Battery soc estimation model based on physical constraints and multi-branch architecture and related device

Abstract

The application provides a battery SOC estimation model based on physical constraints and a multi-branch architecture and related equipment, and relates to the technical field of battery management of energy storage systems. The battery SOC estimation model based on physical constraints and a multi-branch architecture comprises a time sequence feature extraction and battery SOC prediction branch, a physical law embedding and voltage verification branch and an operating mode identification branch. The physical law embedding and voltage verification branch is used for generating an end voltage prediction value based on the estimation value and a measured current. The operating mode identification branch is used for judging the operating state of the current battery through the multi-dimensional time sequence features. The application overcomes the adaptation to complex working conditions and the time-varying characteristics of battery degradation, and provides an effective technical solution for fine battery management of energy storage systems.

Description

Technical Field

[0001] This invention belongs to the field of battery management technology, and particularly relates to a battery SOC estimation model and related equipment based on physical constraints and a multi-branch architecture. Background Technology

[0002] Existing SOC estimation methods can be mainly divided into two categories: physical model-based methods and purely data-driven methods.

[0003] In traditional SOC estimation methods, physical models (such as equivalent circuit models) and data-driven models (such as ordinary neural networks and physical information neural networks PINN) are often disconnected and not deeply integrated. They either rely solely on physical models, whose parameters are easily affected by operating conditions, temperature, and battery aging, resulting in poor adaptability to complex operating conditions and the time-varying characteristics of battery degradation, and insufficient robustness; or they rely solely on data-driven models, lacking the constraints of physical laws. In scenarios such as LTO batteries, where the voltage platform is flat and the SOC is weakly nonlinearly mapped to voltage, large estimation biases and weak generalization ability are likely to occur.

[0004] Traditional models often employ a single-branch architecture design, lacking dedicated branches for embedding physical laws or recognizing operational states. This prevents the rigid integration of core physical laws such as charge conservation into the neural network training process in a differentiable form. Physical consistency can only be verified through posterior methods, leading to estimation results that may violate fundamental physical laws. Consequently, the models suffer from poor interpretability and lack credibility in practical applications.

[0005] Existing technologies cannot achieve physical consistency constraints across the entire time scale, leading to the accumulation of biases in battery SOC estimation during long-term cycling or dynamic operation. The lack of coordinated optimization between parameter estimation and SOC state estimation means that parameter drift directly reduces the accuracy of SOC estimation, making it difficult to support the refined management needs of energy storage systems. Summary of the Invention

[0006] In view of this, the purpose of this invention is to achieve a deep integration of physical laws and data-driven approaches, solving the problems of easy parameter drift in traditional physical models and lack of physical constraints in pure data-driven models, and significantly improving the accuracy, robustness and physical reliability of SOC estimation.

[0007] In a first aspect, embodiments of the present invention provide a battery SOC estimation model based on physical constraints and a multi-branch architecture, the battery SOC estimation model comprising: The branches include: time-series feature extraction and battery SOC prediction, physical law embedding and voltage verification, and operating mode recognition. The temporal feature extraction and battery SOC prediction branch is used to extract the multidimensional temporal features and output a preliminary SOC estimate of the battery. ; The time-series feature extraction and battery SOC prediction branch is specifically configured as follows: obtaining multi-dimensional time series data from the Transformer encoder in the multi-branch neural network model. The multidimensional time series is extracted using a self-attention mechanism. Advanced temporal features High-level temporal features are obtained through a fully connected decoder. Obtain the battery SOC and the advanced timing features The nonlinear mapping relationship between them is used to output a preliminary SOC estimate. ; in, —Input data matrix; —The length of the time series; —Feature dimensions for each time step; —Feature dimensions of the Transformer model; Physical law embedding and voltage verification branch, used based on the estimated value and measured current Generated terminal voltage prediction value ; The predicted value of the generated terminal voltage In the step, based on the estimated value and measured current Generated terminal voltage prediction value : in, ; in, —Boundary open-circuit voltage value and the estimated value Mapping functions between them; —Internal resistance, guaranteed to be positive by the Softplus activation function; —Polarization voltage, limited by Tanh activation range; Run the modal recognition branch to determine the current operating state of the battery through the multidimensional time-series features.

[0008] One possible approach is that the multidimensional temporal features include: Basic physical quantity feature set, rate of change feature set, and operating condition identification feature set; Among them, the basic physical quantity characteristic set includes: instantaneous power, cumulative capacity, and cumulative energy; Rate of change feature set: voltage rate of change, current rate of change, power rate of change; Operating condition identification feature set: charging and discharging step code, number of cycles.

[0009] One possible approach is to optimize the battery SOC estimation model using a multi-task loss function; The multi-task loss function includes: ; in, —The mean squared error loss of the SOC prediction relative to the true label, based on the preliminary SOC estimate. Obtain; —The mean square error loss of the voltage prediction value relative to the true tag, based on the generated voltage prediction value. Obtain; —The cross-entropy loss of running modality recognition is obtained based on the output of the running modality recognition branch; —Loss of physical consistency constraints; —First weighting coefficient; —Second weighting coefficient; —Third weighting coefficient; —Fourth weighting coefficient.

[0010] One possible approach is the loss due to the physical consistency constraint. In the calculation steps, The estimated value based on the output of the multi-branch neural network model Total change The capacity consistency loss is calculated based on the current change rate. ; Based on the SOC observations corresponding to the training labels in the training dataset Battery capacity and the physical consistency loss calculated by the current change rate ; Loss of capacity consistency and physical consistency loss The weighted summation yields the physical consistency constraint loss. .

[0011] One possible approach is the capacity consistency loss. The calculation is as follows: ; in, —Time Window Absolute value of internal current The integral; —The preliminary SOC estimate predicted by the multi-branch neural network model The range of variation.

[0012] One possible approach is the loss of physical consistency. The calculation is as follows: ; in, — The SOC observation value at time step is obtained based on the training labels; —Battery nominal capacity; —From the initial moment arrive The net current integral.

[0013] Secondly, embodiments of this application provide a battery SOC estimation device based on hierarchical physical constraints and a multi-branch architecture, applied to a battery SOC estimation model, including: Training module: Used to build a battery SOC estimation model; Optimization module: used to train and optimize the battery SOC estimation model based on the multi-task loss function to obtain the optimized battery SOC estimation model.

[0014] Thirdly, embodiments of this application provide an electronic device, including: At least one processor; and At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, which can call the program instructions to execute the battery SOC estimation model based on physical constraints and multi-branch architecture as described above.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions that cause the computer to execute the battery SOC estimation model based on physical constraints and a multi-branch architecture as described in any of the preceding claims.

[0016] The embodiments of the present invention bring the following beneficial effects: 1. This invention significantly improves the robustness of the model to complex operating conditions and time-varying characteristics of battery degradation by integrating the equivalent circuit model with the physical information neural network and introducing hierarchical physical constraints. It overcomes the problem of weak nonlinear mapping caused by the flat voltage plateau of LTO batteries and effectively improves the accuracy and generalization ability of battery state of charge estimation.

[0017] 2. This invention adopts a multi-branch collaborative architecture, which rigidly embeds the law of conservation of charge in a differentiable form into the neural network training process, ensuring that the estimation results strictly comply with physical laws, realizing a deep integration of physical laws and data-driven approaches, and enhancing the interpretability of the model and the credibility of practical applications.

[0018] 3. This invention constructs hierarchical constraints through macroscopic capacity consistency loss and microscopic physical consistency loss, and dynamically estimates key parameters such as internal resistance and polarization voltage, thereby achieving coordinated optimization of battery state and parameters, providing strong support for the refined management of energy storage systems.

[0019] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This invention provides a structural diagram of a battery SOC estimation model based on physical constraints and a multi-branch architecture. Figure 2 A flowchart illustrating a battery SOC estimation model construction method based on physical constraints and a multi-branch architecture, provided in an embodiment of the present invention; Figure 3 Connection diagram for the charging and discharging experiment of lithium titanate battery; Figure 4 A flowchart for the invented SOC estimation method; Figure 5 A comparison chart showing the performance results of different constraint models; Figure 6 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] This invention achieves a deep integration of physical laws and data-driven approaches, solving problems such as easy parameter drift in traditional physical models, lack of physical constraints in pure data-driven models, and difficulty in adapting the standard PINN framework to the hybrid dynamic characteristics of batteries. It significantly improves the accuracy, robustness, and physical reliability of SOC estimation, effectively overcomes the problem of weak nonlinear mapping caused by the flat voltage plateau of LTO batteries, adapts to complex operating conditions and time-varying characteristics of battery degradation, and provides an effective technical solution for the refined battery management of energy storage systems. It has good application prospects in scenarios such as high-speed train energy storage systems.

[0025] To facilitate understanding of this embodiment, a detailed description of a battery SOC estimation model based on physical constraints and a multi-branch architecture disclosed in this embodiment of the invention will be provided first, such as... Figure 1-2 The battery SOC estimation model shown includes: The branches include: time-series feature extraction and battery SOC prediction, physical law embedding and voltage verification, and operating mode recognition. The temporal feature extraction and battery SOC prediction branch is used to extract the multidimensional temporal features and output a preliminary SOC estimate of the battery. ; Specifically, in the temporal feature extraction and preliminary SOC prediction branch, a Transformer encoder is used as the backbone network, with a multidimensional time series as input. ,in, Represents the input data matrix. Represents the length of the time series (number of time steps). The feature dimension representing each time step (including multi-dimensional features such as current, voltage, and power).

[0026] Extracting high-level temporal features using a self-attention mechanism ,here This represents the feature dimension of the Transformer model.

[0027] A preliminary SOC estimation is achieved using a fully connected decoder: This module is responsible for learning the complex nonlinear mapping relationship between SOC and the input sequence, extracting high-level temporal features, and outputting a preliminary SOC estimate. .

[0028] Physical law embedding and voltage verification branch, used based on the estimated value and measured current Generated terminal voltage prediction value ; Embedded and Voltage Verification Branch: Constructs a differentiable physical computation module based on an equivalent circuit model, whose core equation is... .

[0029] in The nonlinear relationship with SOC is learned by a multilayer perceptron using an MLP structure. The nonlinear mapping is learned for the input, and the recorded boundary open-circuit voltage values ​​are used. and Strong constraints are applied as boundary conditions to ensure physical realism in fully charged and uncharged states, i.e., when When the value approaches 100% or 0%, the network output approximates the corresponding boundary OCV value. Internal resistance With polarization voltage The data is dynamically estimated by two lightweight MLP networks, with inputs containing Transformer features. Normalization and real-time current , The Softplus activation function ensures a positive value. Activate the limited range using Tanh.

[0030] The output of this branch The difference between the measured value and the actual terminal voltage, calculated strictly using physical formulas, constitutes a physical verification signal.

[0031] The modal recognition branch is used to determine the current operating state of the battery through the multi-dimensional time-series features, and to identify whether the current sequence belongs to the charging or discharging state through the classification network, thus providing the model with operating condition context information.

[0032] For example, an 8-string lithium titanium oxide battery pack with a nominal capacity of 20Ah was selected and subjected to accelerated aging cycle testing in a constant temperature environment of 25°C. The test procedure is as follows: Figure 4The test involved charging at a constant current of 120A to a total voltage of 20.4V, then switching to constant voltage charging until the current dropped to 0.4A (cutoff), followed by a 2-minute rest period. This was then followed by discharging at a constant current of 120A to a total voltage of 15.2V (cutoff), followed by a 5-minute rest period. This cycle was repeated until the battery pack's health status dropped to 80%. During the test, raw sequence data of the total voltage V, total current I, and timestamp t were collected at a sampling frequency of 1Hz. The stable terminal voltage at the end of each cycle's rest period was recorded as a reference value for the boundary open-circuit voltage. and .

[0033] Collect data from the cyclic charge-discharge experiment of lithium titanate batteries. The connection of the lithium titanate battery charge-discharge experiment process is as follows: Figure 3 An 8-string lithium titanium oxide battery pack with a nominal capacity of 20Ah and a nominal voltage of 2.3V was selected as the test object. In a constant temperature 25℃ explosion-proof high and low temperature test chamber (model MGDW-225-40BC), a specific charge-discharge cycle was set and executed using a battery testing system (model CE-6008n-100V200A-H) to accelerate its aging. The specific process was as follows: first, the battery pack was charged at a constant current of 120A (6C) until the total voltage reached 20.4V; then, it was switched to a constant voltage of 20.4V until the current decreased to 0.4A (cutoff), after which it was allowed to rest for 2 minutes; next, it was discharged at a constant current of 120A until the total voltage dropped to 15.2V (cutoff), after which it was allowed to rest for 5 minutes. This charge-discharge cycle was repeated until the battery pack's health status dropped to a predetermined threshold of 80%.

[0034] The raw operational data was collected and stored synchronously. Throughout the experiment, the host computer, using a fixed sampling frequency of 1Hz, synchronously collected and stored the total voltage (V), total current (I), and corresponding timestamp (t) of the battery pack at each sampling moment, forming the raw voltage and current time series dataset for subsequent model building and analysis. The boundary open-circuit voltage reference value was recorded.

[0035] At the end of the resting phase of each charge-discharge cycle, the terminal voltage of the battery pack was recorded. The stable voltage value during the resting phase after charging was recorded as the open-circuit voltage reference value under full charge conditions. The stable voltage value during the resting period after discharge is recorded as the reference value of the open-circuit voltage under venting conditions. This provides key physical boundary constraints for subsequent models.

[0036] In this embodiment, the extraction of temporal features, the embedding of physical laws, and the identification of operating modes are separated yet work together to achieve a deep integration of data-driven and physical laws. This solves the problems of traditional pure data-driven models lacking physical constraints and pure physical model parameters being prone to drift, thereby improving the physical credibility and interpretability of the model estimation results.

[0037] The temporal feature extraction branch uses the self-attention mechanism of the Transformer encoder to extract advanced temporal features, which can accurately capture the complex temporal correlation characteristics of multi-dimensional battery time-series data, adapt to the dynamic changes in battery operation, improve the accuracy of preliminary SOC estimation, and solve the problem that single-time-scale modeling cannot utilize multi-scale correlation information.

[0038] The physical law embedding and voltage verification branch constructs a differentiable physical calculation module based on the equivalent circuit model. It uses Softplus activation to ensure positive internal resistance and Tanh activation to limit the polarization voltage range. At the same time, it combines the mapping relationship between OCV and SOC to generate terminal voltage prediction values, thereby realizing physical verification of the SOC estimation results and constraining the model output to conform to the actual physical characteristics of the battery.

[0039] The modal recognition branch can accurately determine the current operating state of the battery, providing operating condition context information for SOC estimation, enabling the model to adapt to different operating conditions such as charging and discharging, and improving the robustness of the model under complex operating conditions.

[0040] One possible approach is that the multidimensional temporal features include: Basic physical quantity feature set, rate of change feature set, and operating condition identification feature set; Among them, the basic physical quantity characteristic set includes: instantaneous power, cumulative capacity, and cumulative energy; Rate of change feature set: voltage rate of change, current rate of change, power rate of change; Operating condition identification feature set: charging and discharging step code, number of cycles.

[0041] Specifically, based on the original voltage and current time series data, multi-dimensional time series features are derived through calculation, including: instantaneous power, cumulative capacity and energy; rate of change of voltage, current and power; and encoding of charging and discharging steps and cycle count. Using the known SOC state corresponding to the resting voltage at the end of the cycle as a benchmark, the reference SOC value at each sampling time is calculated using the Coulomb integral method as a training label to construct a time series feature training dataset.

[0042] In this embodiment, a multi-dimensional and multi-type time-series feature set is constructed, covering three major categories of features: basic physical quantities, rate of change, and operating condition indicators. This breaks through the limitations of traditional models that only use single voltage and current features, and comprehensively captures the physical state, dynamic change trend, and operating condition background information of the battery. This provides richer and more effective feature support for SOC estimation and improves the effectiveness of feature input.

[0043] The basic physical quantity feature set reflects the core state of the battery from the dimensions of energy and capacity, the rate of change feature set captures the dynamic changes in battery operation, and the operating condition identification feature set marks the battery charging and discharging steps and aging degree. The three types of features complement each other, enabling the model to learn the correlation between battery SOC and operating state more accurately, and further improve the accuracy of SOC estimation.

[0044] Based on the original data, multi-dimensional features are derived, enabling in-depth mining and utilization of the original data without additional hardware acquisition costs. This improves model performance on the basis of existing data and has economic benefits for practical applications.

[0045] In conjunction with the first aspect, the present invention provides a possible implementation of the first aspect, wherein the battery SOC estimation model is optimized using a multi-task loss function; The multi-task loss function includes: ; in, —The mean squared error loss of the SOC prediction relative to the true label, based on the preliminary SOC estimate. Obtain; —The mean square error loss of the voltage prediction value relative to the true tag, based on the generated voltage prediction value. Obtain; —The cross-entropy loss of running modality recognition is obtained based on the output of the running modality recognition branch; —Loss of physical consistency constraints; —First weighting coefficient; —Second weighting coefficient; —Third weighting coefficient; —Fourth weighting coefficient.

[0046] In this embodiment, a multi-task loss function is designed to achieve comprehensive optimization of the model. The losses of SOC prediction, voltage prediction, operating mode recognition, and physical consistency constraints are weighted and fused, allowing the model to optimize multiple task objectives simultaneously during training. This enables collaborative training and performance improvement of each branch, solving the problem that traditional single loss functions only optimize SOC prediction and easily ignore physical constraints and other tasks.

[0047] By introducing different weight coefficients to adjust each loss term, the importance of each task can be flexibly adjusted according to the actual application scenario (such as different battery types and different operating conditions), so that the model can adapt to different application needs and improve the flexibility and versatility of the model.

[0048] Meanwhile, the physical consistency constraint loss is incorporated into the total loss function to achieve rigid guidance of physical laws on model training. This ensures that the model strictly adheres to physical laws while fitting the data, avoiding model outputs that violate the physical characteristics of the battery and improving the reliability of the model estimation results.

[0049] By integrating voltage prediction loss and modal recognition loss, multi-dimensional verification and constraints on SOC estimation results are achieved. Voltage prediction results can verify the physical rationality of SOC estimation, while modal recognition results can ensure the model's adaptability to operating conditions. Multi-dimensional constraints make model training more accurate, thereby improving the accuracy and robustness of the final SOC estimation.

[0050] Based on the foregoing embodiments, the physical consistency constraint loss In the calculation steps, The estimated value based on the output of the multi-branch neural network model Total change The capacity consistency loss is calculated based on the current change rate. ; Based on the SOC observations corresponding to the training labels in the training dataset Battery capacity and the physical consistency loss calculated by the current change rate ; and The two loss terms, weighted and summed, are embedded with the law of charge conservation from the macroscopic integral scale and the microscopic differential scale, respectively, forming a hierarchical physical constraint framework. Through backpropagation, all loss terms are jointly optimized, ensuring that the neural network's parameter updates are guided and corrected by the closed-loop laws of physics while fitting data features.

[0051] In this embodiment, a hierarchical physical consistency constraint loss is constructed, embedding the law of charge conservation from two scales: macroscopic capacity consistency and microscopic physical consistency. This breaks through the limitation of traditional physical constraints modeling only at a single scale, and realizes comprehensive, multi-scale constraints on battery physical laws. This allows the model to satisfy battery physical characteristics at both the macroscopic integral and microscopic differential scales, thereby improving the effectiveness of physical constraints.

[0052] Capacity consistency loss constrains the ratio of SOC change to current consumption at the macroscopic scale, while physical consistency loss constrains the relationship between instantaneous SOC and current integral at the microscopic scale. The combination of the two forms a closed-loop physical constraint framework, ensuring that the SOC value predicted by the model conforms to the law of charge conservation across the entire time scale, thus avoiding macroscopic or microscopic physical deviations in the model.

[0053] By integrating the two types of physical losses through weighted summation, the weights of macroscopic and microscopic constraints can be adjusted according to actual needs, adapting to the characteristics of different batteries and application scenarios, and improving the flexibility of physical constraints.

[0054] Decomposing the physical consistency constraint loss into two sub-losses for separate calculation and fusion allows for more refined and targeted design of physical constraints, facilitating precise control of the physical constraint effect on the model and enabling clearer analysis of the impact of physical constraints at different scales on model performance.

[0055] One possible approach is the capacity consistency loss. The calculation is as follows: This loss is based on the integral form of the Coulomb counting method, which forces the model to predict a physical proportional relationship between the total change in SOC and the integral of the absolute value of the current (total power consumption): ; in, For time window Integral of the absolute value of the internal current Here, 3600 represents the range of SOC variation predicted by the model, and 3600 is the time-to-second conversion factor. By minimizing the variance of this ratio, the model is constrained to maintain consistency in the estimated effective capacity of batteries in the same batch, ensuring the physical consistency between SOC variation and energy consumption on a macroscopic scale.

[0056] In this embodiment, capacity consistency loss is designed based on the coulomb counting method, which forces the model to predict the total change in SOC and the integral of the absolute value of current to satisfy the physical proportional relationship. The law of conservation of charge is strictly followed from the macroscopic integration scale, ensuring that the model's estimation of the effective capacity of batteries in the same batch is consistent, and avoiding the problem of mismatch between SOC change and actual energy consumption.

[0057] By minimizing the variance of the ratio to achieve capacity consistency constraints, the model can maintain a stable capacity ratio relationship in the SOC estimation under different time windows, improve the estimation stability of the model during long-term operation, and solve the problem of capacity estimation drift in traditional models during battery cycling and aging.

[0058] Introducing a second-hour conversion factor ensures unit uniformity, guarantees the physical rationality and numerical accuracy of loss calculation, avoids calculation deviations caused by inconsistent units, and improves the reliability of the loss function.

[0059] The calculation of capacity consistency loss is based on the model-predicted SOC change and measured current data, without the need for additional label data. It achieves macroscopic physical constraints by utilizing existing data, thereby reducing the label dependence of model training.

[0060] One possible approach is the physical consistency loss. The calculation is as follows: ; in, for SOC observations at time (reference label). This refers to the battery's nominal capacity (Ah). From the initial moment arrive The net current integral. By minimizing from any time... The variance of the initial SOC is calculated by inversely inferring the current integral, ensuring that the SOC value predicted by the model satisfies the relationship at every instant.

[0061] In this embodiment, physical consistency loss is designed from the microscopic instantaneous scale to constrain the relationship between the SOC observation value and the net current integral at each moment, ensuring that the SOC predicted by the model conforms to the law of charge conservation at any instant. This makes up for the lack of constraint of macroscopic capacity consistency loss at the microscopic instantaneous scale and achieves full-scale coverage of battery physical laws.

[0062] By minimizing the variance of the back-calculated initial SOC, the model's SOC estimate at any time can maintain a stable physical relationship with the current integral, improving the accuracy and stability of the SOC estimation during dynamic operation (such as when the current changes abruptly), and solving the problem of large estimation deviation in traditional models when the current changes dynamically.

[0063] Loss calculation combines the battery's nominal capacity, SOC observation value, and measured current data, directly linking the battery's core physical parameters with actual operating data. This makes the physical constraints more closely match the battery's actual operating characteristics and improves the effectiveness of the constraints.

[0064] The physical consistency loss is calculated based on the SOC observations of the training labels. It can use the label data to accurately calibrate the microscopic physical constraints of the model, allowing the model to more accurately comply with microscopic physical laws while learning data features, thereby further improving the accuracy of SOC estimation.

[0065] In some embodiments, in step S5, the trained model is applied to similar LTO battery packs. By inputting the voltage and current timing data during operation, a high-precision and physically reliable SOC estimate can be output. The proposed method has achieved significant results in experimental evaluation. Under discharge conditions, the mean absolute error (MAE) and root mean square error (RMSE) of the SOC estimate are as low as 0.8039% and 1.1256%, respectively, demonstrating the model's high-precision ability to capture battery discharge dynamics. Even under charging conditions, where the voltage plateau is flatter and estimation is more difficult, the MAE and RMSE of the SOC estimate remain stable at 1.2785% and 4.3627%, respectively, showing the model's good robustness. Meanwhile, the predicted terminal voltage, which is the core output of the physical branch, has an average absolute error of less than 0.051V and a root mean square error of less than 0.092V under all operating conditions. This strongly verifies the calculation accuracy and the effectiveness of the physical law constraints of the proposed equivalent circuit model branch, indicating that while achieving high-precision SOC estimation, the model ensures that the prediction of key physical quantities conforms to the real dynamic characteristics of the battery.

[0066] pass Figure 5 The ablation experiment radar charts shown provide a systematic analysis of the model, fully validating the effectiveness of the multi-layered technical architecture proposed in this application. After removing the strong OCV boundary constraints, the model's performance degrades in the voltage plateau region, confirming the crucial role of this constraint mechanism in ensuring physical realism under extreme SOC conditions. Furthermore, removing either the microscopic physical consistency loss or the macroscopic capacity consistency loss results in varying degrees of degradation in both SOC accuracy and voltage prediction stability, highlighting the necessity of the designed hierarchical physical constraint framework for maintaining physical self-consistency across different spatiotemporal scales. In contrast, the benchmark model with all physical constraints removed (only boundary conditions retained) performs poorly across the board, not only demonstrating the value of each constraint term but also intuitively showcasing the fundamental advantage of deep embedding of physical laws compared to purely data-driven or posterior-based verification. In summary, the various ablation experiments, from the opposite perspective, prove that the proposed "OCV boundary constraints + dynamic parameter network + hierarchical physical loss" is an organic whole, and its synergistic effect is the core guarantee for achieving high-precision and physically reliable SOC estimation.

[0067] In summary, this invention proposes a complete prediction process that effectively solves the problem of SOC estimation caused by the flat voltage plateau of LTO batteries, and significantly improves the physical reliability and cross-condition generalization ability of the model while ensuring high accuracy.

[0068] This invention addresses the technical challenges of flat voltage plateaus, difficult SOC estimation, complex operating conditions, and high model generalization requirements in LTO batteries of high-speed train energy storage systems. It innovatively integrates an equivalent circuit physical model with a deep neural network. The proposed method constructs a collaborative branch architecture of "data-driven prediction - physical model verification - modal recognition," achieving not only high-precision SOC estimation but also rigidly embedding the law of charge conservation into the learning process in a hierarchical constraint form, significantly enhancing the model's interpretability and physical reliability. Experimental results show that, under charging and discharging conditions, the average absolute error of SOC estimation is as low as 1.2785% and 0.8039%, respectively, with the terminal voltage prediction error remaining within 0.1V. Further verification through systematic ablation experiments confirms that the introduced OCV boundary constraints, dynamic parameter network, and hierarchical physical loss are key components for improving model performance. This method exhibits excellent accuracy and robustness under limited data conditions, providing an effective technical solution for highly reliable and intelligent battery management in high-speed train energy storage systems.

[0069] Secondly, embodiments of this application provide a battery SOC estimation device based on hierarchical physical constraints and a multi-branch architecture, applied to a battery SOC estimation model, including: Training module: Used to build a battery SOC estimation model; Optimization module: used to train and optimize the battery SOC estimation model based on the multi-task loss function to obtain the optimized battery SOC estimation model.

[0070] In this embodiment, a device architecture for the training module and the optimization module is designed to achieve standardized and modular construction and optimization of the battery SOC estimation model. The model construction and training optimization are separated into two core modules, which facilitates the functional decomposition, development and maintenance of the device and improves the engineering feasibility of the device.

[0071] The training module is responsible for building the model, which can quickly build a three-branch neural network architecture, adapt to multi-dimensional feature inputs and multi-branch collaborative model design, and reduce the technical threshold for model building.

[0072] The optimization module trains and optimizes the model based on a multi-task loss function. It can automatically achieve collaborative training of each branch and weighted optimization of multi-objective loss without the need for manual control of the training process, thus improving the automation and efficiency of model training.

[0073] The modular device architecture enables independent upgrades and iterations of each module. For example, the model architecture of the training module or the loss function of the module can be optimized separately without reconstructing the entire device, thus improving the scalability and iterativeness of the device and adapting to subsequent technology upgrades and changes in application requirements.

[0074] Thirdly, embodiments of this application provide an electronic device, including: At least one processor; and At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, which can invoke the program instructions to perform the method as described in the first aspect.

[0075] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions that cause the computer to perform the method described in the first aspect.

[0076] The device provided in this embodiment of the invention has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0077] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0078] Figure 6 A block diagram is shown that is suitable for implementing embodiments of the present invention. Figure 6 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0079] like Figure 6 As shown, the electronic device is represented in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors 410, memory 430, and communication bus 440 connecting different system components (including memory 430 and processing unit 410).

[0080] Communication bus 440 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MAC) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0081] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0082] Memory 430 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 430 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of various embodiments of the present invention.

[0083] A program / utility having a set (at least one) of program modules can be stored in memory 430. Such program modules include—but are not limited to—an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of this invention.

[0084] Processor 410 executes various functional applications and data processing by running programs stored in memory 430, such as implementing embodiments of the present invention. Figure 1 The method provided in the illustrated embodiment.

[0085] This invention provides a non-transitory computer-readable storage medium that stores computer instructions, which cause the computer to execute embodiments of this invention. Figure 1 The method provided in the illustrated embodiment.

[0086] The aforementioned computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0087] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0088] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0089] Computer program code for performing the operations of embodiments of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0090] The foregoing has described specific embodiments of the present invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of the different embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of the present invention, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0093] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0094] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0095] It should be noted that the terminals involved in the embodiments of the present invention may include, but are not limited to, personal computers (PCs), personal digital assistants (PDAs), wireless handheld devices, tablet computers, mobile phones, MP3 players, MP4 players, etc.

[0096] In the several embodiments provided in this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0097] Furthermore, in the various embodiments of the present invention, the functional units can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0098] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A battery SOC estimation model based on physical constraints and a multi-branch architecture, characterized in that, The battery SOC estimation model includes: The branches include: time-series feature extraction and battery SOC prediction, physical law embedding and voltage verification, and operating mode recognition. The temporal feature extraction and battery SOC prediction branch is used to extract the multidimensional temporal features and output a preliminary SOC estimate of the battery. ; The time-series feature extraction and battery SOC prediction branch is specifically configured as follows: obtaining multi-dimensional time series data from the Transformer encoder in a multi-branch neural network model. The multidimensional time series is extracted using a self-attention mechanism. Advanced temporal features High-level temporal features are obtained through a fully connected decoder. Obtain the battery SOC and the advanced timing features The nonlinear mapping relationship between them is used to output a preliminary SOC estimate. ; in, —Input data matrix; —The length of the time series; —Feature dimensions for each time step; —Feature dimensions of the Transformer model; Physical law embedding and voltage verification branch, used based on the estimated value and measured current Generated terminal voltage prediction value ; in, ; in, —Boundary open-circuit voltage value and the estimated value Mapping functions between them; —Internal resistance, guaranteed to be positive by the Softplus activation function; —Polarization voltage, limited by Tanh activation range; Run the modal recognition branch to determine the current operating state of the battery through the multidimensional time-series features.

2. The battery SOC estimation model based on physical constraints and multi-branch architecture according to claim 1, characterized in that, The multidimensional temporal features include: Basic physical quantity feature set, rate of change feature set, and operating condition identification feature set; The basic physical quantity characteristic set includes: instantaneous power, cumulative capacity, and cumulative energy; The rate of change feature set includes: voltage rate of change, current rate of change, and power rate of change; The operating condition identification feature set includes: charging and discharging step codes and cycle count.

3. The battery SOC estimation model based on physical constraints and multi-branch architecture according to claim 1, characterized in that, The battery SOC estimation model is optimized using a multi-task loss function; The multi-task loss function includes: ； in, —The mean squared error loss of the SOC prediction relative to the true label, based on the preliminary SOC estimate. Obtain; —The mean square error loss of the voltage prediction value relative to the true tag, based on the generated voltage prediction value. Obtain; —The cross-entropy loss of running modality recognition is obtained based on the output of the running modality recognition branch; —Loss of physical consistency constraints; —First weighting coefficient; —Second weighting coefficient; —Third weighting coefficient; —Fourth weighting coefficient.

4. The battery SOC estimation model based on physical constraints and multi-branch architecture according to claim 3, characterized in that, The physical consistency constraint loss In the calculation steps, The estimated value based on the output of the multi-branch neural network model Total change The capacity consistency loss is calculated based on the current change rate. ; Based on the SOC observations corresponding to the training labels in the training dataset Battery capacity and the physical consistency loss calculated by the current change rate ; Loss of capacity consistency and physical consistency loss The weighted summation yields the physical consistency constraint loss. .

5. The battery SOC estimation model based on physical constraints and multi-branch architecture according to claim 4, characterized in that, The capacity consistency loss The calculation is as follows: ； in, —Time Window Absolute value of internal current The integral; —The preliminary SOC estimate predicted by the multi-branch neural network model The range of variation.

6. The battery SOC estimation model based on physical constraints and multi-branch architecture according to claim 5, characterized in that, The physical consistency loss The calculation is as follows: ； in, — The SOC observation value at time step is obtained based on the training labels; —Battery nominal capacity; —From the initial moment arrive The net current integral.

7. A battery SOC estimation device based on hierarchical physical constraints and multi-branch architecture, characterized in that, Applied to battery SOC estimation models, including: Training module: Used to build a battery SOC estimation model; Optimization module: used to train and optimize the battery SOC estimation model based on the multi-task loss function to obtain the optimized battery SOC estimation model.

8. An electronic device, characterized in that, include: At least one processor; as well as At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, and the processor can execute the battery SOC estimation model based on physical constraints and multi-branch architecture as described in any one of claims 1 to 7 by calling the program instructions.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause the computer to execute the battery SOC estimation model based on physical constraints and a multi-branch architecture as described in any one of claims 1 to 7.

Images