Battery SOC and SOE joint estimation method based on improved long and short term memory network

Abstract

The invention belongs to the field of battery electric digital data processing, and discloses a battery SOC and SOE joint estimation method based on an improved long-short term memory network, which comprises the following steps: S1, data segmentation, S2, feature extraction, S3, inputting the dimensionality-reduced features obtained in the step S2 into a structure comprising two BiLSTM layers and a full-connection layer, performing sequence modeling by using the BiLSTM layers, and obtaining the SOC and SOE joint estimation of a battery. Based on a BiLSTM model, fitting input discharge time sequence information, and finally taking the obtained full connection layer output as an estimation result of the model to obtain a trained model; and S4, performing model analysis. By using the method, the SOC and SOE of the battery can be predicted by establishing a CNN-BiLSTM model by recording the data change of current, voltage and temperature of the battery in the charging and discharging process through the actual charging and discharging characteristic parameters of the battery under the condition of not knowing the internal construction mechanism and chemical reaction of the battery.

Description

Technical Field

[0001] The invention belongs to the field of battery digital data processing, and in particular relates to a battery SOC and SOE joint estimation method based on an improved long short-term memory network. Background Art

[0002] As the main energy storage component and power source of electric vehicles, lithium batteries directly affect the performance and safety of the entire vehicle. In actual use, batteries have unsafe behaviors such as overcharging and over-discharging, leading to safety accidents such as fire and explosion. Moreover, factors such as long charging time, short battery life, and low estimation accuracy also restrict the development of new energy vehicle technology. In addition, since lithium battery systems have strong coupling and nonlinearity, how to improve the estimation accuracy of BMS (battery management system) for various battery status parameters has become a key issue in improving the safe and stable operation of new energy vehicles.

[0003] SOC (battery state of charge) and SOE (battery remaining energy state) are two important monitoring parameters in BMS. They are two key parameters that describe the battery status. They directly affect the battery's charge and discharge performance and reliability. Accurate estimation of SOC and SOE can optimize the battery's charge and discharge control strategy, improve the battery's service life and efficiency, and is the key to ensuring the efficient and stable operation of the battery system. Therefore, accurate estimation of SOC and SOE is not only the key to ensuring the efficient and stable operation of the battery system, but also an important prerequisite for improving battery service life and user experience.

[0004] At present, the estimation of battery SOC and SOE can be divided into three categories: calculation method, model-based method and data-driven method. The calculation method refers to the use of battery internal parameters such as voltage, current and internal resistance to map the relationship with SOC. The model-based method uses the battery theoretical formula to establish a mathematical model for the battery, and then compares the calculated data of the model with the actual battery observation data. The model parameters are optimized through parameter comparison to achieve the estimation of battery SOC and SOE. The data-driven method does not need to establish a battery principle model, but estimates the battery SOC and SOE by analyzing and processing the actual collected data. Among them, although the calculation method is simple to operate, this method depends on the accuracy of current and voltage sensors, and is easily affected by the cumulative measurement error. Therefore, this method is gradually reduced in the actual SOC and SOE estimation application; the model-based method is one of the current mainstream SOC and SOE estimation methods, with the advantages of high estimation accuracy and little noise influence, but the model established by this method is difficult to change synchronously with the parameters of the actual battery operation, resulting in model distortion and estimation deviation. Data-driven methods have been developed in recent years with the rise of computer technology and artificial intelligence. This method has the advantages of fast estimation speed, high estimation accuracy, and less prior physical knowledge required. However, due to the characteristics of artificial intelligence black box models, data-driven methods require data learning and parameter adjustment, which leads to a long preparation time for this method and makes it difficult to meet real-time requirements. Therefore, a new battery SOC and SOE estimation method is needed. Summary of the invention

[0005] The present invention provides a battery SOC and SOE joint estimation method based on an improved long short-term memory network. The method can use the actual charge and discharge characteristic parameters of the battery, without understanding the internal structure mechanism and chemical reaction of the battery, and by recording the data changes of the current, voltage and temperature of the battery during the charge and discharge process, a CNN-BiLSTM model is established to predict the SOC and SOE of the battery.

[0006] To achieve the above object, the present invention provides the following technical solution: a battery SOC and SOE joint estimation method based on an improved long short-term memory network, the joint estimation method comprising the following steps: S1: Data segmentation: Use sliding window technology to cut the data to form "multi-time domain" input data, and divide the data set into training set and test set; S2: feature extraction, normalizing the data set obtained in step S1, inputting the normalized data into the CNN model, extracting features from the time series data to reduce the dimension of the time series and obtain features after dimensionality reduction; S3: Input the reduced-dimensional features obtained in step S2 into a structure including two BiLSTM layers and one fully connected layer, use the BiLSTM layer to perform sequence modeling, and fit the input discharge time series information based on the BiLSTM model. Finally, use the output of the fully connected layer as the estimation result of the model to obtain a trained model. S4: Model analysis: input the test set in step S1 into the trained model obtained in step S3, and evaluate it using the three indicators of RMSE, MAE and R2.

[0007] Preferably, the specific method of step S1 is: using sliding window technology to divide the continuous charge and discharge data sequence into subsequences of fixed length, and then connecting these subsequences to form a complete time series data as model input; that is, the input variable corresponding to the output value of the battery state at the kth time point is the (k-n+1) input variable between this time point, so that the timing information of the data can be retained and the correlation between historical data and current data can be fully utilized.

[0008] Preferably, the specific method of step S2 is: S2.1: In the CNN convolutional network model, data first enters the convolution layer. During the convolution calculation process, according to the preset step size, the convolution kernel slides from left to right and from top to bottom on the input data to complete the traversal and calculate the data. Finally, the results of multiple convolution calculations are combined to obtain the output of the entire convolution layer. Usually, a convolution layer contains multiple convolution kernels for convolution calculation of the input data. The calculation formula is: ; In the formula, is the output of the convolutional layer. feature maps; is the Sigmoid activation function; is the convolutional layer The weight matrix of the convolution kernel; is the convolution operation; is the convolutional layer The bias of the convolution kernel; For the convolution kernels; Is the input parameter matrix; S2.2: After passing through the convolution layer, the data enters the ReLU function, which is: ; In the formula, Input for activation function; S2.3: The pooling layer uses average pooling, and uses the pooling layer to downsample the data, compress the data and reduce the data dimension, thereby improving the quality of the model features.

[0009] Preferably, the specific method of step S3 is: S3.1: First, the forget gate receives the state of the neuron hidden layer from the previous moment. and the current input , then, by Determine the degree of retention of the input information at the previous moment to reduce the amount of data during model training. The value depends on The current input and Output at any time , which operates according to the following formula: ; in, is the forget gate at time step Output: is the Sigmoid activation function; is the weight matrix of the forget gate; It is the hidden state of the previous moment. and the current input The resulting vector; is the bias of the forget gate; S3.2: Input Gate It mainly determines when to update the cell state and the degree to which the input information is updated. The calculation process of this module consists of two parts. First, the input gate Decide which part of the information needs to be updated; secondly, the hidden layer state and the current input are processed by the Tanh function, and a value between -1 and 1 is output; at the same time, the output value is multiplied by the Sigmoid function to obtain the latest cell state , which operates according to the following formula: ; ; ; in, is the input gate, is the candidate value of the cell state; For the new cell state; is the Sigmoid activation function; is the bitangent function, It is the hidden state of the previous moment. and the current input Get the vector, is the coefficient matrix of the input gate, is the coefficient matrix of candidate cell state values, is the coefficient matrix bias of the input gate, The coefficient matrix bias of the candidate value input gate, is the forget gate at time step Output: S3.3: The output gate determines the output of the current step of the LSTM network. The output gate generates an output vector between 0 and 1 through a Sigmoid activation function based on the state of the memory unit at the current time step, also known as the cell state, and the input of the current time step, to indicate which memory unit information should be output to the output of the current time step. This output vector will be multiplied by the state of the memory unit to generate the final output. The specific technical formula is as follows: ; ; In the formula, is the output of the output gate and the output of the current time step; It is the hidden state of the previous moment. and the current input Get the vector; is the hidden state at the current moment; is the Sigmoid activation function; is the weight matrix of the output gate; For the new cell state; is the bias of the output gate; is the bitangent function.

[0010] Preferably, the specific method of step 4 is: RMSE represents the difference between the actual value and the estimated value of SOE, and is used to measure the discreteness of the sample. When measuring the estimation accuracy of the model, the smaller the RMSE, the higher the estimation accuracy of the model. The RMSE value is calculated by the following formula: ; MAE is an indicator that measures the difference between the actual value and the estimated value of the model. It is used to evaluate the degree of fit of the model on given data. The MAE value is calculated using the following formula: ; R 2 It is an evaluation index to measure the goodness of fit of the model. Its value range is between 0 and 1. The closer it is to 1, the better the estimation performance of the model. R is calculated by the following formula 2 value: ; in, Indicates The actual value of the actual output variable, Indicates The estimated value of the output variable, represents the mean, express Length.

[0011] Compared with the prior art, the present invention has the following beneficial effects: (1) The improved long short-term memory network used in the present invention can efficiently retain effective information of the data and forget redundant information to a certain extent, thus avoiding problems such as the network's inability to remember long-term data, gradient disappearance, and gradient explosion; (2) The present invention processes the time series data through the CNN network before the data enters the BiLSTM model, which can better highlight the sample characteristics and balance the relationship between the model convergence speed and accuracy; Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0013] Figure 1 This is a CNN network structure diagram used in the present invention; Figure 2 This is a diagram of the LSTM network structure used in the present invention; Figure 3 This is a BiLSTM network structure diagram used in the present invention; Figure 4 This is a flow chart of CNN-BiLSTM network model estimation used in the present invention; Figure 5 This is a SOC estimation result diagram used in the present invention; Figure 6 is the SOC estimation error map used in the present invention; Figure 7 This is a graph of SOE estimation results used in the present invention; Figure 8 This is the SOE estimation error diagram used in the present invention. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] Unless otherwise specifically stated, the relative arrangement of the parts and steps described in these embodiments, numerical expressions and numerical values ​​do not limit the scope of the present invention. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the accompanying drawings are not drawn according to the actual proportional relationship. The technology, method and equipment known to ordinary technicians in the relevant field may not be discussed in detail, but in appropriate cases, the technology, method and equipment should be regarded as a part of the authorization specification. In all examples shown and discussed here, any specific value should be interpreted as being merely exemplary, rather than as a limitation. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once a certain item is defined in an accompanying drawing, it does not need to be further discussed in subsequent drawings.

[0016] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used here to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptions used here are interpreted accordingly.

[0017] See also Figure 1-Figure 8 , the present invention provides a technical solution: see Figure 4 The network model estimation flow chart of the present invention, the neural network model of convolution and bidirectional long short-term memory network, namely the CNN-BiLSTM model, the present invention estimates the battery network model according to the following steps: Step S1: Data segmentation, input vector , the sliding window technology is used to cut the data to form "multi-time domain" input data, so as to better mine the discharge history data and improve the correlation between input and output data. The data set is divided into training set and test set. The model is trained and tested on the training set. Finally, the test set is used to verify the model estimation performance.

[0018] Specifically, in step S1, the sliding window technology is used to cut the continuous charge and discharge data sequence into subsequences of fixed length, and then these subsequences are connected to form a complete time series data as the model input; that is, the input variable corresponding to the output value of the battery state at the kth time point is the (k-n+1) input variable between this time point, so that the time series information of the data can be retained and the correlation between historical data and current data can be fully utilized. Figure 4 Where m is the length of the sliding window, k is the starting position of the current sliding window, k+m+1 is the end position of the sliding window, SOC / SOE m is the battery state of charge (SOC) and state of energy (SOE) of the mth sliding window, SOC / SOE m+n is the battery state of charge (SOC) and state of energy (SOE) of the m+nth sliding window, SOC / SOE k+m-1 is the battery state of charge (SOC) and state of energy (SOE) of the k+m-1th sliding window.

[0019] Step S2: Feature extraction, normalization of data, see Figure 4 Normalization formula, where is the normalized value; Is the input parameter matrix; is the minimum value in the initial data; is the maximum value in the initial data. Input the normalized data into the CNN model, see Figure 1 , feature extraction is performed on time series data to reduce the dimension of the time series. The preprocessed battery physical parameters are input into the CNN structure consisting of convolution layer, ReLU activation function, pooling layer and flat layer. The convolution layer extracts the features of the input variables through the convolution kernel, the ReLU function implements nonlinear mapping, and the pooling layer downsamples the extracted features to reduce the dimensionality.

[0020] Specifically, step S2.1: In the CNN convolutional network model, the data first enters the convolution layer. During the convolution calculation process, according to the preset step size, the convolution kernel slides from left to right and from top to bottom on the input data to complete the traversal and calculate the data. Finally, the results of multiple convolution calculations are combined to obtain the output of the entire convolution layer. Usually, a convolution layer contains multiple convolution kernels for convolution calculation of the input data. The calculation formula is: ; In the formula, is the output of the convolutional layer. feature maps; is the Sigmoid activation function; is the convolutional layer The weight matrix of the convolution kernel; is the convolution operation; is the convolutional layer The bias of the convolution kernel; For the convolution kernels; Is the input parameter matrix; Then according to the formula Calculate and output the result C, where c i is the result output by the ith network.

[0021] Step S2.2: After passing through the convolution layer, the data enters the ReLU function, and the ReLU function is: ; In the formula, Input for activation function; Step S2.3: The pooling layer uses average pooling, and the pooling layer is used to downsample the data, compress the data and reduce the data dimension, thereby improving the quality of the model features.

[0022] Step S3: See Figure 2 and Figure 3 Bidirectional long short-term memory neural network (BiLSTM) is a recursive neural network model, which is improved based on the traditional long short-term memory neural network (LSTM). This model is widely used in long time series data. The reduced-dimensional features obtained in step S2 are input into a structure containing two BiLSTM layers and one fully connected layer. The BiLSTM layer is used for sequence modeling, and the input discharge time series information is fitted based on the BiLSTM model. Finally, the fully connected layer output is used as the estimation result of the model to obtain a trained model.

[0023] Specifically, step S3.1: First, Figure 2 As shown, the forget gate receives the state of the hidden layer of neurons from the previous moment. and the current input , and input the cell state C of the previous moment t-1 , then, by Determine the degree of retention of the input information at the previous moment to reduce the amount of data during model training. The value depends on The current input and Output at any time , which operates according to the following formula: ; in, is the forget gate at time step Output: is the Sigmoid activation function; is the weight matrix of the forget gate; It is the hidden state of the previous moment. and the current input The resulting vector; is the bias of the forget gate; Step S3.2: Input Gate It mainly determines when to update the cell state and the degree to which the input information is updated. The calculation process of this module consists of two parts. First, the input gate Decide which part of the information needs to be updated; secondly, the hidden layer state and the current input are processed by the Tanh function, and a value between -1 and 1 is output; at the same time, the output value is multiplied by the Sigmoid function to obtain the latest cell state , which operates according to the following formula: ; ; ; in, is the input gate, is the candidate value of the cell state; For the new cell state; is the Sigmoid activation function; is the bitangent function, It is the hidden state of the previous moment. and the current input Get the vector, is the coefficient matrix of the input gate, is the coefficient matrix of candidate cell state values, is the coefficient matrix bias of the input gate, The coefficient matrix bias of the candidate value input gate, is the forget gate at time step Output: Step S3.3: Figure 3As shown in the figure, the output gate determines the output of the current step of the LSTM network. The output gate generates an output vector between 0 and 1 through a Sigmoid activation function based on the state of the memory unit at the current time step, also called the cell state, and the input of the current time step, to indicate which memory unit information should be output to the output of the current time step. This output vector will be multiplied by the state of the memory unit to generate the final output. The specific technical formula is as follows: ; ; In the formula, is the output of the output gate and the output of the current time step; It is the hidden state of the previous moment. and the current input Get the vector; is the hidden state at the current moment; is the Sigmoid activation function; is the weight matrix of the output gate; For the new cell state; is the bias of the output gate; is the bitangent function.

[0024] Step S4: Model analysis: input the test set in step S1 into the trained model obtained in step S3, and evaluate it using the three indicators of RMSE, MAE and R2.

[0025] Specifically, RMSE represents the difference between the actual value and the estimated value of SOE, and is used to measure the discreteness of the sample. When measuring the estimation accuracy of the model, the smaller the RMSE, the higher the estimation accuracy of the model. The RMSE value is calculated by the following formula: ; MAE is an indicator that measures the difference between the actual value and the estimated value of the model. It is used to evaluate the degree of fit of the model on given data. The MAE value is calculated using the following formula:

[0026] R 2 It is an evaluation index to measure the goodness of fit of the model. Its value range is between 0 and 1. The closer it is to 1, the better the estimation performance of the model. R is calculated by the following formula 2 value: ; in, Indicates The actual value of the actual output variable, Indicates The estimated value of the output variable, represents the mean, express Length.

[0027] Example: Taking into account the diversity of power battery operation scenarios, it is necessary to verify the stability and robustness of the proposed model. This embodiment uses the UDDS operating condition, HWFET operating condition and LA92 operating condition of the 18650PF battery data set under a wide ambient temperature (0°C, 10°C and 25°C) for cross-validation, that is, one operating condition at one temperature is used as the test set each time, and the remaining 8 data sets are used as training data and verification data. Usually 85% of the data is used as training data and 15% of the data is used as verification data.

[0028] The obtained battery data is processed. Since the battery data may be abnormal due to insufficient charge and discharge cycles, power outages, etc. when collecting data, the data needs to be cleaned.

[0029] The processed lithium-ion battery data is input into the long short-term memory network to predict the corresponding lithium-ion battery SOC and SOE data change trends and train the long short-term memory network.

[0030] Taking the UDDS working condition as an example, the SOC and SOE estimation result diagram and estimation error diagram of the CNN-BiLSTM model based on the entire discharge cycle of the UDDS working condition at 25°C are shown in the figure below: Figure 5-8 As shown. Under 25℃ conditions, the MAE of SOC estimation is 0.16%, and the RMSE is 0.21%; the MAE of SOE estimation is 0.15%, and the RMSE is 0.20%. Similarly, under 10℃ conditions, the MAE of SOC estimation is 0.26%, and the RMSE is 0.37%; the MAE of SOE estimation is 0.33%, and the RMSE is 0.41%. Under 0℃ conditions, the MAE of SOC estimation is 0.29%, and the RMSE is 0.38%; the MAE of SOE estimation is 0.28%, and the RMSE is 0.33%. List 1 is as follows: Table 1 SOC and SOE estimation results under different temperature conditions

[0031] After data cleaning, the maximum mean difference is used to calculate the distribution distance between the predicted data and the target data. If the value is small, only the original network needs to be fine-tuned. If it is large, the model structure needs to be adjusted, some layers need to be added or replaced, and the source data needs to be retrained to obtain the final network. The experimental results of real data show that the model of the present invention can effectively improve the efficiency of network learning, with high accuracy and efficiency.

[0032] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A joint estimation method of battery SOC and SOE based on improved long short-term memory network, Features: The joint estimation method comprises the following steps: S1: Data segmentation: Use sliding window technology to cut the data to form "multi-time domain" input data, and divide the data set into training set and test set; S2: feature extraction, normalizing the data set obtained in step S1, inputting the normalized data into the CNN model, extracting features from the time series data to reduce the dimension of the time series and obtain features after dimensionality reduction; S3: Input the reduced-dimensional features obtained in step S2 into a structure including two BiLSTM layers and one fully connected layer, use the BiLSTM layer to perform sequence modeling, and fit the input discharge time series information based on the BiLSTM model. Finally, use the output of the fully connected layer as the estimation result of the model to obtain a trained model. S4: Model analysis: input the test set in step S1 into the trained model obtained in step S3, and evaluate it using the three indicators of RMSE, MAE and R2.

2. According to claim 1, a battery SOC and SOE joint estimation method based on improved long short-term memory network, Features: The specific method of step S1 is: using the sliding window technology to divide the continuous charge and discharge data sequence into subsequences of fixed length, and then connecting these subsequences to form a complete time series data as the model input; that is, the input variable corresponding to the output value of the battery state at the kth time point is the (k-n+1) input variable between this time point, so that the time series information of the data can be retained and the correlation between the historical data and the current data can be fully utilized.

3. According to claim 1, a battery SOC and SOE joint estimation method based on improved long short-term memory network, Features: The specific method of step S2 is: S2.1: In the CNN convolutional network model, data first enters the convolution layer. During the convolution calculation process, according to the preset step size, the convolution kernel slides from left to right and from top to bottom on the input data to complete the traversal and calculate the data. Finally, the results of multiple convolution calculations are combined to obtain the output of the entire convolution layer. Usually, a convolution layer contains multiple convolution kernels for convolution calculation of the input data. The calculation formula is: ； In the formula, is the output of the convolutional layer. feature maps; is the Sigmoid activation function; is the convolutional layer The weight matrix of the convolution kernel; is the convolution operation; is the convolutional layer The bias of the convolution kernel; For the convolution kernels; Is the input parameter matrix; S2.2: After passing through the convolution layer, the data enters the ReLU function, which is: ； In the formula, Input for activation function; S2.3: The pooling layer uses average pooling, and uses the pooling layer to downsample the data, compress the data and reduce the data dimension, thereby improving the quality of the model features.

4. According to claim 1, a battery SOC and SOE joint estimation method based on improved long short-term memory network, Features: The specific method of step S3 is: S3.1: First, the forget gate receives the state of the neuron hidden layer from the previous moment. and the current input , then, by Determine the degree of retention of the input information at the previous moment to reduce the amount of data during model training. The value depends on The current input and Output at any time , which operates according to the following formula: ； in, is the forget gate at time step Output: is the Sigmoid activation function; is the weight matrix of the forget gate; It is the hidden state of the previous moment. and the current input The resulting vector; is the bias of the forget gate; S3.2: Input Gate Mainly to determine when to update the cell state, and the extent to which the input information is updated, The calculation process of consists of two parts. First, the input gate Decide which part of the information needs to be updated; secondly, the hidden layer state and the current input are processed by the Tanh function, and a value between -1 and 1 is output; at the same time, the output value is multiplied by the Sigmoid function to obtain the latest cell state , which operates according to the following formula: ； ； ； in, is the input gate, is the candidate value of the cell state; For the new cell state; is the Sigmoid activation function; is the bitangent function, It is the hidden state of the previous moment. and the current input Get the vector, is the coefficient matrix of the input gate, is the coefficient matrix of candidate cell state values, is the coefficient matrix bias of the input gate, The coefficient matrix bias of the candidate value input gate, is the forget gate at time step Output: S3.3: The output gate determines the output of the current step of the LSTM network. The output gate generates an output vector between 0 and 1 through a Sigmoid activation function based on the state of the memory unit at the current time step, also known as the cell state, and the input of the current time step, to indicate which memory unit information should be output to the output of the current time step. This output vector will be multiplied by the state of the memory unit to generate the final output. The specific technical formula is as follows: ； ； In the formula, is the output of the output gate and the output of the current time step; It is the hidden state of the previous moment. and the current input Get the vector; is the hidden state at the current moment; is the Sigmoid activation function; is the weight matrix of the output gate; For the new cell state; is the bias of the output gate; is the bitangent function.

5. According to claim 1, a battery SOC and SOE joint estimation method based on improved long short-term memory network, Features: The specific method of step 4 is: RMSE represents the difference between the actual value and the estimated value of SOE, and is used to measure the discreteness of the sample. When measuring the estimation accuracy of the model, the smaller the RMSE, the higher the estimation accuracy of the model. The RMSE value is calculated by the following formula: ； MAE is an indicator that measures the difference between the actual value and the estimated value of the model. It is used to evaluate the degree of fit of the model on given data. The MAE value is calculated using the following formula: ； R 2 It is an evaluation index to measure the goodness of fit of the model. Its value range is between 0 and 1. The closer it is to 1, the better the estimation performance of the model. R is calculated by the following formula 2 value: ； in, Indicates The actual value of the actual output variable, Indicates The estimated value of the output variable, represents the mean, express Length.

Images