Battery parameter identification and state-of-charge estimation method based on physical information neural network

Abstract

The invention relates to a battery parameter identification and state-of-charge estimation method based on a physical information neural network, and belongs to the field of battery operation estimation, and the method comprises the following steps: collecting electric vehicle battery operation data; the collected data are preprocessed; the method comprises the following steps: constructing an OCV curve mapping coder-decoder taking a multilayer perceptron as a backbone, carrying out pre-training in a source domain through a voltage-capacity sequence, and outputting an OCV curve by virtue of a moment-by-moment feature extraction function of the coder and the decoder; migration learning is introduced, the characterization capability of a source domain pre-training encoder is reserved, a polynomial obtained by fitting real vehicle discharge data of a target domain is used as a pseudo label, and soft supervision fine tuning is carried out on a decoder; constructing a parameter estimator for dynamic parameter identification and an SOC estimator for data-mechanism fusion; and training two estimators by using the preprocessed data, and finally, identifying battery parameters by using the trained model, and estimating the SOC.

Description

Technical Field

[0001] This invention belongs to the field of battery operation estimation technology, and relates to a method for battery parameter identification and state of charge estimation based on physical information neural network. Background Technology

[0002] To achieve health management of power batteries, researchers have proposed various state indicators, such as State of Charge (SOC), State of Health (SOH), and Remaining Life (RUL). Among them, SOC, as a core indicator measuring the relationship between the battery's current usable capacity and its rated capacity, is closely related to the battery's safe operating boundary, available energy, power output capability, and lifespan characteristics. Accurate SOC estimation is not only a prerequisite for energy management and power distribution but also provides necessary support for battery equalization control, thermal management, and remaining life prediction. Therefore, SOC is widely regarded as a benchmark variable in battery management systems (BMS), playing a central role in power battery state monitoring, health assessment, and intelligent decision-making.

[0003] Whether based on physical models or data-driven approaches, much current research still relies on obtaining open-circuit voltage (OCV) curves through intermittent or low-current charge-discharge tests under static laboratory conditions. These OCV curves are then pre-stored in the battery management system (BMS) in a lookup table format for state-of-the-art (SOC) correction. However, since these OCV curves typically originate from controlled laboratory environments, their generalization ability is limited, making it difficult to comprehensively characterize the true electrochemical characteristics of power batteries under different real-world vehicle operating conditions. This, in turn, limits the accuracy and robustness of SOC estimation. Especially in scenarios where controlled static laboratory testing is not feasible, it is difficult to obtain OCV curves that accurately match actual operating conditions in a timely manner. Therefore, designing an OCV curve mapping encoder-decoder oriented towards real-world operating conditions, capable of obtaining OCV curves with high physical consistency and fitting accuracy even in the absence of controlled laboratory testing conditions, has become a current technical challenge.

[0004] In physical model-based methods, such as equivalent circuit models (ECMs), model parameters are significantly affected by factors such as ambient temperature, driving behavior and road conditions, and battery aging. However, a considerable portion of existing research still assumes that these parameters remain constant throughout the entire operation, obtaining fixed parameter values ​​only through one-time offline experiments. This fails to reflect the time-varying characteristics of the parameters in actual operation and can easily lead to systematic biases in SOC estimation. Although methods based on physical information neural networks achieve dynamic parameter identification to some extent by embedding system control equations into network constraints, these methods require constructing multi-objective coupled loss mechanisms and jointly optimizing hidden states and parameters. The modeling process is complex and computationally expensive, significantly limiting their application in large-scale engineering projects. Therefore, it is necessary to construct a physical information network model with a simpler structure, lower computational cost, and the ability to dynamically identify parameters while maintaining physical constraints.

[0005] In existing data-driven methods, especially neural network methods, the network input typically uses multi-dimensional data such as battery current, voltage, temperature, and even capacity to achieve high SOC estimation accuracy. This reliance on diverse data sources increases dependence on onboard sensors, leading to increased cost and system complexity. Furthermore, the increasing number of onboard sensors exacerbates the risk of data loss due to sensor aging or data transmission line failures, ultimately impacting the performance of neural networks in practical deployments. Therefore, it is crucial to design an SOC estimation model that requires less dimensional input data and is less dependent on onboard sensors, while maintaining SOC estimation accuracy, to mitigate the costs and risks associated with multi-dimensional inputs.

[0006] Therefore, it is urgent to design a physical information network model for battery parameter identification and charge estimation to solve the above problems. Summary of the Invention

[0007] In view of this, the purpose of this invention is to provide a method for battery parameter identification and state of charge estimation based on a physical information neural network.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for battery parameter identification and state of charge estimation based on physical information neural networks includes the following steps: S1: Collects operating data of electric vehicle batteries; S2: Preprocess the collected data; S3: Construct an OCV curve mapping encoder-decoder with a multilayer perceptron as the backbone. In the source domain, it is pre-trained using voltage-capacity sequences. The encoder's time-by-time feature extraction function and the decoder output OCV curve are used together. Transfer learning is introduced to retain the representational ability of the pre-trained encoder in the source domain. The polynomial obtained by fitting the actual vehicle discharge data in the target domain is used as a pseudo-label to perform soft supervision fine-tuning of the decoder. S4: Construct a parameter estimator for dynamic parameter identification and a SOC estimator for data-mechanism fusion; S5: Use the preprocessed data to train the two estimators from step S4, and finally use the trained model to identify battery parameters and estimate SOC.

[0009] Furthermore, the preprocessing described in step S2 specifically includes the following steps: Sort the raw data by timestamp; Missing data is processed by removing sample rows with missing charging status flags, using linear interpolation to fill in short-term missing values ​​in current, voltage, and SOC fields, and discarding time periods with continuous missing values. The mean and standard deviation of the total voltage of the vehicle are calculated, and outlier samples are identified and removed with the mean ± 3 times the standard deviation as the threshold. Based on the charging state, time-ordered samples are labeled as charging, discharging, or idle and aggregated into candidate segments. Constraints, segments, and merges are performed by combining time continuity and SOC monotonicity. An amplitude threshold is applied to the candidate segments, and only valid segments with charging ΔSOC not less than 30% and discharging ΔSOC not less than 50% are retained. Within the effective range, peak detection is performed on the median and standard deviation of the current, voltage, and SOC channels using a sliding window. Anomalies are replaced by median or reconstructed by linear interpolation based on the z-score threshold, and this is implemented synchronously at the same time index of the three channels.

[0010] Furthermore, step S3 specifically includes: Using a multilayer perceptron as the backbone network, an encoder-decoder structure consisting of an encoder and a decoder is constructed. The encoder receives the voltage-capacity time series after preprocessing in step S2, extracts the time series features at each time step and maps them into high-dimensional latent feature vectors. The decoder restores the latent feature vectors to the corresponding OCV curves through nonlinear transformation and linear scaling operations involving three fully connected layers. A pseudo-label-based transfer learning method is introduced to preserve the representational ability of the source domain pre-trained encoder. The OCV polynomial obtained by fitting the target domain real vehicle discharge data is used as a pseudo-label to perform soft supervision fine-tuning of the decoder. The pseudo-label is then embedded in the battery management system (BMS) in the form of a lookup table for SOC correction.

[0011] Furthermore, the parameter estimator for dynamic parameter identification takes the current-time series preprocessed in step S2 as input, uses a Long Short-Term Memory (LSTM) network to extract time-by-time features of the current time series and map them into latent time series features. The parameter mapping head outputs the physically interpretable second-order equivalent circuit model (ECM) parameters in real time, and calculates the terminal voltage in conjunction with the second-order ECM. The error between the calculated terminal voltage and the measured terminal voltage is used as the driving signal to make the ECM parameters adaptively update over time, thereby realizing dynamic parameter identification.

[0012] Furthermore, the parameter estimator for dynamic parameter identification specifically includes: The current-time series is input into a Long Short-Term Memory (LSTM) network at each time step, and the LSTM learns the parameters... Extracting high-dimensional nonlinear latent features from historical current and time information under constraints This feature implicitly includes polarization effects, time constants, and nonlinear dynamic physical information; Construct parameter mapping head pairs for features Linear transformations, normalization, and nonlinear activations are performed to compress the high-dimensional temporal features and project them onto a low-dimensional parameter space, resulting in a normalized parameter vector. ; Through affine transformation The set of physical parameters mapped to the second-order equivalent circuit model ECM And utilize pre-defined physical upper and lower bounds right Constraints are imposed to ensure that the parameters of each resistor and capacitor remain within the physically feasible region. Polarization voltage of the two-branch RC network updated based on second-order ECM The time constant is determined by the branch resistance and capacitance; Substituting the polarization resistance and ohmic resistance into the second-order ECM's terminal voltage calculation formula, the model's terminal voltage is reconstructed. ; By comparison The deviation from the measured terminal voltage is used as a physical loss constraint to update the network parameters of the LSTM and parameter mapping head in reverse, thereby realizing the dynamic identification of battery parameters changing with time and operating conditions, as expressed as: .

[0013] Furthermore, the data-mechanism fusion SOC estimator uses the identified ECM parameters and current-time series to estimate the SOC, and together with the OCV curve generated by the OCV curve mapping encoder-decoder, forms a physical constraint, specifically including: For each sampling time, the current-time data is concatenated with the corresponding dynamic ECM parameter set to construct the SOC estimation input vector. The input vector is decomposed into time series and preprocessed to obtain dynamic components that emphasize fast perturbations. With trend components that characterize slow-changing trends The two types of components are mixed and nonlinearly mapped in the time domain, and then input into a time-mixing network to obtain a fused feature representation. Then, the SOC estimate for the next time step is output through a linear layer and a sigmoid activation function. This constrains the SOC estimation result to a reasonable range of 0-1, expressed as:

[0014]

[0015] The SOC estimate is obtained by interpolation using a pre-trained OCV-SOC curve, resulting in the corresponding open-circuit voltage estimate. It is used together with the ohmic resistance and polarization voltage output by the dynamic parameter identification module for terminal voltage reconstruction, thus coupling SOC estimation and voltage observation at the physical level.

[0016] .

[0017] Furthermore, in step S5, after training is complete, the performance of the model is evaluated using RMSE and MAE to obtain the test loss and test accuracy.

[0018] The beneficial effects of this invention are as follows: This invention draws on the core idea of ​​physical information neural networks, simplifies the traditional PINN to adapt to the computation and data constraints of real-time estimation in vehicles, and reduces the dependence on additional sensors such as voltage, temperature, and capacity while maintaining good physical consistency and interpretability, thereby reducing testing costs and the risks caused by sensor aging or failure. It has the advantages of high accuracy, low computational overhead and easy engineering deployment.

[0019] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1This is a diagram illustrating the overall framework of a physical information network method for battery parameter identification and state of charge estimation. Detailed Implementation

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0023] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0024] Example 1: like Figure 1 As shown, this invention provides a physical information network method for battery parameter identification and state of charge estimation, comprising the following steps: S1: Data Acquisition: This invention uses electric vehicle data collected in a laboratory, including current and time.

[0025] S2: Data Processing: The raw battery operation data obtained in S1 is processed sequentially through time sorting, missing data handling, outlier handling, and charge / discharge data segmentation and cleaning. Specifically, the raw data is first sorted by timestamp, then missing data is processed. Sample rows with missing charging status indicators are removed, short-term missing values ​​such as current, voltage, and SOC are filled using linear interpolation, and time periods with a large number of consecutive missing values ​​are discarded entirely. Subsequently, the mean and standard deviation were calculated based on the total vehicle voltage. Outlier samples were identified and removed using a threshold of "mean ± 3 times standard deviation". Based on this, time-ordered samples were labeled as charging, discharging, or idle according to the charging status and aggregated into candidate segments. Constraints and segmentation / merging were performed by combining time continuity (the time interval between adjacent samples is preferably no more than 3600 s) and SOC monotonicity. An amplitude threshold was applied to the candidate segments, retaining only valid segments with charging ΔSOC no less than 30% and discharging ΔSOC no less than 50%. Finally, within the valid segments, peak detection was performed on the median and standard deviation of the current, voltage, and SOC channels using a sliding window. Outliers were replaced by median or reconstructed by linear interpolation based on the z-score threshold. This was performed synchronously at the same time index of the three channels to ensure the consistency of IV-SOC data.

[0026] S3: Pre-trained Model Construction: An OCV curve mapping encoder-decoder with a multilayer perceptron as its backbone is constructed. Pre-training is performed in the source domain using voltage-capacity sequences, leveraging the encoder's time-by-time feature extraction function and the decoder's output OCV curve. Furthermore, to adapt to real-world vehicle operating conditions, transfer learning is introduced. The representational capabilities of the source domain pre-trained encoder are preserved, and the decoder is soft-supervised fine-tuned using a polynomial fitted from the target domain's real-world vehicle discharge data as a pseudo-label.

[0027] Specifically, using a multilayer perceptron as the backbone network, an encoder-decoder structure is constructed, consisting of an encoder and a decoder. The encoder receives the voltage-capacity time series after preprocessing in step S2, extracts time-series features at each time step, and maps them into high-dimensional latent feature vectors. The decoder uses nonlinear transformation and linear scaling operations, including three fully connected layers, to restore the latent features to the corresponding OCV curves. Thus, OCV can be obtained solely based on voltage-capacity data without laboratory static testing. On this basis, a pseudo-label-based transfer learning method is introduced: the representational ability of the source domain pre-trained encoder is preserved, and the OCV polynomial obtained by fitting the target domain real vehicle discharge data is used as a pseudo-label to perform soft-supervised fine-tuning of the decoder. This makes the obtained OCV curve more consistent with real characteristics under various real vehicle operating conditions, and it can be embedded in the battery management system (BMS) in a lookup table format for SOC correction, improving the fitting accuracy and physical consistency of SOC estimation.

[0028] S4: Model Construction: A parameter estimator and a data-mechanism fusion method for SOC estimation were designed for dynamic parameter identification. The ohmic internal resistance and second-order RC branch parameters are output in real time by a neural network estimator and then coupled with the physical equations to reconstruct the terminal voltage, constructing a terminal voltage loss-guided model for dynamic parameter identification. Based on this, the SOC is estimated by inputting the current-time series data into a TimeMixer.

[0029] The parameter estimator takes the current-time series preprocessed in step S2 as input, uses LSTM to extract time-by-time features of the current time series and maps them to potential time series features. The parameter mapping head outputs physically interpretable second-order equivalent circuit model (ECM) parameters in real time, and calculates the terminal voltage in conjunction with the second-order ECM. The error between the calculated terminal voltage and the measured terminal voltage is used as the driving signal to make the ECM parameters adaptively update over time, thereby realizing the dynamic identification of time-varying effects caused by factors such as temperature, driving behavior and battery aging.

[0030] The data-mechanism fusion SOC estimation method explicitly embeds the basic physical laws of the second-order ECM into the parameter estimator and the SOC estimator. The SOC estimator uses the identified ECM parameters and current-time series to estimate the SOC, and together with the OCV curve generated by the OCV curve mapping encoder-decoder, it forms a physical constraint to ensure the physical consistency and interpretability of the SOC estimation process. During the training phase, there is no need to construct and optimize the multi-objective physical residuals of PINN or perform hidden state-parameter joint optimization, thereby reducing modeling complexity and computational overhead. Furthermore, it reduces the dependence on additional sensors such as voltage, temperature, and capacity when only two-dimensional current and time data are input, thereby reducing costs and mitigating the risk of data quality and missing data caused by sensor aging or failure.

[0031] S5: Input the processed data into the model to complete the model training. Use RMSE and MAE to evaluate the model's performance to obtain the test loss and test accuracy. Finally, use the trained model to identify battery parameters and estimate the state of charge.

[0032] Example 2: In this embodiment, the physical information network method for battery parameter identification and state of charge estimation includes the following steps: 1) This invention uses two electric vehicles equipped with ternary lithium (NMC) power batteries as test subjects. The battery packs have a rated capacity of 150 Ah and are configured as 96 series × 3 parallel. Dataset #1 was collected in a controlled laboratory environment. Through capacity calibration experiments, the battery packs were fully charged from low SOC to 100% SOC and then discharged to the minimum SOC threshold set by the BMS to obtain a usable capacity benchmark under standard conditions. The charging protocol includes three stages: first, a multi-stage constant current fast charge at 0.8C to increase the SOC to approximately 10%; then, a constant current charge at 0.5C for 70 minutes; and finally, a multi-stage constant current fast charge until SOC = 100%. Dataset #2 comes from the actual operating conditions of another electric vehicle over a one-year period. It was collected online by the onboard BMS at a sampling frequency of 0.1 Hz and uploaded to the cloud. Key variables such as timestamps, total vehicle current, total vehicle voltage, SOC, and BMS reference capacity were recorded at 10-second intervals, serving as the basis for real-world operating condition data for subsequent model training and validation.

[0033] 2) To ensure the model can adapt to real-world operating conditions, this invention first pre-trains the OCV curve mapping encoder-decoder based on discharge data from the University of Maryland's 18650 battery. Each discharge data point contains current, voltage, and time information. The cumulative discharge capacity is obtained through ampere-hour integration and aligned with the corresponding voltage sequence to construct a voltage-capacity trajectory characterizing the battery's electrochemical behavior. The encoder uses this trajectory as input for nonlinear mapping and compression, extracting a latent vector representation characterizing the overall discharge characteristics of the battery. The decoder reconstructs the corresponding open-circuit voltage curve based on the latent features, outputting a complete OCV-SOC function within a uniformly divided SOC range. During the pre-training phase, the OCV curve obtained from laboratory HPPC testing is used as the supervision target. By minimizing the error between the predicted curve and the reference curve, the model effectively captures the steady-state OCV information contained in the discharge data. In the transfer learning phase, discharge data from real vehicle operating conditions is used for retraining. Instead of relying on precise experimental OCV curves, approximate soft supervision labels are obtained based on voltage-SOC samples collected from the vehicle through polynomial fitting. Transfer learning starts from the pre-trained weights and, through small-scale iterative updates, makes the OCV curve output by the decoder consistent with the electrochemical characteristics of the new battery.

[0034] 3) The framework integrates mechanistic and data-driven models, enabling dynamic parameter identification and SOC estimation of the battery. It consists of a cascaded ECM parameter identification module and an SOC estimation module.

[0035] The dynamic parameter identification module takes a current-time series as input and consists only of two-dimensional data: current and time. First, the current-time series is input time-by-time into a Long Short-Term Memory (LSTM) network, which then processes the learnable parameters... Extracting high-dimensional nonlinear latent features from historical current and time information under constraints This feature implicitly contains physical information such as polarization effects, time constants, and nonlinear dynamics. Subsequently, a parameter mapping head is constructed for the feature. Linear transformations, normalization, and nonlinear activations are performed to compress the high-dimensional temporal features and project them onto a low-dimensional parameter space, resulting in a normalized parameter vector. Based on this, affine transformation is used to... The set of physical parameters mapped to a second-order equivalent circuit model (ECM) And utilize pre-defined physical upper and lower bounds right Constraints are applied to ensure that the parameters of each resistor and capacitor remain within the physically feasible region, avoiding solutions that are physically meaningless. This is done after obtaining the parameter set. Then, the polarization voltage of the two-branch RC network is updated based on the second-order ECM. The time constant is determined by the branch resistance and capacitance. Substituting the polarization resistance and ohmic resistance into the second-order ECM's terminal voltage calculation formula, the model's terminal voltage is reconstructed. 。 By comparison The deviation from the measured terminal voltage is used as a physical loss constraint to update the network parameters of the LSTM and parameter mapping head in reverse, thereby realizing the dynamic identification of battery parameters changing with time and operating conditions, and thus characterizing the time-varying effects caused by factors such as temperature, driving behavior and battery aging.

[0036]

[0037] 4) The SOC estimation module operates based on the dynamic parameter identification results. For each sampling time, the current-time data is concatenated with the corresponding dynamic ECM parameter set to construct the SOC estimation input vector. The sequence was decomposed and preprocessed over a time series to obtain dynamic components that emphasize rapid perturbations. With trend components that characterize slow-changing trends This allows for the separation of rapid changes and long-term evolutionary trends in SOC. Subsequently, the two types of components are mixed and nonlinearly mapped in the time domain, and then input into a time-mixing network (TimeMixer) to obtain a fused feature representation. Then, the SOC estimate for the next time step is output through a linear layer and a sigmoid activation function. This ensures that the SOC estimation results are constrained within a reasonable range of 0-1. The overall process can be summarized as follows:

[0038]

[0039] Meanwhile, the SOC estimate is interpolated using a pre-trained OCV-SOC curve to obtain the corresponding open-circuit voltage estimate. It is used together with the ohmic resistance and polarization voltage output by the dynamic parameter identification module for terminal voltage reconstruction, coupling SOC estimation and voltage observation at the physical level, further improving the physical consistency and robustness of SOC estimation.

[0040]

[0041] Example 3: An electronic device, including a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method described in Embodiment 1 when executing the computer program.

[0042] Example 4: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in Embodiment 1.

[0043] Example 5: A computer program product includes a computer program that, when executed by a processor, implements the method described in Example 1.

[0044] In the above embodiments, the reference to "this embodiment" in the specification indicates that a specific feature, structure, or characteristic described in connection with the embodiment is included in at least some embodiments, but not necessarily all embodiments. Multiple appearances of "this embodiment" do not necessarily refer to the same embodiment.

[0045] In the above embodiments, although the invention has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory structures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed. The embodiments of the invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims.

[0046] As will be understood by those skilled in the art, the computer-readable storage medium described in this embodiment allows for the implementation of all or part of the steps in the above method embodiments by computer program-related hardware. The aforementioned computer program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0047] The electronic terminal provided in this embodiment includes a processor, a memory, a transceiver, and a communication interface. The memory and the communication interface are connected to the processor and the transceiver and complete communication between them. The memory is used to store computer programs, the communication interface is used to perform communication, and the processor and the transceiver are used to run the computer programs, so that the electronic terminal performs the steps of the above method.

[0048] In this embodiment, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0049] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0050] This invention can be used in a wide range of general-purpose or special-purpose computing system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices, etc.

[0051] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0052] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for battery parameter identification and state of charge estimation based on a physical information neural network, characterized in that: Includes the following steps: S1: Collects operating data of electric vehicle batteries; S2: Preprocess the collected data; S3: Construct an OCV curve mapping encoder-decoder with a multilayer perceptron as the backbone. In the source domain, it is pre-trained using voltage-capacity sequences. The encoder's time-by-time feature extraction function and the decoder output OCV curve are used together. Transfer learning is introduced to retain the representational ability of the pre-trained encoder in the source domain. The polynomial obtained by fitting the actual vehicle discharge data in the target domain is used as a pseudo-label to perform soft supervision fine-tuning of the decoder. S4: Construct a parameter estimator for dynamic parameter identification and a SOC estimator for data-mechanism fusion; S5: Use the preprocessed data to train the two estimators from step S4, and finally use the trained model to identify battery parameters and estimate SOC.

2. The battery parameter identification and state of charge estimation method based on physical information neural network according to claim 1, characterized in that: The preprocessing described in step S2 specifically includes the following steps: Sort the raw data by timestamp; Missing data is processed by removing sample rows with missing charging status flags, using linear interpolation to fill in short-term missing values ​​in current, voltage, and SOC fields, and discarding time periods with continuous missing values. The mean and standard deviation of the total voltage of the vehicle are calculated, and outlier samples are identified and removed with the mean ± 3 times the standard deviation as the threshold. Based on the charging state, time-ordered samples are labeled as charging, discharging, or idle and aggregated into candidate segments. Constraints, segments, and merges are performed by combining time continuity and SOC monotonicity. An amplitude threshold is applied to the candidate segments, and only valid segments with charging ΔSOC not less than 30% and discharging ΔSOC not less than 50% are retained. Within the effective range, peak detection is performed on the median and standard deviation of the current, voltage, and SOC channels using a sliding window. Anomalies are replaced by median or reconstructed by linear interpolation based on the z-score threshold, and this is implemented synchronously at the same time index of the three channels.

3. The battery parameter identification and state-of-charge estimation method based on physical information neural network according to claim 1, characterized in that: Step S3 specifically includes: Using a multilayer perceptron as the backbone network, an encoder-decoder structure consisting of an encoder and a decoder is constructed. The encoder receives the voltage-capacity time series after preprocessing in step S2, extracts the time series features at each time step and maps them into high-dimensional latent feature vectors. The decoder restores the latent feature vectors to the corresponding OCV curves through nonlinear transformation and linear scaling operations involving three fully connected layers. A pseudo-label-based transfer learning method is introduced to preserve the representational ability of the source domain pre-trained encoder. The OCV polynomial obtained by fitting the target domain real vehicle discharge data is used as a pseudo-label to perform soft supervision fine-tuning of the decoder. The pseudo-label is then embedded in the battery management system (BMS) in the form of a lookup table for SOC correction.

4. The battery parameter identification and state-of-charge estimation method based on physical information neural network according to claim 1, characterized in that: The parameter estimator for dynamic parameter identification takes the current-time series preprocessed in step S2 as input, uses a Long Short-Term Memory (LSTM) network to extract time-by-time features of the current time series and maps them to latent time series features. The parameter mapping head outputs the physically interpretable second-order equivalent circuit model (ECM) parameters in real time, and calculates the terminal voltage in conjunction with the second-order ECM. The error between the calculated terminal voltage and the measured terminal voltage is used as the driving signal to make the ECM parameters adaptively update over time, thereby realizing dynamic parameter identification.

5. The battery parameter identification and state-of-charge estimation method based on physical information neural network according to claim 1, characterized in that: The parameter estimator for dynamic parameter identification specifically includes: The current-time series is input into a Long Short-Term Memory (LSTM) network at each time step, and the LSTM learns the parameters... Extracting high-dimensional nonlinear latent features from historical current and time information under constraints This feature implicitly includes polarization effects, time constants, and nonlinear dynamic physical information; Construct parameter mapping head pairs for features Linear transformations, normalization, and nonlinear activations are performed to compress the high-dimensional temporal features and project them onto a low-dimensional parameter space, resulting in a normalized parameter vector. ; Through affine transformation The set of physical parameters mapped to the second-order equivalent circuit model ECM And utilize pre-defined physical upper and lower bounds right Constraints are imposed to ensure that the parameters of each resistor and capacitor remain within the physically feasible region. Polarization voltage of the two-branch RC network updated based on second-order ECM The time constant is determined by the branch resistance and capacitance; Substituting the polarization resistance and ohmic resistance into the second-order ECM's terminal voltage calculation formula, the model's terminal voltage is reconstructed. ; By comparison The deviation from the measured terminal voltage is used as a physical loss constraint to update the network parameters of the LSTM and parameter mapping head in reverse, thereby realizing the dynamic identification of battery parameters changing with time and operating conditions, as expressed as: 。 6. The battery parameter identification and state-of-charge estimation method based on physical information neural network according to claim 1, characterized in that: The data-mechanism fusion SOC estimator uses the identified ECM parameters and current-time series to estimate SOC, and together with the OCV curve generated by the OCV curve mapping encoder-decoder, forms a physical constraint, specifically including: For each sampling time, the current-time data is concatenated with the corresponding dynamic ECM parameter set to construct the SOC estimation input vector. The input vector is decomposed into time series and preprocessed to obtain dynamic components that emphasize fast perturbations. With trend components that characterize slow-changing trends The two types of components are mixed and nonlinearly mapped in the time domain, and then input into a time-mixing network to obtain a fused feature representation. Then, the SOC estimate for the next time step is output through a linear layer and a sigmoid activation function. This constrains the SOC estimation result to a reasonable range of 0-1, expressed as: The SOC estimate is obtained by interpolation using a pre-trained OCV-SOC curve, resulting in the corresponding open-circuit voltage estimate. It is used together with the ohmic resistance and polarization voltage output by the dynamic parameter identification module for terminal voltage reconstruction, thus coupling SOC estimation and voltage observation at the physical level: 。 7. The battery parameter identification and state-of-charge estimation method based on physical information neural network according to claim 1, characterized in that: In step S5, after training is complete, RMSE and MAE are used to evaluate the model's performance to obtain the test loss and test accuracy.

Images