Method and device for evaluating state of charge of energy storage system

Abstract

The invention relates to the technical field of energy storage system state-of-charge evaluation, in particular to an energy storage system state-of-charge evaluation method and device, and the method comprises the following steps: data collection, data processing, establishment of a state-of-charge prediction model, division of a training set, a verification set and a test set, training by using the training set, and verification by using the verification set. And testing by using the test set, and predicting the charge state of the energy storage system by using the charge state prediction model meeting the precision requirement. According to the method, the first bidirectional long-short-term memory neural network layer and the second bidirectional long-short-term memory neural network layer are established, the time sequence of the input data is input into the bidirectional long-short-term memory neural network layer, the original data are analyzed, the accuracy of the prediction result is higher, the multi-head attention mechanism layer is introduced, and the accuracy of the prediction result is improved. The dependency relationship between the load sequence information can be effectively extracted to judge the weight of the influence of each parameter on the SOC output result of the energy storage power station, and the prediction precision is improved.

Description

Technical Field

[0001] This invention relates to the field of energy storage system state of charge assessment technology, specifically to a method and apparatus for energy storage system state of charge assessment. Background Technology

[0002] Traditional energy storage technologies are mainly mechanical, such as flywheel energy storage and compressed air energy storage. However, electrochemical energy storage has a faster response speed, can adapt to various complex environments, and can be distributed. Therefore, as the technology matures, electrochemical energy storage is increasingly being used in practical applications. The mainstream energy storage medium in current technologies is lithium battery. Due to the complexity and nonlinear characteristics of lithium battery energy storage systems, there are many factors affecting its state of charge, and the mapping relationship between each factor and the system is complex. At present, the assessment of state of charge mainly adopts the ampere-hour integral method and the open-circuit voltage method. Although the ampere-hour integral method and the open-circuit voltage method are simple and easy to implement, they have high requirements for environmental conditions, and the results are greatly affected by environmental factors, which will reduce the accuracy of the assessment. Summary of the Invention

[0003] To overcome the technical problem of poor accuracy in assessing the state of charge in existing technologies, this invention provides a method and apparatus for assessing the state of charge of an energy storage system.

[0004] The technical solution adopted by the present invention to achieve the above objectives is: a method for assessing the state of charge of an energy storage system, comprising the following steps:

[0005] S1: Collect energy storage system operation data and corresponding energy storage system state of charge data;

[0006] S2: Normalize the energy storage system operation data and energy storage system state of charge data, and divide them into training set, validation set and test set;

[0007] S3: Construct a state of charge assessment model and initialize the hyperparameters of the state of charge assessment model. The state of charge assessment model includes an input layer, a first bidirectional long short-term memory neural network layer, a second bidirectional long short-term memory neural network layer, a multi-head attention mechanism layer, and an output layer.

[0008] S4: Train the state of charge assessment model using normalized energy storage system operation data and energy storage system state of charge data until the assessment accuracy meets the requirements.

[0009] S5: Use a state of charge prediction model that meets the accuracy requirements to predict the state of charge of the energy storage system.

[0010] Preferably, the energy storage system operation data, including battery stack voltage, stack current, and stack average temperature, is used as input data. The collected energy storage system operation data is normalized so that the normalized data are all within the range [0,1]. The normalized energy storage system operation data and the energy storage system state of charge data are then divided into training, validation, and test sets. The normalization expression is as follows:

[0011]

[0012] Where X represents the collected energy storage system operation data, and X' represents the normalized energy storage system operation data. Max and X Min These are the maximum and minimum values ​​in the collected energy storage system operation data, respectively.

[0013] Preferably, both the first and second bidirectional long short-term memory (LSTM) neural network layers are used to predict the state of charge. Each layer contains multiple bidirectional LSTM neural network units, with the number of units equal to the length of the input data time series. Each LSTM neural network unit is composed of a forward-propagating LSTM neural network and a backward-propagating LSTM neural network. The calculation expression for each LSTM neural network unit is as follows:

[0014] H + =LSTM + (H t-1 ,x t )

[0015] H - =LSTM - (H t-1 ,x t )

[0016] y t =W + ·H + +W - ·H - +b

[0017] Among them, H + and H - The outputs of the forward and backward propagation Long Short-Term Memory (LSTM) neural network layers are respectively the outputs of the forward and backward propagation layers. + and LSTM - These are forward-propagating Long Short-Term Memory (LSTM) neural networks and backward-propagating LSTM neural networks, respectively. t-1 x is the output of the hidden layer of the previous unit.t For the current cell input, y t For the final output, W + and W - , respectively, are the weights of the long short-term memory neural network layers that propagate forward and backward, and b is the bias;

[0018] The expression for the Long Short-Term Memory (LSTM) neural network in the forward-propagating LSM and backward-propagating LSM neural networks is as follows:

[0019]

[0020] f t =σ(W f [H t-1 ,x t ]+b f )

[0021] i t =σ(W i [H t-1 ,x t ]+b i )

[0022]

[0023] O t =σ(W o [H t-1 ,x t ]+b o )

[0024] H t =O t *tanh(C t )

[0025] Among them, C t C t-1 , These represent the current cell state, the previous cell state, and the output layer cell state, respectively. t i t O t The outputs of the forget, input, and output structures in gate control are W, respectively. f W i W c W o These are the weight parameters, b f b i b c b o For bias, H t-1 x is the output of the hidden layer of the previous unit. t The input is the current cell, and σ and tanh are both activation functions.

[0026] Preferably, the multi-head attention mechanism layer processing flow includes the following steps:

[0027] The input dataset X is linearly mapped to a query matrix Q, a key matrix K, and a value matrix V, using the following formula:

[0028] Q = W i Q X

[0029] K = W i K X

[0030] V = W i V X

[0031] Among them, W i Q W i K W i V These are weight matrices used to map queries, keys, and values, respectively.

[0032] By learning the matrix, the multi-head attention mechanism layer adjusts the representation of the input data sequence in different mapping spaces, and the scaling dot product attention mechanism adjusts the network structure. Its expression is:

[0033]

[0034] Where, d k Let I be the dimension of matrix K. att (·) is used to calculate the attention weight matrix, softmax represents the normalized exponential function, and T represents the transpose;

[0035] The output of each bidirectional long short-term memory neural network unit is calculated individually, and then the results are linearly combined to obtain the output value, which is expressed as follows:

[0036] MultiHead(Q,K,V)=Concat(head1,...,head h W 0

[0037] head i (Q,K,V)=I att (QW i Q ,KW i K VW i V )

[0038] Where MultiHead is the multi-head attention matrix, Concat is the matrix concatenation operation, and head i Let head be the attention matrix for the i-th head. h Let W be the attention matrix for the h-th head. 0 The output weight matrix is ​​h, where h is the number of self-attention modules selected, i.e., the number of heads.

[0039] Preferably, the state of charge (SCC) assessment model is trained using normalized energy storage system operating data and energy storage system SCC data, including the following steps:

[0040] S401: Input the training set into the state of charge assessment model and train the state of charge assessment model using the gradient descent method until the assessment accuracy meets the requirements.

[0041] S402: Input the validation set into the trained state of charge assessment model, evaluate the state of charge of the energy storage system in the validation set, and calculate its evaluation accuracy. If the evaluation accuracy requirement is not met, readjust the hyperparameters in the state of charge assessment model and repeat steps S401-S402 until the evaluation accuracy meets the requirements.

[0042] S403: Input the test set into the validated state of charge (SOC) assessment model, assess the SOC of the energy storage system in the test set, calculate its assessment accuracy, and if the assessment accuracy requirement is not met, readjust the hyperparameters in the SOC assessment model and repeat steps S401-S403 until the assessment accuracy meets the requirements.

[0043] Preferably, in step S401, the root mean square error (RMSE) is used to evaluate the assessment accuracy. The accuracy requirement is that the difference between the RMSEs of the state-of-charge assessment models after two consecutive training iterations is less than 0.001. The formula for calculating the RMSE is:

[0044]

[0045] Where RMSE is the root mean square error, n is the total sample size, i is the sample batch, and y is the mean square error. i For the true value, These are predicted values.

[0046] Preferably, in steps S402 and S403, the evaluation accuracy is calculated using the mean absolute percentage error (MAE), with an accuracy requirement of less than 0.05. The formula for calculating the MAE is as follows:

[0047]

[0048] Where MAPE is the mean absolute percentage error, n is the total sample size, i is the sample batch, and y is the mean absolute percentage error. iFor the true value, These are predicted values.

[0049] Preferably, the hyperparameters of the state of charge assessment model include the hidden vector dimension of the first bidirectional long short-term memory neural network layer and the number of attention mechanism heads of the second bidirectional long short-term memory neural network layer. The hidden vector dimension of the first bidirectional long short-term memory neural network layer is 64, and the number of attention mechanism heads is 4. The hidden vector dimension of the second bidirectional long short-term memory neural network layer is 32, and the number of attention mechanism heads is 4.

[0050] A state-of-charge assessment device for an energy storage system includes a memory and a processor. The memory stores a computer program that, when loaded by the processor, executes the method described above.

[0051] A computer-readable storage medium storing a computer program adapted to perform the above-described method when loaded by a processor.

[0052] The beneficial effects of this invention are that by establishing a first bidirectional long short-term memory neural network layer and a second bidirectional long short-term memory neural network layer, the time series of input data is input into the bidirectional long short-term memory neural network layer to analyze the original data, resulting in higher accuracy of prediction results. Furthermore, this invention introduces a multi-head attention mechanism layer, which can effectively extract the dependencies between load sequence information to determine the weight of each parameter's influence on the SOC output of the energy storage power station, thereby improving prediction accuracy. Attached Figure Description

[0053] Figure 1 This is a flowchart of a method for assessing the state of charge of an energy storage system according to the present invention.

[0054] Figure 2 This is a schematic diagram of the state of charge assessment model in a specific embodiment of the present invention.

[0055] Figure 3 This is a comparison chart of the predicted values ​​and the actual values ​​in the training set in a specific embodiment of the present invention.

[0056] Figure 4 This is a comparison chart of the predicted values ​​and the actual values ​​of the validation set in a specific embodiment of the present invention. Detailed Implementation

[0057] The present invention provides a method for assessing the state of charge (SOC) of an energy storage system, comprising the following steps:

[0058] S1: Collect energy storage system operation data and corresponding energy storage system state of charge data. The energy storage system operation data is used as input data, including battery stack voltage, stack current and stack average temperature.

[0059] S2: Normalize the energy storage system operation data and energy storage system state of charge data, and divide them into training set, validation set and test set;

[0060] The collected energy storage system operation data is normalized so that the normalized data are all within the range [0,1]. The normalization expression is as follows:

[0061]

[0062] Where X represents the collected energy storage system operation data, and X' represents the normalized energy storage system operation data. Max and X Min These are the maximum and minimum values ​​in the collected energy storage system operation data, respectively.

[0063] S3: Construct a state of charge assessment model and initialize the hyperparameters of the state of charge assessment model. The state of charge assessment model includes an input layer, a first bidirectional long short-term memory neural network layer, a second bidirectional long short-term memory neural network layer, a multi-head attention mechanism layer, and an output layer.

[0064] Both the first and second bidirectional long short-term memory (LSTM) neural network layers are used to predict the state of charge. Each layer contains multiple bidirectional LSTM units, with the number of units equal to the length of the input data time series. Each LSTM unit is composed of a forward-passing LSTM neural network and a backward-passing LSTM neural network. The processing flow for the input data in both layers is as follows: The input data is processed in the corresponding LSTM unit according to the time series input value. Forward calculation is performed first, and the output is passed to the next LSTM unit for further forward calculation. This process continues until the last LSTM unit performs forward calculation, followed by backward calculation. The output is then passed to the previous LSTM unit for further backward calculation, and so on, until the first LSTM unit performs backward calculation and outputs the result.

[0065] The calculation expression for a bidirectional long short-term memory neural network unit is:

[0066] H + =LSTM + (H t-1 ,x t )

[0067] H -=LSTM - (H t-1 ,x t )

[0068] y t =W + ·H + +W - ·H - +b

[0069] Among them, H + and H - The outputs of the forward and backward propagation Long Short-Term Memory (LSTM) neural network layers are respectively the outputs of the forward and backward propagation layers. + and LSTM - These are forward-propagating Long Short-Term Memory (LSTM) neural networks and backward-propagating LSTM neural networks, respectively. t-1 x is the output of the hidden layer of the previous unit. t For the current cell input, y t For the final output, W + and W - , respectively, are the weights of the long short-term memory neural network layers that propagate forward and backward, and b is the bias;

[0070] The expression for the Long Short-Term Memory (LSTM) neural network in the forward-propagating LSM and backward-propagating LSM neural networks is as follows:

[0071]

[0072] f t =σ(W f [H t-1 ,x t ]+b f )

[0073] i t =σ(W i [H t-1 ,x t ]+b i )

[0074]

[0075] O t =σ(W o [H t-1 ,x t ]+b o )

[0076] H t =O t *tanh(C t )

[0077] Among them, C t C t-1 , These represent the current cell state, the previous cell state, and the output layer cell state, respectively. t i t O t The outputs of the forget, input, and output structures in gate control are W, respectively. f W i W c W o These are the weight parameters, b f b i b c b o For bias, H t-1 x is the output of the hidden layer of the previous unit. t The input is the current cell, where σ and tanh are both activation functions;

[0078] The output of each bidirectional long short-term memory (LSTM) neural network unit in the first and second bidirectional LSM neural network layers is input to the multi-head attention mechanism layer for processing. The processing flow of the multi-head attention mechanism layer includes the following steps:

[0079] The input dataset X is linearly mapped to a query matrix Q, a key matrix K, and a value matrix V, using the following formula:

[0080] Q = W i Q X

[0081] K = W i K X

[0082] V = W i V X

[0083] Among them, W i Q W i K W i V These are weight matrices used to map queries, keys, and values, respectively.

[0084] By learning the matrix, the multi-head attention mechanism layer adjusts the representation of the input data sequence in different mapping spaces, and the scaling dot product attention mechanism adjusts the network structure. Its expression is:

[0085]

[0086] Where, d k Let I be the dimension of matrix K. att(·) is used to calculate the attention weight matrix, softmax represents the normalized exponential function, and T represents the transpose;

[0087] The output of each bidirectional long short-term memory neural network unit is calculated individually, and then the results are linearly combined to obtain the output value, which is expressed as follows:

[0088] MultiHead(Q,K,V)=Concat(head1,...,head h W 0

[0089] head i (Q,K,V)=I att (QW i Q ,KW i K VW i V )

[0090] Where MultiHead is the multi-head attention matrix, Concat is the matrix concatenation operation, and head i Let head be the attention matrix for the i-th head. h Let W be the attention matrix for the h-th head. 0 The output weight matrix is ​​h, which is the number of self-attention modules selected, i.e., the number of heads.

[0091] The hyperparameters of the state of charge assessment model include the hidden vector dimension of the first bidirectional long short-term memory neural network layer and the number of attention mechanism heads of the second bidirectional long short-term memory neural network layer. The hidden vector dimension of the first bidirectional long short-term memory neural network layer is 64, the number of attention mechanism heads is 4, and the hidden vector dimension of the second bidirectional long short-term memory neural network layer is 32, the number of attention mechanism heads is 4.

[0092] S4: Train the state of charge assessment model using normalized energy storage system operation data and energy storage system state of charge data until the assessment accuracy meets the requirements.

[0093] The state of charge (SCC) assessment model is trained using normalized energy storage system operating data and energy storage system SCC data, including the following steps:

[0094] S401: Input the training set into the state of charge (SCC) assessment model and train the SCC model using gradient descent until the assessment accuracy meets the requirements; evaluate the assessment accuracy using root mean square (RMS) error. The accuracy requirement is that the difference between the RMS errors of the SCC model after two consecutive training iterations is less than 0.001. The formula for calculating the RMS error is:

[0095]

[0096] Where RMSE is the root mean square error, n is the total sample size, i is the sample batch, and y is the mean square error. i For the true value, This is a predicted value;

[0097] S402: Input the validation set into the trained state of charge (SOC) assessment model, evaluate the SOC of the energy storage system on the validation set, and calculate its evaluation accuracy. If the evaluation accuracy requirement is not met, readjust the hyperparameters in the SOC assessment model and repeat steps S401-S402 until the evaluation accuracy meets the requirement. Calculate the evaluation accuracy using the mean absolute percentage error (MAE), which must be less than 0.05. The formula for calculating the MAE is:

[0098]

[0099] Where MAPE is the mean absolute percentage error, n is the total sample size, i is the sample batch, and y is the mean absolute percentage error. i For the true value, This is a predicted value;

[0100] S403: Input the test set into the validated state of charge (SOC) assessment model, assess the SOC of the energy storage system in the test set, calculate its assessment accuracy, and if the accuracy requirement is not met, readjust the hyperparameters in the SOC assessment model and repeat steps S401-S403 until the assessment accuracy meets the requirements; calculate the assessment accuracy using the mean absolute percentage error (MAE), with an accuracy requirement of less than 0.05. The formula for calculating the MAE is:

[0101]

[0102] Where MAPE is the mean absolute percentage error, n is the total sample size, i is the sample batch, and y is the mean absolute percentage error. i For the true value, This is a predicted value;

[0103] S5: Use a state of charge prediction model that meets the accuracy requirements to predict the state of charge of the energy storage system. That is, input the battery stack voltage, stack current and stack average temperature at any time into the model to obtain the predicted state of charge value.

[0104] A state-of-charge assessment device for an energy storage system includes a memory and a processor. The memory stores a computer program that, when loaded by the processor, executes the method described above.

[0105] A computer-readable storage medium storing a computer program adapted to perform the method of claim 1 when loaded by a processor.

[0106] The prediction performance of the training set is as follows Figure 3 As shown, the prediction results on the validation set are as follows: Figure 4 As shown, Figure 3 and Figure 4 It can be seen that the above method has high overall accuracy in assessing the state of charge (SOC). However, when the SOC value at the discharge cutoff is small and the SOC value at the charge cutoff is large, the predicted value has a relatively large error compared to the actual value. This is because the actual SOC value is too small or too large. During each charge process, as the SOC value gradually increases, the predicted value gradually approaches the actual value. Figure 3 It can be seen that the predicted value deviates significantly from the actual value during fluctuating charge and discharge. This is because the actual SOC value changes continuously during fluctuating charge and discharge. Therefore, it can be seen that the SOC estimation value of the present invention has a low error under various operating conditions.

[0107] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A method for assessing the state of charge (SOC) of an energy storage system, characterized in that, Includes the following steps: S1: Collect energy storage system operation data and corresponding energy storage system state of charge data; S2: Normalize the energy storage system operation data and energy storage system state of charge data, and divide them into training set, validation set and test set; S3: Construct a state of charge assessment model and initialize the hyperparameters of the state of charge assessment model. The state of charge assessment model includes an input layer, a first bidirectional long short-term memory neural network layer, a second bidirectional long short-term memory neural network layer, a multi-head attention mechanism layer, and an output layer. S4: Train the state of charge assessment model using normalized energy storage system operation data and energy storage system state of charge data until the assessment accuracy meets the requirements. S5: Use a state of charge prediction model that meets the accuracy requirements to predict the state of charge of the energy storage system.

2. The method for assessing the state of charge of an energy storage system according to claim 1, characterized in that, The energy storage system operation data, including battery stack voltage, stack current, and average stack temperature, is used as input data. The collected energy storage system operation data is normalized so that the normalized data are all within the range [0,1]. The normalized energy storage system operation data and the energy storage system state of charge data are then divided into training, validation, and test sets. The normalization expression is as follows: Where X represents the collected energy storage system operation data, and X' represents the normalized energy storage system operation data. Max and X Min These are the maximum and minimum values ​​in the collected energy storage system operation data, respectively.

3. The method for assessing the state of charge of an energy storage system according to claim 1, characterized in that, Both the first and second bidirectional long short-term memory (LSTM) neural network layers are used to predict the state of charge. Each layer contains multiple bidirectional LSTM neural network units, with the number of units equal to the length of the input data time series. Each LSTM unit is composed of a forward-propagating LSTM neural network and a backward-propagating LSTM neural network. The calculation expression for each LSTM unit is as follows: H + =LSTM + (H t-1 ,x t ) H - =LSTM - (H t-1 ,x t ) y t =W + ·H + +W - ·H - +b Among them, H + and H - The outputs of the forward and backward propagation Long Short-Term Memory (LSTM) neural network layers are respectively the outputs of the forward and backward propagation layers. + and LSTM - These are forward-propagating Long Short-Term Memory (LSTM) neural networks and backward-propagating LSTM neural networks, respectively. t-1 x is the output of the hidden layer of the previous unit. t For the current cell input, y t For the final output, W + and W - , respectively, are the weights of the long short-term memory neural network layers that propagate forward and backward, and b is the bias; The expression for the Long Short-Term Memory (LSTM) neural network in the forward-propagating LSM and backward-propagating LSM neural networks is as follows: f t =σ(W f [H t-1 ,x t ]+b f ) i t =σ(W i [H t-1 ,x t ]+b i ) The t =σ(W o [H t-1 ,x t ]+b o ) H t = O t *fish(C) t ) Among them, C t C t-1 , These represent the current cell state, the previous cell state, and the output layer cell state, respectively. t i t O t The outputs of the forget, input, and output structures in gate control are W, respectively. f W i W c W o These are the weight parameters, b f b i b c b o For bias, H t-1 x is the output of the hidden layer of the previous unit. t The input is the current cell, and σ and tanh are both activation functions.

4. The method for assessing the state of charge of an energy storage system according to claim 1, characterized in that, The multi-head attention mechanism layer processing flow includes the following steps: The input dataset X is linearly mapped to a query matrix Q, a key matrix K, and a value matrix V, using the following formula: in, These are weight matrices used to map queries, keys, and values, respectively. By learning the matrix, the multi-head attention mechanism layer adjusts the representation of the input data sequence in different mapping spaces, and the scaling dot product attention mechanism adjusts the network structure. Its expression is: Where, d k Let I be the dimension of matrix K. att (·) is used to calculate the attention weight matrix, softmax represents the normalized exponential function, and T represents the transpose; The output of each bidirectional long short-term memory neural network unit is calculated individually, and then the results are linearly combined to obtain the output value, which is expressed as follows: MultiHead(Q,K,V)=Concat(head1,...,head h )W 0 Where MultiHead is the multi-head attention matrix, Concat is the matrix concatenation operation, and head i Let head be the attention matrix for the i-th head. h Let W be the attention matrix for the h-th head. 0 The output weight matrix is ​​h, where h is the number of self-attention modules selected, i.e., the number of heads.

5. The method for assessing the state of charge of an energy storage system according to claim 2, characterized in that, The state of charge (SCC) assessment model is trained using normalized energy storage system operating data and energy storage system SCC data, including the following steps: S401: Input the training set into the state of charge assessment model and train the state of charge assessment model using the gradient descent method until the assessment accuracy meets the requirements. S402: Input the validation set into the trained state of charge assessment model, evaluate the state of charge of the energy storage system in the validation set, and calculate its evaluation accuracy. If the evaluation accuracy requirement is not met, readjust the hyperparameters in the state of charge assessment model and repeat steps S401-S402 until the evaluation accuracy meets the requirements. S403: Input the test set into the validated state of charge (SOC) assessment model, assess the SOC of the energy storage system in the test set, calculate its assessment accuracy, and if the assessment accuracy requirement is not met, readjust the hyperparameters in the SOC assessment model and repeat steps S401-S403 until the assessment accuracy meets the requirements.

6. The method for assessing the state of charge of an energy storage system according to claim 5, characterized in that, In step S401, the root mean square error (RMSE) is used to evaluate the assessment accuracy. The accuracy requirement is that the difference between the RMSEs of the state-of-charge assessment models after two consecutive training iterations is less than 0.

001. The formula for calculating the RMSE is as follows: Where RMSE is the root mean square error, n is the total sample size, i is the sample batch, and y is the mean square error. i For the true value, These are predicted values.

7. The method for assessing the state of charge of an energy storage system according to claim 5, characterized in that, In steps S402 and S403, the evaluation accuracy is calculated using the mean absolute percentage error (MAE). The accuracy requirement is that the MAE is less than 0.

05. The formula for calculating the MAE is as follows: Where MAPE is the mean absolute percentage error, n is the total sample size, i is the sample batch, and y is the mean absolute percentage error. i For the true value, These are predicted values.

8. The method for assessing the state of charge of an energy storage system according to claim 4, characterized in that, The hyperparameters of the state of charge assessment model include the hidden vector dimension of the first bidirectional long short-term memory neural network layer and the number of attention mechanism heads of the second bidirectional long short-term memory neural network layer. The hidden vector dimension of the first bidirectional long short-term memory neural network layer is 64, and the number of attention mechanism heads is 4. The hidden vector dimension of the second bidirectional long short-term memory neural network layer is 32, and the number of attention mechanism heads is 4.

9. A device for assessing the state of charge of an energy storage system, characterized in that, It includes a memory and a processor, the memory being used to store a computer program, the computer program being used to execute the method of any one of claims 1-8 when loaded by the processor.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that is adapted to execute the method described in any one of claims 1-8 when loaded by a processor.

Images