Battery energy storage characteristic evaluation method and system based on MM-LSTM-ATTENTION, computer readable storage medium and processor

Abstract

The invention relates to the technical field of energy storage system reliability evaluation, in particular to a battery energy storage characteristic evaluation method and system based on MM-LSTM-ATTENTION, a computer readable storage medium and a processor. Aiming at the processing of original collected data, a morphological filtering method is adopted, and the processed data are recombined into a new data set for prediction. According to the method, the problem of data interference can be effectively solved, and the prediction performance of the model is further improved. According to the method, the LSTM network and the attention mechanism are combined and applied to energy storage battery performance evaluation, and the strategy effectively improves the precision and efficiency of the model during time sequence data processing.

Description

Technical Field

[0001] The present invention relates to the technical field of reliability evaluation of energy storage systems, and particularly to a method, a system, a computer-readable storage medium and a processor for evaluating battery energy storage characteristics based on MM-LSTM-ATTENTION. Background Art

[0002] As an important means to solve the volatility and intermittency of renewable energy power generation, energy storage technology has received extensive attention and development globally in recent years. Especially on the grid side, energy storage power stations can not only meet the application requirements such as peak shaving, frequency modulation, and voltage regulation of regional power grids, but also provide functions such as emergency response and black start, providing reliable guarantees for the safe and stable operation of regional power grids. With the continuous progress of ion battery technology and the gradual reduction of costs, its application in grid-side energy storage power stations is becoming increasingly widespread.

[0003] However, the performance evaluation and detection of energy storage batteries face many challenges. On the one hand, during long-term use, energy storage batteries will experience performance degradation, such as capacity attenuation and increased internal resistance. These degradation phenomena will directly affect the operation efficiency and safety of energy storage power stations. On the other hand, existing energy storage battery test standards and detection methods mainly focus on the electrical performance of batteries, and the research on aspects such as battery consistency manufacturing technology, battery grouping technology, and battery thermal management technology is not deep enough. In addition, with the rapid development of the energy storage industry, the original standardized pattern can no longer meet the new requirements of current energy storage battery technology research and application. In response to the above problems, scholars and research institutions at home and abroad have successively carried out research on energy storage battery performance evaluation and detection methods. Among them, the data-driven energy storage battery performance evaluation method has attracted much attention due to its advantages such as high precision, high efficiency, and scalability. By collecting and analyzing the information of the energy storage battery degradation dataset, key features affecting battery performance can be extracted, and then a model that can accurately predict battery performance degradation can be constructed. However, existing methods basically only focus on the health state of energy storage power station batteries and pay less attention to other indicators. Summary of the Invention

[0004] In view of the problems in the prior art, the present invention provides a method, a system, a computer-readable storage medium and a processor for evaluating battery energy storage characteristics based on MM-LSTM-ATTENTION. The specific technical solutions are as follows: A method for evaluating battery energy storage characteristics based on MM-LSTM-ATTENTION includes the following steps: Step S1, collecting an energy storage battery degradation dataset, where the dataset includes a historical battery degradation dataset or and an operation dataset; Step S2: Numerically filter the energy storage battery degradation dataset using the mathematical morphology filtering method to obtain the filtered energy storage battery degradation dataset; Step S3: Extract the features affecting the energy storage battery based on the filtered energy storage battery degradation dataset, construct the eigenvalue input vector X and the evaluation label Y, and use the historical battery degradation dataset as the training data; Step S4: Preprocess the eigenvalue input vector X to obtain the preprocessed eigenvalue input vector X; Step S6: Input the operating data of the energy storage battery into the trained LSTM-ATTENTION model to obtain the evaluation label of the energy storage battery. Step S6: Input the operating data of the energy storage battery into the trained LSTM-ATTENTION model to obtain the evaluation label of the energy storage battery.

[0005] Preferably, the mathematical morphology filtering method in step S2 is specifically as follows: ; wherein, is the opening operation symbol, is the closing operation symbol, is the dilation operation symbol; is f k ( n ); f k ( n ) represents the k-th dimensional data in the energy storage battery degradation dataset F, F = { f 1( n ), f 2( n ), … f m ( n )}, and there are m-dimensional data in total; is g ( n ), g ( n ) is a one-dimensional signal representing the structural element.

[0006] Preferably, the eigenvalue input vector X = [SV, IC, DTV, TVC, T, R, C], where SV represents the singular value of the original voltage matrix A of the energy storage battery, IC represents the battery capacity increment of the continuous voltage increment of the energy storage battery, DTV represents the change in temperature and voltage of the energy storage battery, TVC represents the change rate of the terminal voltage of the energy storage battery, T represents the temperature of the energy storage battery, and R and C represent the charging and discharging power of the energy storage battery.

[0007] Preferably, the evaluation label Y = [rd, sλ, cd, st, rλ, cλ, ec, rt]; where rd represents the cycle life of the energy storage battery, sλ represents the system efficiency of the energy storage battery, cd represents the calendar life of the energy storage battery, st represents the energy storage duration of the energy storage battery, rλ and cλ represent the charge and discharge efficiency of the energy storage battery, ec represents the capacity of the energy storage battery, and rt represents the response time of the energy storage battery.

[0008] Preferably, the preprocessing of the eigenvalue input vector X in step S4 is specifically to normalize the eigenvalue input vector X using min-max normalization, as follows: ; Where X std represents the normalized data value, x represents the unnormalized data value, x max and x min represent the maximum and minimum values in the unnormalized data, respectively.

[0009] Preferably, step S5 specifically includes the following steps: Step S51, construct an LSTM model, the LSTM model contains 3 gates, namely the input gate, the forget gate and the output gate; Step S52, input the preprocessed eigenvalue input vector X and the evaluation label Y into the LSTM model, introduce the attention mechanism, and obtain the trained LSTM-ATTENTION model when the evaluation index of the model reaches the preset value.

[0010] Preferably, the attention mechanism specifically includes the following steps: (1) Calculate the attention score, calculate the correlation between each input vector and the query vector through the attention scoring function, and the most commonly used scoring function is the dot product function, as follows: ; Where represents the attention score of the LSTM model at time step t, represents the output vector of the LSTM model at time step t, is the query vector; (2) Calculate the attention distribution weight, use the Softmax function to normalize the attention score of the th eigenvalue input vector to the range [0,1], as follows: ; Among them, α t represents the attention distribution weight at time step t ; S t represents the attention score at time step t , and the denominator represents the sum of exp( from time step t to time step S k ); (3) Calculate the final attention output vector. According to the attention distribution weight α, perform weighted averaging on the input data to obtain the final output, specifically as follows: ; where y t+1 represents the final predicted output at time step t + 1; αk represents the attention weight at time step k, indicating the degree of attention to historical information; is the output of the LSTM model at time step k.

[0011] A battery energy storage characteristic evaluation system based on MM-LSTM-ATTENTION, applying the described method, includes: A data acquisition module, used to acquire the energy storage battery degradation data set, and the data set includes the historical battery degradation data set or and the operation data set; A filtering module, used to perform digital filtering on the energy storage battery degradation data set by using the mathematical morphology filtering method to obtain the filtered energy storage battery degradation data set; A training data construction module, used to extract the features affecting the energy storage battery according to the filtered energy storage battery degradation data set, and construct the eigenvalue input vector X and the evaluation label Y, and use the historical battery degradation data set as the training data; A preprocessing module, used to preprocess the eigenvalue input vector X to obtain the preprocessed eigenvalue input vector X; A model training module, used to construct an LSTM-ATTENTION model, and use the evaluation label Y in the training data and the preprocessed eigenvalue input vector X to train the LSTM-ATTENTION model to obtain a trained LSTM-ATTENTION model; An evaluation module, used to input the operation data of the energy storage battery into the trained LSTM-ATTENTION model to obtain the evaluation label of the energy storage battery.

[0012] A computer-readable storage medium, the computer-readable storage medium including a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the battery energy storage characteristic evaluation method based on MM-LSTM-ATTENTION.

[0013] A processor, the processor being used to run a program, wherein when the program runs, it executes the battery energy storage characteristic evaluation method based on MM-LSTM-ATTENTION.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: For the processing of the original collected data, the present invention adopts a morphological filtering method, recombines the processed data into a new data set and then performs prediction. This method can effectively solve the data interference problem and further improve the prediction performance of the model. The present invention combines the LSTM network with the attention mechanism and applies it to the performance evaluation of energy storage batteries. This strategy effectively improves the accuracy and efficiency of the model when processing time series data. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts do not necessarily draw according to the actual ratio.

[0016] Figure 1 It is the method flow chart of the present invention.

[0017] Figure 2 It is the schematic diagram of the LSTM model.

[0018] Figure 3 It is the architecture diagram of LSTM + attention mechanism.

[0019] Figure 4 It is the comparison result diagram of the present invention and other algorithms.

[0020] Figure 5 It is the rd prediction result diagram.

[0021] Figure 6 It is the system schematic diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be understood that when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0024] It should also be understood that the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0025] It should be further understood that the term " / and" used in the specification of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0026] Embodiment 1: As Figure 1As shown in the figure, this embodiment provides a method for evaluating the energy storage characteristics of batteries based on MM-LSTM-ATTENTION. First, digital filtering is performed on the actually collected energy storage battery degradation data set using mathematical morphology to filter out interference and retain the original features. Then, according to the information of the extracted and processed battery degradation data set, combined with parameters such as voltage, charge and discharge power, and temperature, a set of characteristic values that can reflect the energy storage characteristics of the battery is constructed. Secondly, the overall characteristic data is normalized to eliminate the influence of the dimension and numerical differences between different variables. Then, an LSTM-ATTENTION model is constructed. The structural advantages of the LSTM model are used to capture the time series characteristics in the process of battery performance degradation, and at the same time, the attention mechanism is combined to weight the key features to improve the prediction accuracy of the model. Finally, regression model evaluation indexes, such as mean absolute error, mean relative error, and mean square error, are set to evaluate the trained LSTM-ATTENTION model. By comparing the differences between the regression data and the actual data, the model parameters and structure are continuously optimized to obtain the optimal energy storage battery performance evaluation model, and based on this, a comprehensive analysis of the batteries in the energy storage power station is carried out. The method proposed by the present invention can make full use of the information of the energy storage battery degradation data set, extract the key features affecting the battery performance, and construct a high-precision performance evaluation model. This method provides strong technical support for the battery performance evaluation and detection work of the grid-side energy storage power station, and helps to promote the healthy development of the energy storage industry. Specifically, it includes the following steps: Step S1, collect the energy storage battery degradation data set, and the data set includes the historical battery degradation data set or and the operation data set.

[0027] Step S2, perform digital filtering on the energy storage battery degradation data set using the mathematical morphology filtering method to obtain the filtered energy storage battery degradation data set.

[0028] The mathematical morphology filtering method in Step S2 is specifically as follows: ; Among them, is the opening operation symbol, is the closing operation symbol, is the dilation operation symbol; is f k ( n ) f k ( n ) represents the k-th dimension data in the energy storage battery degradation data set F, F = { f 1( n ) f 2( n ) f m (n )}, with a total of m - dimensional data, including battery operation data information such as temperature, current, and voltage; is g ( n ), g ( n ) is a one - dimensional signal representing a structuring element.

[0029] (1) Basic morphological operation operator - opening operation Using the structuring element g ( n ) to perform the opening operation on the signal f ( n ) is defined as: ; In the formula: represents the erosion operation.

[0030] (2) Basic morphological operation operator - closing operation Using the structuring element g ( n ) to perform the closing operation on the signal f ( n ) is defined as: ; (3) Basic morphological operation operator - dilation Using the structuring element g ( n ) to perform dilation on the data f ( n ) in all the collected overall datasets: ; Among them, D f and D g respectively represent f ( n ) and g ( n )'s domains.

[0031] (4) Basic morphological operation operator - erosion Using the structuring element g ( n ) to perform erosion on the data f ( n ) in all the collected overall datasets: .

[0032] Perform the above-mentioned MM filtering process on all the collected energy storage battery degradation datasets F to obtain the filtered energy storage battery degradation datasets, providing clean data for subsequent calculations.

[0033] Step S3: Extract the features affecting the energy storage battery according to the filtered energy storage battery degradation datasets, construct the eigenvalue input vector X and the evaluation label Y, and use the historical battery degradation datasets as the training data.

[0034] The eigenvalue input vector X = [SV, IC, DTV, TVC, T, R, C], where SV represents the singular value of the original voltage matrix A of the energy storage battery, IC represents the battery capacity increment of the continuous voltage increment of the energy storage battery, DTV represents the change in temperature and voltage of the energy storage battery, TVC represents the change rate of the terminal voltage of the energy storage battery, T represents the temperature of the energy storage battery, and R, C represent the charging and discharging power of the energy storage battery.

[0035] (1) The calculation method of the singular value (Singular value, SV) of the original voltage matrix A of the energy storage battery is as follows: As a basic matrix decomposition technique, the singular value decomposition method compresses the information in a complex matrix with a simpler matrix, achieving data compression while maintaining the information characteristics. The singular value decomposition of the original voltage matrix A can be expressed as: ; where A is the original voltage input matrix of size m×n, m represents the time points, and n represents the columns of voltage measurement values at different battery stages; P is the left singular matrix of size m×m of the original voltage matrix A; Q is the right singular matrix of size n×n of the original voltage matrix A; E is the identity matrix of size m×n, that is, all matrix elements are 0 except for the elements on the main diagonal. Extract the values in the singular matrix as SV, which is another voltage feature representation of the battery.

[0036] (2) The calculation method of the battery capacity increment IC (Incremental Capacity, IC) of the continuous voltage increment of the energy storage battery is as follows: The IC curve, as a common feature analysis method for describing the battery degradation process, can characterize the battery degradation from the electrode level and has a high resolution for the battery charge and discharge plateau region. Define the IC curve as the battery capacity increment of the continuous voltage increment. During the battery degradation process, the IC curve changes continuously, and the degradation characteristics of the battery can be obtained. Its mathematical representation is: ; In the formula, Q, I, V, and t are the discharge capacity, current, terminal voltage, and sampling time of the energy storage battery, respectively.

[0037] (3)The calculation method of the change amount DTV (Differential Thermal Voltammetry) of the temperature and voltage of the energy storage battery is as follows: Temperature and terminal voltage have a profound impact on the degree of battery deterioration. Based on this, the first typical battery characteristic is obtained by using differential thermal voltammetry (DTV). The use of the DTV method allows the degree and type of phase change within the electrode material to be identified by analyzing the changes in temperature and voltage. This information is crucial for understanding the entropy change in the system, which determines the degradation stage of the battery. The calculation formula for the DTV characteristic value: ; In the formula, V is the terminal voltage; T is the temperature; t is the sampling time.

[0038] (4)The calculation method of the terminal voltage change rate TVC (Terminal Voltage Characteristic) of the energy storage battery is as follows: ; In the formula: V L is the low cut-off voltage value of the battery, V H is the high cut-off voltage value; t L and t H respectively represent V L and V H the end time and start time of the discharge process between.

[0039] The evaluation labels are Y = [rd, sλ, cd, st, rλ, cλ, ec, rt]; where rd represents the cycle life of the energy storage battery, sλ represents the system efficiency of the energy storage battery, cd represents the calendar life of the energy storage battery, st represents the energy storage duration of the energy storage battery, rλ and cλ represent the charge and discharge efficiencies of the energy storage battery, ec represents the capacity of the energy storage battery, and rt represents the response time of the energy storage battery.

[0040] The SV reflects the structural changes in voltage data during the battery degradation process (such as electrode phase transformation), and is directly related to the capacity (ec) attenuation of the energy storage battery. The decrease in the IC peak directly reflects the reduction in available capacity, and the change in the IC curve shape directly indicates the changes in the charge and discharge efficiency (rλ, cλ) of the energy storage battery. The DTV combines temperature and voltage changes to reveal the internal entropy change and phase transformation process of the battery. The entropy change affects the thermodynamic efficiency and then affects the system efficiency (sλ) of the energy storage battery, while the material aging caused by the phase transformation accelerates the calendar life (cd) attenuation of the energy storage battery. The rate of change of the terminal voltage is related to the dynamic response ability (rt) of the energy storage battery, and the shortening of the voltage platform reflects the energy storage ability (st) of the energy storage battery. The R, C, and T of the energy storage battery directly affect the chemical reaction rate and aging mechanism, and high temperature accelerates the shortening of the cycle life (rd) of the energy storage battery, and temperature fluctuations directly affect the decrease in the power transmission efficiency (sλ) of the energy storage battery.

[0041] From the above relationships, it can be seen that there is a strong correlation between the input and output constructed by the present invention, but its display expression is difficult to directly give. Therefore, the present invention uses an intelligent algorithm to map the relationship between the input and the output Y.

[0042] Step S4: Preprocess the eigenvalue input vector X to obtain the preprocessed eigenvalue input vector X.

[0043] The dimensions of different variables are different and the numerical differences are large. Considering the input and output ranges of the non-linear activation function in the model, and also to equally handle the influence of each variable on the prediction result, it is necessary to normalize each variable and normalize all data to the interval [0, 1]. Therefore, the specific preprocessing of the eigenvalue input vector X is to normalize the eigenvalue input vector X using the minimum-maximum normalization, as follows: ; Where, X std represents the normalized data value, x represents the unnormalized data value, x max and x min represent the maximum and minimum values in the unnormalized data respectively.

[0044] Step S5: Construct an LSTM-ATTENTION model, and use the evaluation label Y in the training data and the preprocessed eigenvalue input vector X to train the LSTM-ATTENTION model to obtain a trained LSTM-ATTENTION model. Specifically, it includes the following steps: Step S51: Construct an LSTM model. The LSTM model is shown in Figure 2As shown in the figure, the LSTM model contains three gates, namely the input gate, the forget gate, and the output gate; the three gates control the flow of information between the tuple and the network. i t , o t , f t Represent the state values ​​of the input gate, output gate and forget gate respectively.

[0045] Generate the information that needs to be updated and store it in the cell state, which includes 2 steps: i) The result of the input gate passing through the sigmoid layer i t To update information; ii) New candidate values ​​generated by the tanh layer C t will be added to the cell state, C t Multiply the old cell state to forget the unnecessary information and the new candidate information i t * C t Add together.

[0046] ; The sigmoid layer of the forget gate determines the old cell state C t-1 Forget the information, the input is the input of the current layer x t And the output of the previous layer h t-1 , at this moment the cell state output is: ; The output information is determined by the output gate. First, the initial output is obtained through the sigmoid layer. The tanh layer is used to scale the value of the cell state to between [-1, 1] and multiply it pairwise with the output obtained by sigmoid to obtain the output. h t .

[0047] ; Where: , , , Communication x t The weight matrix with the input gate, forget gate, output gate and tuple input of the tuple; , , , are respectively connections h t-1 and the weight matrices of the input gate, forget gate, output gate, and tuple input of the tuple; , , , are the bias vectors of the input gate, forget gate, output gate, and tuple input; σ represents the sigmoid activation function.

[0048] Take the feature data obtained in the previous step as the input of the LSTM network model, and the output is: ; In the formula, is the value of the battery characteristic index predicted by the output of the LSTM network model at time t+1, ··· etc. are the historical data of the evaluation labels, and x is the eigenvalue input data, that is, the constructed input eigenvalue set X = [SV, IC, DTV, TVC, T, R, C].

[0049] Step S52, input the preprocessed eigenvalue input vector X and the evaluation label Y into the LSTM model, introduce the attention mechanism, and obtain the trained LSTM-ATTENTION model when the evaluation index of the model reaches the preset value, as Figure 3 shown.

[0050] The attention mechanism specifically includes the following steps: (1) Calculate the attention score, calculate the correlation between each input vector and the query vector through the attention scoring function, and the most commonly used scoring function is the dot product function, specifically as follows: ; Among them, represents the attention score of the LSTM model at time step t, represents the output vector of the LSTM model at time step t, is the query vector; (2) Calculate the attention distribution weight, use the Softmax function to normalize the attention score of the th eigenvalue input vector to the range of [0,1], specifically as follows: ; Among them, α t represents the attention distribution weight at time step t ; S t represents the attention distribution weight at time step tThe attention score at that time, where the denominator represents the cumulative sum from time step to time step t of exp( S k ); (3) Calculate the final attention output vector. According to the attention distribution weight α, perform weighted averaging on the input data to obtain the final output, specifically as follows: ; where y t+1 represents the final predicted output at time step t + 1; αk represents the attention weight at time step k, indicating the degree of attention to historical information; is the output of the LSTM model at time step k.

[0051] Perform regression using the LSTM-ATTENTION model based on the processed data. In addition, to establish the evaluation index of the mapping model for the characteristics analysis of energy storage batteries, compare the regression data y with the actual data y*, and use the mean absolute error, mean relative error, and mean square error as the model evaluation indexes to evaluate the training model and obtain the optimal trained LSTM-ATTENTION model. The following are the three evaluation error indexes: ① Mean absolute error ; ② Mean relative error ; ③ Root mean square error ; In the above formulas: y is the output value; y* is the actual value; n represents the number of samples.

[0052] Step S6: Input the operation data of the energy storage battery into the trained LSTM-ATTENTION model to obtain the evaluation label of the energy storage battery.

[0053] In this embodiment, the NASA Oxford battery degradation dataset in the United States is used as the original feature data. The original data is processed by morphological filtering to strip out all clean data information containing battery characteristics. All feature sequences are sampled at intervals of every 12h as sampling points, and the overall data samples collected are divided into 3000 groups of data.

[0054] According to the terminal voltage ( V ), battery temperature (T), sampling time ( t ), discharge capacity (Q), battery low cut-off voltage value ( V L ), high cut-off voltage value ( VH ) Calculate DTV, IC, TVC, SV, and calculate the characteristics of classic energy storage batteries: ; In addition, combining the charge and discharge power (R, C) and temperature (T) together constitutes a set of 7 characteristic values X = [SV, IC, DTV, TVC, V, I, T, R, C] that can reflect the energy storage characteristics of the battery.

[0055] Similarly, the battery characteristic analysis indicators: cycle life (rd), system efficiency (sλ), calendar life (cd), energy storage duration (st), charge and discharge efficiency (rλ, cλ), capacity (ec), response time (rt).

[0056] Subsequently, 3000 groups of data characteristics X and labels Y are formed, and then they are divided according to 8:2 to obtain the training set and the test set, and the optimal MM-LSTM-ATTENTION model is trained according to the evaluation indicators.

[0057] Figure 4 Shows the accuracy comparison between the MM-LSTM-ATTENTION model and other methods when predicting each battery characteristic index. It can be seen from Figure 4 that the MM-LSTM-ATTENTION model proposed by the present invention can achieve the best prediction effect in all indicators. Taking the rd prediction implemented by the method of the present invention as an example, the result Figure 5 can be seen that the present invention can accurately predict the battery rd, indicating the changing trend of the battery characteristic index predicted by the present invention over time, and clearly showing the performance degradation law of the battery under different working environments. Through the method of the present invention, the battery performance can be accurately predicted, providing a scientific basis for the operation and maintenance decision-making of the energy storage power station.

[0058] The present invention innovatively combines the LSTM network with the attention mechanism and applies it to the performance evaluation of energy storage batteries. This strategy effectively improves the accuracy and efficiency of the model in processing time series data. The performance evaluation of energy storage batteries is a complex and crucial task, which requires the model to accurately capture the degradation characteristics of the battery during long-term use and make reliable predictions based on these characteristics. Traditional machine learning methods often struggle to fully utilize the time series information in historical data when dealing with such problems, resulting in inaccurate prediction results. The LSTM neural network, with its unique gated memory cell structure, can better handle load characteristics, namely time series data, and fully extract the key information therein. However, the LSTM network also has certain limitations when dealing with certain types of data, such as noisy raw data, which can lead to a slower convergence speed and affected accuracy when the LSTM network processes such data. To solve this problem, the present invention introduces the attention mechanism on the basis of the LSTM network. The attention mechanism enables the model to focus more on the key information that has a greater impact on load changes (specifically referring to the key feature changes during the performance degradation process of energy storage batteries) during the training process and assigns a greater weight to it. In this way, the model can more effectively extract the features that have an important impact on the prediction results, thereby improving the prediction accuracy. At the same time, for the processing of the original collected data, the present invention adopts a morphological filtering method to recombine the processed data into a new data set and then make predictions. This method can effectively solve the data interference problem and further improve the prediction performance of the model. This innovative method not only improves the accuracy and efficiency of the model in processing complex time series data but also provides new ideas and technical support for the performance evaluation of energy storage batteries. By introducing the attention mechanism, the model can more accurately capture the key features during the battery degradation process, providing a more reliable basis for the maintenance and management of the battery.

[0059] Embodiment 2: Based on the same inventive concept as Embodiment 1, this embodiment provides a battery energy storage characteristic evaluation system based on MM-LSTM-ATTENTION, applying the method described above, including: A data acquisition module for acquiring an energy storage battery degradation data set, where the data set includes a historical battery degradation data set or and an operation data set; A filtering module for digitally filtering the energy storage battery degradation data set using a mathematical morphological filtering method to obtain a filtered energy storage battery degradation data set; A training data construction module for extracting the features affecting the energy storage battery according to the filtered energy storage battery degradation data set and constructing a feature value input vector X and an evaluation label Y, using the historical battery degradation data set as the training data; A preprocessing module for preprocessing the eigenvalue input vector X to obtain a preprocessed eigenvalue input vector X; A model training module for constructing an LSTM-ATTENTION model and training the LSTM-ATTENTION model with the evaluation label Y in the training data and the preprocessed eigenvalue input vector X to obtain a trained LSTM-ATTENTION model; An evaluation module for inputting the operation data of the energy storage battery into the trained LSTM-ATTENTION model to obtain an evaluation label of the energy storage battery.

[0060] Embodiment 3: Based on the same inventive concept as Embodiment 1, this embodiment provides a computer-readable storage medium, which includes a stored program. When the program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned method for evaluating the battery energy storage characteristics based on MM-LSTM-ATTENTION.

[0061] Embodiment 4: Based on the same inventive concept as Embodiment 1, this embodiment provides a processor for running a program. When the program runs, it executes the above-mentioned method for evaluating the battery energy storage characteristics based on MM-LSTM-ATTENTION.

[0062] Those of ordinary skill in the art can realize that the modules of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0063] In the embodiments provided by the present invention, it should be understood that the division of modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules can be combined into one module, one module can be split into multiple modules, or some features can be ignored, etc.

[0064] In addition, the functional modules in each embodiment of the present invention can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0065] If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0066] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention, and they should all be covered by the scope of the claims and the description of the present invention.

Claims

1. A method for evaluating the energy storage characteristics of a battery based on MM-LSTM-ATTENTION, characterized in that, It includes the following steps: Step S1, collect the energy storage battery degradation data set, where the data set includes the historical battery degradation data set or and the operation data set; Step S2, perform digital filtering on the energy storage battery degradation data set by using the mathematical morphology filtering method to obtain the filtered energy storage battery degradation data set; Step S3, extract the features affecting the energy storage battery according to the filtered energy storage battery degradation data set, and construct the eigenvalue input vector X and the evaluation label Y, and use the historical battery degradation data set as the training data; The eigenvalue input vector X = [SV, IC, DTV, TVC, T, R, C], where SV represents the singular value of the original voltage matrix A of the energy storage battery, IC represents the battery capacity increment of the continuous voltage increment of the energy storage battery, DTV represents the change in temperature and voltage of the energy storage battery, TVC represents the terminal voltage change rate of the energy storage battery, T represents the temperature of the energy storage battery, and R, C represent the charge and discharge power of the energy storage battery; The evaluation label Y = [rd, sλ, cd, st, rλ, cλ, ec, rt]; where rd represents the cycle life of the energy storage battery, sλ represents the system efficiency of the energy storage battery, cd represents the calendar life of the energy storage battery, st represents the energy storage duration of the energy storage battery, rλ, cλ represent the charge and discharge efficiency of the energy storage battery, ec represents the capacity of the energy storage battery, and rt represents the response time of the energy storage battery; Step S4, preprocess the eigenvalue input vector X to obtain the preprocessed eigenvalue input vector X; the preprocessing of the eigenvalue input vector X is specifically to normalize the eigenvalue input vector X by using the minimum-maximum normalization, as follows: ； Among them, X std represents the normalized data value, x represents the unnormalized data value, x max and x min respectively represent the maximum and minimum values in the unnormalized data; Step S5, construct an LSTM-ATTENTION model, and use the evaluation label Y and the preprocessed eigenvalue input vector X in the training data to train the LSTM-ATTENTION model to obtain a trained LSTM-ATTENTION model; Step S6, input the operation data of the energy storage battery into the trained LSTM-ATTENTION model to obtain the evaluation label of the energy storage battery.

2. The method for evaluating the energy storage characteristics of a battery based on MM-LSTM-ATTENTION according to claim 1, characterized in that, The mathematical morphology filtering method in step S2 is specifically as follows: ； Among them, is the opening operation symbol, is the closing operation symbol, is the dilation operation symbol; is f k ( n ); f k ( n ) represents the k - th dimensional data in the energy storage battery degradation dataset F, F = { f 1( n ), f 2( n ), … f m ( n )}, and there are m - dimensional data in total; is g ( n ), g ( n ) is a one - dimensional signal, representing the structuring element.

3. The method for evaluating the energy storage characteristics of a battery based on MM-LSTM-ATTENTION according to claim 1, characterized in that, Step S5 specifically includes the following steps: Step S51, construct an LSTM model, and the LSTM model includes 3 gates, namely the input gate, the forget gate, and the output gate; Step S52, input the preprocessed eigenvalue input vector X and the evaluation label Y into the LSTM model, introduce the attention mechanism, and obtain a trained LSTM-ATTENTION model when the evaluation index of the LSTM-ATTENTION model reaches a preset value.

4. The method for evaluating the energy storage characteristics of a battery based on MM-LSTM-ATTENTION according to claim 3, characterized in that, The attention mechanism specifically includes the following steps: (1) Calculate the attention score, and calculate the correlation between each input vector and the query vector through the attention scoring function, where the attention scoring function uses the dot product function, specifically as follows: ； wherein, denotes the attention score of the LSTM model at time step t, denotes the output vector of the LSTM model at time step t, is the query vector; (2)Calculate the attention distribution weights, and use the Softmax function to normalize the attention scores of the input vector of the feature values to the range [0, 1] as follows: ； Among them, α t represents the attention distribution weight at time step t . S t represents the attention score at time step t , and the denominator represents the sum of exp( from time step t to time step S k ). (3) Calculate the final attention output vector, and perform weighted averaging on the input data according to the attention distribution weight α to obtain the final output, specifically as follows: ； Among them, y t+1 represents the final predicted output at time step t +1; αk represents the attention weight at time step k, indicating the degree of attention to historical information; is the output of the LSTM model at time step k.

5. A system for evaluating the energy storage characteristics of a battery based on MM-LSTM-ATTENTION, characterized in that, Applying the method according to any one of claims 1 to 4 includes: A data acquisition module, which is used to acquire a degradation data set of an energy storage battery, and the data set includes a historical battery degradation data set or and an operation data set; A filtering module, which is used to perform digital filtering on the degradation data set of the energy storage battery by using a mathematical morphology filtering method to obtain a filtered degradation data set of the energy storage battery; A training data construction module, which is used to extract features affecting the energy storage battery according to the filtered degradation data set of the energy storage battery, construct a feature value input vector X and an evaluation label Y, and use the historical battery degradation data set as training data; A preprocessing module, which is used to preprocess the feature value input vector X to obtain a preprocessed feature value input vector X; A model training module, which is used to construct an LSTM-ATTENTION model, and use the evaluation label Y in the training data and the preprocessed feature value input vector X to train the LSTM-ATTENTION model to obtain a trained LSTM-ATTENTION model; An evaluation module, which is used to input the operation data of the energy storage battery into the trained LSTM-ATTENTION model to obtain an evaluation label of the energy storage battery.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute a battery energy storage characteristic evaluation method according to any one of claims 1 to 4.

7. A processor, characterized in that, The processor is used to run the program, wherein when the program runs, it executes a battery energy storage characteristic evaluation method according to any one of claims 1 to 4.

Images