Lithium battery charge state estimation method and system based on linear layer and convolutional neural network

Abstract

The invention provides a lithium battery state-of-charge estimation method and system based on a linear layer and a convolutional neural network. The method comprises the steps of obtaining historical operation data of a lithium battery and performing data preprocessing; a state of charge estimation model is constructed, a battery time sequence is decomposed into a long-term trend part and a local fluctuation part, a global linear trend feature and a multi-scale local fluctuation feature are extracted through linear branch simulation, the sensitivity to real-time dynamic is enhanced through multi-scale convolution and differential branches, and adaptive fusion is performed on the features, so that the real-time dynamic sensitivity is improved. According to the method, the charge state estimation basic model is constructed, the loss function is established, the basic model is trained in combination with the optimization algorithm, the charge state estimation value of the lithium battery is output, the charge state change characteristics of the battery under the complex dynamic working condition can be comprehensively and accurately captured, and thus the charge state estimation accuracy is remarkably improved. The lithium battery state-of-charge estimation system provided by the invention can realize the lithium battery state-of-charge estimation method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of battery management system, and particularly relates to a lithium battery state of charge estimation method and system based on linear layer and convolutional neural network. BACKGROUND

[0002] Lithium iron phosphate (LiFePO4) batteries have been widely used in electric vehicles (EVs) and stationary energy storage systems due to their excellent safety performance, long cycle life, and relatively low cost. Accurate estimation of the state of charge (SOC) of the battery is crucial for ensuring the safe operation of the battery system, optimizing energy management strategies, prolonging the service life of the battery, and ensuring the reliability of electric vehicles and energy storage systems.

[0003] However, the SOC of LiFePO4 batteries exhibits a global linear trend combined with superimposed local fluctuations within each charge-discharge cycle, which is mainly due to its internal complex electrochemical processes such as polarization effect, ion diffusion, temperature gradient, etc. For example, the polarization effect can cause voltage hysteresis and uneven current distribution, while the temperature difference between individual cells in the battery pack can accelerate capacity decay and lead to uneven discharge. In addition, the formation and evolution of the solid electrolyte interface film (SEI) and the decomposition of the electrolyte will introduce additional nonlinearities and uncertainties, especially during the battery aging process. These factors make the SOC trajectory highly dynamic and sensitive to internal and external influences, making it difficult to accurately estimate the SOC in real-world application scenarios. In addition, electrode heterogeneity and operating conditions (such as high-rate cycling or deep discharge) can amplify local fluctuations and accelerate aging mechanisms.

[0004] Existing SOC estimation methods can be mainly divided into model-based methods and data-driven methods. Model-based methods, such as open-circuit voltage (OCV) method and electrochemical models, provide interpretable results using physical principles. The OCV method is widely used due to its direct relationship with SOC, but it is limited by hysteresis and sensitive to measurement noise and parameter drift. Electrochemical models can capture detailed battery dynamics, but they usually require extensive parameter calibration and are computationally intensive, limiting their application in real-time battery management systems. Importantly, these traditional methods often have limited ability to extract both the global linear trend and local fluctuation characteristics of the SOC, especially under complex and dynamic operating conditions.

[0005] With the development of artificial intelligence technology, data-driven SOC estimation methods have received increasing attention. Deep learning models, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), have shown strong performance in feature extraction and time series modeling. Models such as Informer and TimesNet further enhance the ability to model long-range dependencies and complex temporal patterns. However, traditional deep learning models often struggle to capture both global linear trends and high-frequency local fluctuations in SOC estimation, which can lead to reduced estimation accuracy in diverse operating scenarios. Attention mechanism-based models, such as Transformer and its variants, introduce mechanisms to capture relationships between sequence elements by assigning dynamic weights to different time steps. Although these models have achieved success in various time series prediction tasks, they may not fully exploit the explicit temporal order that is crucial for modeling the continuous evolution of battery SOC. This limitation is particularly evident when predicting sudden local SOC fluctuations caused by rapid load changes or internal state transitions.

[0006] Therefore, there is an urgent need for a new method that can effectively decompose and process linear trends and local fluctuation components in battery time series, thereby achieving high accuracy and strong robustness in SOC estimation under various complex operating conditions. SUMMARY

[0007] To overcome the shortcomings of the prior art, the present application aims to provide a lithium battery state of charge estimation method and system based on linear layer and convolutional neural network to improve the accuracy and robustness of state of charge estimation.

[0008] The present application realizes the estimation of the state of charge of the lithium battery by the following technical solutions: The present application provides a lithium battery state of charge estimation method based on linear layer and convolutional neural network, which comprises the following steps: Step 1: Obtain historical operation data of the lithium battery and perform data preprocessing; Step 2: decompose the preprocessed historical running data into a long-range trend component and a short-range fluctuation component through a data decomposition mechanism, extract a global linear trend feature from the long-range trend component and the short-range fluctuation component through a linear module, and extract a multi-scale local fluctuation feature from the preprocessed historical running data through a fluctuation module, calculate a first-order time difference of the preprocessed historical running data through a difference feature extraction module, extract a sensitivity feature by taking the first-order time difference as input, and extract an instant dynamic feature from a latest time step data segment of the preprocessed historical running data through a local enhancement module; fuse the global linear trend feature, the multi-scale local fluctuation feature, and the sensitivity feature through a main feature fusion module to generate a main state of charge estimation, weight and fuse the main state of charge estimation and the instant dynamic feature to output a final state of charge estimation, and complete the construction of the state of charge estimation model; Step 3: based on the preprocessed historical running data as input, establish a loss function and train the state of charge estimation model combined with an optimization algorithm to obtain a trained state of charge estimation model; Step 4: input real-time or near real-time historical running data of the lithium battery, output the state of charge estimation value of the lithium battery through the trained state of charge estimation model.

[0009] In one embodiment, in step 1, the historical running data of the lithium battery includes: time series data of voltage, current, and temperature.

[0010] In one embodiment, in step 2, the preprocessed historical running data is decomposed into a long-range trend component and a short-range fluctuation component, including: decomposing the preprocessed historical running data into a long-range trend component and a short-range fluctuation component; setting the long-range trend component as the input preprocessed historical running data, and calculating the short-range fluctuation component as the input preprocessed historical running data minus the moving average.

[0011] In one embodiment, in step 2, the global linear trend feature is extracted from the long-range trend component and the short-range fluctuation component through a linear module, including: respectively performing independent linear transformation on the long-range trend component and the short-range fluctuation component through a fully connected layer to respectively obtain hidden representations of the long-range trend component and the short-range fluctuation component; combining the hidden representations of the long-range trend component and the short-range fluctuation component to extract the global linear trend feature.

[0012] In one embodiment, in step 2, the multi-scale local fluctuation feature is extracted from the preprocessed historical running data through a fluctuation module, including: processing the preprocessed historical running data of the original input through at least two parallel one-dimensional convolution layers, wherein each of the parallel one-dimensional convolution layers has a preset different convolution kernel size; and splicing the output feature maps of the at least two parallel one-dimensional convolution layers to extract multi-scale local fluctuation features.

[0013] In one embodiment, in step 2, the sensitivity features are extracted by taking the first-order time difference as input, including: processing the first-order time difference through at least one one-dimensional convolution network comprising a stack of one-dimensional convolution layers and an activation function to extract sensitivity features reflecting the rate of change of the state of charge of the lithium battery.

[0014] In one embodiment, in step 2, the real-time dynamic features are extracted from the latest time step data segment of the preprocessed historical running data by the local enhancement module, including: extracting a latest time step continuous data segment of a predetermined length from the preprocessed historical running data; extracting real-time dynamic features through a fully connected layer based on the latest time step continuous data segment.

[0015] In one example, in step 2, the final state of charge estimate is output by weighted fusion of the main state of charge estimate and the real-time dynamic features, and the following weighted formula is used for fusion: , wherein, represents the final state of charge estimate, represents the main state of charge estimate generated by the main feature module, represents the real-time dynamic features generated by the local enhancement module, is a weight parameter.

[0016] In one embodiment, in step 3, the loss function is established and the state of charge estimation model is trained by combining an optimization algorithm, including: using forward propagation to calculate the predicted state of charge estimate value by taking the preprocessed historical running data as input, and establishing a loss function, wherein the loss function is the absolute mean square error or the root mean square error , calculated as follows: , , wherein, is the true state of charge estimate value of the i-th sample, is the predicted state of charge estimate value of the i-th sample, is the predicted state of charge estimate value of the i-th sample, is the predicted state of charge estimate value of the i-th sample, the number of samples for the batch or the entire dataset; The gradient of the loss function on each parameter of the state of charge estimation model is calculated by a back propagation algorithm, and these parameters are updated using an optimization algorithm based on gradient descent to minimize the loss function, and hyperparameters are tuned by a hyperparameter tuning strategy to train the state of charge estimation model; wherein the parameters of the state of charge estimation model include the weights and biases of the state of charge estimation model.

[0017] In another aspect, the application also provides a lithium battery state of charge estimation system based on linear layers and convolutional neural networks, which applies the lithium battery state of charge estimation method based on linear layers and convolutional neural networks, comprising: a data acquisition module for acquiring historical operation data of the lithium battery and performing data preprocessing; a state of charge estimation model construction module for decomposing the preprocessed historical operation data into long-range trend components and short-range fluctuation components through a data decomposition mechanism, extracting global linear trend features from the long-range trend components and the short-range fluctuation components through a linear module, extracting multi-scale local fluctuation features from the preprocessed historical operation data through a fluctuation module, calculating a first-order time difference of the preprocessed historical operation data through a difference feature extraction module, extracting sensitivity features with the first-order time difference as input, and extracting real-time dynamic features from the latest time step data segment of the preprocessed historical operation data through a local enhancement module; the global linear trend features, the multi-scale local fluctuation features and the sensitivity features are fused through a main feature fusion module to generate a main state of charge estimation, the main state of charge estimation and the real-time dynamic features are weighted and fused to output a final state of charge estimation, and the construction of the state of charge estimation model is completed; a state of charge estimation model training module for establishing a loss function based on the preprocessed historical operation data as input and training the state of charge estimation model combined with an optimization algorithm to obtain a trained state of charge estimation model; a state of charge estimation prediction module for inputting real-time or near real-time historical operation data of the lithium battery, outputting a state of charge estimation value of the lithium battery through the trained state of charge estimation model.

[0018] Compared with the prior art, the application has at least the following beneficial effects: (1) Improved estimation accuracy under complex working conditions: the state of charge estimation model architecture proposed by the application can more comprehensively and accurately capture the state of charge change characteristics of the battery under complex dynamic working conditions by explicitly decomposing the battery time series into linear trend and local fluctuation components and processing them by special modules (linear module, fluctuation module, difference feature extraction module, local enhancement module), thereby significantly improving the accuracy of state of charge estimation.

[0019] (2) Enhanced modeling capability for multi-scale features: The fluctuation module uses multi-core convolution to extract wave information of different time scales in parallel, the difference module focuses on the rate of change, the linear module focuses on the overall trend, and the local enhancement module focuses on recent dynamics. This multi-angle and multi-scale feature extraction and fusion mechanism enables the model to better understand and represent the complex electrochemical processes inside the battery and its external performance.

[0020] (3) Improved robustness and generalization of the model: Through the collaborative work of each module and the final adaptive weighted fusion, the model not only learns the main state of charge evolution law, but also effectively captures various disturbances and mutations, thereby maintaining good estimation performance and robustness under different battery types, aging levels and working conditions.

[0021] (4) The model structure has certain interpretability: Compared with some pure "black box" deep learning models, the modular design of the state of charge estimation model (such as distinguishing trends and fluctuations) enables different parts of the model to have relatively clear functional orientation, which helps to understand the model decision-making process and provides ideas for further optimization of the model BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0023] Figure 1 The flowchart of the lithium battery state of charge estimation method based on linear layer and convolutional neural network provided by the present application.

[0024] Figure 2 The structure diagram of the state of charge estimation model provided by the embodiment of the present application.

[0025] Figure 3 The structure diagram of the lithium battery state of charge estimation system based on linear layer and convolutional neural network provided by the embodiment of the present application. DETAILED DESCRIPTION

[0026] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purpose, the technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0027] It should be understood that the terms "comprises" and "comprising," when used in this specification and accompanying claims, indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0028] It should also be understood that the terms used herein are for the purpose of describing particular embodiments and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

[0029] It should further be understood that the term "and / or" as used herein refers to any one or more of the associated listed items, and all possible combinations of the items, and includes these combinations.

[0030] The concept of the present application is to provide a lithium battery state of charge estimation method and system based on linear layer and convolutional neural network, which is used to decompose the battery time series into linear trend and local fluctuation components explicitly, design linear module, fluctuation module, difference feature extraction module, and local enhancement module for feature extraction, then perform preliminary fusion of features through the main feature fusion module, and output the final state of charge estimation after secondary weighted fusion, so that the model can better understand and represent the complex electrochemical process inside the battery and its external performance, thereby more comprehensively and accurately capturing the state of charge change characteristics of lithium batteries under complex dynamic working conditions, and significantly improving the accuracy of state of charge estimation.

[0031] The scheme of the embodiment of the present application is as shown in Figure 1 A lithium battery state of charge estimation method based on linear layer and convolutional neural network, comprising the following steps: S1, obtaining historical running data of the lithium battery and performing data preprocessing. Specifically as follows: S11, in this embodiment, the goal of this stage is to collect historical time series data related to the performance and state of the lithium battery, which is the basis for subsequent model learning and prediction. Specifically, it includes the voltage sequence, current sequence, temperature sequence of the lithium battery and the corresponding time sequence data. These data constitute the input sequence , wherein is a preset review window length (sequence length), is the number of input features (for example, voltage, current, temperature, etc.).

[0032] S12, obtain several groups of battery historical operation data, and perform data integration and preprocessing to obtain a sample set required for training. Specifically, from the historical data of one or more batteries (for example, LiFePO4 batteries) running under different working conditions (such as different temperatures, different charge-discharge rates, and different aging degrees), time series segments containing voltage, current, temperature, and the like, and corresponding real SOC values are extracted. The original data is preprocessed as necessary, such as data cleaning (removing outliers), data alignment, and data normalization or standardization (for example, scaling the data to [0, 1] or a distribution with a mean of 0 and a variance of 1) to improve the efficiency and stability of model training. The preprocessed data is divided into a training set, a validation set, and a test set to form a sample set for model training, parameter tuning, and performance evaluation.

[0033] S2, decompose the historical operation data into a long-range trend component and a short-range fluctuation component through a data decomposition mechanism, extract global linear trend features from the long-range trend component and the short-range fluctuation component through a linear module, extract multi-scale local fluctuation features from the historical operation data through a fluctuation module, calculate and process the first-order time difference of the historical operation data through a difference feature extraction module, extract sensitivity features using the first-order time difference as input, and extract real-time dynamic features from the latest time step data segment of the historical operation data through a local enhancement module; the global linear trend features, the multi-scale local fluctuation features, and the sensitivity features are fused through a main feature module to generate a main state of charge estimate, the main state of charge estimate and the real-time dynamic features are weighted and fused to output a final state of charge estimate, and the construction of the state of charge estimation model is completed. Specifically as follows: The preprocessed historical operation data is input into the pre-trained state of charge estimation model (LF-Net) for SOC estimation at the current time to obtain an estimated SOC result. In this embodiment, the estimated SOC result refers to the predicted value of the state of charge of the lithium battery at the current time. The LF-Net state of charge estimation model is the core, and its detailed architecture is shown in Figure 2 The model is constructed based on the analysis of the global linear trend and local fluctuation features existing in the lithium battery operation data at the same time, and is trained by using historical battery operation data as a sample set by using a deep learning training method.

[0034] S21, construct the LF-Net model, the architecture of which is shown in Figure 2 The LF-Net model mainly includes a data decomposition mechanism, a linear module, a fluctuation module, a difference feature extraction module, a local enhancement module, a main feature fusion network, and a final SOC output calculation. In this embodiment, the specific composition of the LF-Net model is as follows, and the design aims to cooperatively capture the multi-scale time characteristics of the battery: The data decomposition mechanism receives the input preprocessed historical operation data sequence As Figure 2 the original data . This mechanism first decomposes the input sequence to distinguish long-term trends and short-term fluctuations. For example, the original data is taken as the long-term trend component , and the original data minus its moving average is taken as the short-term fluctuation component , as shown in the following formula: , , where denotes the moving average operation. At the same time, the first-order difference of the original data is calculated for the difference feature extraction module.

[0035] Linear module: This module receives the decomposed long-term trend component and short-term fluctuation component as input. Through respective independent fully connected layers, linear transformation is performed on these two components to extract and model the global linear trend of the data, represented as follows: , , where , , , are the weights and biases of the corresponding fully connected layers, respectively. Then the transformed features and are fused, and the feature output by the linear module represents the global evolution trend of the SOC.

[0036] Fluctuation module: This module receives the original input sequence (i.e., the original data ) as input. Its internal structure contains multiple parallel one-dimensional convolutional layers, each with different convolution kernel sizes. This multi-scale convolutional structure aims to extract local fluctuation characteristics of different time scales from the input original data : , for example Figure 2 in , denotes the convolutional layer index, and the fluctuation features extracted by each convolutional layer are subsequently concatenated to form a comprehensive fluctuation feature vector in Figure 2 with [concatenated dimension] as the output of the fluctuation module. This module focuses on capturing high-frequency disturbances and local dynamic changes in the battery charging and discharging process.

[0037] Differential feature extraction module: This module receives the first-order time difference of the raw data As input, specifically includes voltage , current , temperature , time . The differential sequence is first processed by a stacked one-dimensional convolutional layer (which may contain activation functions such as GELU) to enhance the sensitivity to rapid changes and transient dynamics of the data. The processing process can be represented as: , The feature vector corresponding to the last time step is selected, that is, the last time step feature shown in Figure 2 as the output of this module. This module helps the model to capture the rate information of SOC change.

[0038] Local enhancement module: This module focuses on analyzing the latest data segment of the input sequence . It extracts the data of the last time step in the input sequence, denoted as , flattens the data segment, and then processes it through a fully connected layer to obtain the local output : , Where FC represents the fully connected layer, represents the flattening operation; this module aims to capture the recent dynamic information that most directly affects the current SOC.

[0039] Main feature fusion module: This module receives the global linear trend feature from the linear module, the multi-scale local fluctuation feature from the fluctuation module, and the sensitivity feature extracted by the differential feature extraction module. These features at different levels are first concatenated into a combined feature vector (“Fusion module input [concatenated dimension]” in Figure 2 is shown): , Then, the combined feature vector is processed through a special fusion network (FusionNet) including two fully connected layers for feature fusion, to realize the deep interaction and information integration of the features, and finally generate the main state of charge estimation , that is, the main output in Figure 2 : .

[0040] Final state of charge estimation output calculation: the final SOC estimation result is obtained by weighting and combining the main state of charge estimation generated by the main feature fusion module and the "local output generated by the local enhancement module: represents the final state of charge estimation, and the weight coefficient is a learnable parameter that is dynamically adjusted through the model training process to adaptively balance the contribution of global / multi-scale features and recent local features to the final prediction, thereby completing the construction of the state of charge estimation model.

[0041] S3, based on the preprocessed historical running data as input, establish a loss function and train the state of charge estimation model combined with the optimization algorithm to obtain the trained state of charge estimation model. Specifically as follows: Based on the sample set constructed in S12, the LF-Net basic model is trained using the forward propagation, back propagation algorithm and hyperparameter tuning strategy to obtain the final state of charge estimation model. In this embodiment, the prepared training set data is used to train the LF-Net model end-to-end. The training process usually uses gradient descent-based optimization algorithms (such as Adam, SGD, etc.).

[0042] In the forward propagation phase, the training sample is input into the model to calculate the predicted SOC value. Then, the difference between the predicted SOC value and the true SOC value is quantified by the loss function. In this embodiment, the mean absolute error MAE or the root mean square error RMSE is used for quantification, and the calculation formulas of MAE and RMSE are as follows:

[0043] The loss function is calculated by the back propagation algorithm to calculate the loss function of each parameter (network weight and bias, including the learnable weight ​​​​​​​​​​The gradient of the model is calculated, and these parameters are updated using an optimization algorithm to minimize the loss function. During training, the model's performance is monitored using validation set data, and hyperparameters are tuned. Hyperparameters may include learning rate, batch size, lookback window length L, the number of convolutional layers and kernel size in each module, the number of units in fully connected layers, the window size of the moving average, and the lookback length of local augmentation modules. Hyperparameter tuning strategies may include grid search, random search, or Bayesian optimization. Training terminates when the model's performance on the validation set reaches a preset convergence criterion (e.g., the loss no longer decreases significantly or the performance metric reaches the target value) or after completing a preset number of training rounds. The model with the best validation performance is saved as the final state-of-charge estimation model.

[0044] The design of the entire model reflects a comprehensive consideration of the global trend of battery SOC and multi-scale local fluctuations, aiming to achieve high-precision SOC estimation through the collaborative work of various modules and the effective fusion of features.

[0045] S4. Using real-time or near-real-time historical operating data of the lithium battery as input, and through the trained state-of-charge estimation model, output the estimated state-of-charge value of the lithium battery. Details are as follows: In this embodiment, the trained and finalized LF-Net state-of-charge (SOC) estimation model is deployed to a practical application (such as a server or embedded BMS). During runtime, the model receives real-time or near real-time historical battery operating data (time series such as voltage, current, and temperature) and outputs an estimate of the current SOC. This estimation result can be used for various purposes within the battery management system, including energy management, safety warnings, range prediction, and user interface display.

[0046] On the other hand, the embodiment also provides a lithium battery state of charge estimation system based on linear layers and convolutional neural networks, such as Figure 3 As shown, it includes: The data acquisition module is used to acquire historical operating data of lithium batteries and perform data preprocessing. The state of charge estimation model construction module is configured to decompose the preprocessed historical operation data into a long-range trend component and a short-range fluctuation component through a data decomposition mechanism, extract a global linear trend feature from the long-range trend component and the short-range fluctuation component through a linear module, extract a multi-scale local fluctuation feature from the preprocessed historical operation data through a fluctuation module, calculate a first-order time difference of the preprocessed historical operation data through a difference feature extraction module, extract a sensitivity feature by taking the first-order time difference as an input, and extract an instant dynamic feature from a latest time step data segment of the preprocessed historical operation data through a local enhancement module; the global linear trend feature, the multi-scale local fluctuation feature, and the sensitivity feature are fused through a main feature fusion module to generate a main state of charge estimation, the main state of charge estimation is fused with the instant dynamic feature to output a final state of charge estimation, and the construction of the state of charge estimation model is completed; The state of charge estimation model training module is configured to take the preprocessed historical operation data as an input, establish a loss function, and train the state of charge estimation model in combination with an optimization algorithm to obtain a trained state of charge estimation model. The state of charge estimation prediction module is configured to take real-time or near real-time lithium battery historical operation data as an input, output a state of charge estimation value of the lithium battery through the trained state of charge estimation model.

[0047] Those skilled in the art can understand that all or part of the processes in the method of the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiment of the method.

[0048] Therefore, the application further provides a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program is executed by a processor to make the processor execute the lithium battery state of charge estimation method based on a linear layer and a convolutional neural network as described in any one of the above-mentioned method embodiments. Specifically, when the computer program is executed by the processor, the following steps can be implemented: obtaining battery historical operation data, including voltage sequence, current sequence, temperature sequence, and other time series data; inputting the battery historical operation data into a pre-trained LF-Net state of charge estimation model to estimate the state of charge of the lithium battery at the current time to obtain an estimated SOC result; and outputting the SOC estimation result; wherein the LF-Net state of charge estimation model is obtained by training a historical battery operation data sample set using a deep learning training method, and the architecture of the LF-Net state of charge estimation model includes a data decomposition mechanism, a linear module, a fluctuation module, a difference feature extraction module, a local enhancement module, a main feature fusion network, and a final SOC output mechanism.

[0049] The storage medium can be a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and various computer readable storage media that can store program codes.

[0050] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in a general manner. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0051] In several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of each module is only a logical function division, and actual implementation can have another division manner. For example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0052] The steps in the method embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The modules in the system embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically, or two or more modules can be integrated in one module.

[0053] The integrated module, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part of the prior art that makes a contribution, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, a terminal or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0054] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements shall be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. A method for estimating the state of charge of a lithium battery based on linear layers and convolutional neural networks, characterized in that, The lithium battery state-of-charge estimation method includes the following steps: Step 1: Obtain historical operating data of the lithium battery and perform data preprocessing; Step 2: The preprocessed historical operating data is decomposed into long-range trend components and short-range fluctuation components through a data decomposition mechanism. A linear module extracts global linear trend features from these components, and a fluctuation module extracts multi-scale local fluctuation features. A differential feature extraction module calculates the first-order time difference of the preprocessed historical operating data, using this first-order time difference as input to extract sensitivity features. A local enhancement module extracts real-time dynamic features from the latest time step data segment of the preprocessed historical operating data. A main feature fusion module fuses the global linear trend features, multi-scale local fluctuation features, and sensitivity features to generate a main charge state estimate. The main charge state estimate is then weighted and fused with the real-time dynamic features to output the final charge state estimate, completing the construction of the charge state estimation model. Step 3: Based on the preprocessed historical running data as input, establish a loss function and combine it with an optimization algorithm to train the state of charge estimation model, and obtain the trained state of charge estimation model; Step 4: Using real-time or near-real-time historical operating data of the lithium battery as input, the trained state of charge estimation model outputs the estimated state of charge of the lithium battery.

2. The lithium battery state-of-charge estimation method based on linear layers and convolutional neural networks according to claim 1, characterized in that, In step 1, the historical operating data of the lithium battery includes time series data of voltage, current, and temperature.

3. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 1, characterized in that, Step 2, which involves decomposing the preprocessed historical operational data into long-range trend components and short-range fluctuation components, includes: The preprocessed historical operational data is decomposed into long-range trend components and short-range fluctuation components. The long-range trend component is set as the input preprocessed historical operating data, and the short-range fluctuation component is calculated as the input preprocessed historical operating data minus the moving average.

4. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 3, characterized in that, Step 2, which involves extracting global linear trend features from the long-range trend component and the short-range fluctuation component using a linear module, includes: The long-range trend component and the short-range fluctuation component are independently linearly transformed through a fully connected layer to obtain the hidden representations of the long-range trend component and the short-range fluctuation component, respectively. By combining the hidden representations of long-range trend components and short-range fluctuation components, global linear trend features are extracted.

5. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 2, characterized in that, Step 2, which involves extracting multi-scale local fluctuation features from the preprocessed historical operational data using the fluctuation module, includes: The preprocessed historical running data of the original input is processed by at least two parallel one-dimensional convolutional layers, wherein each of the parallel one-dimensional convolutional layers has a preset different convolutional kernel size; And by splicing the output feature maps of the at least two parallel one-dimensional convolutional layers, multi-scale local fluctuation features are extracted.

6. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 2, characterized in that, In step 2, the process of extracting sensitivity features from the first-order time difference as input includes: processing the first-order time difference through at least one one-dimensional convolutional network containing stacked one-dimensional convolutional layers and activation functions to extract sensitivity features that reflect the rate of change of the state of charge of the lithium battery.

7. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 2, characterized in that, Step 2, which involves extracting real-time dynamic features from the latest time step data segment of the preprocessed historical running data using the local enhancement module, includes: Extract the latest time step continuous data segment of a predetermined length from the preprocessed historical running data; Based on the latest continuous data segments at the latest time step, real-time dynamic features are extracted through a fully connected layer.

8. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 1, characterized in that, In step 2, when the principal charge state estimate is weighted and fused with the instantaneous dynamic features to output the final charge state estimate, the following weighting formula is used for fusion: ， in, This represents the final state of charge estimate. This represents the principal charge state estimate generated by the main feature module. This represents the real-time dynamic features generated by the local enhancement module. These are the weight parameters.

9. The lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks according to claim 1, characterized in that, Step 3, which involves establishing a loss function and training a state-of-charge estimation model using an optimization algorithm, includes: Forward propagation is employed, using preprocessed historical operating data as input to calculate the predicted state of charge estimate, and a loss function is established, where the loss function is the absolute mean square error. or root mean square error The calculation is as follows: ， ， in, For the first The true state-of-charge estimate of each sample For the first The predicted state of charge estimate for each sample, The number of samples in this batch or the entire dataset; The gradient of the loss function with respect to each parameter of the state of charge estimation model is calculated using the backpropagation algorithm, and these parameters are updated using a gradient descent-based optimization algorithm to minimize the loss function. Hyperparameters are then tuned using a hyperparameter tuning strategy to train the state of charge estimation model. The parameters of the state of charge estimation model include the weights and biases of the model.

10. A lithium battery state-of-charge estimation system based on linear layers and convolutional neural networks, characterized in that, The lithium battery state-of-charge estimation method based on linear layers and convolutional neural networks as described in any one of claims 1 to 9 includes: The data acquisition module is used to acquire historical operating data of lithium batteries and perform data preprocessing. The charge state estimation model construction module decomposes preprocessed historical operating data into long-range trend components and short-range fluctuation components through a data decomposition mechanism. A linear module extracts global linear trend features from these components, and a fluctuation module extracts multi-scale local fluctuation features. A difference feature extraction module calculates the first-order time difference of the preprocessed historical operating data, using this first-order time difference as input to extract sensitivity features. A local enhancement module extracts real-time dynamic features from the latest time step data segment of the preprocessed historical operating data. A main feature fusion module fuses the global linear trend features, multi-scale local fluctuation features, and sensitivity features to generate the main charge state estimate. Finally, a weighted fusion of the main charge state estimate and the real-time dynamic features outputs the final charge state estimate, completing the construction of the charge state estimation model. The charge state estimation model training module is used to establish a loss function and combine an optimization algorithm to train the charge state estimation model based on preprocessed historical running data as input, so as to obtain the trained charge state estimation model. The state of charge estimation prediction module is used to take real-time or near-real-time historical operating data of lithium batteries as input, and output the state of charge estimate of lithium batteries through a trained state of charge estimation model.

Images