A data-driven based temperature adaptive SOC estimation method and apparatus

Abstract

The application discloses a data-driven temperature self-adaptive SOC estimation method and device, and relates to the field of electric vehicle SOC estimation. The method comprises the following steps: acquiring target power battery working condition data; performing feature extraction on the target power battery working condition data to obtain low-temperature SOC features; inputting the low-temperature SOC features into a trained low-temperature SOC estimation model to obtain a low-temperature SOC estimation result; inputting the target power battery working condition data into a trained normal-temperature SOC estimation model to obtain a normal-temperature SOC estimation result; and determining a final SOC estimation result according to the low-temperature SOC estimation result and the normal-temperature SOC estimation result, so that high-precision lithium battery SOC estimation is realized.

Description

Technical Field

[0001] This application relates to the field of electric vehicle SOC estimation, and in particular to a data-driven temperature-adaptive SOC estimation method and apparatus. Background Technology

[0002] In recent years, electric vehicles have received widespread attention due to their significant advantages in technology, environmental protection, and energy conservation. Lithium-ion batteries, as the power source for electric vehicles, are widely used due to their superior performance, including high energy density, high voltage, and long lifespan. The state of charge (SOC) of a lithium battery reflects its remaining capacity and is one of the important evaluation indicators for battery packs. Accurate and stable SOC estimation is crucial for extending battery life and ensuring the safe operation of electric vehicles. However, due to the influence of temperature, unknown noise, and uncertain outliers, accurate SOC estimation of lithium-ion batteries is often difficult. Currently, SOC estimation for lithium-ion batteries mainly falls into three categories: ampere-hour integration method, model-based methods, and data-driven methods.

[0003] The ampere-hour integration method monitors the battery's real-time SOC based on the initial SOC by integrating the current over time. It is easy to implement and is commonly used in automotive BMS systems to estimate battery SOC. Model-based methods primarily estimate SOC by establishing battery models such as equivalent circuit models, fractional-order models, and electrochemical models, identifying parameters, and finally using observation algorithms such as Kalman filters. Model-based methods are closed-loop systems and can reduce uncertainty interference through self-calibration. Data-driven methods establish machine learning or deep learning models, inputting measurable battery signals into the model for training, and using the trained model to estimate SOC. This eliminates the need for complex battery models and is often used in practice for cloud-based SOC estimation based on big data.

[0004] The drawbacks of the ampere-hour integration method are that, as an open-loop method, its estimation results are easily affected by the initial SOC offset and cumulative error, requiring continuous correction using other methods. The drawbacks of model-based methods are that building accurate battery models is usually quite complex, and the model parameters of a single model vary greatly under different temperatures and material systems, making it difficult to guarantee estimation accuracy under different operating environments. The drawbacks of data-driven methods are that they rely entirely on data, lack interpretability, and the estimation accuracy is highly dependent on data quality and quantity, and they also have generalization problems under different operating conditions and temperatures. Summary of the Invention

[0005] The purpose of this application is to provide a data-driven temperature-adaptive SOC estimation method and device, which can achieve high-precision lithium battery SOC estimation. It solves the problem that when performing SOC estimation in practical applications, the temperature difference under different operating environments will cause changes in the battery's internal parameters and aggravate side reactions (changing the electrode reaction rate and causing material phase transitions), which will lead to data distribution distortion. Therefore, it is difficult to guarantee the accuracy of SOC estimation at low temperatures when using data-driven methods.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a data-driven temperature-adaptive SOC estimation method, including: Acquire target power battery operating condition data; Feature extraction is performed on the target power battery operating condition data to obtain low-temperature SOC features; The low-temperature SOC features are input into the trained low-temperature SOC estimation model to obtain the low-temperature SOC estimation result; The target power battery operating condition data is input into the trained room temperature SOC estimation model to obtain the room temperature SOC estimation result; The final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result.

[0007] Optionally, the data-driven temperature-adaptive SOC estimation method further includes: The final SOC estimation result is filtered using a noise adaptive square root unscented Kalman filter to obtain the filtered SOC estimation result.

[0008] Optionally, the target power battery operating condition data includes voltage, current, and temperature.

[0009] Optionally, feature extraction is performed on the target power battery operating condition data to obtain low-temperature SOC features, specifically including: Based on the Arrhenius equation, the current in the target power battery operating condition data is normalized to obtain the normalized current. The voltage-temperature differential feature is obtained by differentiating the voltage and temperature in the target power battery operating condition data; the normalized current and voltage-temperature differential features constitute the low-temperature SOC feature.

[0010] Optionally, the room-temperature SOC estimation model includes a first feature extraction module, a BiGRU model, and an attention mechanism module; The first feature extraction module is used to extract features from the target power battery operating condition data to obtain power battery features; The BiGRU model is used to extract time-series features from power battery characteristics; The attention mechanism module is used to perform weighted fusion processing on temporal features to obtain room temperature SOC estimation results.

[0011] Optionally, the low-temperature SOC estimation model includes a second feature module and a Transformer model; The second feature module is used to extract features from the low-temperature SOC features to obtain low-temperature intermediate features; The Transformer model is used to extract time-series features from intermediate features at low temperatures to obtain the low-temperature SOC estimation results.

[0012] Optionally, the final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result, specifically including: The final SOC estimation result is obtained by weighted summing of the low-temperature SOC estimation result and the room-temperature SOC estimation result. Alternatively, the low-temperature SOC estimation result and the room-temperature SOC estimation result can be input into the trained meta-model to obtain the final SOC estimation result; the trained meta-model is a model trained with the low-temperature SOC estimation result and the room-temperature SOC estimation result corresponding to the sample power battery operating condition data as input, and the true SOC value corresponding to the sample power battery operating condition data as the label.

[0013] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the above-described data-driven temperature adaptive SOC estimation method.

[0014] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described data-driven temperature adaptive SOC estimation method.

[0015] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described data-driven temperature adaptive SOC estimation method.

[0016] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a data-driven, temperature-adaptive SOC estimation method and apparatus. The method involves acquiring target power battery operating condition data; extracting features from the target power battery operating condition data to obtain low-temperature SOC features; inputting the low-temperature SOC features into a trained low-temperature SOC estimation model to obtain a low-temperature SOC estimation result; inputting the target power battery operating condition data into a trained room-temperature SOC estimation model to obtain a room-temperature SOC estimation result; and determining the final SOC estimation result based on the low-temperature and room-temperature SOC estimation results. By fusing the low-temperature and room-temperature SOC estimation results to determine the final SOC estimation result, temperature-adaptive SOC estimation is achieved at different temperatures, while maintaining high accuracy in SOC estimation at both low and room temperatures. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an application environment diagram of a data-driven temperature adaptive SOC estimation method according to an embodiment of this application.

[0019] Figure 2 This is a flowchart illustrating a data-driven temperature adaptive SOC estimation method provided in one embodiment of this application.

[0020] Figure 3 This is a schematic diagram illustrating the specific process of a data-driven temperature adaptive SOC estimation method provided in an embodiment of this application.

[0021] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] The data-driven temperature adaptive SOC estimation method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send target power battery operating condition data to server 104. After receiving the target power battery operating condition data, server 104 performs feature extraction on the target power battery operating condition data to obtain low-temperature SOC features. These low-temperature SOC features are then input into a trained low-temperature SOC estimation model to obtain a low-temperature SOC estimation result. The target power battery operating condition data is then input into a trained room-temperature SOC estimation model to obtain a room-temperature SOC estimation result. Based on the low-temperature and room-temperature SOC estimation results, a final SOC estimation result is determined. Server 104 can then feed back the final SOC estimation result to terminal 102. In addition, in some embodiments, the data-driven temperature-adaptive SOC estimation method can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly perform SOC estimation based on the target power battery operating condition data, or the server 104 can obtain the target power battery operating condition data from the data storage system and perform SOC estimation based on the target power battery operating condition data.

[0025] The terminal 102 can be, but is not limited to, various desktop computers and laptops. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0026] In one exemplary embodiment, such as Figure 2 As shown, a data-driven temperature-adaptive SOC estimation method is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 205.

[0027] Step 201: Obtain the target power battery operating condition data.

[0028] Step 202: Extract features from the target power battery operating condition data to obtain low-temperature SOC features.

[0029] Step 203: Input the low-temperature SOC features into the trained low-temperature SOC estimation model to obtain the low-temperature SOC estimation result; Step 204: Input the target power battery operating condition data into the trained room temperature SOC estimation model to obtain the room temperature SOC estimation result. Step 205: Determine the final SOC estimation result based on the low-temperature SOC estimation result and the room-temperature SOC estimation result.

[0030] Implementing steps 201 to 205 addresses the problem that temperature differences under different operating environments can lead to changes in internal battery parameters and exacerbate side reactions (altering electrode reaction rates and causing material phase transitions) during SOC estimation in practical applications. This results in distorted data distribution and makes it difficult to guarantee SOC estimation accuracy at low temperatures when using data-driven methods. This method is characterized by strong practicality, high estimation accuracy, and robustness. It can be used for SOC estimation of power batteries in new energy vehicles and has diverse application prospects in electric vehicles. This method can achieve high-accuracy SOC estimation at both low and room temperature.

[0031] like Figure 3 As shown, the data-driven temperature adaptive SOC estimation method provided in this application mainly consists of the following four steps: data preprocessing, feature extraction, SOC estimation model construction, and temperature adaptive SOC estimation based on model fusion.

[0032] First, dynamic operating condition data of the power battery is acquired through various channels, including the vehicle battery management system, cloud-based battery monitoring platform, manufacturer-provided data, and public datasets. This dynamic operating condition data includes current, voltage, temperature, and actual SOC data. Next, the collected current, voltage, and temperature data undergo data cleaning (i.e., data preprocessing). Then, two data-driven SOC estimation models are constructed, with different model structures and input features, suitable for high-precision SOC estimation at normal and low temperatures, respectively. The normal temperature SOC estimation model directly estimates the SOC using the current, voltage, and temperature data from the power battery dynamic operating condition data. The low-temperature SOC estimation model extracts features from the current, voltage, and temperature signals collected from the power battery dynamic operating condition data, and then uses the extracted features as input to the low-temperature SOC estimation model. During training, hyperparameter optimization techniques are used to optimize the training process, and gradient descent algorithm is used to train the network. Finally, through model fusion, the two SOC estimation models are fused based on the current actual temperature to achieve temperature-adaptive SOC estimation at different temperatures. The estimation results are further processed through an online filter to obtain the final high-precision estimation result.

[0033] Step 1: Data preprocessing.

[0034] Taking the driving data of the A123 18650 lithium iron phosphate battery as an example, the dynamic operating condition data of the power battery is first separated into operating conditions. The static data and constant current-constant voltage charging data from the experiment are removed, and the dynamic operating condition data at different temperatures are separated into independent data files to facilitate subsequent model training and validation. The test data includes data at -25℃, -10℃, 0℃, 10℃, and 25℃, with each temperature containing three operating conditions: Dynamic Stress Test (DST), US06 Driving Plan, and Federal City Driving Plan (FUDS). These three operating conditions differ significantly. To fully verify the generalization ability of the proposed method, the DST operating condition is selected as the training set, and the US06 and FUDS operating conditions are selected as the test sets to verify the network's generalization. For SOC estimation, due to the complex and variable dynamic operating conditions and extremely high real-time requirements in practical applications, the data is not preprocessed extensively; feature extraction and model development are directly performed based on the original current, voltage, and temperature data.

[0035] Step 2: Feature extraction.

[0036] For the low-temperature SOC estimation model, the input data of the model is obtained by feature extraction based on voltage, current and temperature signals. Two features are extracted based on the current, voltage and temperature data, including the current data after Arrhenius current normalization (normalized current) and the voltage differential data with respect to temperature (voltage-temperature differential feature). These two features are used as low-temperature SOC features.

[0037] Arrhenius current normalization, based on the Arrhenius equation, equates the current at any temperature to the current that produces the same ion flux at 25°C, thereby eliminating the influence of low-temperature environments on electrochemical reaction rates and effectively characterizing the actual dynamic properties of batteries at low temperatures. The Arrhenius equation can generally be described as: (1); in, and It is the concentration-based reaction rate constant; and This is absolute temperature, measured in Kelvin (K). Activation energy; is the molar gas constant.

[0038] Based on the Arrhenius equation, the current is normalized: (2); in, The normalized current, To measure current, This is the actual temperature. For reference temperature, room temperature (25℃) is usually selected.

[0039] The voltage-temperature differential characteristic captures the amplified entropic and polarization effects at low temperatures, enabling the low-temperature SOC estimation model to accurately capture battery state information at low temperatures. Entropic and polarization effects are significantly amplified at low temperatures, and are difficult to reflect through measurable macroscopic signals at room and high temperatures. Therefore, the voltage-temperature differential characteristic is a unique feature of low-temperature environments and can effectively reflect the dynamic characteristics of lithium batteries at low temperatures. The voltage-temperature differential characteristic is calculated as follows: (3); in, It is a voltage-temperature differential characteristic. and These represent voltage and temperature, respectively.

[0040] The input features of the final low-temperature SOC estimation model include normalized current (Arrhenius current normalization feature) and voltage-temperature differential feature.

[0041] For the room temperature SOC estimation model, the entropy-heat effect and polarization effect are far less significant at room temperature than at low temperature, especially the entropy-heat effect. If voltage-temperature differential features are extracted at room temperature, the feature signal-to-noise ratio will be too low due to the insignificant entropy-heat effect, which will lead to noise overwhelming the features. At room temperature, the battery dynamic characteristics are relatively stable, the dynamic response is fast and the polarization is weak. Therefore, the original data of current, voltage and temperature are directly used as the input of the room temperature SOC estimation model.

[0042] In step 202, feature extraction is performed on the target power battery operating condition data to obtain low-temperature SOC features. Specifically, this includes: normalizing the current in the target power battery operating condition data based on the Arrhenius equation to obtain the normalized current; differentiating the voltage in the target power battery operating condition data with respect to temperature to obtain voltage-temperature differential features; the normalized current and voltage-temperature differential features constitute the low-temperature SOC features.

[0043] Step 3: SOC estimation model construction.

[0044] 1. The room temperature SOC estimation model is constructed using a deep learning model. At room temperature, the battery has a high ion migration rate, low internal resistance, short polarization voltage decay time constant, and an approximately linear voltage-SOC relationship. Therefore, based on the characteristics of fast dynamic response and weak polarization at room temperature, it is relatively easy to achieve high-precision SOC estimation. The lightweight and generalization of the model are the main goals of the room temperature SOC estimation model. The model is constructed with a convolutional neural network (CNN)-bidirectional gated recurrent unit (BiGRU)-attention mechanism (AM) model (CNN-BiGRU-AM) as the core.

[0045] CNNs possess powerful feature extraction capabilities, automatically extracting features to address the complex and dynamic conditions and high real-time requirements often encountered in SOC estimation. CNNs feature local connectivity and weight sharing, significantly reducing the number of model parameters, accelerating training, and improving generalization performance. In ambient temperature scenarios with weak polarization effects, CNNs can efficiently capture local current-voltage dynamics. One-dimensional CNNs are primarily used for time series problems and consist of three layers. The first feature extraction module includes an input layer, convolutional layers, max-pooling layers, and an output layer. The input layer transmits raw data to the network. Input data passes through the input layer and then undergoes convolutional computation in the convolutional layers. The convolutional layers are the core of the CNN, extracting local features from higher-level inputs and passing all information to lower layers to obtain more complex and abstract features. (4); Among them, for the first One input and one output, Given the input sequence, For convolution kernel weights, The kernel size is [size]. For bias terms, This is the output of the convolutional layer.

[0046] The output of the convolutional layer is then non-linearly activated by an activation function; this application selects the ReLU function as the activation function. (5); in, The input to the ReLU function, i.e., the output of the convolutional layer. ; These are the features after processing by the activation function.

[0047] Then, a max pooling layer is added after the convolutional layer to reduce the dimensionality of the data, thereby further performing feature extraction and reducing the computational burden of the model: assuming the input of the pooling layer (the input of the pooling layer is the feature processed by the activation function) has a length of... The sequence, with a pooling window size of First, zero values ​​are added to both ends of the input sequence for padding. After padding, the length of the input sequence becomes... The length of the output sequence after pooling is: (6); in, The length of the output sequence after pooling. For adjustable fill amount parameters, This is the stride size of the pooling window. This indicates rounding down to the nearest integer.

[0048] The first output sequence element It is determined by the maximum value within the corresponding window in the input sequence, as shown in the formula: (7); Finally, the pooled output sequence is passed through the output layer and output to the BiGRU model. The output layer is usually composed of fully connected layers. (8); in, The output of the first feature extraction module is the power battery feature. For activation functions, the Sigmoid function is typically chosen. and These are the weights and biases of the fully connected layer, respectively.

[0049] The CNN is followed by a BiGRU model to capture temporal relationships. The GRU model is a deep learning model suitable for solving sequence problems; it's a variant of the Recurrent Neural Network (RNN). By introducing a gating mechanism, it can selectively control the updating and resetting of information, allowing the network to learn long-term dependencies more effectively and avoiding gradient vanishing and exploding problems. Simultaneously, the model structure is relatively lightweight and has higher efficiency. BiGRU, on the other hand, stacks a reverse GRU in parallel with the GRU model. The bidirectional GRU performs temporal processing in different directions simultaneously, effectively capturing temporal dependencies in both forward and backward directions. The GRU mainly consists of update gates, reset gates, candidate hidden state layers, and final hidden state layers.

[0050] For a single GRU, the data is first updated by the gate to control the retention of historical information: (9); in, For activation functions, the Sigmoid function is typically chosen. The input to the GRU is the same as the output of the CNN. , To update the weights of the gates, To update the door offset, The old state; To update the output of the gate.

[0051] Then, the impact of resetting the gate control history information on the candidate state is as follows: (10); in, For activation functions, the Sigmoid function is typically chosen. A value close to 0 indicates that historical states should be ignored. A value close to 1 indicates a combination of historical states. To reset the weight of the door, To reset the door offset.

[0052] Then, a temporary new state is generated through the candidate hidden state layer: (11); In the formula, The temporary new state generated for the candidate hidden state layer. It is the hyperbolic tangent function. and These represent the weights and biases of the candidate hidden state layer, respectively.

[0053] Finally, the old and new states are merged through the final hidden state layer: (12); in, This is the output of the final hidden state layer; These are logical operators.

[0054] The inverse GRU processes the sequence in reverse (from t=L to t=1). The parameters of the two GRUs are independent. Finally, the outputs of the two GRUs are concatenated to obtain the final BiGRU model for extracting temporal features. (13); in, The time-series features output by the BiGRU model; This is the output of the positive GRU. This is the output of the inverted GRU.

[0055] At the end of the model, an attention mechanism (AM) module is connected to automatically focus on key dynamic regions in the data, especially the voltage plateau region of lithium iron phosphate batteries. AM calculates the correlation between each input sequence and assigns weights to each input sequence, enabling the model to pay more attention to the important parts of the sequence. The input sequence obtains feature vectors through a fully connected layer, and then compares them with themselves to obtain their respective weights. Important sequences have higher weights, and unimportant sequences have lower weights. Finally, the input sequence is multiplied by the weights to obtain a new input sequence, which is the room-temperature SOC estimation result. Its calculation can be represented as follows: First, the weights are obtained through a fully connected layer: (14); in, This is the output of the fully connected layer; and These are the weights and biases, respectively. This is the input sequence of the fully connected layer, i.e., the output of the BiGRU model.

[0056] Then, the weights are normalized using the Softmax function to obtain the final weight coefficients for each sequence: (15); in, For sequence Weighting coefficients; For the first The weights of each sequence itself; It is a natural exponential function.

[0057] Finally, the original input sequence is multiplied by the weights to obtain a new input sequence: (16); in, The results are the SOC estimates at room temperature.

[0058] The room-temperature SOC estimation model includes a first feature extraction module, a BiGRU model, and an attention mechanism module. The first feature extraction module is used to extract features from the target power battery operating condition data to obtain power battery features. The BiGRU model is used to extract time-series features from the power battery features. The attention mechanism module is used to perform weighted fusion processing on the time-series features to obtain the room-temperature SOC estimation result.

[0059] 2. The low-temperature SOC estimation model is constructed using a deep learning model. At low temperatures, the battery's internal resistance surges, polarization voltage establishes slowly, and the voltage-SOC relationship exhibits strong nonlinearity. Therefore, based on the characteristics of strong polarization and long-term temporal dependencies at low temperatures, capturing long-term temporal dependencies and strong nonlinear mapping capabilities is the main objective of the low-temperature SOC estimation model. The model is built around a CNN-Transformer model with a built-in CNN, utilizing the powerful feature extraction capabilities of CNNs to automatically extract features, and using the Transformer model to capture temporal relationships. The Transformer model is a deep learning model suitable for solving sequence problems. The Transformer model consists of an encoder and a decoder, internally including pointwise, stacked self-attention mechanisms and fully connected layers. The core mechanism of the Transformer model is the self-attention mechanism, which can model dependencies without considering distances in the input or output sequences, directly calculating the dependencies between any two positions, better capturing global information, and possessing higher parallelism and computational efficiency. Since state estimation is a many-to-one regression problem, which estimates battery state based on feature sequences rather than sequence-to-sequence transformations or generation, only the encoder part of the Transformer is used. The encoder part of the Transformer model is excellent at modeling time series data and capturing long-term dependencies within the time series. The Transformer encoder model consists of an input layer, a self-attention mechanism layer, a residual connection and normalization layer, a feedforward neural network layer, and an output layer.

[0060] First, in the input layer, the input low-temperature SOC features (battery feature data) are positionally encoded. Positional encoding uses trigonometric functions to inject positional information into the time series, generating a unique code for each vector at each position. This code is then added to the input time series data. The trigonometric function positional encoding can be described as follows: (17); (18); in, It is a position vector; Indicates location, Indicates the feature dimension.

[0061] Then, the self-attention mechanism layer is entered, where, for a sequence, the calculation is performed for each element. Query, Key, and Value: (19); (20); (twenty one); in, , , respectively query ,key Sum The linear transformation weight matrix.

[0062] Then, the scaling dot product attention is calculated, each result is divided by a normalization factor, and then processed using the softmax activation function: (twenty two); in, To scale the attention features of the dot product, This is the normalization factor.

[0063] The multi-head attention mechanism is calculated based on the scaled dot product attention mechanism described above. It concatenates the attention outputs of each head and performs a linear transformation by multiplying them by the projection matrix, ultimately yielding the output of the multi-head attention mechanism. (twenty three); (twenty four); in, For the attention output after splicing; Indicates a splicing operation; This is the output of the multi-head attention mechanism; This is the projection matrix.

[0064] Then, we proceed to the residual connection and normalization layer: (25); (26); in, The input to the residual block is the output of the multi-head attention mechanism. For the output of the residual block, The residual function is typically composed of multiple convolutional or fully connected layers. For the output of the normalization layer, Presentation layer normalization operation.

[0065] The layer normalization formula is as follows: (27); in, and These are the scaling parameter and the offset parameter, respectively, both of which are learnable parameters; The mean of the input; The variance of the input; It is a very small stability constant, typically 10.-5 To prevent the denominator from being 0.

[0066] Then, the nonlinear transformation is performed in the feedforward neural network layer: (28); in, and These are the weights and biases of the first feedforward neural network layer, respectively. and These are the weights and biases of the second feedforward neural network layer, respectively. This is the output of the feedforward neural network layer.

[0067] The feedforward neural network layer is followed by another residual connection and normalization layer. The processing of the residual connection and normalization layer is shown in formulas (25)-(27), and will not be described in detail here.

[0068] Finally, the input is fed into the output layer, and the low-temperature SOC estimation result is output by the final fully connected layer. A sigmoid function is added as the activation function in the fully connected layer. (29); in, This is the input to the fully connected layer, which is also the final output of the previous layer. and These represent the weights and biases of the fully connected layer, For the activation function, the Sigmoid function is usually chosen.

[0069] The low-temperature SOC estimation model includes a second feature module and a Transformer model. The second feature module is used to extract features from the low-temperature SOC features to obtain intermediate low-temperature features. The Transformer model is used to extract time-series features from the intermediate low-temperature features to obtain the low-temperature SOC estimation result.

[0070] 3. The Nadam algorithm is used when training the low-temperature SOC estimation model and the room-temperature SOC estimation model. Methods for training deep learning models include basic gradient descent, momentum optimization (an improvement on this), and further improved adaptive learning rate optimization algorithms. The Nadam algorithm is an adaptive learning rate optimization algorithm, its predecessor being the Adam algorithm. The Adam algorithm dynamically adjusts the learning rate of each parameter through the estimation of the first and second moments of the gradient, while introducing bias correction to ensure that the learning rate has a defined range for each iteration, thus stabilizing the parameters and significantly improving convergence speed and accuracy. The Nadam algorithm introduces Nesterov momentum improvement, updating the gradient one step ahead with the current momentum direction before calculating the gradient, making the gradient estimation more accurate. Its principle is as follows: First-moment estimation of the gradient: (30); (31); in, This represents the gradient of the loss function with respect to the model parameters. The momentum decay coefficient; for The first moment estimate of the gradient at time step 1, i.e., the exponential moving average of the gradient; for First-moment estimate of the gradient at time step; This is the first moment estimate of the initial gradient.

[0071] Second-moment estimation of the gradient: (32); (33); in, The gradient of the loss function with respect to the model parameters. for The second moment estimate of the gradient at time intervals, i.e., the exponential moving average of the squared gradient. The second-order moment attenuation coefficient, for Second-moment estimation of the gradient at time t. This is the second moment estimate of the initial gradient.

[0072] The deviation correction formula is shown below: (34); (35); Among them, superscript express The power of this power represents the current iteration step number; The first moment estimate of the corrected gradient; This is the second moment estimate of the corrected gradient.

[0073] Nesterov Advanced Update: (36); in, This is the first moment estimate of the gradient for the lead-up update.

[0074] Parameter update: (37); in, For model parameters, This is the current iteration step number. For learning rate, To prevent small constants with a denominator of 0.

[0075] During model training, the hyperparameters were determined manually using a grid search method. The final hyperparameters were as follows: for the low-temperature SOC estimation model, the number of convolutional kernels was 128, the kernel size was 10, the number of neurons in the fully connected layer was 64, and the learning rate was 0.0006; for the room-temperature SOC estimation model, the number of convolutional kernels was 64, the kernel size was 10, the number of neurons in the GRU network was 128, and the learning rate was 0.0012.

[0076] The training set includes dynamic data of power battery operating conditions. The current, voltage and temperature in the dynamic data of power battery operating conditions are used as inputs, and the actual SOC value in the dynamic data of power battery operating conditions is used as the label to estimate the SOC at room temperature, thus obtaining a trained SOC estimation model at room temperature.

[0077] Feature extraction is performed on current, voltage, and temperature in the dynamic data of power battery operating conditions to obtain low-temperature SOC features. Using the low-temperature SOC features as input and the actual SOC value in the dynamic data of power battery operating conditions as labels, a low-temperature SOC estimation model is obtained, resulting in a trained low-temperature SOC estimation model.

[0078] The target power battery operating condition data in step 201 includes voltage, current, and temperature.

[0079] Step 4: Temperature-Adaptive SOC Estimation Based on Model Fusion. During SOC estimation, both the ambient temperature SOC estimation model and the low temperature SOC estimation model are performed simultaneously, and the estimation results of the two models are fused. Considering the differences in actual application scenarios, two model fusion schemes are proposed.

[0080] In step 205, the final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result. This can be achieved through either of the following two schemes.

[0081] Option 1: The low-temperature SOC estimation result and the room-temperature SOC estimation result are weighted and summed to obtain the final SOC estimation result.

[0082] Option 1 employs an adaptive dynamic weighting method based on actual temperature. It determines dynamic weights according to the difference between the current actual temperature and the ambient temperature (25℃), and determines the final SOC estimation result through weighted summation, thus achieving temperature distance adaptation: First, the absolute value of the temperature difference is calculated: (38); in, This is the absolute value of the difference between the current actual temperature and the normal temperature (25℃). The current actual temperature is the temperature from the target power battery operating condition data. Then, the dynamic weight value is calculated: (39); in, These are dynamic weight values; The linter function is used to adjust the shape of the weight curve so that the weight values ​​quickly approach 1 at low temperatures.

[0083] Based on the calculation results of the dynamic weight values, the estimation results of the low-temperature SOC estimation model and the normal-temperature SOC estimation model are weighted and fused: (40); in, This is the final SOC estimation result; The results are the SOC estimation results at low temperature; The results represent the SOC estimation at room temperature. When the weight values ​​approach 0, the model primarily relies on the room temperature SOC estimation model; when the weight values ​​approach 1, the model primarily relies on the low temperature SOC estimation model.

[0084] In the final verification, a temperature point within the range of 25℃ to -30℃ was randomly selected, and the SOC was estimated using the above method under dynamic operating conditions. The result was then compared with the actual value to evaluate the estimation effect. This scheme has clear physical interpretability and low computational cost. Compared with the usual branch switching strategy, it can effectively prevent abrupt changes in SOC estimation results, making it suitable for application in automotive environments and scenarios with high safety requirements.

[0085] Option 2: Input the low-temperature SOC estimation result and the room-temperature SOC estimation result into the trained meta-model to obtain the final SOC estimation result; the trained meta-model is a model trained with the low-temperature SOC estimation result and the room-temperature SOC estimation result corresponding to the sample power battery operating condition data as input, and the true SOC value corresponding to the sample power battery operating condition data as the label.

[0086] Scheme 2 employs an ensemble learning approach, training a meta-model to fuse the estimation results of the low-temperature SOC estimation model and the room-temperature SOC estimation model. First, a meta-model is constructed based on multiple fully connected layers. Then, the SOC estimation results from the low-temperature and room-temperature SOC estimation models are integrated into meta-features. These meta-features are input into the meta-model for training and validation to obtain the final SOC estimation result. The meta-model is constructed based on a stack of multiple fully connected layers, with each fully connected layer calculated as follows: (41); in, and These are the weights and biases of the meta-model, respectively. This is the output of the fully connected layer; The input to the fully connected layer is the meta-feature composed of the low-temperature SOC estimation result and the normal-temperature SOC estimation result.

[0087] When training and validating the meta-model, multiple temperature ranges are first defined: [-30℃, -10℃), [-10℃, 10℃], and (10℃, 25℃). Then, two temperature points are randomly selected from each of these ranges, for a total of six temperature points. Low-temperature SOC estimation and normal-temperature SOC estimation models are used to estimate the SOC under dynamic operating conditions at these selected temperature points. The estimation results are concatenated as meta-features, and the corresponding true SOC values ​​are used as labels to construct the training set for the meta-model. For final validation, a temperature point within the range of 25℃ to -30℃ is randomly selected, and the SOC under its dynamic operating conditions is estimated using the above method. The estimation results are then compared with the true values ​​to evaluate the estimation performance. This data-driven approach is suitable for cloud-based big data platforms and the development of novel batteries with unknown internal mechanisms.

[0088] The data-driven temperature-adaptive SOC estimation method further includes: using a noise-adaptive square root unscented Kalman filter to filter the final SOC estimation result, thereby obtaining a filtered SOC estimation result.

[0089] The final SOC estimation result is further processed using a noise-adaptive square root unscented Kalman filter (SRAUKF). The Kalman filter (KF) is an efficient recursive filter for linear dynamic systems based on a Bayesian filtering algorithm, capable of estimating the system state from a series of incomplete and noisy measurements. The UKF is a variant of the KF, using unscented transformations to apply the Kalman filter architecture to nonlinear systems. The AUKF, building upon the UKF, optimizes the fixed process noise Q and observation noise R into real-time updated techniques, further overcoming the time-varying noise characteristics of battery systems. SRAUKF, based on the AUKF, maintains the non-negativity and numerical stability of the covariance matrix through Cholesky decomposition.

[0090] SRAUKF makes the traditional KF system applicable to nonlinear systems, and further overcomes the time-varying noise characteristics of the battery system by updating the process noise Q and observation noise R online. Furthermore, it maintains the non-negativity and numerical stability of the covariance matrix by introducing Cholesky decomposition. After fusing the estimation results of the two models, the output is further processed based on SRAUKF to further suppress the inherent random noise of the deep learning model, thereby further improving the estimation accuracy. When processing the estimation results based on SRAUKF, prior estimation of the state is performed using the ampere-hour integral method. Then, the fused estimation result of the deep learning model is treated as a measurement and updated posteriorly based on SRAUKF. Specifically, the state-space equations of the system are first established: The state equation is: (42); in, For the first Step State value; For the first Step State value, For the first Step current, For rated capacity, To estimate noise; This represents the time interval step between adjacent sampling points.

[0091] The measurement equation is: (43); in, The final fusion output of the deep learning model is the SOC estimation result. For measuring noise.

[0092] According to state-space theory, for the convenience of expressing subsequent steps, the state-space equations constructed above are expressed in the standard form of the following state-space equations: The state equation is: (44); The measurement equation is: (45); in, Represents SOC, Representing DL, Then it means item.

[0093] (1) Time update: First, perform an unscented transformation. For each state, generate Sigma points (key sampling points): (46); (47); (48); (49); in, For the first The 0th Sigma point of the step; For the first State estimates of the step state equations; For the first Step 1 One Sigma point; The dimension of the system state vector; This is the expansion factor for the Sigma points, controlling the dispersion of the Sigma points relative to the mean. This is a scaling parameter used to adjust the distribution characteristics of the Sigma points; The scaling parameter controls the distribution range of the Sigma points around the mean, and is usually taken as a small positive number; As a secondary scaling parameter, it helps adjust the distribution of Sigma points to make the covariance better match the prior characteristics; for The lower triangular matrix (Cholesky factor) of the step is used to represent the square root of the covariance matrix: (50); in, for The covariance matrix of the step, for The lower triangular matrix of the step; superscript Indicates transpose Then, the Sigma points are input into the state equation to calculate the predicted state: (51); (52); in, For the first Step 1 The intermediate value of the state estimate; For the first The current of the step; For the first The predicted state of the step; In order to be with the first Efficiency parameters associated with each Sigma point; This refers to the battery's nominal capacity. For the first The mean weight of each Sigma point.

[0094] Where the weight W is: (53); in, , , These are the intermediate values ​​for the 0th, 1st, and 2nd states, respectively.

[0095] Calculate the square root of the predicted covariance: (54); in, To predict the square root of the covariance; For the first The process noise covariance of the step; For the first Covariance weights for each Sigma point.

[0096] The weights are: (55); in, It is a constant.

[0097] (2) Observation update: The predicted Sigma points are updated using the measurement equation: (56); in, For the first Step 1 The observed predicted value is the observation result obtained by substituting the predicted state Sigma point into the measurement equation; For the first Step 1 The intermediate predicted state is the intermediate predicted state calculated from the state equation.

[0098] Calculation of observation and prediction: (57); in, For the first The average prediction result at the observation level is obtained by weighted summation of all observed predictions.

[0099] Calculate the square root of the observed covariance: (58); in, For the first The square root of the observed covariance of the step; For the first The measurement noise covariance of the step.

[0100] Calculate the cross-covariance between the state and the observation: (59); in, For the first The cross-covariance between the state of the step and the observation.

[0101] Then, the Kalman gain is calculated based on the square root of the observation covariance and the cross-covariance between the state and the observations: (60); in, For the first Kalman gain of the step.

[0102] Perform the final covariance square root update and state update to obtain the final SOC estimate, and use it as the initial value for the next iteration of SRAUKF: Covariance square root update: (61); in, For the first The square root of the updated covariance; To predict the square root of the covariance.

[0103] Status Update: (62); in, For the first The status after the update; For the first The actual observed values ​​of the step; For the first The mean of the prior observations predicted in the step.

[0104] During the SRAUKF calculation, the noise matrix is ​​updated online using the Sage-Husa method.

[0105] First, calculate the residuals: (63); Update observation noise variance: (64); in, For the first Observation noise variance after step update; For the first Observation noise variance after step update; This is the forgetting factor, used to adjust the degree of influence of new and old observation data on the update of observation noise variance; For the first The residual of the first step, i.e. the first step Step actual observation value =Average of prior observations predicted The difference reflects the degree of deviation between the observation prediction and the actual observation.

[0106] Noise variance during update process: (65); (66); in, For the first Step-by-step update of process noise variance; This is an intermediate quantity; For the first Step-by-step update of process noise variance; This is the forgetting factor, typically ranging from 0.95 to 0.99.

[0107] This application has the following advantages: (1) In view of the characteristics of fast dynamic response and weak polarization of lithium-ion batteries at room temperature, a lightweight room temperature SOC estimation model was constructed with CNN-BiGRU-AM model as the core. By automatically extracting features through CNN, capturing bidirectional temporal relationships through BiGRU, and automatically focusing on key dynamic regions in the data, high-precision SOC estimation at room temperature is achieved with higher efficiency while ensuring accuracy. This solves the problem of how to further optimize the lightweighting of deep learning models based on achieving high-precision SOC estimation in specific application scenarios.

[0108] (2) In view of the characteristics of strong polarization and long time-series dependence at low temperature, a low temperature SOC estimation model with a CNN-Transformer model with a CNN in front is constructed as the core to capture long time-series dependence and strong nonlinear mapping capability. By adding a CNN in front of the Transformer network, the powerful feature extraction capability is used to automatically extract features, which solves the problem of poor estimation accuracy caused by the difficulty in feature extraction due to the complex and ever-changing dynamic working conditions and high real-time requirements when performing SOC estimation. At the same time, the time-series modeling based on the Transformer network solves the problem that the strong polarization and long time-series dependence characteristics of lithium battery are difficult to capture when performing SOC estimation at low temperature, thus achieving high-precision SOC estimation at low temperature.

[0109] (3) For low-temperature environments, Arrhenius current normalization and voltage-temperature differential features were used for feature extraction. Arrhenius current normalization was used to eliminate the influence of low temperature on ion diffusion rate, and voltage-temperature differential features were used to capture the amplified entropic heat effect and polarization effect unique to low temperatures. Through feature extraction under low-temperature conditions, the actual dynamic characteristics of the battery can be effectively characterized. At the same time, the proposed features are based on real-time signals, which can be performed without accumulating sufficient data, and can effectively adapt to SOC estimation, an application with extremely high real-time requirements. This solves the problem of difficulty in extracting effective features when performing SOC estimation, and also solves the problem that the strong polarization and entropic heat effect of lithium batteries at low temperatures make it difficult for the original signal to truly reflect the dynamic characteristics of the battery, thus making it difficult to achieve high-precision SOC estimation at low temperatures.

[0110] (4) Two parallel schemes for integrating the base model are proposed, namely the adaptive dynamic weight method based on actual temperature and the ensemble learning method. The two methods are suitable for different scenarios. The adaptive dynamic weight method based on actual temperature has clear physical interpretability and low computational consumption, and is suitable for vehicle environment and scenarios with high safety requirements. The ensemble learning method is based on data-driven construction and is suitable for cloud big data platform and new battery R&D scenarios with unknown mechanisms. It solves the problem that specific methods are difficult to fully adapt to specific scenarios due to the complexity and variability of application scenarios in practical applications, as well as the problem of SOC estimation results jumping caused by a single hard switching method.

[0111] (5) The final SOC estimation results of the two deep learning models were further processed by SRAUKF technology to further improve the estimation accuracy. SRAUKF can be applied to nonlinear systems. At the same time, it overcomes the time-varying noise characteristics of the battery system by updating the process noise Q and observation noise R online. In addition, by introducing Cholesky decomposition to maintain the non-negative definiteness and numerical stability of the covariance matrix, it solves the problems of numerical instability and difficulty in overcoming the time-varying noise characteristics of the battery system in the traditional UKF. At the same time, it solves the problem that the estimation accuracy of the deep learning model is limited due to the inherent random noise.

[0112] This application also provides an application scenario in which the above-mentioned data-driven temperature-adaptive SOC estimation method is applied. Specifically, the data-driven temperature-adaptive SOC estimation method provided in this embodiment can be applied to the SOC estimation scenario of a power battery. The power battery SOC estimation scenario includes a data acquisition stage and a power battery SOC estimation stage; the target power battery operating condition data enters the power battery SOC estimation stage from the data acquisition stage, and the final SOC estimation result is obtained through human-machine collaboration. The data-driven temperature-adaptive SOC estimation method provided in this embodiment belongs to the power battery SOC estimation stage. Specifically, in the power battery SOC estimation stage for the target power battery operating condition data, features can be extracted from the target power battery operating condition data to obtain low-temperature SOC features. The low-temperature SOC features are input into a trained low-temperature SOC estimation model to obtain a low-temperature SOC estimation result. The target power battery operating condition data are input into a trained room-temperature SOC estimation model to obtain a room-temperature SOC estimation result. The final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result.

[0113] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores State of Charge (SOC) estimation data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a data-driven, temperature-adaptive SOC estimation method.

[0114] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0115] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0116] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0117] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0118] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0119] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0120] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0121] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0122] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A data-driven based temperature adaptive SOC estimation method, characterized in that, The data-driven temperature adaptive SOC estimation method includes: Acquire target power battery operating condition data; Feature extraction is performed on the target power battery operating condition data to obtain low-temperature SOC features; The low-temperature SOC features are input into the trained low-temperature SOC estimation model to obtain the low-temperature SOC estimation result; The target power battery operating condition data is input into the trained room temperature SOC estimation model to obtain the room temperature SOC estimation result; The final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result.

2. The data-driven temperature adaptive SOC estimation method according to claim 1, characterized in that, The data-driven temperature adaptive SOC estimation method further includes: The final SOC estimation result is filtered using a noise adaptive square root unscented Kalman filter to obtain the filtered SOC estimation result.

3. The data-driven temperature adaptive SOC estimation method according to claim 1, characterized in that, The target power battery operating condition data includes voltage, current, and temperature.

4. The data-driven temperature adaptive SOC estimation method according to claim 3, characterized in that, Feature extraction is performed on the target power battery operating condition data to obtain low-temperature SOC features, specifically including: Based on the Arrhenius equation, the current in the target power battery operating condition data is normalized to obtain the normalized current. The voltage-temperature differential feature is obtained by differentiating the voltage and temperature in the target power battery operating condition data; the normalized current and voltage-temperature differential features constitute the low-temperature SOC feature.

5. The data-driven temperature adaptive SOC estimation method according to claim 1, characterized in that, The room-temperature SOC estimation model includes a first feature extraction module, a BiGRU model, and an attention mechanism module; The first feature extraction module is used to extract features from the target power battery operating condition data to obtain power battery features; The BiGRU model is used to extract time-series features from power battery characteristics; The attention mechanism module is used to perform weighted fusion processing on temporal features to obtain room temperature SOC estimation results.

6. The data-driven temperature adaptive SOC estimation method according to claim 1, characterized in that, The low-temperature SOC estimation model includes a second feature module and a Transformer model; The second feature module is used to extract features from the low-temperature SOC features to obtain low-temperature intermediate features; The Transformer model is used to extract time-series features from intermediate features at low temperatures to obtain the low-temperature SOC estimation results.

7. The data-driven temperature adaptive SOC estimation method according to claim 1, characterized in that, The final SOC estimation result is determined based on the low-temperature SOC estimation result and the room-temperature SOC estimation result, specifically including: The final SOC estimation result is obtained by weighted summing of the low-temperature SOC estimation result and the room-temperature SOC estimation result. Alternatively, the low-temperature SOC estimation result and the room-temperature SOC estimation result can be input into the trained meta-model to obtain the final SOC estimation result; the trained meta-model is a model trained with the low-temperature SOC estimation result and the room-temperature SOC estimation result corresponding to the sample power battery operating condition data as input, and the true SOC value corresponding to the sample power battery operating condition data as the label.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the data-driven temperature adaptive SOC estimation method according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the data-driven temperature-adaptive SOC estimation method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the data-driven temperature-adaptive SOC estimation method according to any one of claims 1-7.

Images