data preprocessing of the independent variable data for which invalid data is deleted further includes:

Stratified normalization of the independent variable data for different forecast durations.

According to the system according to the second aspect of the present invention, the third processing module 103 is configured such that the establishment of a unit operating state prediction model based on a bidirectional LSTM neural network includes:

Use a multi-layer stacked bidirectional LSTM neural network to establish a unit operating status prediction model;

The number of neurons in the input layer of the unit operating state prediction model is equal to the number of dimensions of the independent variable data.

The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps in the method for predicting the operating state of the pumping station unit in any one of the first aspects of the present disclosure are realized.

FIG. 6 is a structural diagram of an electronic device according to an embodiment of the present invention. As shown in FIG. 6 , the electronic device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Wherein, the processor of the electronic device is used to provide calculation and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, an operator network, near field communication (NFC) or other technologies. The display screen of the electronic device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device may be a touch layer covered on the display screen, or a button, a trackball or a touch pad provided on the housing of the electronic device , and can also be an external keyboard, touchpad, or mouse.

Those skilled in the art can understand that the structure shown in Figure 6 is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic equipment to which the solution of the present application is applied. Devices may include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.

A fourth aspect of the present invention discloses a computer readable storage medium. A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps in the steps in the method for predicting the operating state of the pumping station unit in any one of the first aspects of the present disclosure are realized.

Please note that the various technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, all possible combinations of the various technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features , should be considered as within the scope of this specification. The above examples only express several implementation modes of the present application, and the description thereof is relatively specific and detailed, but should not be construed as limiting the scope of the patent for the invention. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the scope of protection of the patent application should be based on the appended claims.

Claims (4)

1. A method for predicting an operation state of a pump station unit, the method comprising:

step S1, based on time sequence data of unit operation, acquiring self-variable data and operation state actual measurement data of unit operation state prediction, and deleting invalid data in the self-variable data;

the self-variable data includes: the temperature of a thrust bearing bush of the unit, the temperature of an upper guide shoe, the temperature of a stator winding, the temperature of a lower guide shoe, current, voltage, active power, reactive power, power factor and vibration swing degree of the unit; the operation state actual measurement data includes: the vibration and swing degree states of the unit;

the method for deleting invalid data in the self-variable data comprises the following steps:

sampling and diluting the self-variable data, deleting an abnormal value which is empty or out of range from the sampled and diluted self-variable data, and recording a time stamp of the abnormal value;

generating a deletion matrix of data points and time stamps of the deleted abnormal values in the self-variable data of the deleted abnormal values, and marking the data points and the time stamps as data anomalies;

s2, performing data preprocessing on the self-variable data with invalid data deleted to obtain self-variable data after data preprocessing;

the method for preprocessing the self-variable data for deleting the invalid data comprises the following steps:

carrying out weighted average fusion on the values of the self-variable data of the multiple acquisition points of the same equipment;

sample instance normalization is carried out on the self-variable data containing the data points with abnormal labeling data;

carrying out layer normalization on the self-variable data with different prediction time lengths;

s3, building a unit running state prediction model based on a bidirectional LSTM neural network; training and testing the unit running state prediction model by applying the self-variable data and the running state actual measurement data which are subjected to data preprocessing; the method for establishing the unit running state prediction model based on the bidirectional LSTM neural network comprises the following steps:

establishing a unit operation state prediction model by using a multi-layer stacked bidirectional LSTM neural network;

the number of neurons of an input layer of the unit running state prediction model is equal to the number of dimensions of the self-variable data;

MAE and MSE errors in the training process can be reduced along with the progress of the iterative process;

;

where n is the batch size,as predicted value, y _i Is a true value;

；

select R ² MAPE and single-batch time consumption are used as an evaluation index system of the model; r is R ² The calculation formula of (2) is as follows:

；

the MAPE calculation formula is as follows:

；

calculating the error between the predicted value and the true value by using a mode of superposition of multiple errors of MSE error and N times of MAE error, and feeding back to optimize the training of the model; algorithm selection R ² MAPE and single-batch time consumption are used as an evaluation index system of the model;

and S4, predicting the change trend of the real-time running state of the pump station unit by applying the trained unit running state prediction model.

2. A system for predicting the operational status of a pump station assembly, the system comprising:

the first processing module is configured to collect self-variable data and running state actual measurement data predicted by the running state of the unit based on time sequence data of the running of the unit, and delete invalid data in the self-variable data;

the self-variable data includes: the temperature of a thrust bearing bush of the unit, the temperature of an upper guide shoe, the temperature of a stator winding, the temperature of a lower guide shoe, current, voltage, active power, reactive power, power factor and vibration swing degree of the unit; the operation state actual measurement data includes: the vibration and swing degree states of the unit;

the method for deleting invalid data in the self-variable data comprises the following steps:

sampling and diluting the self-variable data, deleting an abnormal value which is empty or out of range from the sampled and diluted self-variable data, and recording a time stamp of the abnormal value;

generating a deletion matrix of data points and time stamps of the deleted abnormal values in the self-variable data of the deleted abnormal values, and marking the data points and the time stamps as data anomalies;

the second processing module is configured to perform data preprocessing on the self-variable data with invalid data deleted to obtain self-variable data after data preprocessing;

the method for preprocessing the self-variable data for deleting the invalid data comprises the following steps:

carrying out weighted average fusion on the values of the self-variable data of the multiple acquisition points of the same equipment;

sample instance normalization is carried out on the self-variable data containing the data points with abnormal labeling data;

carrying out layer normalization on the self-variable data with different prediction time lengths;

the third processing module is configured to establish a unit running state prediction model based on the bidirectional LSTM neural network; training and testing the unit running state prediction model by applying the self-variable data and the running state actual measurement data which are subjected to data preprocessing;

the method for establishing the unit running state prediction model based on the bidirectional LSTM neural network comprises the following steps:

establishing a unit operation state prediction model by using a multi-layer stacked bidirectional LSTM neural network;

the number of neurons of an input layer of the unit running state prediction model is equal to the number of dimensions of the self-variable data;

MAE and MSE errors in the training process can be reduced along with the progress of the iterative process;

;

where n is the batch size,as predicted value, y _i Is a true value;

；

select R ² MAPE and single-batch time consumption are used as an evaluation index system of the model; r is R ² The calculation formula of (c) is shown as follows,

；

the MAPE calculation formula is as follows:

；

calculating the error between the predicted value and the true value by using a mode of superposition of multiple errors of MSE error and N times of MAE error, and feeding back to optimize the training of the model; algorithm selection R ² MAPE and single-batch time consumption are used as an evaluation index system of the model;

and the fourth processing module is configured to apply the trained unit running state prediction model to predict the change trend of the real-time running state of the pump station unit.

3. An electronic device comprising a memory and a processor, the memory storing a computer program, the processor implementing the steps in a pump station unit operation state prediction method according to claim 1 when executing the computer program.

4. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of a pump station set operation state prediction method according to claim 1.