LSTM (Long Short Term Memory) model generation method, fault prediction method and device

Abstract

The invention provides an LSTM model generation method, a fault prediction method and a fault prediction device. The generation method comprises the following steps: acquiring a plurality of historical times of an energy storage power station and historical feature information corresponding to each historical time; and generating an input sequence and an output sequence, creating an LSTM model, and training the LSTM model based on the input sequence and the output sequence. Faults of the energy storage power station can be predicted based on the LSTM model and the SVM model.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage, and particularly to a method for generating an LSTM model, a method for fault prediction, and a device thereof. Background Art

[0002] As an important energy regulation and storage facility in the power grid, the safe and stable operation of an energy storage power station is crucial for the stability and reliability of the power grid. Lithium-ion batteries are widely used in energy storage power stations due to their high energy density, long life, and other advantages. However, lithium-ion batteries may experience performance degradation or faults during operation due to reasons such as overcharging, over-discharging, and abnormal temperature, and even cause safety accidents. Therefore, developing an effective fault warning system is of great significance for ensuring the safe operation of energy storage power stations. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for generating an LSTM model, a method for fault prediction, and a device thereof.

[0004] To achieve one of the above-mentioned invention purposes, an embodiment of the present invention provides a method for generating an LSTM model for fault prediction of an energy storage power station, including the following steps: obtaining Num1 historical times Time1, Time2,..., Time Num1 , and the corresponding historical feature information TState i for each historical time Time i , where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time k , and the intervals between adjacent historical times are equal; generating a training set start = 1, and continuously perform the following operations: if start + A + B ≤ Num1, then use TState start , TState start+1 ,..., TState start+A as the input sequence, use TState start+B , TState start+1+B ,..., TState start+A+B as the output sequence, add the input sequence and the output sequence as a binary tuple to the training set S, increase the value of start by 1, and perform subsequent operations; if start + A + B > Num1, stop performing subsequent operations; where A and B are both natural numbers; creating an LSTM model, using the input sequence in each element of the training set S as the input data, and the output sequence as the corresponding output data, and training the LSTM model based on each input data and the corresponding output data.

[0005] As a further improvement of an implementation mode of the present invention, the energy storage power station is provided with a plurality of energy storage batteries of the same specification; the historical characteristic information includes: the voltage, current, temperature and SOC value of all energy storage batteries of the energy storage power station.

[0006] As a further improvement of an embodiment of the present invention, the “obtaining Num1 historical times Time1, Time2, ..., Time Num1 , and each historical time Time i Corresponding historical feature information TState i "Specifically includes: obtaining Num1 historical times Time1, Time2, ..., Time Num1 , and each historical time Time i Corresponding historical feature information TState i , then, the historical feature information TState1, TState2, ..., TState Num1 All were normalized.

[0007] As a further improvement of an embodiment of the present invention, the “processing of historical feature information TState1, TState2, ..., TState Num1 All historical feature information TState i They all include: the eigenvalue corresponding to feature State1, the eigenvalue corresponding to feature State2, ..., feature State Num2 The corresponding eigenvalue, where Num2 is a natural number; calculate X max (j)=max(state(1,j),state(2,j),...,state(Num1,j)),X min (j) = min(state(1,j),state(2,j),...,state(Num1,j)), where j and Num2 are both natural numbers, 1≤j≤Num2; for any historical feature information TState i Any eigenvalue State in j The eigenvalues ​​of state(i,j) are processed as follows: Calculate Then, set the eigenvalue state(i,j) to X norm .

[0008] The embodiment of the present invention also provides a generating device for an LSTM model for fault prediction of an energy storage power station, including the following modules: an information acquisition module, configured to acquire Num1 historical times Time1, Time2,..., Time of the energy storage power station Num1 , and the historical feature information TState corresponding to each historical time Time i , where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time i , and the intervals between adjacent historical times are equal; an information processing module, configured to generate a training set start = 1, and continuously perform the following operations: If start + A + B ≤ Num1, then use TState start , TState start+1 ,..., TState start+A as the input sequence, use TState start+B , TState start+1+B ,..., TState start+A+B as the output sequence, use the input sequence and the output sequence as a binary tuple and add it to the training set S, increase the value of start by 1, and perform subsequent operations; If start + A + B > Num1, stop performing subsequent operations; where A and B are both natural numbers; a model generation module, configured to create an LSTM model, use the input sequence in each element in the training set S as input data, use the output sequence as the corresponding output data, and train the LSTM model based on each input data and the corresponding output data.

[0009] As a further improvement of an embodiment of the present invention, a plurality of energy storage batteries of the same specification are provided in the energy storage power station; the historical feature information includes: the voltage, current, temperature, and SOC value of all energy storage batteries in the energy storage power station.

[0010] As a further improvement of an embodiment of the present invention, the information acquisition module is further configured to: acquire Num1 historical times Time1, Time2,..., Time of the energy storage power station Num1 , and the historical feature information TState corresponding to each historical time Time i , and then perform normalization processing on the historical feature information TState1, TState2,..., TState i . Num1

[0011] ​​As a further improvement of an embodiment of the present invention, the information acquisition module is further configured to: for each historical feature information TState i All contain: the feature value corresponding to feature State1, the feature value corresponding to feature State2,..., the feature Num2 value corresponding to State, where Num2 is a natural number; calculate X max (j) = max(state(1,j), state(2,j),..., state(Num1,j)), X min (j) = min(state(1,j), state(2,j),..., state(Num1,j)), where j and Num2 are both natural numbers, 1 ≤ j ≤ Num2; for any feature value State i in any historical feature information TState j perform the following processing on the feature value state(i,j): calculate Then, set the feature value state(i,j) to X norm .

[0012] An embodiment of the present invention also provides a fault prediction method for an energy storage power station, including the following steps: execute the above generation method to obtain an LSTM model, and obtain a trained SVM model; obtain A + 1 historical times Time′1, Time′2,..., Time′ A+1 , and each historical time Time′ m corresponding historical feature information TState′ m , where m and n are natural numbers, 1 ≤ m, n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ n , and the intervals between adjacent historical times are equal; input the A + 1 historical feature information TState′1, TState′2,..., TState′ A+1 into the LSTM model, and obtain A + 1 target historical feature information from the model; use the SVM model to process all the A + 1 target historical feature information, and receive the fault information of the energy storage power station from the SVM model.

[0013] An embodiment of the present invention also provides a fault prediction device for an energy storage power station, including the following models: a preprocessing module, configured to execute the above generation method to obtain an LSTM model, and obtain a trained SVM model; a historical data acquisition module, configured to obtain A + 1 historical times Time′1, Time′2,..., Time′ A+1 , and each historical time Time′m The corresponding historical feature information TState′ m , where m and n are natural numbers, 1 ≤ m, n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ n , and the intervals between adjacent historical times are equal; a prediction module, configured to input A + 1 pieces of historical feature information TState′1, TState′2,..., TState′ A+1 into the LSTM model, and obtain A + 1 pieces of target historical feature information from the model; process all of the A + 1 pieces of target historical feature information using the SVM model, and receive the fault information of the energy storage power station from the SVM model.

[0014] Compared with the prior art, the technical effect of the present invention is that: the embodiments of the present invention provide a method for generating an LSTM model, a fault prediction method and an apparatus thereof. The method for generating includes the following steps: obtaining a plurality of historical times of the energy storage power station, and the historical feature information corresponding to each historical time; generating an input sequence and an output sequence, creating an LSTM model, and training the LSTM model based on the input sequence and the output sequence. Based on this LSTM model and the SVM model, the fault of the energy storage power station can be predicted. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a schematic flowchart of the method for generating an LSTM model in an embodiment of the present invention;

[0016] Figure 2 and Figure 3 is an experimental result diagram of the method for generating an LSTM model in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The present invention will be described in detail below with reference to the various embodiments shown in the drawings. However, these embodiments do not limit the present invention, and any structural, method, or functional transformation made by those of ordinary skill in the art based on these embodiments is included in the protection scope of the present invention.

[0018] Spatial relative position terms such as "upper", "above", "lower", "below", etc. used in this text are for the purpose of facilitating description to describe the relationship of one unit or feature relative to another unit or feature as shown in the accompanying drawings. Spatial relative position terms may be intended to include different orientations of the device in use or operation other than the orientation shown in the figures. For example, if the device in the figure is flipped, the unit described as being "below" or "beneath" another unit or feature will be "above" the other unit or feature. Therefore, the exemplary term "below" can encompass both the upper and lower orientations. The device may be oriented in other ways (rotated 90 degrees or other orientations), and the spatially related descriptive terms used in this text can be interpreted accordingly.

[0019] Moreover, it should be understood that although terms such as first, second, etc. may be used in this text to describe various elements or structures, the objects described should not be limited by these terms. These terms are only used to distinguish these described objects from each other. For example, the first end may be referred to as the second end, and similarly the second end may also be referred to as the first end, which does not deviate from the protection scope of this application.

[0020] Embodiment 1 of the present invention provides a method for generating an LSTM model for fault prediction of an energy storage power station, as Figure 1 shown, including the following steps:

[0021] Step 101: Obtain Num1 historical times Time1, Time2,..., Time Num1 of the energy storage power station, and the corresponding historical feature information TState i for each historical time Time i , where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time k , and the intervals between adjacent historical times are equal;

[0022] Step 102: Generate a training set start = 1, and continuously perform the following operations: If start + A + B ≤ Num1, then use TState start , TState start+1 ,..., TState start+A as the input sequence, and use TState start+B , TState start+1+B ,..., TState start+A+BAs the output sequence, the input sequence and the output sequence are taken as a tuple and added to the training set S, the value of start is increased by 1, and subsequent operations are performed; if start+A+B>Num1, the subsequent operations are stopped; wherein A and B are both natural numbers;

[0023] Step 103: Create an LSTM model, take the input sequence in each element in the training set S as input data, and the output sequence as corresponding output data, and train the LSTM model based on each input data and corresponding output data.

[0024] The LSTM (Long Short-Term Memory) model is a special recurrent neural network (RNN) that can learn and memorize long-term dependencies. The LSTM model can predict the feature information of the time to be predicted by learning the changing trends of multiple historical feature information of the energy storage power station.

[0025] In this embodiment, the energy storage power station is provided with a plurality of energy storage batteries of the same specification; the historical characteristic information includes: the voltage, current, temperature and SOC value of all the energy storage batteries of the energy storage power station.

[0026] Here, the voltage may be the output voltage of the energy storage battery, the current may be the output current of the energy storage battery, the temperature may be the ambient temperature of the energy storage battery, and the full name of SOC in English is State of Charge, and its Chinese name is Charge Characteristics.

[0027] Here, the data can be collected by BMS (Battery Management System).

[0028] Here, the parameters of the corresponding second-order equivalent circuit model can be identified through the Hybrid Pulse Power Characterizationtest (HPPC) data, and the state of charge (SOC) of the corresponding energy storage battery can be calculated accordingly.

[0029] In this embodiment, the “obtaining Num1 historical times Time1, Time2, ..., Time Num1 , and each historical time Time i Corresponding historical feature information TState i "Specifically includes: obtaining Num1 historical times Time1, Time2, ..., Time Num1, and each historical time Time i Corresponding historical feature information TState i , then, the historical feature information TState1, TState2, ..., TState Num1 All were normalized.

[0030] Here, it can be understood that the dimensions of voltage, current, temperature and SOC value are generally different. Therefore, this will affect the results of data analysis. In order to eliminate the dimensional impact between indicators, data standardization can be performed first to solve the comparability between data indicators.

[0031] In this embodiment, the “historical feature information TState1, TState2, ..., TState Num1 All are normalized” specifically includes:

[0032] Each historical feature information TState i They all include: the eigenvalue corresponding to feature State1, the eigenvalue corresponding to feature State2, ..., feature State Num2 The corresponding eigenvalue, where Num2 is a natural number; calculate X max (j)=max(state(1,j),state(2,j),...,state(Num1,j)),X min (j)=min(state(1,j),state(2,j),...,state(Num1,j)), where j and Num2 are both natural numbers, 1≤j≤Num2;

[0033] For any historical feature information TState i Any eigenvalue State in j The eigenvalues ​​of state(i,j) are processed as follows: Calculate Then, set the eigenvalue state(i,j) to X norm .

[0034] It can be understood that after the above processing, the voltage, current, temperature and SOC values ​​are all in the range of [-0.5, 0.5]. Since the activation functions used by the "gate" of the LSTM model are sigmoid and tanh, the value range is also [-0.5, 0.5], so they can be automatically connected.

[0035] Figure 2 The actual and predicted temperature results for a single battery module are shown in the figure. Figure 2 It shows that LSTM can better predict the voltage of the battery module. Figure 3This is the loss convergence of the LSTM model training. The prediction error mse of the model is.

[0036] Model performance visualization: Figure 2 The comparison between the LSTM model’s prediction of the battery module voltage and the actual measurement results is shown. This visualization helps us intuitively understand the model’s prediction performance.

[0037] Prediction accuracy evaluation: By comparing the degree of fit between the actual voltage curve and the predicted voltage curve, the prediction accuracy of the model can be evaluated. If the two curves are highly consistent, it means that the prediction accuracy of the model is high; conversely, if the difference is large, it indicates that the prediction performance of the model may need to be improved.

[0038] Time series trend capture: The figure also shows the LSTM model's ability to capture the trend of battery voltage changes over time. If the model can accurately predict the trend of voltage changes, it means that it has a good learning effect on the battery's time series characteristics.

[0039] Figure 3 The figure shows how the loss function changes during the model training process. The loss function is used to measure the error between the model's predicted value and the actual value, so its change reflects the progress of the model training.

[0040] Convergence evaluation: By observing the changes in the loss curve, you can evaluate the convergence of the model. If the loss curve gradually decreases and tends to be stable, it means that the model is gradually optimizing and tending to converge; if the loss curve fluctuates greatly or does not decrease, it means that the model may have underfitting or overfitting problems, or there are other problems in the training process.

[0041] Training effect evaluation: The final stable value of the loss curve can help evaluate the training effect of the model. A lower loss value usually means that the model has a smaller prediction error on the training data and a better training effect.

[0042] Figure 2 Help evaluate the prediction performance of LSTM models in practical applications, especially the prediction accuracy and trend capture ability of battery module voltage.

[0043] Figure 3 Help monitor and evaluate the training process and effect of the LSTM model to ensure that the model can converge effectively and optimize the prediction performance.

[0044] Together, these two plots provide a comprehensive evaluation of the LSTM model performance, ensuring the reliability and accuracy of the model in predicting battery characteristics.

[0045] Embodiment 2 of the present invention provides a device for generating an LSTM model for fault prediction of an energy storage power station, comprising the following modules:

[0046] An information acquisition module, configured to acquire Num1 historical times Time1, Time2, ..., Time of the energy storage power station Num1 , and the historical feature information TState corresponding to each historical time Time i , where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time i , and the intervals between adjacent historical times are equal; k

[0047] An information processing module, configured to generate a training set start = 1, and continuously perform the following operations: If start + A + B ≤ Num1, then use TState start , TState start+1 , ..., TState start+A as the input sequence, use TState start+B , TState start+1+B , ..., TState start+A+B as the output sequence, use the input sequence and the output sequence as a binary tuple and add it to the training set S, increase the value of start by 1, and perform subsequent operations; If start + A + B > Num1, stop performing subsequent operations; where A and B are both natural numbers;

[0048] A model generation module, configured to create an LSTM model, use the input sequence in each element in the training set S as input data, and use the output sequence as the corresponding output data, and train the LSTM model based on each input data and the corresponding output data.

[0049] In this embodiment, a plurality of energy storage batteries of the same specification are provided in the energy storage power station; the historical feature information includes: the voltage, current, temperature, and SOC values of all the energy storage batteries in the energy storage power station.

[0050] In this embodiment, the information acquisition module is further configured to: acquire Num1 historical times Time1, Time2, ..., Time of the energy storage power station Num1 , and the historical feature information TState corresponding to each historical time Time i , and then perform normalization processing on the historical feature information TState1, TState2, ..., TState i Num1

[0051] ​​​In this embodiment, the information acquisition module is further configured to: for each piece of historical feature information TState i All include: the feature value corresponding to feature State1, the feature value corresponding to feature State2,..., the feature Num2 corresponding feature value, where Num2 is a natural number; calculate X max (j) = max(state(1,j), state(2,j),..., state(Num1,j)), X min (j) = min(state(1,j), state(2,j),..., state(Num1,j)), where both j and Num2 are natural numbers, 1 ≤ j ≤ Num2; for any feature value State i in any piece of historical feature information TState j perform the following processing on the feature value state(i,j): calculate Then, set the feature value state(i,j) to X norm .

[0052] Embodiment 3 of the present invention provides a fault prediction method for an energy storage power station, including the following steps:

[0053] Step 1: Execute the generation method in Embodiment 1 to obtain an LSTM model, and obtain a trained SVM model;

[0054] Step 2: Obtain A + 1 historical times Time′1, Time′2,..., Time′ A+1 , and each historical time Time′ m corresponding historical feature information TState′ m , where m and n are natural numbers, 1 ≤ m,n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ n , and the intervals between adjacent historical times are equal;

[0055] Step 3: Input the A + 1 pieces of historical feature information TState′1, TState′2,..., TState′ A+1 into the LSTM model, and obtain A + 1 target historical feature information from the model; use the SVM model to process all the A + 1 target historical feature information, and receive the fault information of the energy storage power station from the SVM model.

[0056] SVM (Support Vector Machine) is a generalized linear classifier that performs binary classification of data in a supervised learning manner. Its decision boundary is the maximum margin hyperplane solved for the learning samples. SVM uses the hinge loss function to calculate the empirical risk and adds a regularization term to the solution system to optimize the structural risk. It is a classifier with sparsity and robustness. SVM can perform nonlinear classification through kernel methods. The SVM model is used for fault detection, identifying potential battery failures by analyzing the deviation between the predicted target historical feature information and the ideal feature information, as well as other key features.

[0057] In large-scale energy storage power stations, due to the huge amount of data, fault detection for battery modules (such as overvoltage, overtemperature, overcurrent and SOC jump) is particularly important. Faults are usually obvious in the voltage of the battery module, so monitoring and analyzing battery parameters such as voltage, current and temperature are key. However, usually, sample data can be clearly marked as positive samples (faults) or negative samples (safety) after a comprehensive inspection in the laboratory. To achieve this goal, the predictive model must provide an acceptable level of accuracy.

[0058] In the process of using support vector machine (SVM) for battery fault detection, the key is to find the best classification hyperplane to maximize the interval between data points of different categories. This process can be explained in more detail as follows:

[0059] SVM aims to find a hyperplane w·x+b=0 to maximize the interval between two categories of data points. The hyperplane is the decision boundary and the interval is the minimum distance from the nearest data point to the hyperplane. The goal of the optimization problem is to minimize This reflects the need to maximize the margin. At the same time, the constraints for all data points need to be satisfied: i (w·x i +b)≥1, ensuring that the data points are correctly classified. By introducing the Lagrange multiplier λ i , integrating the constraints into the optimization problem. This can transform the original problem into a dual problem, which can be solved by Lagrange multipliers.

[0060] The dual problem is of the form: The solution to this problem involves choosing an optimal set of λ i When the data is not linearly separable, SVM uses kernel techniques to handle nonlinear problems. Kernel functions (such as radial basis functions) allow SVM to find the best hyperplane in high-dimensional space, thereby effectively handling complex data patterns.

[0061] The fault detection based on SVM is as follows:

[0062] 1. Data collection and preprocessing: First, collect relevant data of the batteries in the energy storage power station (such as voltage, current, temperature). Then clean, standardize or normalize these data for subsequent processing.

[0063] 2. Feature selection: Select features closely related to the battery health status, such as the discharge capacity, charging time, temperature fluctuation of the battery, etc. This is crucial for the accuracy of the subsequent model.

[0064] 3. Train the SVM model: Use the labeled training data to train the SVM classifier. According to the characteristics of the data, select an appropriate kernel function to handle the linearly inseparable situation.

[0065] 4. Model testing and evaluation: Use the test data set to evaluate the model, and pay attention to performance indicators such as accuracy and recall rate.

[0066] 5. Fault detection: Input the real-time or newly collected data into the trained SVM model, and the model will predict whether the battery has a fault according to the data features.

[0067] Through these steps, the SVM can effectively distinguish the normal and faulty states of the battery, and can maintain high classification ability even in the case of complex data patterns or non-linearly separable situations. This makes the SVM an ideal choice for battery fault detection.

[0068] Embodiment 4 of the present invention provides a fault prediction device for an energy storage power station, including the following models:

[0069] A preprocessing module, configured to execute the generation method in Embodiment 1 to obtain an LSTM model, and acquire a trained SVM model;

[0070] A historical data acquisition module, configured to acquire A + 1 historical times Time′1, Time′2,..., Time′ A+1 , and the corresponding historical feature information TState′ m for each historical time Time′ m , where m and n are natural numbers, 1 ≤ m, n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ n , and the intervals between adjacent historical times are equal;

[0071] A prediction module, configured to use the A + 1 historical feature information TState′1, TState′2,..., TState′ A+1The information is input into the LSTM model, and A+1 target historical feature information is obtained from the model; the A+1 target historical feature information is processed using the SVM model, and the fault information of the energy storage power station is received from the SVM model.

[0072] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0073] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0074] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0075] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0076] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention.

Claims

1. A method for generating an LSTM model for fault prediction of an energy storage power station, characterized in that: The following steps are involved: Obtain Num1 historical times Time1, Time2, ..., Time of the energy storage power station k , and the historical feature information TState corresponding to each historical time Time i where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time i , and the intervals between adjacent historical times are equal; k ​ Generate training set start=1, continue to perform the following operations: if start+A+B≤Num1, then TState start , TState start+1 , ..., TState start+A As an input sequence, TState start+B , TState start+1+B , ..., TState start+A+B As the output sequence, the input sequence and the output sequence are taken as a tuple and added to the training set S, the value of start is increased by 1, and subsequent operations are performed; if start+A+B>Num1, the subsequent operations are stopped; wherein A and B are both natural numbers; Create an LSTM model, use the input sequence in each element in the training set S as input data, and the output sequence as corresponding output data, and train the LSTM model based on each input data and corresponding output data.

2. The generation method according to claim 1, characterized in that: The energy storage power station is provided with a plurality of energy storage batteries of the same specifications; The historical characteristic information includes: the voltage, current, temperature and SOC value of all energy storage batteries in the energy storage power station.

3. The generation method according to claim 2, characterized in that: The "obtaining Num1 historical times Time1, Time2, ..., Time Num1 , and each historical time Time i Corresponding historical feature information TState i Specifically include: Get Num1 historical times Time1, Time2, ..., Time of the energy storage power station Num1 , and each historical time Time i Corresponding historical feature information TState i , then, the historical feature information TState1, TState2, ..., TState Num1 All were normalized.

4. The generation method according to claim 3, characterized in that: The "historical feature information TState1, TState2, ..., TState Num1 All are normalized” specifically includes: Each historical feature information TState i They all include: the eigenvalue corresponding to feature State1, the eigenvalue corresponding to feature State2, ..., feature State Num2 The corresponding eigenvalue, where Num2 is a natural number; calculate X max (j)=max(state(1,j),state(2,j),...,state(Num1,j)),X min (j)=min(state(1,j),state(2,j),...,state(Num1,j)), where j and Num2 are both natural numbers, 1≤j≤Num2; For any historical feature information TState i Any eigenvalue State in j The eigenvalues ​​of state(i,j) are processed as follows: Calculate Then, set the eigenvalue state(i,j) to X norm .

5. A device for generating an LSTM model for fault prediction of an energy storage power station, characterized in that: Includes the following modules: An information acquisition module, configured to acquire Num1 historical times Time1, Time2, ..., Time of the energy storage power station Num1 , and the historical feature information TState corresponding to each historical time Time i , where i, k, and Num1 are natural numbers, 1 ≤ i, k ≤ Num1, and when i < k, Time i is earlier than Time i , and the intervals between adjacent historical times are equal; k ​ Information processing module, used to generate training sets start=1, continue to perform the following operations: if start+A+B≤Num1, then TState start , TState start+1 , ..., TState start+A As an input sequence, TState starr+B , TState start+1+B , ..., TState start+A+B As the output sequence, the input sequence and the output sequence are taken as a tuple and added to the training set S, the value of start is increased by 1, and subsequent operations are performed; if start+A+B>Num1, the subsequent operations are stopped; wherein A and B are both natural numbers; The model generation module is used to create an LSTM model, taking the input sequence in each element in the training set S as input data and the output sequence as corresponding output data, and training the LSTM model based on each input data and corresponding output data.

6. The generating device according to claim 5, characterized in that The energy storage power station is provided with a plurality of energy storage batteries of the same specifications; The historical characteristic information includes: the voltage, current, temperature and SOC value of all energy storage batteries in the energy storage power station.

7. The generating device according to claim 6, characterized in that: The information acquisition module is also used for: Get Num1 historical times Time1, Time2, ..., Time of the energy storage power station Num1 , and each historical time Time i Corresponding historical feature information TState i , then, the historical feature information TState1, TState2, ..., TState Num1 All were normalized.

8. The generating device according to claim 7, characterized in that The information acquisition module is also used for: Each historical feature information TState i They all include: the eigenvalue corresponding to feature State1, the eigenvalue corresponding to feature State2, ..., feature State Num2 The corresponding eigenvalue, where Num2 is a natural number; calculate X max (j)=max(state(1,j),state(2,j),...,state(Num1,j)),X min (j)=min(state(1,j),state(2,j),...,state(Num1,j)), where j and Num2 are both natural numbers, 1≤j≤Num2; For any historical feature information TState i Any eigenvalue State in j The eigenvalues ​​of state(i,j) are processed as follows: Calculate Then, set the eigenvalue state(i,j) to X norm .

9. A fault prediction method for an energy storage power station, characterized in that: The following steps are involved: Execute the generation method in any one of claims 1 to 4 and obtain the LSTM model, and obtain the trained SVM model; Obtain A + 1 historical times Time′1, Time′2, ..., Time′ A+1 , and the historical feature information TState′ corresponding to each historical time Time′ m where m and n are natural numbers, 1 ≤ m, n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ m , and the intervals between adjacent historical times are equal;​​ A+1 pieces of historical feature information TState′1, TState′2, ..., TState′ A+1 The information is input into the LSTM model, and A+1 target historical feature information is obtained from the model; the A+1 target historical feature information is processed using the SVM model, and the fault information of the energy storage power station is received from the SVM model.

10. A fault prediction device for an energy storage power station, characterized in that: Includes the following models: A preprocessing module, used to execute the generation method in any one of claims 1 to 4 and obtain an LSTM model, and obtain a trained SVM model; Historical data acquisition module, configured to acquire A + 1 historical times Time′1, Time′2, ..., Time′ A+1 , and the historical feature information TState′ corresponding to each historical time Time′ m , where m and n are natural numbers, 1 ≤ m, n ≤ A + 1, and when m < n, Time′ m is earlier than Time′ m , and the intervals between adjacent historical times are equal;​​ The prediction module is used to transform A+1 historical feature information TState′1, TState′2, ..., TState′ A+1 The information is input into the LSTM model, and A+1 target historical feature information is obtained from the model; the A+1 target historical feature information is processed using the SVM model, and the fault information of the energy storage power station is received from the SVM model.

Images