Lithium battery health state prediction method and related equipment

Abstract

The invention provides a lithium battery health state prediction method and related equipment. The method comprises the following steps: acquiring power data and battery capacity of a lithium battery in a plurality of charge-discharge cycles; performing feature extraction on the power data to obtain a plurality of feature sequences, and determining a health state value sequence of the lithium battery based on the battery capacity; screening a target sequence from the feature sequences based on correlation coefficients of the feature sequences and the health state value sequences; obtaining a first state prediction value based on the target sequence and the health state value sequence; correcting the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value; determining a dynamic weight based on the first state prediction value, the modified prediction value, a residual between the first state prediction value and the modified prediction value, and a trace of a covariance matrix; and fusing the first state prediction value and the corrected prediction value based on the dynamic weight to obtain a health state prediction result of the lithium battery in the target time period.

Description

Technical Field

[0001] The present disclosure relates to the field of battery health status prediction, and in particular to a method for predicting the health status of a lithium battery and related equipment. Background Art

[0002] Lithium-ion batteries are widely used in electric vehicles and renewable energy storage systems due to their high energy density, long cycle life, low self-discharge, and cost-effectiveness. Accurately predicting the state of health (SOH) of lithium batteries, a widely used metric for evaluating battery performance, has become a core requirement for battery management systems. SOH indicates the degree of battery capacity decay and directly impacts the safety and service life of the device.

[0003] Due to the nonlinear degradation, time-varying, and complex dynamics of lithium batteries, SOH cannot be directly derived from observational measurements. Current SOH prediction techniques fall into two categories: model-based and data-driven. Model-based methods simulate battery degradation behavior by constructing precise battery models, such as equivalent circuit models, electrochemical models, and empirical models. These methods are generally applicable only to specific types of lithium-ion batteries, limiting their generalization across different battery types and resulting in inaccurate predictions. Data-driven methods directly learn aging patterns from historical data and use algorithms to map health features to SOH. Prominent data-driven techniques include support vector machines, Gaussian process regression, and neural network algorithms. The prediction results of data-driven models are poorly interpretable, and their accuracy depends largely on the quality of the selected health features and the learning ability of the training algorithm. High-quality training data is required to ensure the robustness and stability of the trained models. During battery service, as sensors age, the collected data gradually introduces errors due to environmental changes and improper operation, which can affect the results of data-driven models and lead to inaccurate predictions. Summary of the Invention

[0004] The present disclosure proposes a method for predicting the health status of a lithium battery and related equipment to at least partially solve the above technical problems to a certain extent.

[0005] In a first aspect, the present disclosure provides a method for predicting the health status of a lithium battery, comprising:

[0006] Obtain power data and battery capacity of lithium batteries during multiple charge and discharge cycles;

[0007] Extracting features from the power data to obtain a plurality of feature sequences, and determining a health status value sequence of the lithium battery based on the battery capacity;

[0008] Filtering a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence;

[0009] Obtaining a first state prediction value based on the target sequence and the health state value sequence;

[0010] Correcting the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value;

[0011] determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix;

[0012] The first state prediction value and the revised prediction value are fused based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period.

[0013] In a second aspect of the present disclosure, a device for predicting the health status of a lithium battery is provided, comprising:

[0014] An acquisition module is used to obtain power data and battery capacity of the lithium battery during multiple charge and discharge cycles;

[0015] a feature extraction module, configured to extract features from the power data to obtain a plurality of feature sequences, and determine a health status value sequence of the lithium battery based on the battery capacity;

[0016] a feature screening module, configured to screen a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence;

[0017] A first prediction module, configured to obtain a first state prediction value based on the target sequence and the health state value sequence;

[0018] A second prediction module is used to correct the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value;

[0019] a weight module for determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix;

[0020] A fusion module is used to fuse the first state prediction value and the corrected prediction value based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period.

[0021] In a third aspect of the present disclosure, an electronic device is provided, comprising one or more processors, a memory; and one or more programs, wherein the one or more programs are stored in the memory and executed by the one or more processors, and the programs include instructions for executing the method described in the first aspect.

[0022] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium containing a computer program is provided. When the computer program is executed by one or more processors, the processors are caused to execute the method described in the first aspect.

[0023] In a fifth aspect of the present disclosure, a computer program product is provided, comprising computer program instructions, which, when executed on a computer, enable the computer to execute the method described in the first aspect.

[0024] As can be seen from the above, the present disclosure provides a method and related equipment for predicting the health status of lithium batteries. The method extracts the characteristic sequence of power data in the charge and discharge cycle of lithium batteries, and generates a health status (SOH) value sequence in combination with the battery capacity. The correlation between the characteristics and SOH is used to screen key target sequences to reduce data redundancy. Subsequently, a preliminary prediction is performed based on the target sequence to obtain a first state prediction value. The prediction error is then corrected in real time through residual detection technology to generate a corrected prediction value and its covariance matrix to quantify uncertainty. Finally, the preliminary prediction and the corrected prediction results are integrated through a dynamic weight allocation mechanism, in which the weight is dynamically adjusted according to the residual size and the trace of the covariance matrix, so that the model automatically increases the weight of the corrected value when the data fluctuates, and relies more on the preliminary prediction when the data is stable, thereby effectively suppressing noise interference and improving prediction robustness. It can reduce the SOH prediction error of lithium batteries, especially in the late stage of battery aging, and can significantly capture the capacity mutation characteristics, providing a more reliable status assessment basis for the battery management system (BMS). BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 Schematic diagram of a lithium battery health status prediction architecture according to an embodiment of the present disclosure.

[0027] Figure 2 Schematic diagram of the hardware structure of an exemplary electronic device according to an embodiment of the present disclosure.

[0028] Figure 3Schematic diagram of the flow of the method for predicting the health status of a lithium battery according to an embodiment of the present disclosure.

[0029] Figure 4 Schematic diagram of the principle of the method for predicting the health status of a lithium battery according to an embodiment of the present disclosure.

[0030] Figure 5 Schematic diagram of the principle of correction based on residual detection in an embodiment of the present disclosure.

[0031] Figure 6 Schematic diagram of a device for predicting the health status of a lithium battery according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the present disclosure is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0033] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should have the usual meanings understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the described object changes, the relative position relationship may also change accordingly.

[0034] It is understandable that before using the technical solutions disclosed in the various embodiments of this disclosure, the type, scope of use, usage scenarios, etc. of the personal information involved in this disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0035] For example, in response to a user's active request, a prompt message is sent to the user to clearly inform the user that the operation requested will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the electronic device, application, server, storage medium, or other software or hardware that performs the operations of the disclosed technical solution based on the prompt message.

[0036] It is understandable that the above notification and user authorization process are merely illustrative and do not limit the implementation of the present disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of the present disclosure.

[0037] Figure 1 A schematic diagram of a lithium battery health status prediction architecture according to an embodiment of the present disclosure is shown. Figure 1 The lithium battery health status prediction architecture 100 may include a server 110, a terminal 120, and a network 130 that provides a communication link. The server 110 and the terminal 120 may be connected via a wired or wireless network 130. The server 110 may be an independent physical server, a server cluster or a distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, security services, and CDN.

[0038] Terminal 120 can be implemented in hardware or software. For example, when implemented in hardware, terminal 120 can be any electronic device with a display screen that supports page display, including but not limited to smartphones, tablet computers, e-book readers, laptop computers, and desktop computers. When terminal 120 is implemented in software, it can be installed in the electronic devices listed above; it can be implemented as multiple software or software modules (such as software or software modules used to provide distributed services), or it can be implemented as a single software or software module, and no specific limitations are given here.

[0039] It should be noted that the method for predicting the health status of a lithium battery provided in the embodiment of the present application can be executed by the terminal 120 or by the server 110. It should be understood that Figure 1 The number of terminals, networks, and servers in the embodiment is for illustration only and is not intended to limit the number of terminals, networks, and servers.

[0040] Figure 2 FIG. 2 shows a schematic diagram of the hardware structure of an exemplary electronic device 200 provided in an embodiment of the present disclosure. Figure 2 As shown, electronic device 200 may include: processor 202, memory 204, network module 206, peripheral interface 208 and bus 210. Processor 202, memory 204, network module 206 and peripheral interface 208 are connected to each other through bus 210 in communication with each other within electronic device 200.

[0041] The processor 202 may be a central processing unit (CPU), a neural network processor (NPU), a microcontroller (MCU), a programmable logic device, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or one or more integrated circuits. The processor 202 may be used to perform functions related to the technology described in this disclosure. In some embodiments, the processor 202 may also include multiple processors integrated into a single logical component. For example, Figure 2 As shown, the processor 202 may include a plurality of processors 202a, 202b, and 202c.

[0042] The memory 204 may be configured to store data (eg, instructions, computer code, etc.). Figure 2 As shown, the data stored in the memory 204 may include program instructions (for example, program instructions for implementing the method for predicting the health status of a lithium battery in an embodiment of the present disclosure) and data to be processed (for example, the memory may store configuration files of other modules, etc.). The processor 202 may also access the program instructions and data stored in the memory 204, and execute the program instructions to operate on the data to be processed. The memory 204 may include a volatile storage device or a non-volatile storage device. In some embodiments, the memory 204 may include a random access memory (RAM), a read-only memory (ROM), an optical disk, a magnetic disk, a hard disk, a solid-state drive (SSD), a flash memory, a memory stick, etc.

[0043] The network module 206 can be configured to provide the electronic device 200 with communication with other external devices via a network. The network can be any wired or wireless network capable of transmitting and receiving data. For example, the network can be a wired network, a local wireless network (e.g., Bluetooth, WiFi, near field communication (NFC)), a cellular network, the Internet, or a combination thereof. It will be appreciated that the type of network is not limited to the specific examples above. In some embodiments, the network module 206 can include any number of network interface controllers (NICs), radio frequency modules, transceivers, modems, routers, gateways, adapters, cellular network chips, and the like.

[0044] The peripheral interface 208 can be configured to connect the electronic device 200 to one or more peripheral devices to implement information input and output. For example, the peripheral devices can include input devices such as a keyboard, a mouse, a touchpad, a touch screen, a microphone, and various sensors, and output devices such as a display, a speaker, a vibrator, and an indicator light.

[0045] The bus 210 can be configured to transmit information between the various components of the electronic device 200 (e.g., the processor 202, the memory 204, the network module 206, and the peripheral interface 208), such as an internal bus (e.g., a processor-memory bus), an external bus (USB port, PCI-E bus), etc.

[0046] It should be noted that although the architecture of the electronic device 200 shown above only shows the processor 202, the memory 204, the network module 206, the peripheral interface 208, and the bus 210, in a specific implementation, the architecture of the electronic device 200 may also include other components necessary for normal execution. In addition, it will be understood by those skilled in the art that the architecture of the electronic device 200 may also include only the components necessary to implement the embodiments of the present disclosure, and does not necessarily include all the components shown in the figure.

[0047] In related art, SOH prediction methods based on the Transformer model use a multi-layer Transformer encoder to process time series data, capturing long-term dependencies through a self-attention mechanism. A fully connected layer outputs a predicted SOH value. The Transformer model is a neural network structure designed for processing sequential data. Through self-attention and positional encoding, it efficiently combines global dependencies in time series with sequential information, significantly outperforming traditional data-driven methods in prediction tasks. The Transformer model generally adopts an encoder-decoder architecture. The encoder layer consists of multiple identical layers, each containing two sublayers: the first sublayer implements a multi-head self-attention mechanism, and the second sublayer is a simple, positionally arranged, fully connected feedforward network. Both sublayers utilize residual connections and layer normalization. The decoder layer has a similar architecture to the encoder layer, but also includes an additional multi-head self-attention mechanism to process the encoder output sequence. SOH prediction is a many-to-one regression problem based on multiple features, so SOH prediction can be achieved using only the encoder layer of the Transformer model. This method directly learns battery aging patterns from historical data and achieves capacity prediction by mapping health features to SOH. However, there are also significant limitations: First, its prediction results are poorly interpretable, and the accuracy of the model is highly dependent on the quality of the selected health features and the learning ability of the training algorithm. A large amount of high-quality training data is required to ensure the robustness and stability of the model. Secondly, this method is extremely sensitive to measurement noise. During the long-term service of the battery, as the sensor performance degrades, the collected data will gradually introduce systematic errors due to environmental changes and improper operation, resulting in a significant decrease in prediction accuracy. In addition, due to the complete lack of physical mechanism constraints in the model, its prediction results may violate the basic physical laws of battery degradation, such as predicting irreversible processes such as capacity rebound. Finally, this method is difficult to adapt online to the aging differences between individual batteries, and cannot dynamically adjust the model parameters according to real-time data to adapt to the aging characteristics of different batteries, limiting its application value in complex practical scenarios.

[0048] SOH prediction methods based on empirical models and traditional EKF are model-based methods that simulate battery degradation behavior by building accurate battery models. These established models are usually combined with filtering techniques to achieve accurate SOH estimation. Empirical models are used to fit the relationship between battery aging and cycle number to characterize the battery aging process. The single exponential empirical model is as follows:

[0049] SOH k =SOH k-1 -βe -γk ; where β is the decay speed and γ is the decay rate.

[0050] Establish discrete state equations and observation equations based on the empirical model:

[0051]

[0052] z k =H·x k +v k ,H=[1 0 0]

[0053] Among them, the state vector x k =[SOH(k),β(k),γ(k)] T , process noise w=[w1,w2,w3] T , z k is the SOH observation value output by Transformer, v k is the observation noise.

[0054] The covariance matrix Q is:

[0055] Linearize the nonlinear state equation and calculate the Jacobian matrix F:

[0056]

[0057] This method relies heavily on accurate model parameter initialization and is extremely sensitive to the initial parameter settings. Slight parameter deviations may lead to filtering divergence and prediction failure. Secondly, the traditional EKF framework uses fixed noise covariance parameters, which makes it difficult to dynamically adapt to the changes in degradation characteristics of the battery at different aging stages, resulting in a decrease in state estimation accuracy as the number of cycles increases.

[0058] Therefore, how to improve the accuracy of the prediction results of the health status of lithium batteries has become an urgent problem that needs to be solved.

[0059] In view of this, the present disclosure provides a lithium battery health status prediction method and related equipment. By extracting the characteristic sequence of power data in the charge and discharge cycle of the lithium battery and combining it with the battery capacity to generate a health status (SOH) value sequence, the correlation between the characteristics and SOH is used to screen key target sequences to reduce data redundancy. Subsequently, a preliminary prediction is performed based on the target sequence to obtain a first state prediction value. The prediction error is then corrected in real time through residual detection technology to generate a corrected prediction value and its covariance matrix to quantify uncertainty. Finally, the preliminary prediction and the corrected prediction results are integrated through a dynamic weight allocation mechanism, in which the weight is dynamically adjusted according to the residual size and the trace of the covariance matrix. This allows the model to automatically increase the weight of the corrected value when the data fluctuates and rely more on the preliminary prediction when the data is stable, thereby effectively suppressing noise interference and improving prediction robustness. It can reduce the SOH prediction error of lithium batteries, especially in the late stage of battery aging, and can significantly capture the capacity mutation characteristics, providing a more reliable status assessment basis for the battery management system (BMS).

[0060] See also Figure 3 , Figure 3 A schematic flow chart of a method for predicting the health status of a lithium battery according to an embodiment of the present disclosure is shown. The method for predicting the health status of a lithium battery according to an embodiment of the present disclosure can be deployed on a terminal or a server. Figure 3 In the method 300 for predicting the health status of a lithium battery, the method 300 may further include the following steps.

[0061] In step S310 , power data and battery capacity of the lithium battery during multiple charge and discharge cycles are obtained.

[0062] A charge-discharge cycle refers to the complete process of a lithium battery discharging from a fully charged state to a cutoff voltage and then recharging to a fully charged state. Power data refers to the electrical parameters recorded during the charge-discharge process of a lithium battery, reflecting the battery's real-time operating status. Specifically, power data can include voltage and voltage-related data, such as charge-discharge curves: voltage changes over time / capacity in characteristic phases (e.g., constant current charging, constant voltage charging). Sudden voltage changes (e.g., overcharge / overdischarge) can accelerate battery aging, and the shape of the voltage curve can indicate battery health. Power data can also include current and current-related data, such as the charge-discharge rate (C-rate): For example, 1C represents a full charge / discharge of the battery's capacity in 1 hour. A high C-rate can exacerbate battery heating and polarization. Current fluctuations (e.g., during rapid acceleration / braking in electric vehicles) affect the instantaneous load on the battery, necessitating data monitoring to prevent overload. Power data can also include temperature and temperature-related data, such as thermal management: the optimal operating temperature for lithium batteries is 20-40°C. High temperatures accelerate side reactions, while low temperatures cause capacity fading and increase internal resistance. Temperature data is used to modify battery model parameters (such as internal resistance and capacity) to improve prediction accuracy. Power data can include internal resistance, AC internal resistance (ACIR) and DC internal resistance (DCIR): measured by small-signal AC excitation or high-current pulses, respectively. Increased internal resistance is a direct indicator of battery aging and is strongly correlated with capacity decay. Power data can include energy and power, for example, energy efficiency: the ratio of discharge energy to charge energy, which reflects battery cycle loss. Energy / power data is used to evaluate battery performance under specific operating conditions.

[0063] Battery capacity can refer to the amount of charge that a battery can release under specific conditions (such as temperature and discharge rate), and is measured in ampere-hours (Ah) or milliampere-hours (mAh).

[0064] In step S320, feature extraction is performed on the power data to obtain a plurality of feature sequences, and a health status value sequence of the lithium battery is determined based on the battery capacity.

[0065] Among them, in the prediction of the health status of lithium batteries, state characterization can be achieved through multi-dimensional signal analysis and aging index quantification: time domain analysis (such as statistical mean, variance, and range to capture dynamic fluctuations) is used for power data such as voltage, current, and temperature during the charging and discharging process to generate a multi-dimensional feature sequence reflecting the internal electrochemical state of the battery; at the same time, the SOH value is calculated based on the ratio of the actual battery capacity and the rated capacity measured synchronously to form a continuous aging state label sequence. The raw power data is converted into interpretable physical and chemical indicators through the feature sequence, and the data-state mapping relationship is established in combination with the SOH sequence. Compared with directly using the original data, the convergence speed of the prediction model can be increased and the prediction error can be reduced.

[0066] Specifically, the SOH value is calculated based on the battery capacity, and the calculation formula is:

[0067]

[0068] Among them, Q t Q is the maximum capacity of the battery currently available, n This is the factory rated capacity of the battery.

[0069] In some embodiments, the power data is subjected to feature extraction to obtain multiple feature sequences, including the following:

[0070] Extracting a first duration of constant current charging based on a first current range in a plurality of the charge and discharge cycles to obtain a first characteristic sequence;

[0071] extracting a second duration of constant voltage discharge in a plurality of the charge and discharge cycles to obtain a second characteristic sequence;

[0072] Extracting a first slope of a charging voltage varying with time during constant current charging within a first current range in a plurality of the charge-discharge cycles to obtain a third characteristic sequence;

[0073] extracting a second slope of the discharge voltage changing with time during constant voltage discharge in a plurality of the charge and discharge cycles to obtain a fourth characteristic sequence;

[0074] extracting a first average voltage within a first capacity range during a charging phase of a plurality of the charge-discharge cycles to obtain a fifth characteristic sequence;

[0075] extracting a second average voltage within the first capacity range during the constant voltage discharge phase of the plurality of charge and discharge cycles to obtain a sixth characteristic sequence;

[0076] Extracting minimum values ​​of the IC curve in the constant current discharge stage of the plurality of charge and discharge cycles within the first voltage range to obtain a seventh characteristic sequence;

[0077] extracting peak values ​​of the IC curves in the constant current discharge stage of the plurality of charge and discharge cycles to obtain an eighth characteristic sequence;

[0078] The discharge voltage corresponding to the peak value of the IC curve in the constant current discharge stage in the multiple charge and discharge cycles is extracted to obtain a ninth characteristic sequence.

[0079] Specifically, see Figure 4 , Figure 4 A schematic diagram illustrating a lithium battery health status prediction according to an embodiment of the present disclosure is shown. The battery degradation characteristics include the duration t of the charge and discharge voltage curves within the same voltage range, the curve slope g, the average voltage Vaverage within the 10%-90% capacity range, the minimum value of the IC curve within the 3.8V-4.1V range, the peak voltage, and the peak position. Specifically, one or more of the following sub-steps may be included:

[0080] The duration of the charging stage at a constant current of 1.5A is collected for each charge and discharge cycle of the lithium battery, and all the time constitutes the decay characteristic sequence 1;

[0081] Collect the duration of the constant voltage discharge stage under each charge and discharge cycle of the lithium battery, and all the time constitutes the decay characteristic sequence 2;

[0082] Collect the slopes of the Vt curves of the lithium battery at the 1.5A constant current charging stage under each charge and discharge cycle. All slopes constitute the third decay characteristic sequence;

[0083] Collect the slopes of the Vt curves during the constant voltage discharge phase of each charge and discharge cycle of the lithium battery. All slopes constitute the fourth decay characteristic sequence.

[0084] Collect the average voltage Vaverage in the range of 10%-90% capacity during the charging stage of each charge and discharge cycle of the lithium battery. All the average voltages constitute the decay characteristic sequence 5;

[0085] Collect the average voltage Vaverage in the range of 10%-90% capacity during the constant voltage discharge stage of each charge and discharge cycle of the lithium battery. All the average voltages constitute the decay characteristic sequence 6;

[0086] Collect the minimum values ​​of the IC curve in the range of 3.8V-4.1V during the constant current discharge stage of each charge and discharge cycle of the lithium battery. All minimum values ​​form characteristic sequence seven;

[0087] Collect the peak values ​​of the IC curve during the constant current discharge stage of each charge and discharge cycle of the lithium battery, and all peak values ​​form a characteristic sequence eight;

[0088] The discharge voltage corresponding to the peak value of the IC curve during the constant current discharge stage of each charge and discharge cycle of the lithium battery is collected, and all voltage values ​​form a characteristic sequence nine.

[0089] Among them, the capacity Q is expressed as:

[0090]

[0091] Where Q is the battery charge capacity, t1 is the charge start time, t2 is the charge end time, and I is the current.

[0092] Among them, the IC curve calculation formula is:

[0093]

[0094] In step S330 , a target sequence is selected from the feature sequence based on the correlation coefficient between the feature sequence and the health status value sequence.

[0095] Among them, the correlation strength between the feature sequence and the state of health (SOH) value sequence can be quantified through statistical correlation analysis, thereby screening out the key features that are most sensitive to battery aging. First, the Pearson correlation coefficient or Spearman rank correlation coefficient of each feature sequence (with the SOH sequence) is calculated to measure the closeness of its linear or monotonic relationship; then a correlation threshold is set (such as |r|>0.8), and feature sequences that are highly correlated (positively or negatively correlated) with SOH are retained as target sequences, and redundant or noisy features are eliminated. This can reduce the data dimension and eliminate the interference of irrelevant features on the prediction model, while retaining the features with the most aging characterization capabilities, so that the subsequent prediction model can maintain high accuracy while improving computational efficiency, and has stronger generalization when migrating across battery models due to the clear physical meaning of the features.

[0096] In some embodiments, screening a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence includes:

[0097] The correlation coefficient between the feature sequence and the health status value sequence is calculated based on the Pearson correlation coefficient; wherein the Pearson correlation coefficient r xy Expressed as:

[0098]

[0099] Among them, x i is the i-th element of the feature sequence, is the average value of the characteristic sequence, y i is the i-th element of the health status value sequence, is the average value of the health status value sequence;

[0100] The characteristic sequence whose correlation coefficient meets the requirements is determined as the target sequence.

[0101] Specifically, r xy The correlation coefficient is between -1 and 1. When the absolute value of the correlation coefficient is greater than 0.8, it indicates that the extracted features are strongly correlated with the SOH and can effectively reflect the degree of battery degradation. The extracted multiple features are input into the data-driven model for training and testing. The synergy of multiple features helps further improve the robustness and reliability of the model.

[0102] In step S340, a first state prediction value is obtained based on the target sequence and the health state value sequence.

[0103] Among them, by constructing a feature-state mapping model, using the screened target sequence (such as voltage fluctuation rate, internal resistance change rate and other features that are strongly correlated with the health state) as input, machine learning or physical models can be used to regress and predict the health state value sequence: specifically, algorithms such as support vector regression (SVR), long short-term memory network (LSTM) or equivalent circuit model can be selected. The nonlinear relationship between features and SOH is learned during the training phase, and the first state prediction value is output based on the real-time feature sequence during the prediction phase. In this way, by focusing on key aging features, the convergence speed and generalization ability of the prediction model are significantly improved. At the same time, because the physical meaning of the target sequence is clear, the model's interpretability is enhanced, and the dominant factors of battery capacity attenuation can be clearly identified, providing a high-confidence initial prediction benchmark for subsequent residual correction and dynamic fusion.

[0104] Specifically, if Figure 4 As shown in Figure 1, the feature sequence and health status value sequence can be combined into a new dataset and divided into a training set and a test set in an appropriate ratio. The feature sequence and the corresponding health status value sequence in the training set are input into the Transformer model for preliminary training and the Transformer preliminary prediction value is obtained.

[0105] In some embodiments, obtaining a first state prediction value based on the target sequence and the health state value sequence includes:

[0106] Obtaining an input feature sequence based on the fusion of the target sequence and the health status value sequence;

[0107] Generating a position encoding matrix of the input feature sequence based on trigonometric functions;

[0108] Combining the position encoding matrix and the input feature sequence to obtain an input feature sequence with position encoding;

[0109] Performing self-attention feature extraction on the position-encoded input feature sequence to obtain an attention feature;

[0110] The first state prediction value is determined based on the attention feature.

[0111] Specifically, if Figure 4 As shown in the figure, the Transformer model can add positional encoding information to the input feature sequence and generate positional encoding values ​​through sine and cosine functions; positional encoding enables the model to recognize the temporal relationship of features in the sequence. For example, positional encoding can be generated by sine and cosine functions: for input position Pos and dimension i:

[0112] Among them, d model is the hidden layer dimension of the entire model, d model =h×d k , h is the number of attention heads. The positional encoding is added to the input embedding vector: X'=X+PE.

[0113] The feature sequence can be processed by the multi-head self-attention mechanism. Specifically, the input sequence X is linearly transformed into a query matrix (Query), a key matrix (Key), and a value matrix (Value). Q = W Q X, K = W K X, V = W V ·X, where W Q 、W K and W V Is a trainable weight matrix. Calculate the scaled dot product attention: Among them, d k is the dimension of the key matrix K.

[0114] For multi-head attention processing, the output of each head is concatenated and linearly projected. The dimension of each head is d k :

[0115] MultiHead(Q,K,V)=Concat(head1,...,head h )W O .

[0116] Among them, W O Is a trainable projection matrix. In the multi-head attention mechanism, W O is a trainable weight matrix used to project the concatenated multi-head output back to the original dimension. O The shape is [h×d k ,d model ] matrix, the spliced ​​multi-head output (dimension h×d k ) linearly transform back to the original dimension d model .

[0117] The attention output is further processed by the Feed Forward Network (FFN):

[0118] FFN(x)=ReLU(xW1+b1)W2+b2, where W1 and W2 are weight matrices, and b1 and b2 are biases. FFN(x) can be a component in each encoder layer. The model's prediction result is obtained by passing the features of the last time step of the encoder's final output through a projection network. A physical constraint mechanism can be introduced by adding a single exponential empirical model βe to the loss function. -kγ The penalty item,

[0119]

[0120] Among them, λ = 0.1 is the constraint strength coefficient, k i Represents the current cycle number. This constraint ensures that the Transformer prediction value conforms to the physical law of exponential battery capacity decay, avoiding prediction results that violate the battery degradation mechanism (such as capacity recovery). When training the Transformer model, the trainable parameters in the joint model (primarily the parameters of the Transformer and MLP dynamic weight network, including W1 and W2) can be optimized through gradient descent to minimize the loss function composed of prediction error and physical consistency error.

[0121] In step S350, the first state prediction value is corrected based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value.

[0122] Among them, residual analysis is used to quantify the deviation between the first-state predicted value and the actual state of health (SOH), and a statistical model is used to dynamically correct the error. First, the residual sequence between the predicted value and the actual SOH is calculated. The mean, variance, and other parameters of the residual are statistically analyzed through a sliding window to construct a residual distribution model (such as a Gaussian mixture model or a Kalman filter noise model). Then, based on the statistical characteristics of the residual model (such as the confidence interval), a correction term is generated to compensate the first predicted value to obtain a corrected predicted value. At the same time, the statistical uncertainty of the correction value is quantified through the covariance matrix (for example, the diagonal elements represent the prediction variance at each moment, and the off-diagonal elements reflect time correlation). This can improve the robustness of the prediction. Residual detection can reduce the prediction error. The covariance matrix provides an error confidence basis for subsequent dynamic weight fusion, allowing the model to automatically increase the weight of the correction value when the data fluctuates and rely more on the initial prediction when the data is stable, ultimately narrowing the fluctuation range of the prediction error.

[0123] In some embodiments, the first state prediction value is corrected based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value, including:

[0124] Constructing a state equation and an observation equation for a single exponential empirical model based on the first state prediction value;

[0125] Obtaining the corrected predicted value based on the state equation and the observation equation;

[0126] determining an exponential moving average based on a residual between the first state prediction value and the revised prediction value;

[0127] updating process noise covariance and measurement noise based on the residual and the exponential moving average;

[0128] The corrected prediction value is updated based on the process noise covariance and the measurement noise, and the covariance matrix is ​​obtained.

[0129] Specifically, if Figure 5 As shown, Figure 5 A schematic diagram of the principle of correction based on residual detection according to an embodiment of the present disclosure is shown. For example, AEKF correction based on residual detection can dynamically adapt to different battery individuals by adjusting β and γ parameters in real time, including the following sub-steps:

[0130] The initial prediction value z output by Transformer k As the observation value, it is input into the AEKF module for correction. First, the state equation based on the single exponential empirical model is established:

[0131]

[0132] The observation equation is: k =H·x k +v k , H=[1 0 0].

[0133] Executing the prediction yields:

[0134]

[0135] Calculate the observation value z k and EKF predictions The residual between:

[0136] The residual reflects the instantaneous difference of the model and drives the Kalman gain K k Calculation is used to adjust the weights of predicted values ​​and observed values. The change in residuals can be used to determine whether the system state has changed suddenly.

[0137] Exponential Moving Average (EMA) is a time series smoothing method that assigns exponentially decaying weights to historical data, making recent data have a greater impact on the current average, while the impact of distant data gradually weakens.

[0138] Calculate the exponential moving average Δ′ of the residual at the current moment k :Δ′ k =β·Δ k +(1-β)·Δ′ k-1 .

[0139] Among them, △ k is the residual at the current moment, β is the smoothing coefficient that controls the weight of new data, β∈

[0140] (0, 1). The smaller β is, the more historical data is included, and the stronger the smoothing effect is; the larger β is, the more sensitive it is to fluctuations in recent data, and the weaker the smoothing effect is.

[0141] Process noise covariance update:

[0142]

[0143] Regarding process noise, since SOH decays exponentially, in the early stages of battery aging, SOH decays slowly and process noise is low. In the later stages of aging, SOH decay accelerates and uncertainty increases, requiring a larger noise term. Therefore, the overall noise grows logarithmically with the number of cycles k, with a rapid increase initially and a gradual decrease later. Large prediction residuals indicate that the model's β and γ parameters deviate from reality, requiring increased degrees of freedom for adjusting β and γ. When the residuals are small, maintaining low noise stabilizes the parameters. γ is generally more stable than β, so a smaller base noise is used.

[0144] Measurement Noise Update:

[0145]

[0146] Use EMA residual sequence for smoothing to identify mutation points. Then set the residual fluctuation range. When the residual is in the abnormal state range, R immediately increases 100 times to quickly respond to abnormal changes; when the residual is in the normal fluctuation range, R remains unchanged; when 0<|Δ k |<0.1Δ′ k When , the model consistency is considered high, and R is slowly reduced. Using this AEKF as a post-processing module for the Transformer output, whether the system is in a stable state or subjected to abnormal interference, not only can it detect abnormal moments, resist transient interference, quickly smooth noise, and reduce the impact of sensor noise on predictions, but it can also correct prediction deviations through empirical models to obtain predictions that conform to long-term stable attenuation trends.

[0147] Perform the update:

[0148]

[0149] Dynamic estimation of β and γ can dynamically track the changes in battery aging characteristics. Kalman gain K kAccording to the predicted covariance P pred and observation noise R k Adaptive adjustment: if the Transformer output error is small, it relies more on the observed value correction state; if the model prediction error is large, it trusts the observed data more.

[0150] In step S360, a dynamic weight is determined based on the first state prediction value, the revised prediction value, the residual between the first state prediction value and the revised prediction value, and the trace of the covariance matrix.

[0151] Among them, a dynamic weight allocation mechanism based on uncertainty quantification can be constructed by fusing the statistical characteristics of multi-source prediction information. First, the residual sequence of the first-state prediction value and the revised prediction value is used, combined with the trace of the covariance matrix (characterizing the spatiotemporal correlation of the error of the revised prediction value), to calculate the relative credibility index of the two (such as the prediction consistency measure based on the Mahalanobis distance or the Kalman gain coefficient); then the weight allocation strategy is dynamically adjusted according to the current operating condition of the battery. When the data is stable and the model credibility is high (such as the diagonal element value of the covariance matrix is ​​small), the weight of the revised prediction value is increased to eliminate the systematic bias of the initial model; when the data is abnormal or the model mismatch is large (such as a sudden increase in the residual or a significant non-diagonal element of the covariance matrix), the weight of the initial prediction value is increased to avoid overfitting noise. A dynamic balance between prediction robustness and accuracy is achieved.

[0152] In some embodiments, determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix includes:

[0153] obtaining an input eigenvector based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix;

[0154] Performing a nonlinear transformation on the input feature vector based on a first activation function to obtain a transformation feature;

[0155] The transformed features are mapped based on a second activation function to obtain the dynamic weight.

[0156] Specifically, the trace of the covariance in the AEKF framework is set as the uncertainty indicator, and the dynamic weight is generated by MLP to obtain the fusion prediction result. The specific steps are as follows:

[0157] First, construct an input feature vector containing four key elements. The vector consists of: the initial prediction value z output by the Transformer module k , the predicted value after correction by AEKF module The residual error between two predicted values ​​|Δ k|, and the trace of the AEKF covariance matrix tr(P k The trace of the covariance matrix reflects the uncertainty of the state estimation. The larger the trace value, the higher the uncertainty of the state estimation. Therefore, the trace of the AEKF covariance tr(P k ) as the uncertainty indicator of the model. These four-dimensional features are combined into the input vector N:

[0158]

[0159] The feature vector is input into the designed multi-layer perceptron network. The MLP network adopts a two-layer structure: the first layer is a hidden layer containing 64 neurons, and the ReLU activation function is used for nonlinear transformation:

[0160] h (1) =W1·N T +b1

[0161] a (1) =ReLU(h (1) )=max(0,h (1) )

[0162] Where W1 is a 4×16 weight matrix and b1 is the corresponding bias vector. The ReLU function can effectively handle nonlinear relationships while avoiding the vanishing gradient problem.

[0163] The output of the hidden layer is then passed to the second output layer.

[0164] h (2) =W2·a (1) +b2,

[0165]

[0166] Where W2 is a 16×1 dimensional weight matrix and b2 is the corresponding bias vector. The weight α is obtained by mapping the output to (0, 1) using the sigmoid function. k for:

[0167]

[0168] In step S370, the first state prediction value and the revised prediction value are fused based on the dynamic weight to obtain a health state prediction result of the lithium battery in the target time period.

[0169] Among them, through the dynamic weight allocation mechanism, the real-time credibility of the first state prediction value and the revised prediction value is comprehensively considered to achieve the optimal fusion of the two. Specifically, based on the covariance matrix obtained by residual detection (characterizing the error distribution of the revised prediction value) and the model confidence of the first prediction value (such as sliding window statistics based on historical prediction errors), a dynamic weight function is constructed (such as using the information fusion formula in the Kalman filter or a weighting strategy based on the Mahalanobis distance) so that the weight is automatically adjusted as the battery state changes. When the residual fluctuation is small and the confidence of the revised value is high (such as the stable aging stage of the battery), the weight of the revised value is increased to suppress the systematic deviation of the initial model; when the residual is abnormal or the data noise is large (such as sudden temperature shock or sensor error), the weight of the initial prediction value is increased to avoid over-correction, which can improve the prediction accuracy.

[0170] In some embodiments, the first state prediction value and the revised prediction value are fused based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period, including:

[0171] Among them, SOH k is the health status prediction result of the kth charge and discharge cycle in the target time period, α k is the dynamic weight, z k is the first state prediction value of the kth charge and discharge cycle, is the corrected predicted value of the kth charge and discharge cycle.

[0172] MLP is used to learn the instantaneous differences and uncertainties in AEKF predictions and generate real-time dynamic weights that reflect model uncertainty. When the covariance trajectory is small, the residual is low, indicating that the prediction results are consistent, in line with the degradation trend of the pre-fitted empirical model, and the system is in a stable state. Therefore, the weight α k tends to 0, the fusion prediction result is prioritized by the Transformer prediction value. On the contrary, when the covariance trajectory is large and the residual is large, the Transformer prediction deviates seriously from the empirical model and the weight α k The fusion prediction results are mainly based on the AEKF correction value. The MLP-based dynamic weight fusion method combines the Transformer prediction results with the AEKF correction value, and can use the accuracy of the empirical model to suppress the interference of abnormal results.

[0173] As can be seen, according to the method of the embodiment of the present disclosure, a hybrid model prediction framework for lithium battery SOH is proposed by dynamically weighting the Transformer model based on multi-feature extraction and the AEKF based on the empirical model. First, multiple aging features are extracted from the voltage curve and IC curve, and a single exponential empirical model is used as a physical constraint to obtain a Transformer prediction result that conforms to the physical trend. Then, the AEKF based on residual recognition is used as the post-processing module of the Transformer, and the empirical model is used as the state equation in the AEKF to correct the initial Transformer prediction result and reduce the impact of sensor noise on the prediction. Finally, the trace of the covariance in the AEKF framework is used as the uncertainty indicator of the model to weigh the confidence of the model prediction and the observed data. The Transformer pre-estimation result, the AEKF correction result, the residual of the two, and the trace of the covariance are used as features. Dynamic weights are generated through the MLP, and the fusion prediction result of the Transformer prediction value and the AEKF correction value is output. The proposed lithium battery SOH prediction framework combines the advantages of model-based and data-driven methods to improve the robustness and accuracy of the prediction model in noisy environments. It solves the problems of data errors caused by sensor performance degradation, environmental changes and improper operation; the defects of pure data-driven methods that are easily disturbed by data errors and lack physical constraints in prediction results; and the problem that traditional EKF fixed noise parameters are not suitable for battery nonlinear degradation.

[0174] It should be noted that the method of the embodiments of the present disclosure can be performed by a single device, such as a computer or server. The method of the embodiments of the present disclosure can also be applied in a distributed scenario, where multiple devices cooperate to perform the method. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiments of the present disclosure, and the multiple devices will interact with each other to complete the method.

[0175] It should be noted that the above description is limited to some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0176] Based on the same technical concept, corresponding to any of the above embodiments and methods, the present disclosure also provides a device for predicting the health status of a lithium battery, see Figure 6 The device for predicting the health status of a lithium battery comprises:

[0177] An acquisition module is used to obtain power data and battery capacity of the lithium battery during multiple charge and discharge cycles;

[0178] a feature extraction module, configured to extract features from the power data to obtain a plurality of feature sequences, and determine a health status value sequence of the lithium battery based on the battery capacity;

[0179] a feature screening module, configured to screen a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence;

[0180] A first prediction module, configured to obtain a first state prediction value based on the target sequence and the health state value sequence;

[0181] A second prediction module is used to correct the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value;

[0182] a weight module for determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix;

[0183] A fusion module is used to fuse the first state prediction value and the corrected prediction value based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period.

[0184] For the convenience of description, the above devices are described as being functionally divided into various modules. Of course, when implementing the present disclosure, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0185] The device of the above embodiment is used to implement the corresponding lithium battery health status prediction method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0186] Based on the same technical concept, corresponding to any of the above-mentioned embodiment methods, the present disclosure also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the lithium battery health status prediction method as described in any of the above embodiments.

[0187] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0188] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the method for predicting the health status of a lithium battery as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0189] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. Within the scope of the present disclosure, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of clarity.

[0190] In addition, to simplify the description and discussion, and so as not to obscure the embodiments of the present disclosure, known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, devices may be shown in the form of block diagrams to avoid obscuring the embodiments of the present disclosure, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure are to be implemented (i.e., these details should be fully within the purview of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it will be apparent to those skilled in the art that the embodiments of the present disclosure may be implemented without these specific details or with variations in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0191] Although the present disclosure has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0192] The embodiments of the present disclosure are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A method for predicting the health status of a lithium battery, characterized in that: include: Obtain power data and battery capacity of lithium batteries during multiple charge and discharge cycles; Extracting features from the power data to obtain a plurality of feature sequences, and determining a health status value sequence of the lithium battery based on the battery capacity; Filtering a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence; Obtaining a first state prediction value based on the target sequence and the health state value sequence; Correcting the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value; determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix; The first state prediction value and the revised prediction value are fused based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period.

2. The method according to claim 1, characterized in that The power data is subjected to feature extraction to obtain multiple feature sequences, including the following: Extracting a first duration of constant current charging based on a first current range in a plurality of the charge and discharge cycles to obtain a first characteristic sequence; extracting a second duration of constant voltage discharge in a plurality of the charge and discharge cycles to obtain a second characteristic sequence; Extracting a first slope of a charging voltage varying with time during constant current charging within a first current range in a plurality of the charge-discharge cycles to obtain a third characteristic sequence; extracting a second slope of the discharge voltage changing with time during constant voltage discharge in a plurality of the charge and discharge cycles to obtain a fourth characteristic sequence; extracting a first average voltage within a first capacity range during a charging phase of a plurality of the charge-discharge cycles to obtain a fifth characteristic sequence; extracting a second average voltage within the first capacity range during the constant voltage discharge phase of the plurality of charge and discharge cycles to obtain a sixth characteristic sequence; Extracting minimum values ​​of the IC curve in the constant current discharge stage of the plurality of charge and discharge cycles within the first voltage range to obtain a seventh characteristic sequence; extracting peak values ​​of the IC curves in the constant current discharge stage of the plurality of charge and discharge cycles to obtain an eighth characteristic sequence; The discharge voltage corresponding to the peak value of the IC curve in the constant current discharge stage in the multiple charge and discharge cycles is extracted to obtain a ninth characteristic sequence.

3. The method according to claim 1, characterized in that Screening a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence includes: The correlation coefficient between the feature sequence and the health status value sequence is calculated based on the Pearson correlation coefficient; wherein the Pearson correlation coefficient r xy Expressed as: Among them, x i is the i-th element of the feature sequence, is the average value of the characteristic sequence, y i is the i-th element of the health status value sequence, is the average value of the health status value sequence; The characteristic sequence whose correlation coefficient meets the requirements is determined as the target sequence.

4. The method according to claim 1, wherein Obtaining a first state prediction value based on the target sequence and the health state value sequence includes: Obtaining an input feature sequence based on the fusion of the target sequence and the health status value sequence; Generating a position encoding matrix of the input feature sequence based on trigonometric functions; Combining the position encoding matrix and the input feature sequence to obtain an input feature sequence with position encoding; Performing self-attention feature extraction on the position-encoded input feature sequence to obtain an attention feature; The first state prediction value is determined based on the attention feature.

5. The method according to claim 1, wherein The first state prediction value is corrected based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value, including: Constructing a state equation and an observation equation for a single exponential empirical model based on the first state prediction value; Obtaining the corrected predicted value based on the state equation and the observation equation; determining an exponential moving average based on a residual between the first state prediction value and the revised prediction value; updating process noise covariance and measurement noise based on the residual and the exponential moving average; The corrected prediction value is updated based on the process noise covariance and the measurement noise, and the covariance matrix is ​​obtained.

6. The method according to claim 1, wherein Determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix includes: obtaining an input eigenvector based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix; Performing a nonlinear transformation on the input feature vector based on a first activation function to obtain a transformation feature; The transformed features are mapped based on a second activation function to obtain the dynamic weight.

7. The method according to claim 1, characterized in that The first state prediction value and the revised prediction value are integrated based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period, including: Among them, SOH k is the health status prediction result of the kth charge and discharge cycle in the target time period, α k is the dynamic weight, z k is the first state prediction value of the kth charge and discharge cycle, is the corrected predicted value of the kth charge and discharge cycle.

8. A device for predicting the health status of a lithium battery, characterized in that: include: An acquisition module is used to obtain power data and battery capacity of the lithium battery during multiple charge and discharge cycles; a feature extraction module, configured to extract features from the power data to obtain a plurality of feature sequences, and determine a health status value sequence of the lithium battery based on the battery capacity; a feature screening module, configured to screen a target sequence from the feature sequence based on a correlation coefficient between the feature sequence and the health status value sequence; A first prediction module, configured to obtain a first state prediction value based on the target sequence and the health state value sequence; A second prediction module is used to correct the first state prediction value based on residual detection to obtain a corrected prediction value and a covariance matrix of the corrected prediction value; a weight module for determining a dynamic weight based on the first state prediction value, the revised prediction value, a residual between the first state prediction value and the revised prediction value, and a trace of the covariance matrix; A fusion module is used to fuse the first state prediction value and the corrected prediction value based on the dynamic weight to obtain a health state prediction result of the lithium battery in a target time period.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 7.

Images