Lithium-ion battery state estimation method, system, device, and medium

Abstract

This application belongs to the field of lithium-ion battery technology and discloses a method, system, device, and medium for lithium-ion battery state estimation. The method includes: acquiring time-series feature data containing at least voltage, current, and temperature during battery operation; processing the data using either of two neural network models: a first model is a convolutional neural network-long short-term memory network model, which outputs the state of energy (SOE) in response to the input time-series feature data; the second model is a convolutional neural network-gated recurrent unit joint estimation model based on model-independent meta-learning optimization, which outputs both SOE and state of health (SOH) in response to the input time-series feature data and a set of indirect health factors. This application achieves dynamic SOE estimation or co-estimation of SOE and SOH through two models, improving the accuracy and adaptability of battery state estimation.

Description

Technical Field

[0001] This application relates to the field of lithium-ion battery technology, and in particular to a method, system, device and medium for estimating the state of a lithium-ion battery. Background Technology

[0002] Typically, a battery is considered unsuitable for vehicle use and enters a phase of degradation when its capacity decays to 80% of its rated capacity. Forecasts indicate that the scale of power batteries will reach approximately 2.186 million tons by 2030. This massive scale not only risks resource waste and environmental pollution but also poses a serious challenge to sustainable socio-economic development. Against this backdrop, accurate estimation of the battery's State of Energy (SOE) is particularly crucial. SOE estimation can further deduce the battery's performance under low power demand, high power demand, and dynamic power demand, and predict its sustainable operating time. Especially in transient conditions such as braking and acceleration of electric vehicles, the energy and power provided by the battery directly determine the safety and stability of vehicle operation; therefore, accurate SOE assessment is of great significance. Furthermore, as an important parameter for vehicle energy management and optimization, SOE enables reasonable energy allocation, effectively extending the driving range of electric vehicles and improving battery utilization efficiency while meeting vehicle power performance requirements. On the other hand, the State of Health (SOH), another key parameter, reflects the degree of battery performance degradation and remaining lifespan. Researchers typically employ various direct or indirect methods to assess State of Harm (SOH). Battery aging primarily stems from irreversible changes in the properties of the electrolyte, positive electrode, and negative electrode materials, as well as the gradual deterioration of internal component structures. These changes often lead to increased internal impedance and operating temperature, thereby accelerating performance degradation. Summary of the Invention

[0003] To address the aforementioned issues, this application provides a method, system, device, and medium for estimating the state of lithium-ion batteries. It acquires time-series data on voltage, current, and temperature, as well as capacity decay curves from several lithium-ion batteries; designs a convolutional neural network-long short-term memory network (CNN-LSTM) fusion architecture to achieve dynamic SOE estimation at low and normal temperatures; and establishes a CNN-gated recurrent unit (GRU) joint estimation model optimized based on model-independent meta-learning (MAML) to achieve joint estimation of SOE and SOH of lithium-ion batteries.

[0004] According to the first aspect of this application, a method for estimating the state of a lithium-ion battery is provided, the method comprising: Acquire time-series characteristic data during the operation of a lithium-ion battery, wherein the time-series characteristic data includes at least voltage, current, and temperature; The time-series feature data is processed using either the first model or the second model to output battery state estimation results; wherein: The first model is a convolutional neural network-long short-term memory network model. In response to the input time-series feature data, it is processed into a normalized time-series feature matrix. The normalized time-series feature matrix is ​​then processed by the convolutional neural network to obtain a time-series deep feature vector. The time-series deep feature vector is then processed by the long short-term memory network to obtain the energy state of the battery SOE estimate. The second model is a joint estimation model of convolutional neural network and gated recurrent unit based on model-independent meta-learning optimization. In response to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's state of energy (SOE) estimate and state of health (SOH) estimate.

[0005] Furthermore, the convolutional neural network in the first model includes convolutional layers, pooling layers, and fully connected layers, and its computation process is expressed as follows: In the formula: For the first i The output of a convolutional layer; These are the activation functions for the convolutional layer and the fully connected layer, respectively. X This is the normalized time-series feature matrix; This is a convolution operation; W i For the first i The weight matrix of the convolutional layer; b i For the first i The bias of layer convolution; This is the output of the pooling layer; pool This is a pooling operation; b j This is the deviation of the pooling layer; H c This is a temporal depth feature vector; W k This is the weight matrix of the fully connected layer; b k For deviations in fully connected layers; The time-series depth feature vector H c After being expanded by time step, it is compared with the normalized temporal feature matrix. X The corresponding feature vectors in the sequence are concatenated and used as the input sequence for the Long Short-Term Memory network.

[0006] Furthermore, the long short-term memory network in the first model includes a forget gate, an input gate, and an output gate, and its computation process is expressed as follows: In the formula: f t The output of the forget gate; i t The output of the input gate; g t The output of the input node; O t This is the output of the output gate; σ It is the sigmoid function; C t This represents the current state of the cell. h t This represents the intermediate state at the current moment; x t This is the input for the current moment; h t-1 This is the intermediate state of the previous moment; C t-1 This represents the cell state at the previous moment; , , , These correspond to the forget gate, input gate, input node, and output gate, respectively. x t The weight matrix; U f , U i , U g , U o The forget gate, input gate, input node, and output gate correspond to the intermediate states of the previous time step, respectively. h t-1 The weight matrix; , , , These represent the biases for the forget gate, input gate, input node, and output gate, respectively; ⊙ represents the Hadamard product. Current input x tThis includes the feature vector corresponding to the time step in the temporal feature matrix, and the feature sub-vector aligned with that time step after the temporal depth feature vector is expanded by the time step; The intermediate state at the last moment of the LSTM is mapped by a fully connected layer to output an estimated value of the State of Energy (SOE).

[0007] Furthermore, when the second model is used, the set of indirect health factors is extracted based on battery capacity test and internal resistance test data.

[0008] Furthermore, the processing procedure of the second model includes: Based on the aforementioned set of indirect health factors, estimate the current health status. ; The voltage at the current moment Current ,temperature and the health status The inputs are fed into the same joint estimation model to estimate the energy state at the current time. ; Output health status and energy state The joint estimation results.

[0009] Furthermore, the second model performs meta-learning across different battery aging states using the MAML algorithm during the training phase to adapt to the degradation patterns across samples.

[0010] Furthermore, it also includes a data construction step: by conducting aging tests, capacity tests, and internal resistance tests on multiple lithium iron phosphate single cells, and combining them with publicly available datasets, a multi-dimensional battery aging database is constructed for training the first model or the second model.

[0011] According to the second technical solution of this application, a lithium-ion battery state estimation system is provided, the system comprising: The data acquisition module is configured to acquire time-series characteristic data during the operation of a lithium-ion battery, wherein the time-series characteristic data includes at least voltage, current and temperature; The state estimation module is configured to process the time-series feature data using a first model or a second model, and output the battery state estimation result; wherein: The first model is a convolutional neural network-long short-term memory network model. In response to the input time-series feature data, it is processed into a normalized time-series feature matrix. The normalized time-series feature matrix is ​​then processed by the convolutional neural network to obtain a time-series deep feature vector. The time-series deep feature vector is then processed by the long short-term memory network to obtain the energy state of the battery SOE estimate. The second model is a joint estimation model of convolutional neural network and gated recurrent unit based on model-independent meta-learning optimization. In response to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's state of energy (SOE) estimate and state of health (SOH) estimate.

[0012] According to the third technical solution of this application, an electronic device is provided, the electronic device comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the method described above.

[0013] According to the fourth technical solution of this application, a non-transitory computer-readable storage medium storing instructions is provided, which, when executed by a processor, performs the method described above.

[0014] The lithium-ion battery state estimation methods, systems, devices, and media according to the various schemes in this application have at least the following technical advantages: 1) The lithium-ion battery state estimation method provided in this application adopts a convolutional neural network-long short-term memory network (CNN-LSTM) fusion architecture. It takes the voltage, current and temperature time series feature data of the battery as input, and can effectively extract time series deep features to achieve dynamic and accurate estimation of the state of energy (SOE) under low temperature and normal temperature conditions, thus solving the problem of estimation inaccuracy caused by battery performance differences.

[0015] 2) This application also provides a joint estimation model of convolutional neural network-gated recurrent unit (CNN-GRU) optimized based on model-independent meta-learning (MAML), which takes time-series feature data and a set of indirect health factors as input to achieve synergistic estimation of state of energy (SOE) and state of health (SOH), thereby improving the comprehensiveness and adaptability of battery state estimation.

[0016] 3) This application provides two optional neural network model architectures to meet the different application requirements of SOE single estimation and SOE and SOH joint estimation, respectively. Users can choose flexibly according to the actual scenario, which has high practicality and promotion value.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a lithium-ion battery state estimation method according to an embodiment of this application. Figure 2This is an initial capacity distribution diagram of a battery in one embodiment of this application; Figure 3 This is a diagram showing the internal resistance distribution of a battery in one embodiment of this application; Figure 4 This is a NASA battery capacity decay curve in one embodiment of this application; Figure 5 This is a laboratory battery capacity decay curve in one embodiment of this application; Figure 6 This is a flowchart of SOE estimation using CNN-LSTM in one embodiment of this application; Figure 7 This is a comparison chart of SOE estimation for a NASA battery in one embodiment of this application. Figure 8 A comparison chart of SOE estimations for a laboratory dataset in one embodiment of this application; Figure 9 This is a flowchart of SOE and SOH estimation based on M-CG in one embodiment of this application; Figure 10 This is a comparison chart of SOH estimates for a NASA battery in one embodiment of this application. Figure 11 A comparison chart of SOH estimates for a laboratory dataset in one embodiment of this application; Figure 12 This is a schematic diagram of the structure of a lithium-ion battery state estimation system according to one embodiment of this application. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0020] This application provides a lithium-ion battery state estimation method based on temperature measurement data and deep learning. Please refer to [link to relevant documentation]. Figure 1 This is a flowchart of a lithium-ion battery state estimation method provided in an embodiment of this application. The method includes the following steps S10 and S20.

[0021] S10: Acquire timing characteristic data during the operation of the lithium-ion battery. The timing characteristic data includes at least voltage, current, and temperature.

[0022] In some embodiments, 100 single 1100mAh / 3.65V lithium iron phosphate cells of model 18650 are used to conduct aging tests on the batteries to obtain the decline in the health status of the lithium-ion batteries. The batteries are then subjected to capacity tests and internal resistance tests in sequence to obtain the required laboratory dataset, which includes time-series characteristic data of the lithium-ion batteries during operation.

[0023] Specifically, step S10 includes the following steps S101-S105.

[0024] S101: Battery selection. 100 lithium iron phosphate (LiFePO4) batteries were used as experimental subjects. A multi-dimensional battery aging database was constructed using NASA open datasets.

[0025] S102: Battery capacity test.

[0026] According to relevant national standards, before cascade utilization, the batteries must undergo capacity testing to screen out batteries with low capacity or those that are damaged, and retain the batteries for research.

[0027] S103: Battery internal resistance test.

[0028] The ohmic internal resistance of lithium-ion batteries was tested using the hybrid pulse power characteristic method (HPPC) under operating conditions. The response voltage and current relationship during discharge, rest, and charging was established to obtain the corresponding internal resistance values, and batteries with high internal resistance and large differences in internal resistance were screened out.

[0029] S104: After testing, the selected batteries undergo aging experiments. Since batteries may encounter various situations during actual use, high-rate charge-discharge cycles are added to the standard cycle conditions to accelerate aging.

[0030] S105: Construct the experimental dataset.

[0031] The required data was constructed using both NASA and laboratory datasets. From the extensive NASA dataset, test data for four batteries (B0045, B0046, B0047, and B0048) were selected. The aging and degradation of these batteries were conducted at a low temperature of 4°C to broaden the temperature range for degradation. Laboratory data was obtained from a battery data center. After capacity and internal resistance testing of 100 batteries, 20 batteries had excessively low capacity and 5 batteries had excessively high internal resistance. Ultimately, 75 batteries were selected as the data source for battery state estimation.

[0032] S20: Process the time-series feature data using either the first model or the second model to output the battery state estimation results. The first model is a convolutional neural network-long short-term memory network model. Responding to the input time-series feature data, it processes the data into a normalized time-series feature matrix. The convolutional neural network then processes this normalized matrix to obtain a time-series deep feature vector. The long short-term memory network is then used to process this deep feature vector to obtain the battery's state of energy (SOE) estimate. The second model is a convolutional neural network-gated recurrent unit joint estimation model based on model-independent meta-learning optimization. Responding to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's SOE estimate and state of health (SOH) estimate.

[0033] In this embodiment, the first model is selected when only dynamic monitoring of the battery's state of energy (SOE) is required; the second model is selected when the battery's state of health (SOH) needs to be evaluated simultaneously for lifespan management and energy allocation optimization. The purpose of this design is to provide two optional neural network model architectures to meet the different application requirements of single SOE estimation and joint SOE and SOH estimation, respectively. Users can flexibly choose the most suitable model according to their actual application scenario, thereby ensuring estimation accuracy while balancing computational efficiency and information completeness, thus improving the adaptability and practicality of the battery management system.

[0034] For example, when the application scenario is real-time energy management and range prediction for electric vehicles, the first model is selected. This scenario requires the system to obtain the current remaining energy (SOE) of the battery within a millisecond response time to support driver range alerts, air conditioning, and power distribution of the power system. The first model only requires easily acquired signals such as voltage, current, and temperature, has a low computational burden, and fast inference speed, which can meet the stringent real-time requirements of the vehicle controller. When the application scenario is screening for the secondary use of power batteries, the second model is selected. This scenario not only requires knowing the current remaining energy (SOE) of the battery to determine its usable capacity, but also needs to accurately assess the battery's state of health (SOH) to determine whether it is suitable for retirement and use in secondary applications such as energy storage power stations and low-speed electric vehicles. The second model, by introducing indirect health factors and using the MAML algorithm for cross-sample learning, can simultaneously output SOE and SOH, providing a complete state assessment basis for secondary use. When the application scenario is long-term health monitoring and lifespan prediction of energy storage systems, the second model is selected. Energy storage systems typically operate for several to several decades, and their aging patterns are complex. The second model, which collaboratively estimates SOE and SOH, can track the trends of battery capacity decay and internal resistance growth, provide early warnings of faulty batteries, and formulate balanced maintenance strategies, thereby extending the overall system lifespan. When the application scenario involves rapid battery state verification in a fast-charging station, the first model is selected. Fast charging requires determining the battery's current energy state within a very short time to adjust the charging power to prevent overcharging or overheating. The first model is computationally simple and can quickly provide SOE estimates during charging intervals, assisting in the safe and efficient execution of charging strategies.

[0035] In some embodiments, a combined model based on convolutional neural networks and long short-term memory networks (i.e., the first model CNN-LSTM) is proposed to estimate the SOE of lithium-ion batteries. Convolutional neural networks (CNN) are good at extracting features from input data, while long short-term memory networks (LSTM) are able to process time series data. The advantages of the combined model are suitable for dealing with the complex nonlinear behavior of batteries caused by aging.

[0036] like Figure 6The diagram shows the SOE estimation flowchart of CNN-LSTM. First, battery test data is input. A health factor is extracted from the input battery test data. After the health factor extraction, the data is standardized. The standardized data is then divided into training and test sets. Next, the CNN-LSTM model parameters are initialized. After parameter initialization, the CNN-LSTM model is built, and the model's loss function is calculated. The CNN-LSTM model parameters are updated based on the calculated loss function. After the update, it is determined whether training is complete. If training is not complete, the next set of training data is used, and the process returns to the step of building the CNN-LSTM model and repeats. If training is complete, the CNN-LSTM estimation model is output. Based on the output CNN-LSTM estimation model, the battery energy for a certain period in the test set is estimated. After completing a single battery energy estimation, it is determined whether the estimation is complete. If the estimation is not complete, the battery energy for the next time step is estimated, and the process returns to the step of estimating the battery energy for a certain period in the test set and repeats. If the estimation is complete, the SOE estimation result is output, and the entire SOE estimation process ends.

[0037] Specifically, the first model can be constructed through the following steps S21~S23.

[0038] S21: Data processing, receiving the normalized time series feature matrix: in The time step length, F For feature dimensions (including at least voltage) Current ,temperature It can also expand internal resistance, (and other derived features).

[0039] S22: Model parameter initialization, initializing the relevant parameters of the LSTM model, the maximum number of iterations and the iteration termination condition, as well as the initial relevant parameters of the CNN.

[0040] S23: Establish a CNN-LSTM model. A CNN mainly consists of convolutional layers, pooling layers, and fully connected layers. Convolutional layers extract local features through sliding kernel calculations, automatically capturing key features in the battery data. Pooling layers sample and compress feature dimensions, preserving salient features of the data. Fully connected layers map high-level abstract features to the target's output space. The main formulas of the CNN network are as follows: In the formula: For the first i The output of a convolutional layer; These are the activation functions for the convolutional layer and the fully connected layer, respectively. X This is the normalized time-series feature matrix; This is a convolution operation; W i For the first i The weight matrix of the convolutional layer; b i For the first i The bias of layer convolution; This is the output of the pooling layer; pool This is a pooling operation; b j This is the deviation of the pooling layer; H c This is a temporal depth feature vector; W k This is the weight matrix of the fully connected layer; b k This represents the deviation of the fully connected layer.

[0041] Temporal deep feature vector H c After being expanded by time step, it is compared with the normalized temporal feature matrix. X The corresponding feature vectors in the sequence are concatenated and used as the input sequence for the Long Short-Term Memory network.

[0042] The LSTM algorithm comprises three gating units: a forget gate, an input gate, and an output gate. This mechanism dynamically adjusts the information flow, enabling the model to autonomously choose between updating or forgetting information, thereby capturing long-range correlation features in time-series data. The main formulas of the LSTM network are as follows: In the formula: f t The output of the forget gate; i t The output of the input gate; g t The output of the input node; O t This is the output of the output gate; σ It is the sigmoid function; C t This represents the current state of the cell. h t This represents the intermediate state at the current moment;x t This is the input for the current moment; h t-1 This is the intermediate state of the previous moment; C t-1 This represents the cell state at the previous moment; , , , These correspond to the forget gate, input gate, input node, and output gate, respectively. x t The weight matrix; U f , U i , U g , U o The forget gate, input gate, input node, and output gate correspond to the intermediate states of the previous time step, respectively. h t-1 The weight matrix; , , , These represent the biases for the forget gate, input gate, input node, and output gate, respectively; ⊙ represents the Hadamard product.

[0043] Current input x t This includes the feature vector corresponding to the time step in the temporal feature matrix, and the feature sub-vector aligned with that time step after being expanded by the temporal depth feature vector according to the time step; the intermediate state of the last time step of the LSTM is mapped by the fully connected layer to output the estimated value of the energy state SOE.

[0044] The steps for predicting the SOE of a single cell based on the first model include: after building the model, adjusting the parameters of the model using the training set, adjusting the model parameters according to the magnitude of the estimation error; validating the model using the test set, and outputting the estimation results.

[0045] Since the long-term nonlinear degradation characteristics of lithium batteries are difficult to capture effectively by traditional time series models, an improved CNN-GRU model algorithm is proposed. Simultaneously, a meta-learning algorithm is introduced to adapt to cross-sample degradation patterns, achieving joint estimation of SOE and SOH. In some embodiments, such as... Figure 9 As shown, the specific data processing steps of the second model include: S201. Initialize model parameters: Initialize the relevant parameters of the M-CG model, such as the kernel size of the convolutional layer, the number of GRU units, and the selection of activation functions.

[0046] S202. Determine the termination condition: Set the maximum number of iterations and the estimation error accuracy.

[0047] S203, Lithium-ion Battery State of Health (SOH) Estimation: Five indirect health factors are used as input parameters for the state of health. These indirect health factors include constant-voltage charging time, voltage change rate, constant-current charging time, standard deviation of charging current, and temperature integral during the constant-current charging phase. First, the model parameters are initialized. Health data is extracted from a self-built laboratory dataset and NASA data. The extracted health factors are used as input features and fed into the M-CG model. The model is trained, and its parameters are updated in real time. The M-CG estimation model is output, providing the SOH estimation results. .

[0048] S204, Lithium-ion Battery SOE Estimation: After obtaining the SOH estimation result of the battery, the voltage at time t is obtained from the original input feature. Current ,temperature Add to the basis The introduction of SOH features can update the battery's aging state in real time. By feeding the four input features into the M-CG estimation model, the model is trained and its initial parameters are updated to obtain the energy state estimation result at time t. As output.

[0049] S205. Determine if the termination condition has been met: If the number of iterations reaches the preset value, stop the iteration and finally obtain the joint estimation result of SOE and SOH at time t.

[0050] In some embodiments, the performance of the first model or the second model can be evaluated by setting evaluation metrics, including error values ​​and root mean square error values.

[0051] To further illustrate the feasibility and feasibility of the method proposed in this application, specific experimental data will be used to illustrate it below.

[0052] Figure 2 and Figure 3 The capacity and internal resistance distribution diagrams of 100 retired batteries are obtained by conducting multiple charge-discharge cycle experiments until the batteries reach the retired state, and by conducting capacity and internal resistance tests.

[0053] Figure 4 Originating from NASA's Center for Estimation Excellence, this data pertains to a commercial 18650 battery model. The battery underwent cyclic charge-discharge testing at a low temperature (4°C) until it reached its end-of-life condition. The capacity decay map is generated by capturing the capacity change after each cycle.

[0054] Figure 5Based on 100 batteries, 20 retired batteries with excessively low capacity and 5 batteries with excessively high internal resistance were removed. The remaining 75 batteries were renumbered (1-75) and used as the main source of batteries for state estimation and consistency sorting. Test data from 8 randomly selected batteries were used, and the graph shows their capacity degradation curves throughout the testing process.

[0055] Figure 6 This is a flowchart of SOE estimation based on a CNN-LSTM neural network model, which fully leverages the synergistic advantages of CNN and LSTM. The process involves data processing to obtain the model's input features, initializing the parameters of the LSTM and CNN models, continuously training the CNN-LSTM model to update its parameters, and finally obtaining the SOE estimation result through the determined CNN-LSTM model.

[0056] Figure 7 Based on the NASA low-temperature aging dataset, the CNN-LSTM model demonstrates significant advantages in SOE estimation across individual cells: compared to the LSTM model, the maximum deviation of the estimated trajectory from the true value is reduced by 42.8%, and the average RMSE is reduced by 41.8%, especially in SOE estimation of cell B0047, where R² is improved by 16.3%. The CNN-LSTM model, which integrates convolutional features, effectively improves the accuracy and robustness of SOE estimation by enhancing local feature extraction capabilities.

[0057] Figure 8 To validate the model based on a self-built laboratory dataset of decommissioned batteries at room temperature, a stratified random sampling method was used to divide the dataset: six battery groups numbered 1, 12, 38, 53, 62, and 75 were selected to construct the training set; simultaneously, eight battery groups numbered 8, 15, 23, 35, 49, 57, 66, and 72 were selected as the test set. These samples are located at the four corners, edges, and center of the battery module, respectively, effectively validating the model's adaptability to batteries with different electrochemical characteristics. The CNN-LSTM model showed significant advantages over the baseline LSTM model. Specifically, in the eight samples of the test set, the CNN-LSTM model reduced the mean absolute error by 42.7%, the root mean square error by 38.9%, and the goodness of fit by 18.53%, indicating that it has better estimation accuracy and curve fitting ability.

[0058] Figure 9 The flowchart for SOE and SOH estimation based on M-CG is as follows: In the data processing, health factors are extracted from battery test data and the data is divided. The parameters of the model are initialized and the iteration termination condition of the model is set. The input features of SOH and SOE are put into the model, and the model parameters are continuously trained and updated in real time to obtain the joint estimation results of SOE and SOH.

[0059] Figure 10 To illustrate the SOH estimation comparison based on the NASA low-temperature dataset, battery B0045 was selected as the training set, whose complete aging trajectory provides the learning foundation for typical capacity degradation patterns. The test set covers the full life cycle charge-discharge cycle data of three heterogeneous batteries (B0046, B0047, and B0048). Through comparative analysis, the following conclusions were drawn: The M-CG model significantly outperforms the traditional recurrent neural network architecture in terms of cross-device error convergence and the ability to capture the degradation inflection point. The M-CG model's SOH estimates for different battery cells have a higher fit to the true values. Especially near the capacity degradation inflection point, the M-CG model's estimation results are closer to the true values. The M-CG model also demonstrates superior performance in error control (MAE, RMSE) and trend fitting (R²). 2 The optimized model demonstrates significant advantages in all aspects, indicating that it effectively improves the accuracy and robustness of lithium battery health state estimation.

[0060] Figure 11 To compare the SOH estimation results based on a self-built laboratory dataset of decommissioned batteries at room temperature, the training set includes six battery groups (numbered 1, 12, 38, 53, 62, and 75), whose capacity decay rates cover the overall distribution of the dataset (19%-37%), ensuring that the model can learn diverse capacity decay patterns. The test set includes eight battery groups (numbered 8, 15, 23, 35, 49, 57, 66, and 72) to evaluate the model's generalization performance on unseen data. The proposed M-CG model demonstrates superior performance in the SOH estimation task for all eight decommissioned lithium batteries. Compared to the LSTM and GRU benchmark models, M-CG significantly reduces the estimation error on the test set (MAE reduction of over 30%, RMSE reduction of over 25%), and significantly improves the dynamic fit between the estimated curve and the true value (R²). 2 The mean value is greater than 0.95, which fully verifies the advanced nature of the model architecture and the advantage of estimation accuracy.

[0061] Another aspect of this application provides a lithium-ion battery state estimation system, such as... Figure 12 The diagram shown is a structural diagram of a lithium-ion battery state estimation system provided in an embodiment of this application. The lithium-ion battery state estimation system includes: The data acquisition module 1201 is configured to acquire time-series characteristic data during the operation of a lithium-ion battery, wherein the time-series characteristic data includes at least voltage, current and temperature; The state estimation module 1202 is configured to process the time-series feature data using a first model or a second model, and output the battery state estimation result; wherein: The first model is a convolutional neural network-long short-term memory network model. In response to the input time-series feature data, it is processed into a normalized time-series feature matrix. The normalized time-series feature matrix is ​​then processed by the convolutional neural network to obtain a time-series deep feature vector. The time-series deep feature vector is then processed by the long short-term memory network to obtain the energy state of the battery SOE estimate. The second model is a joint estimation model of convolutional neural network and gated recurrent unit based on model-independent meta-learning optimization. In response to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's state of energy (SOE) estimate and state of health (SOH) estimate.

[0062] It should be noted that the lithium-ion battery state estimation device provided in the above embodiments and the lithium-ion battery state estimation method provided in the foregoing embodiments belong to the same concept. The specific way in which each module and unit performs operations has been described in detail in the method embodiments, and will not be repeated here.

[0063] Another aspect of this application provides an electronic device, including: a controller; and a memory for storing one or more programs, which, when executed by the controller, perform the methods described in the various embodiments above.

[0064] Another aspect of this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0065] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various embodiments described above.

[0066] According to one aspect of the embodiments of this application, a computer system is also provided, including a Central Processing Unit (CPU), which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) or a program loaded from storage into random access memory (RAM), such as performing the methods described above. Various programs and data required for system operation are also stored in the RAM. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0067] For example, a computer system includes a Central Processing Unit (CPU), which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) or loaded from storage into random access memory (RAM), such as executing the methods described in the above embodiments. The RAM also stores various programs and data required for system operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0068] The following components are connected to the I / O interface: input components including keyboards, mice, etc.; output components including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage components including hard drives; and communication components including network interface cards such as LAN (Local Area Network) cards and modems. The communication components perform communication processing via networks such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage components as required.

[0069] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs various functions defined in the system of this application.

[0070] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: 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), 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 application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. The transmitted data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0071] 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 this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated 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 a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0072] The module units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0073] The above embodiments are only used to illustrate this application and are not intended to limit this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this application. Therefore, all equivalent technical solutions also fall within the scope of this application, and the patent protection scope of this application should be defined by the claims.

Claims

1. A method for estimating the state of a lithium-ion battery, characterized in that, The method includes: Acquire time-series characteristic data during the operation of a lithium-ion battery, wherein the time-series characteristic data includes at least voltage, current, and temperature; The time-series feature data is processed using either the first model or the second model to output battery state estimation results; wherein: The first model is a convolutional neural network-long short-term memory network model. In response to the input time-series feature data, it is processed into a normalized time-series feature matrix. The normalized time-series feature matrix is ​​then processed by the convolutional neural network to obtain a time-series deep feature vector. The time-series deep feature vector is then processed by the long short-term memory network to obtain the energy state of the battery SOE estimate. The second model is a joint estimation model of convolutional neural network and gated recurrent unit based on model-independent meta-learning optimization. In response to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's state of energy (SOE) estimate and state of health (SOH) estimate.

2. The lithium-ion battery state estimation method according to claim 1, characterized in that, The convolutional neural network in the first model includes convolutional layers, pooling layers, and fully connected layers, and its computation process is represented as follows: In the formula: For the first i The output of a convolutional layer; These are the activation functions for the convolutional layer and the fully connected layer, respectively. X This is the normalized time-series feature matrix; This is a convolution operation; W i For the first i The weight matrix of the convolutional layer; b i For the first i The bias of layer convolution; This is the output of the pooling layer; pool This is a pooling operation; b j This is the deviation of the pooling layer; H c For time-series deep feature vectors; i>W k This is the weight matrix of the fully connected layer; b k For deviations in fully connected layers; The time-series depth feature vector H c After being expanded by time steps, it is compared with the normalized temporal feature matrix. X The corresponding feature vectors in the sequence are concatenated and used as the input sequence for the Long Short-Term Memory network.

3. The lithium-ion battery state estimation method according to claim 2, characterized in that, The long short-term memory network in the first model includes a forgetting gate, an input gate, and an output gate, and its calculation process is expressed as follows: In the formula: f t The output of the forget gate; i t The output of the input gate; g t The output of the input node; O t This is the output of the output gate; σ It is the sigmoid function; C t This represents the current state of the cell. h t This represents the intermediate state at the current moment; x t This is the input for the current moment; h t-1 This is the intermediate state of the previous moment; C t-1 This represents the cell state at the previous moment; , , , These correspond to the forget gate, input gate, input node, and output gate, respectively. x t The weight matrix; U f , U i , U g , U o The forget gate, input gate, input node, and output gate correspond to the intermediate states of the previous time step, respectively. h t-1 The weight matrix; , , , These represent the biases for the forget gate, input gate, input node, and output gate, respectively; ⊙ represents the Hadamard product. Current input x t This includes the feature vector corresponding to the time step in the temporal feature matrix, and the feature sub-vector aligned with that time step after the temporal depth feature vector is expanded by the time step; The intermediate state at the last moment of the LSTM is mapped by a fully connected layer to output an estimated value of the State of Energy (SOE).

4. The lithium-ion battery state estimation method according to claim 1, characterized in that, When the second model is used, the set of indirect health factors is extracted based on battery capacity test and internal resistance test data.

5. The lithium-ion battery state estimation method according to claim 4, characterized in that, The processing steps of the second model include: Based on the aforementioned set of indirect health factors, estimate the current health status. ; The voltage at the current moment Current ,temperature and the health status The inputs are fed into the same joint estimation model to estimate the energy state at the current time. ; Output health status and energy state The joint estimation results.

6. The lithium-ion battery state estimation method according to claim 5, characterized in that, The second model performs meta-learning across different battery aging states using the MAML algorithm during the training phase to adapt to the degradation patterns across samples.

7. The lithium-ion battery state estimation method according to any one of claims 1 to 6, characterized in that, It also includes a data construction step: by conducting aging tests, capacity tests, and internal resistance tests on multiple lithium iron phosphate cells, and combining them with publicly available datasets, a multi-dimensional battery aging database is constructed for training the first model or the second model.

8. A lithium-ion battery state estimation system, characterized in that, The system includes: The data acquisition module is configured to acquire time-series characteristic data during the operation of a lithium-ion battery, wherein the time-series characteristic data includes at least voltage, current and temperature; The state estimation module is configured to process the time-series feature data using a first model or a second model, and output the battery state estimation result; wherein: The first model is a convolutional neural network-long short-term memory network model. In response to the input time-series feature data, it is processed into a normalized time-series feature matrix. The normalized time-series feature matrix is ​​then processed by the convolutional neural network to obtain a time-series deep feature vector. The time-series deep feature vector is then processed by the long short-term memory network to obtain the energy state of the battery SOE estimate. The second model is a joint estimation model of convolutional neural network and gated recurrent unit based on model-independent meta-learning optimization. In response to the input time-series feature data and a set of indirect health factors extracted from the battery data, it outputs the battery's state of energy (SOE) estimate and state of health (SOH) estimate.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing instructions, characterized in that, When the instructions are executed by the processor, the method according to any one of claims 1 to 7 is performed.

Images