A method and system for estimating battery state of health based on segmented charging data

Abstract

This invention belongs to the field of battery evaluation technology and provides a method and system for estimating battery health status based on segment charging data. The method calculates the corresponding IC and IE segments from the real-time acquired charging data of the target segment and extracts the peak / valley information for each segment. If the acquired target segment data contains multiple sets of joint features corresponding to IC and IE curves, each set of joint features is input into the corresponding pre-trained sub-model for processing. The outputs of each sub-model are adaptively weighted and fused using an attention mechanism to obtain the final battery health status estimate. If the target segment data contains only a single set of joint features, that set of features is input into the corresponding sub-model to directly obtain the battery health status estimate. This invention improves the robustness and generalization ability of SOH estimation.

Description

Technical Field

[0001] This invention belongs to the field of battery evaluation technology, specifically relating to a method and system for estimating battery health status based on segmented charging data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the rapid development of new energy electric vehicles, energy storage systems, and portable electronic devices, lithium-ion batteries have become the mainstream energy storage device due to their advantages such as high energy density, long cycle life, and low self-discharge rate. During long-term use, the capacity and performance of batteries will inevitably degrade due to factors such as electrochemical reactions, accumulation of side reactions, and structural changes. Therefore, accurate assessment of the State of Health (SOH) of lithium-ion batteries is a crucial prerequisite for ensuring the safe operation of battery systems, optimizing energy management strategies, and extending battery life.

[0004] Among existing SOH estimation methods, the incremental capacity (IC) curve-based analysis method has attracted widespread attention due to its ability to reflect the characteristics of internal electrochemical reactions in batteries. By differentiating the voltage-capacity curve, characteristic peak information in the IC curve can be obtained, which can then be used to characterize the degree of battery aging. Meanwhile, with the development of battery mechanical property testing methods, the volume expansion behavior of batteries during charge and discharge processes is increasingly being used to characterize changes in their internal structure. The incremental expansion (IE) curve, constructed based on expansion characteristics, also provides a new dimension of information for battery health status assessment.

[0005] However, most existing SOH estimation methods based on IC or IE curves rely on complete charge-discharge cycle data, typically requiring continuous sampling over a wide voltage range. This prerequisite is difficult to meet in practical applications, especially during the online operation of electric vehicles and energy storage systems, where batteries are often incomplete charge-discharge or stochastic operating conditions, resulting in fragmented and discontinuous partial charging data. Furthermore, some existing methods utilize only a single electrochemical or mechanical characteristic, failing to fully exploit the complementary information between different physical features, thus limiting the accuracy and robustness of SOH estimation under complex operating conditions. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a method and system for estimating battery state of health (SOH) based on segmented charging data. This invention improves the robustness and generalization ability of SOH estimation.

[0007] According to some embodiments, the present invention adopts the following technical solution:

[0008] A method for estimating battery health status based on segmented charging data includes the following steps:

[0009] Acquire battery parameter data during the charge-discharge test cycle and preprocess the acquired data;

[0010] IC curves and IE curves are calculated based on the preprocessed data. Within the same voltage range, the peak / valley features in the corresponding IC curves and IE curves are matched and combined as a set of joint features. The battery health status is used as the output.

[0011] Build sub-models, and train each sub-model using the extracted joint features and the corresponding battery health status values ​​to obtain the trained sub-models.

[0012] For the charging data of the target segment acquired in real time, the corresponding IC and IE segments are calculated, and the peak / valley information of each segment is extracted;

[0013] If the data of the target segment contains multiple sets of joint features corresponding to IC curves and IE curves, then each set of joint features is input into the corresponding pre-trained sub-model for processing. The output results of each sub-model are adaptively weighted and fused using an attention mechanism to obtain the final battery health status estimate. If the data of the target segment contains only a single set of joint features, then that set of features is input into the corresponding sub-model to directly obtain the battery health status estimate.

[0014] As an alternative implementation, the battery parameter data includes battery stress, current, voltage, and capacity data.

[0015] As an alternative implementation method, the process of calculating the IC curve and IE curve based on the preprocessed data includes: The specific calculation formulas for the IC and IE curves are as follows:

[0016] ;

[0017] ;

[0018] Where Q represents battery capacity, V represents voltage, E represents battery expansion stress, and ΔX represents the change in characteristic X between adjacent sampling points, where X is Q, V, or E.

[0019] As an alternative implementation, the sub-model is structured based on a BP neural network, comprising an input layer, a hidden layer, and an output layer connected sequentially. The training process of the sub-model includes initializing the neural network, determining the number of neurons in the output layer, hidden layer, and output layer, initializing the connection weights between each layer using a random initialization method, determining the activation function, correcting the sub-model error through backpropagation with a threshold, correcting the threshold using gradient descent based on the output error, determining whether the sub-model training has reached the expected level, and if not, returning to the hidden layer to continue iterative training until the expected level is met, and outputting the SOH estimate.

[0020] As a further defined implementation method, the output calculation formula for the hidden layer is as follows:

[0021] ;

[0022] in, w ij These are the weight coefficients between the input layer and the hidden layer. f For the activation function, the Tanh activation function is chosen here. x i For the output vector, a j The threshold of the hidden layer neurons. n Represents the number of nodes in the input layer;

[0023] The formula for calculating the output layer is as follows:

[0024] ;

[0025] in, b k For the threshold, w jk These are the weighting coefficients between the hidden layer and the output layer; h j Represents the output of the hidden layer. l This represents the number of neurons in the hidden layer;

[0026] The mean squared error is used to calculate the neural network error, and the formula is as follows:

[0027] ;

[0028] in, This represents the estimated SOH value, where SOH is the expected true value of the sub-model's output.

[0029] As a further defined implementation, the process of correcting the threshold using gradient descent based on the output error includes:

[0030] ;

[0031] ;

[0032] ;

[0033] ;

[0034] in, η For learning rate, x i As input features, h j Represents the output of the hidden layer. e k Represents network error. m This represents the number of nodes in the output layer.

[0035] As an alternative implementation, the process of matching the peak / valley features in the corresponding IC curve and IE curve within the same voltage range and combining them as a set of joint features includes: considering the peak / valley combinations of detectable IE and IC curves within different voltage ranges as joint features, and constructing a sub-model with the same structure but independent parameters for them.

[0036] As an alternative implementation method, the process of adaptively weighted fusion of the output results of each sub-model using an attention mechanism includes: introducing an attention mechanism to automatically learn the latent features of all sub-model outputs; dynamically adjusting the contribution of different peaks through normalized weights to adapt to changes in the number of peaks and incomplete data; inputting the weighted fused global health features into the regression layer to obtain the battery health status estimation result; calculating the estimation error; and outputting the battery health status result if the expected error is met; otherwise, updating the weights of each neuron in the corresponding sub-model until the expected result is met.

[0037] As a further defined implementation, the process of introducing an attention mechanism to automatically learn the latent features of all sub-model outputs includes: performing a linear transformation on each latent feature, projecting the BP sub-model outputs onto a unified attention space, and compressing the mapped features into a scalar score.

[0038] ;

[0039] in, e i Representing the i Attention score for each sub-model, W a Represents the attention weight matrix, b a represents the attention bias, v is the learnable attention vector, and tanh(·) represents the hyperbolic tangent activation function, used to perform nonlinear mapping on the linear transformation result;

[0040] The normalized weights are calculated as follows:

[0041] ;

[0042] In order to adapt to changes in the number of peaks and valleys, a mask variable is introduced. m i Ruofeng i If it exists, then m i =1, if peak i If it does not exist, then m i =0, N The number of peaks / valleys contained in the segment curve. α i For the first i Attention weights for each sub-model;

[0043] The hidden features output by the BP network are weighted and summed using attention weights, and then input into the regression layer to obtain the SOH estimate:

[0044] ;

[0045] in, h represents the estimated value of SOH. i The output represents the sub-model. g (·) represents a fully connected network.

[0046] As an alternative implementation method, during the training of the sub-model, a physical loss function is constructed based on prior physical knowledge of the battery aging process, and combined with the data loss to construct the model's total loss function. When the loss function has not yet converged or has not reached the preset accuracy requirement, the iterative optimization process of forward propagation and backpropagation is continuously carried out until the convergence criterion is met and the training is completed.

[0047] As a further defined implementation, mean squared error is used as the data-driven loss term to measure the deviation between the model's predicted values ​​and the true values. Its expression is as follows:

[0048] ;

[0049] in, Y k Represents the true SOH value. Represents the estimated SOH value. m Represents the number of samples;

[0050] Based on the monotonic evolution relationship between the voltage values ​​corresponding to the IC and IE curves and the SOH, a physical information-driven loss term is constructed:

[0051] ;

[0052] in, ΔV k This represents the change in peak / valley characteristic voltage values ​​corresponding to adjacent segments of the curve. δ represents the change in the estimated SOH value in adjacent data segments, and is the tolerance coefficient. It is used to emphasize the monotonic trend of the voltage value corresponding to each peak / valley feature increasing as the battery SOH decreases, while allowing local non-strict monotonic changes.

[0053] The model's total loss function is composed of a weighted average of the data-driven loss term and the physical information constraint term.

[0054] A system for estimating battery health status based on segmented charging data, comprising:

[0055] The data acquisition module is configured to acquire battery parameter data during the charge-discharge test cycle of the battery and preprocess the acquired data.

[0056] The curve feature extraction module is configured to calculate the IC curve and IE curve based on the preprocessed data. Within the same voltage range, the peak / valley features in the corresponding IC curve and IE curve are matched and combined as a set of joint features, with the battery health status as the output.

[0057] The sub-model building module is configured to build sub-models and train each individual sub-model using the extracted joint features and the corresponding battery health status values ​​to obtain the trained sub-models.

[0058] The feature extraction module is configured to calculate the corresponding IC and IE segments from the charging data of the target segment acquired in real time, and extract the peak / valley information corresponding to each segment.

[0059] The battery health state estimation module is configured to, if the data of the target segment contains multiple sets of joint features corresponding to IC curves and IE curves, input each set of joint features into the corresponding pre-trained sub-model for processing, and use an attention mechanism to adaptively weight and fuse the output results of each sub-model to obtain the final battery health state estimate; if the data of the target segment contains only a single set of joint features, input that set of features into the corresponding sub-model to directly obtain the battery health state estimate.

[0060] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0061] (1) The present invention can accurately estimate the SOH of the battery by relying only on the segment charging data, without the need for complete charge and discharge cycle data support; at the same time, it does not require that the length of each segment data be consistent, and only that the effective peak features can be extracted to achieve SOH estimation, which significantly improves the applicability of the method under actual working conditions.

[0062] (2) This invention integrates the characteristic information of IC curve and IE curve, and simultaneously characterizes the electrochemical behavior and mechanical expansion characteristics of the battery, which can comprehensively reflect the multidimensional change information during the battery aging process; the peak / valley feature extraction method adopted is simple to calculate and easy to implement, and is convenient for engineering application and online deployment.

[0063] (3) The present invention constructs corresponding sub-models for different peak / valley features and introduces an attention mechanism to adaptively weight and fuse the output results of each sub-model, so that the model can dynamically adjust the weight allocation according to the effective feature information contained in the fragment data, thereby improving the robustness and generalization ability of SOH estimation.

[0064] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0065] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0066] Figure 1 This is a flowchart of a battery SOH estimation method based on segment charging data, according to one embodiment.

[0067] Figure 2 This is a schematic diagram of the basic structure of a BP neural network according to one embodiment;

[0068] Figure 3 This is a schematic diagram of a SOH estimation model based on fragment data, according to one embodiment.

[0069] Figure 4 This is a schematic diagram showing the changes in IC, IE, and SOC during the charging process of one embodiment.

[0070] Figure 5 This is a battery SOH estimation result of one embodiment, wherein (a) is the SOH estimation result of the first group of batteries, and (b) is the SOH estimation result of the second group of batteries. Detailed Implementation

[0071] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0072] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0073] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0074] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0075] Example 1

[0076] A battery SOH estimation method based on segmented charging data, such as Figure 1 As shown, it includes the following steps:

[0077] (1) Perform charge and discharge tests on the battery and record data such as battery stress, current, voltage and capacity during the cycle;

[0078] (2) Preprocess and clean the collected data;

[0079] (3) Calculate and plot the IC curve and IE curve. In the same voltage range, match the peak / valley features in the corresponding IC curve and IE curve, and combine them as a set of joint features. Further extract the peak / valley parameters in each segment curve, including the vertical coordinate value corresponding to the peak / valley and the corresponding horizontal coordinate voltage value. Use the features as model input parameters and SOH as input features.

[0080] The specific formulas for calculating IC and IE curves are as follows:

[0081] (1)

[0082] (2)

[0083] Where Q represents battery capacity, V represents voltage, and E represents battery expansion stress. ΔX represents the change in characteristic X between adjacent sampling points, where X is Q, V, or E.

[0084] (4) Build sub-models based on BP neural network, and train each independent sub-model using the extracted input features and output SOH values ​​to obtain the trained sub-models;

[0085] (5) Calculate and obtain the corresponding IC and IE segments from the newly acquired segment charging data in real time, and extract the peak / valley information corresponding to the segment;

[0086] (6) If the fragment data contains multiple sets of peak / valley joint features corresponding to IC curves and IE curves, then each set of peak / valley features is input into the corresponding trained sub-model for processing, and an attention mechanism module is introduced at the output of each sub-model to adaptively weight and fuse the output results of each sub-model to obtain the final SOH estimate; if the fragment data contains only a single set of peak / valley features, then the set of features is input into the corresponding sub-model and the SOH estimate result is directly output.

[0087] Among them, the sub-model network structure based on BP neural network is as follows: Figure 2 As shown, the main training process for the sub-model begins with neural network initialization, determining the number of neurons in the output layer, hidden layers, and output layer. Then, random initialization is used to initialize the connection weights between layers and determine the activation functions.

[0088] The formula for calculating the hidden layer output is as follows:

[0089] (3)

[0090] in, w ij These are the weight coefficients between the input layer and the hidden layer. f For the activation function, the Tanh activation function is chosen here. x i For the output vector, a j is the threshold of the hidden layer neurons, and n represents the number of nodes in the input layer.

[0091] The formula for calculating the output layer is as follows:

[0092] (4)

[0093] in, b k For the threshold, w jk h represents the weight coefficients between the hidden layer and the output layer. j Represents the output of the hidden layer. l This represents the number of neurons in the hidden layer.

[0094] The mean squared error is used to calculate the neural network error, and the formula is as follows:

[0095] (5)

[0096] in, This represents the estimated SOH value, where SOH is the expected true value of the sub-model's output.

[0097] For the error backpropagation correction threshold of the BP neural network, the threshold is corrected using the gradient descent method based on the output error, as shown in the following formula.

[0098] (6)

[0099] (7)

[0100] (8)

[0101] (9)

[0102] in, η For learning rate, x i For input features. h j Represents the output of the hidden layer. e k Represents network error. m This represents the number of nodes in the output layer.

[0103] Finally, determine whether the training of the BP sub-model has met expectations. If not, return to the hidden layer and continue iterative training until the expectations are met, and output the SOH estimate.

[0104] Furthermore, multiple BP network sub-models are fused to build a SOH estimation model, with the framework as follows: Figure 3 As shown in the diagram. Specifically, the peak / valley combinations of the detectable IE and IC curves within the corresponding voltage range are first treated as independent local degradation feature units, and a BP neural network sub-model with the same structure but independent parameters is constructed for each unit. If the obtained segment contains only a single set of IC and IE peak / valley features, the output is the SOH estimate output by the corresponding BP sub-model. If the segment contains multiple sets of peak / valley features, the BP sub-model does not directly output the SOH estimate, but instead uses the output of its last hidden layer as the health characterization corresponding to that peak, to characterize the indicative ability of this local electrochemical process to the overall health status.

[0105] Building upon this foundation, an attention mechanism is introduced to adaptively weight and fuse all latent features. The attention module automatically learns the relative importance of the latent features output by each sub-model and dynamically adjusts the contribution of different peaks through normalized weights, thus naturally adapting to variations in the number of peaks and incomplete data. The weighted and fused global health features are then input into the regression layer to obtain the SOH estimation result. Finally, the estimation error is calculated. If the expected error is met, the SOH result is output; otherwise, the weights of each neuron in the corresponding sub-model are updated until the expected result is achieved.

[0106] Specifically, the attention mechanism works as follows: First, a linear transformation is performed on each latent feature, the output of the BP sub-model is projected onto a unified attention space, and the mapped features are compressed into scalar scores.

[0107] (10)

[0108] in, e i Representing the i Attention score for each sub-model, W a Represents the attention weight matrix, b a represents the attention bias, v is the learnable attention vector, and tanh(·) represents the hyperbolic tangent activation function, used to perform nonlinear mapping on the results of linear transformations.

[0109] Therefore, the normalized weights are calculated as follows:

[0110] (11)

[0111] In order to adapt to changes in the number of peaks and valleys, a mask variable is introduced. m i Ruofeng i If it exists, then m i =1, if peak i If it does not exist, then m i =0. N This represents the number of peaks / valleys contained in the segment curve. α i For the first i Attention weights for each sub-model.

[0112] Finally, the hidden features output by the BP network are weighted and summed using attention weights, and then input into the regression layer to obtain the SOH estimate:

[0113] (12)

[0114] in, h represents the estimated value of SOH. i The output represents the sub-model. g (·) represents a fully connected network.

[0115] In the loss function section, to enhance the physical interpretability of the model, this embodiment transforms prior physical knowledge of the battery aging process into constraints on the neural network training process. These constraints originate from observed electrochemical and mechanical evolution laws, which not only effectively improve the model's estimation accuracy but also enhance its physical consistency and cross-condition generalization ability. Specifically, based on the monotonic evolution relationship between the peak / valley characteristics in the IC and IE curves and the battery's SOH, this embodiment embeds this type of physical prior information into the training loss function of the neural network. L train This further guides and constrains the model parameter update process.

[0116] Based on this, mean squared error is used as the data-driven loss term to measure the deviation between the model's predicted values ​​and the true values. Its expression form is as follows:

[0117] (13)

[0118] in, Y k Represents the true SOH value. Represents the estimated SOH value. m Represents the number of samples.

[0119] To comprehensively incorporate physical constraints, based on the monotonic evolution relationship between the voltage values ​​and SOH corresponding to the discovered IC and IE curve features, a physical information-driven loss term was further constructed:

[0120] (14)

[0121] in, ΔV k This represents the change in peak / valley characteristic voltage values ​​corresponding to adjacent segments of the curve. This represents the change in the estimated SOH value between adjacent data segments. δ is a tolerance coefficient, emphasizing the monotonic trend of increasing voltage values ​​corresponding to each peak / valley characteristic as the battery SOH decreases, while allowing for local non-strict monotonic changes, with a value of 0.001.

[0122] Based on the above design, the model's total loss function, composed of a data-driven loss term and a physical information constraint term, can be expressed as:

[0123] (15)

[0124] in, αThese are weighting coefficients used to balance the relative importance of data fitting error and physical constraints.

[0125] This embodiment conducts charge-discharge experiments on a pouch lithium-ion battery and collects data such as battery stress, voltage, and current. Then, it calculates and plots the IC, IE, and SOC curves as a function of voltage during the battery charging process, as shown below. Figure 4 As shown.

[0126] according to Figure 4 As can be seen, during the charging process, the IC and IE curves exhibit multiple distinct peaks and troughs, with these peak-trough characteristics ranging from 4% to 86% SOC, consistent with the charging and discharging habits of most users. Furthermore, individual peak / trough characteristics within the IC and IE ranges can be captured using only battery data within the 0.1V range, making data acquisition and implementation easy. Therefore, combining the IC and IE peak and trough information within the same voltage range can help estimate the battery's health status.

[0127] Since the battery charging process is random in actual use, in order to achieve SOH estimation based solely on fragmented data, a framework is constructed for each peak and trough in the curve, such as... Figure 2 The sub-model is shown. Then, the peak / valley features of the corresponding IC and IE curves are extracted within the voltage range of the segment, and the corresponding IC and IE peak / valley features within the same voltage range are matched and combined. For example, A1 and A2 form a set of joint features, B1 and B2, C1 and C2, D1 and D2, E1 and E... 2以此类推 For each set of joint features, the peak / valley amplitude corresponding to the vertical axis and the voltage value corresponding to the horizontal axis are extracted to form the input feature vector. The SOH is used as the output target to train the corresponding sub-model independently.

[0128] In this embodiment, a total of 5 BP neural network sub-models are constructed. The initial learning rate of each sub-model is set to 0.01, the number of input layer nodes is 4, the number of output layer nodes is 1, the network contains 3 hidden layers, and each hidden layer has 12 neurons.

[0129] Finally, the peak / valley feature parameters of the IC and IE curves are calculated and extracted using the collected random charging segment data, and these features are input into the pre-trained corresponding sub-models for testing. When the segment data contains multiple sets of peak / valley features, an attention mechanism is introduced into the hidden layer state vectors output by multiple sub-models for adaptive weighted fusion to obtain the final SOH estimate; when only a single set of peak / valley features is contained, the corresponding sub-model directly outputs the SOH estimate.

[0130] The SOH estimation results of the two sets of batteries obtained in this embodiment are as follows: Figure 5As shown, the results demonstrate that the extracted IC and IE peak / valley features can effectively characterize battery aging status, and each sub-model can accurately capture the trend of health status changes. Furthermore, the constructed estimation model can automatically adjust the fusion method according to the number of features contained in the data fragment, making it suitable for both single-feature input scenarios and multi-feature joint input scenarios. This better reflects the data characteristics under actual random charging conditions, exhibiting high estimation accuracy and good generalization ability.

[0131] Example 2

[0132] A system for estimating battery health status based on segmented charging data, comprising:

[0133] The data acquisition module is configured to acquire battery parameter data during the charge-discharge test cycle of the battery and preprocess the acquired data.

[0134] The curve feature extraction module is configured to calculate the IC curve and IE curve based on the preprocessed data. Within the same voltage range, the peak / valley features in the corresponding IC curve and IE curve are matched and combined as a set of joint features, with the battery health status as the output.

[0135] The sub-model building module is configured to build sub-models and train each individual sub-model using the extracted joint features and the corresponding battery health status values ​​to obtain the trained sub-models.

[0136] The feature extraction module is configured to calculate the corresponding IC and IE segments from the charging data of the target segment acquired in real time, and extract the peak / valley information corresponding to each segment.

[0137] The battery health state estimation module is configured to, if the data of the target segment contains multiple sets of joint features corresponding to IC curves and IE curves, input each set of joint features into the corresponding pre-trained sub-model for processing, and use an attention mechanism to adaptively weight and fuse the output results of each sub-model to obtain the final battery health state estimate; if the data of the target segment contains only a single set of joint features, input that set of features into the corresponding sub-model to directly obtain the battery health state estimate.

[0138] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROMIt takes the form of a computer program product implemented on (such as optical memory, etc.).

[0139] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0140] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0141] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0142] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for estimating battery state of health based on segmented charging data, characterized in that, Includes the following steps: Acquire battery parameter data during the charge-discharge test cycle and preprocess the acquired data; IC curves and IE curves are calculated based on the preprocessed data. Within the same voltage range, the peak / valley features in the corresponding IC curves and IE curves are matched and combined as a set of joint features. The battery health status is used as the output. Build sub-models, and train each sub-model using the extracted joint features and the corresponding battery health status values ​​to obtain the trained sub-models. For the charging data of the target segment acquired in real time, the corresponding IC and IE segments are calculated, and the peak / valley information of each segment is extracted; If the data of the target segment contains multiple sets of joint features corresponding to IC curves and IE curves, then each set of joint features is input into the corresponding pre-trained sub-model for processing. The output results of each sub-model are adaptively weighted and fused using an attention mechanism to obtain the final battery health status estimate. If the data of the target segment contains only a single set of joint features, then that set of features is input into the corresponding sub-model to directly obtain the battery health status estimate. The process of adaptively weighted fusion of the outputs of each sub-model using an attention mechanism includes: introducing an attention mechanism to automatically learn the latent features of all sub-model outputs; dynamically adjusting the contribution of different peaks through normalized weights to adapt to changes in the number of peaks and incomplete data; inputting the weighted fused global health features into the regression layer to obtain the battery health status estimation result; calculating the estimation error; and outputting the battery health status result if the expected error is met; otherwise, updating the weights of each neuron in the corresponding sub-model until the expected result is met.

2. The method for estimating battery health status based on segmented charging data as described in claim 1, characterized in that, The battery parameter data includes battery stress, current, voltage, and capacity data; The process of calculating the IC curve and IE curve based on the preprocessed data includes: The specific calculation formulas for the IC and IE curves are as follows: ； ； Where Q represents battery capacity, V represents voltage, E represents battery expansion stress, and ΔX represents the change in characteristic X between adjacent sampling points, where X is Q, V, or E.

3. The method for estimating battery health status based on segmented charging data as described in claim 1, characterized in that, The sub-model is based on a backpropagation (BP) neural network and includes an input layer, a hidden layer, and an output layer connected in sequence. The training process of the sub-model includes initializing the neural network, determining the number of neurons in the output layer, hidden layer, and output layer, initializing the connection weights between each layer using a random initialization method, determining the activation function, backpropagating the error of the sub-model to correct the threshold, using gradient descent to correct the threshold based on the output error, determining whether the sub-model training has reached the expected level, and if not, returning to the hidden layer to continue iterative training until the expected level is met, and outputting the SOH estimate.

4. The method for estimating battery health status based on segmented charging data as described in claim 1, characterized in that, The formula for calculating the output of the hidden layer is as follows: ； in, w ij These are the weight coefficients between the input layer and the hidden layer. f For the activation function, the Tanh activation function is chosen here. x i For the output vector, a j The threshold of the hidden layer neurons. n Represents the number of nodes in the input layer; The formula for calculating the output layer is as follows: ； in, b k For the threshold, w jk The weight coefficients between the hidden layer and the output layer h j Represents the output of the hidden layer. l This represents the number of neurons in the hidden layer; The mean squared error is used to calculate the neural network error, and the formula is as follows: ； in, This represents the estimated SOH value, where SOH is the expected true value of the sub-model's output. The process of correcting the threshold using gradient descent based on the output error includes: ； ； ； ； in, η For learning rate, x i As input features, h j Represents the output of the hidden layer. e k Represents network error. m Represents the number of output layer nodes, superscript All of these represent corrected data.

5. The method for estimating battery health status based on segmented charging data as described in claim 1, characterized in that, Within the same voltage range, the process of matching the peak / valley features of the corresponding IC curve and IE curve and combining them as a set of joint features includes: taking the peak / valley combinations of detectable IE and IC curves within different voltage ranges as joint features and constructing a sub-model with the same structure but independent parameters for them.

6. The method for estimating battery health based on segmented charging data as described in claim 1, characterized in that, it introduces... The attention mechanism automatically learns the latent features from the outputs of all sub-models, including: performing a linear transformation on each latent feature, projecting the outputs of the BP sub-models onto a unified attention space, and compressing the mapped features into scalar scores. ； in, e i Representing the i Attention score for each sub-model, W a Represents the attention weight matrix, b a represents the attention bias, v is the learnable attention vector, and tanh(·) represents the hyperbolic tangent activation function, used to perform nonlinear mapping on the linear transformation result; The normalized weights are calculated as follows: ； In order to adapt to changes in the number of peaks and valleys, a mask variable is introduced. m i Ruofeng i If it exists, then m i =1, if peak i If it does not exist, then m i =0, N The number of peaks / valleys contained in the segment curve. α i For the first i Attention weights for each sub-model; The hidden features output by the BP network are weighted and summed using attention weights, and then input into the regression layer to obtain the SOH estimate: ； in, h represents the estimated value of SOH. i The output represents the sub-model. g (·) represents a fully connected network.

7. The method for estimating battery health status based on segmented charging data as described in claim 1, characterized in that, During the training of the sub-model, a physical loss function is constructed based on prior physical knowledge of the battery aging process, and combined with the data loss to construct the model's total loss function. When the loss function has not yet converged or has not reached the preset accuracy requirement, the forward propagation and back propagation iterative optimization process is continuously carried out until the convergence criterion is met and the training is completed.

8. The method for estimating battery health status based on segmented charging data as described in claim 7, characterized in that, Mean squared error is used as the data-driven loss term to measure the deviation between the model's predicted values ​​and the actual values. Its expression is as follows: ； in, Y k Represents the true SOH value. Represents the estimated SOH value. m Represents the number of samples; Based on the monotonic evolution relationship between the voltage values ​​corresponding to the IC and IE curves and the SOH, a physical information-driven loss term is constructed: ； in, ΔV k This represents the change in peak / valley characteristic voltage values ​​corresponding to adjacent segments of the curve. δ represents the change in the estimated SOH value in adjacent data segments, and is the tolerance coefficient. It is used to emphasize the monotonic trend of the voltage value corresponding to each peak / valley feature increasing as the battery SOH decreases, while allowing local non-strict monotonic changes. The model's total loss function is composed of a weighted average of the data-driven loss term and the physical information constraint term.

9. A system for estimating battery health status based on segmented charging data, the method of any one of claims 1-8, characterized in that, include: The data acquisition module is configured to acquire battery parameter data during the charge-discharge test cycle of the battery and preprocess the acquired data. The curve feature extraction module is configured to calculate the IC curve and IE curve based on the preprocessed data. Within the same voltage range, the peak / valley features in the corresponding IC curve and IE curve are matched and combined as a set of joint features, with the battery health status as the output. The sub-model building module is configured to build sub-models and train each individual sub-model using the extracted joint features and the corresponding battery health status values ​​to obtain the trained sub-models. The feature extraction module is configured to calculate the corresponding IC and IE segments from the charging data of the target segment acquired in real time, and extract the peak / valley information corresponding to each segment. The battery health state estimation module is configured to, if the data of the target segment contains multiple sets of joint features corresponding to IC curves and IE curves, input each set of joint features into the corresponding pre-trained sub-model for processing, and use an attention mechanism to adaptively weight and fuse the output results of each sub-model to obtain the final battery health state estimate; if the data of the target segment contains only a single set of joint features, input that set of features into the corresponding sub-model to directly obtain the battery health state estimate.

Images