A method for combined prediction of state of charge and state of power in combination with state of health of a battery

Abstract

This invention provides a method for jointly predicting the state of charge (SOH) and state of power (SOC) of a battery by combining the battery's health status. First, a parameter self-learning skip mode decomposition and optimization algorithm is used to decompose the original SOH sequence into multiple oscillating and skipping components. Next, an input adaptive twin physical information diffusion model is constructed, taking the original battery data and the decomposed SOH components as input, to simultaneously predict both SOH and SOC. Then, using SOC as a constraint, combined with the battery's internal resistance and voltage, the sustained peak charge / discharge current required for SOP is estimated. Finally, using the AEPPKF model, with voltage and SOC as input, the final SOP is estimated. This method innovatively achieves dynamic joint estimation of these three parameters, enabling a more comprehensive and accurate understanding of the true state of the lithium battery, facilitating timely adjustments to usage or maintenance, effectively extending battery life, and ensuring vehicle safety.

Description

Technical Field

[0001] This invention relates to the field of battery health management technology, and specifically to a method for jointly predicting the state of charge and state of power of a battery by combining the battery's health status. Background Technology

[0002] Currently, multi-dimensional collaborative prediction and judgment of battery status has attracted widespread attention. As the requirements for safety, efficiency, and lifespan in battery applications continue to increase, single battery status estimation can no longer meet the needs of current battery health management systems and battery-based devices. Comprehensive assessment of the current state of the battery plays a significant role in battery maintenance and long-term health management.

[0003] However, current joint predictions of battery state of operation (SOP) mostly rely on deep learning joint models to improve prediction accuracy. These models are time-consuming to train and predict, highly dependent on data, and have limited practical applicability. Furthermore, most current joint predictions lack consideration of SOP, while real-world battery applications involve fluctuating power demands and diverse environmental conditions, thus deviating from actual power requirements.

[0004] Given the strong coupling between SOH, SOC, and SOP of a battery and the aforementioned problems, this invention proposes a joint prediction method for SOC and SOP that considers the battery's health status, enabling coordinated prediction of all three and better meeting actual production needs. The adaptive domain transition-tent chaotic cooperative snake optimization algorithm (TSO) is used to optimize the core decomposition parameters, thereby improving the decomposition quality of the parameter self-learning jump mode decomposition method (JMD) for battery SOH data and its adaptability to SOH data from batteries under different operating conditions. For the physics-based PIDM model, an adaptive mechanism is added during its forward noise addition process, constructing an input adaptive twin physical information diffusion model (APIDM), and achieving coordinated prediction of SOH and SOC. This method has stronger interpretability and extrapolation capabilities, lower data requirements, and stronger model generalization ability. Summary of the Invention

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing a method for jointly predicting the state of charge and state of power of a battery by combining the battery health status. This method can jointly predict the SOC and SOP of a lithium battery, which has important practical significance for improving the efficiency and safety of lithium batteries.

[0006] The method includes the following steps:

[0007] Step 1: Based on the parameter self-learning jump mode decomposition method JMD (Jump Mode Decomposition), the healthy state SOH sequence is decomposed. The core parameters of JMD are optimized by the adaptive domain transition-tent-chaotic snake optimization algorithm TSO (Transfer-domain Self-adaptive Tent-chaotic Snake Optimization), which decomposes the SOH signal into multiple amplitude-frequency modulation AM-FM (Amplitude Modulation-Frequency Modulation) oscillation components and jump components.

[0008] Step 2: Construct an Input Adaptive Twin Physical Information Diffusion Model (APIDM): a first input adaptive twin physical information diffusion sub-model APIDM1 and a second input adaptive twin physical information diffusion sub-model APIDM2. The SOH oscillation component and jump component obtained from Step 1 are used as inputs to the first input adaptive twin physical information diffusion sub-model APIDM1 to predict the SOH of the battery. Combining the actual current, temperature and voltage data of the battery, the predicted SOH result is used as the synchronous input to the second input adaptive twin physical information diffusion sub-model APIDM2 to output the predicted SOC result.

[0009] Step 3: Using the SOC output in Step 2 as a constraint, and combining parameters such as battery internal resistance and voltage, estimate the continuous peak charge and discharge current required for predicting the battery power state SOP.

[0010] Step 4: Using voltage and SOC as inputs, the SOP is estimated using the Adaptive Extended Particle Kalman Filter (AEPKF) model.

[0011] In step 1, the parameter-based self-learning skip mode decomposition method JMD decomposes the SOH sequence, including:

[0012] JMD decomposition is performed on the SOH sequence, which decomposes the sequence into AM-FM oscillation components. Jump components and center frequency The updated formula is:

[0013] ,

[0014] ,

[0015] ,

[0016] in, The Fourier transform of the Kth oscillation component in the (i+1)th iteration; For the Fourier transform of the SOH sequence; For the Fourier transform of the jump component in the i-th iteration; α is the Fourier transform of the Lagrange multipliers in the i-th iteration; α is the parameter for the bandwidth of the equilibrium oscillation component; For frequency; Let K be the center frequency of the Kth oscillation component in the i-th iteration; This represents the jump component after the (i+1)th iteration update; This represents the center frequency of the k-th oscillation component in the (i+1)-th iteration; The penalty parameter is D; the first derivative matrix is ​​T; the transpose is I; the identity matrix is ​​d; and the differential symbol is d. For the Lagrange multipliers related to the auxiliary variables in the i-th iteration, the update formula is: ; As an auxiliary variable; Let be the Lagrange multiplier in the i-th iteration. Repeat the iteration until convergence to obtain the oscillatory and jumping components of the decomposed SOH sequence.

[0017] In step 1, the optimization of the core parameters of JMD using the Adaptive Domain Transition-Tent Chaotic Cooperative Snake Optimization Algorithm (TSO) includes:

[0018] Initialize the core parameters to be optimized in JMD, including the number of modalities M, bandwidth penalty parameter α, and jump regularization parameter β, and construct the initialization parameter vector. The TSO algorithm iteratively optimizes the parameters. The parameter update formula is determined by the parameters Q and Temp, and the update formula is as follows:

[0019] ,

[0020] ,

[0021] in, and This represents the current parameter vector and random parameter vector values ​​of JMD; This represents the updated value of the j-th dimension parameter of the i-th individual in the (t+1)-th iteration; This indicates that the j-th dimension parameter value of an individual is randomly selected during the t-th iteration; It is a random number; and The corresponding fitness values ​​are respectively and ; Represents the optimal parameter vector values; The corresponding fitness value is , and These are the lower and upper bounds for the algorithm's search, respectively. and is a constant; exp is the natural exponential function; Q and Temp represent the algorithm's search power and update strategy adjustment, respectively, with the following formulas:

[0022] ,

[0023] ,

[0024] Where b and B represent the maximum number of iterations and the current number of iterations, respectively; Represents a constant;

[0025] Generating random numbers using tent mapping (Tent) :

[0026] ,

[0027] Where rand represents a random number.

[0028] Step 2 includes:

[0029] The Input Adaptive Twin Physical Information Diffusion Model (APIDM) consists of two APIDMs connected in parallel: the first Input Adaptive Twin Physical Information Diffusion Submodel APIDM1 and the second Input Adaptive Twin Physical Information Diffusion Submodel APIDM2.

[0030] First, the AM-FM oscillation components after JMD decomposition and jump components The first input to the adaptive twin physical information diffusion sub-model APIDM1 , Then, a forward noise addition-backward noise reduction process is performed on the model: the input of the first input is the adaptive twin physical information diffusion sub-model APIDM1. Gradually add Gaussian noise, as shown in the following formula:

[0031] ,

[0032] in, This indicates that, given an initial sample x0, the first... Noisy samples at each diffusion time step The conditional probability distribution; For diffusion time steps; This is the cumulative noise figure; For the first Noisy samples at each diffusion time step; It follows a Gaussian distribution; , For noise variance;

[0033] right By adding an exponential decay factor, the improved formula is:

[0034] ,

[0035] in, This indicates the first time under the condition of initial input sample x0. The noise variance parameter of the diffusion book is adaptively adjusted in a simplified step; Indicates the regulating factor; Indicates input data The physical residual, the formula is: , The attenuation coefficient is... Indicates the number of battery cycles. That is, brand new, unused batteries; then the improved ones The formula is: ;

[0036] Next, the reverse denoising process is performed, and the specific formula is as follows:

[0037] ,

[0038] ,

[0039] in, Indicates that in the given first Noisy samples in each diffusion step Conditions, No. -1 diffusion non-sample The conditional probability distribution; This indicates that the first input adaptive twin physical information diffusion submodel APIDM1 is in the... Each diffusion time step is based on noisy samples. The generated mean prediction function; The noise in the model prediction;

[0040] Loss function for generating the first input adaptive twin physical information diffusion submodel APIDM1 Train the model and predict its output:

[0041] ,

[0042] in, Indicates the weight of each time step; Represents the regularization coefficient; For the expectation, that is, for , and Take the average of all combinations; The model parameters represent the first input adaptive twin physical information diffusion sub-model APIDM1; This represents the SOH physical residual corresponding to the input data;

[0043] The model is iterated based on the loss function of the constructed first input adaptive twin physical information diffusion sub-model APIDM1, while simultaneously calculating the loss function until the loss function converges, at which point the predicted value is output. The predicted SOH value is obtained by summing the results.

[0044] The predicted output SOH of the first input adaptive twin physical information diffusion sub-model APIDM1 is simultaneously used as the input of the second input adaptive twin physical information diffusion sub-model APIDM2 to construct the input variables of the second input adaptive twin physical information diffusion sub-model APIDM2. ,in For time steps, , , Representing the voltage, current, and SOC values ​​at time k, respectively, the formulas for the forward denoising, reverse denoising, and loss function of the second input adaptive twin physical information diffusion sub-model APIDM2 are as follows:

[0045] ,

[0046] ,

[0047] ,

[0048] in, , ; These are the model parameters for APIDM2; The regularization coefficient is used. This represents the input variables of the adaptive twin physical information diffusion submodel APIDM2 given the second input. Noisy sample added at the τth diffusion time step under the condition The conditional probability distribution; This indicates adding noisy samples at the given τ-th diffusion time step. Sample at the (τ-1)th diffusion time step under the condition The conditional probability distribution; This represents the loss function of the second input adaptive twin physical information diffusion submodel APIDM2; Let represent the noisy sample at the τ-th diffusion time step; This represents the cumulative noise figure of the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step; This represents the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step, combined with the health state SOH condition and based on the noisy samples. The generated mean prediction function; This represents the noise predicted by the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step, combined with the SOH condition of the health state. This indicates the input variables under the condition of SOH (Healthy State). The corresponding State of Charge (SOC) physical residual is calculated; the iteration continues according to the mathematical formula of the second input adaptive twin physical information diffusion sub-model APIDM2, while calculating the loss function until the loss function converges, and the SOC estimate is output.

[0049] Step 3 includes: First, calculate the peak charging and discharging currents before predicting the SOP (Start of Production). Then, apply the maximum and minimum SOC (State of Charge) values ​​to the peak currents as constraints. The charging current should be constrained by the maximum SOC, and the discharging current by the minimum SOC. The formulas for calculating the minimum charging current and maximum discharging current at SOP with SOC as the constraint are as follows:

[0050] ,

[0051] in, , These represent the minimum charging current and the maximum discharging current, respectively. Indicates the duration of charging or discharging; This represents the state of charge at time t; Indicates the coulombic efficiency of the battery; Battery capacity; Indicates the upper limit of the allowed state of charge; This indicates the lower limit of the permissible state of charge.

[0052] Step 4 includes:

[0053] The state-space equations and observation equations of the AEPPKF are constructed using the following formulas:

[0054] ,

[0055] ,

[0056] in, This represents the state vector at time t, where e is the natural constant; This represents the voltage of the first polarized branch at time t; This represents the voltage of the second polarization branch at time t; Indicates the sampling time interval; and It is a time constant; and Indicates the battery's internal resistance; Represents the battery current at time t; , , The system process noise follows a Gaussian distribution; Represents the observed value at time t; Let be the open-circuit voltage at time t; This indicates the ohmic internal resistance of the battery; This indicates observation noise.

[0057] Step 4 also includes:

[0058] Perform state vector Update, resampling, and state estimation;

[0059] The state update formula is:

[0060] ,

[0061] in, This represents the i-th updated state vector; Indicates the filter gain; This indicates the deviation between the observed and predicted values; This represents the observation noise covariance at time t; Indicates the forgetting factor; Represents the residual; Represents the observation matrix; This represents the covariance matrix before the update.

[0062] Step 4 also includes:

[0063] Weight calculation and normalization:

[0064] ,

[0065] ,

[0066] in, Let represent the unnormalized weight of the i-th particle at time k; This represents the normalized weight of the i-th particle at time k; Indicates the observed likelihood; Indicates the state transition probability; This represents the probability density of importance.

[0067] The state estimation formula is:

[0068] ,

[0069] ,

[0070] in, Represents the SOC state update value at time t; N represents the total number of particles;

[0071] The final battery SOP expression under SOC constraints is estimated to be:

[0072] ,

[0073] in, This represents the sustained peak charging power of the battery under SOC-limited conditions at time k+L; This represents the sustained peak discharge power of the battery under SOC-limited conditions at time k+L; This represents the minimum continuous charging current under the SOC upper limit constraint; This represents the maximum continuous discharge current under the SOC lower limit constraint; State of charge (SOC) at time t t Corresponding open-circuit voltage; SOC t t represents the state of charge at time t; k represents the start time of discrete prediction; L represents the prediction step size; k+L represents the end time of prediction.

[0074] In summary, this invention achieves coordinated estimation and dynamic constraint control of battery operating states by constructing a progressive joint prediction framework for State of Health (SOH), State of Charge (SOC), and State of Power (SOP). This method balances prediction accuracy, physical interpretability, and engineering feasibility, providing reliable support for the safe management, lifespan extension, and efficient operation of power batteries.

[0075] The present invention also provides an electronic device, including a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method.

[0076] The present invention also provides a storage medium storing a computer program or instructions that, when the computer program or instructions are run on a computer, execute the steps of the method described.

[0077] Beneficial effects: This invention proposes a joint prediction framework with state transitivity. Instead of modeling SOH, SOC, and SOP independently, it first decomposes and predicts SOH, then uses the predicted health state as the conditional input for SOC prediction, and finally uses the predicted SOC to constrain and solve SOP at time k+L, forming a progressive prediction link of "SOH→SOC→SOP", which enhances the ability to describe multi-state coupling.

[0078] This invention addresses the problem of slow decay, local abrupt changes, and non-stationary perturbations in battery SOH sequences. It utilizes JMD to decompose the original health state signal into oscillating and jumping components, and uses TSO to adaptively optimize the number of modes, bandwidth penalty coefficient, and jumping regularization parameter, so that the decomposition result retains both degradation trend and abrupt change information, thereby improving the feature effectiveness of subsequent state prediction.

[0079] This invention constructs an input adaptive twin physical information diffusion model for SOH and SOC prediction, and combines an adaptive update mechanism of polarization voltage, ohmic internal resistance and observation noise in the SOP estimation stage. It introduces AEPKF to realize state correction, so that the data-driven prediction results and the physical constraints of the equivalent circuit work together to output an executable continuous peak charge and discharge power under the condition of satisfying the SOC safety boundary. Attached Figure Description

[0080] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, and the advantages of the present invention in the above and / or other aspects will become clearer.

[0081] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation

[0082] like Figure 1 As shown, this embodiment of the invention provides a method for jointly predicting the state of charge and state of power of a battery by combining the battery's health status, including the following steps:

[0083] Step 1: Based on the parameter self-learning jump mode decomposition method JMD, the SOH sequence is decomposed. The adaptive domain transition-tent chaos cooperative snake optimization algorithm TSO is used to optimize the core parameters of JMD, and the SOH signal is decomposed into multiple AM-FM oscillation components and jump components.

[0084] Step 2: Construct the input adaptive twin physical information diffusion model APIDM—APIDM1 and APIDM2. The SOH oscillation component and jump component obtained from Step 1 are used as the first input adaptive twin physical information diffusion sub-model APIDM1 to predict the SOH of the battery. Combining the actual current, temperature and voltage data of the battery, the predicted SOH result is used as the synchronous input of the second input adaptive twin physical information diffusion sub-model APIDM2 to output the predicted SOC result.

[0085] Step 3: Using the SOC output in Step 2 as a constraint, and combining parameters such as battery internal resistance and voltage, estimate the continuous peak charge and discharge current required for battery SOP prediction.

[0086] Step 4: Using voltage and SOC as inputs, the SOP is estimated using the Adaptive Extended Kalman Particle Filter (AEPKF) model.

[0087] In step 1, the SOH sequence decomposition based on the parameter self-learning skip mode decomposition method JMD is as follows:

[0088] JMD decomposition is performed on the SOH sequence, which decomposes the sequence into AM-FM oscillation components. Jump components and its center frequency The updated formula is:

[0089] ,

[0090] ,

[0091] ,

[0092] in, The Fourier transform of the Kth oscillation component in the (i+1)th iteration; For the Fourier transform of the SOH sequence; For the Fourier transform of the jump component in the i-th iteration; α is the Fourier transform of the Lagrange multipliers in the i-th iteration; α is the parameter for the bandwidth of the equilibrium oscillation component; For frequency; The center frequency of the Kth oscillation component in the (i+1)th iteration; D is the penalty parameter; D is the first derivative matrix; I is the identity matrix; For the Lagrange multipliers related to the auxiliary variables in the i-th iteration, the update formula is: ; As an auxiliary variable; Let be the Lagrange multiplier in the i-th iteration. Repeat the above iterations until convergence to obtain the oscillatory and jumping components of the decomposed SOH sequence.

[0093] In step 1, the specific steps for optimizing the core parameters of JMD using the Adaptive Domain Transition-Tent Chaotic Cooperative Snake Optimization Algorithm (TSO) are as follows:

[0094] Initialize the core parameters to be optimized in JMD, including: number of modalities M, bandwidth penalty parameter α, and jump regularization parameter β, and construct the initialization parameter vector. Each parameter's initial value is randomly determined to ensure a more comprehensive search range for JMD parameters. The TSO algorithm iteratively optimizes the parameters, and its parameter update formula is determined by parameters Q and Temp. The update formula is as follows:

[0095] ,

[0096] ,

[0097] In the formula, and This represents the current parameter vector and random parameter vector values ​​of JMD, with corresponding fitness values ​​as follows: and , This represents the optimal parameter vector value, and its corresponding fitness value. , and These are the lower and upper bounds for the algorithm's search, respectively. and Let Q and Temp be constants, representing the algorithm's search power and update strategy adjustment, respectively, with the following formulas:

[0098] ,

[0099] ,

[0100] Where t and T represent the maximum number of iterations and the current number of iterations, respectively; Represents a constant;

[0101] Generating random numbers using tent mapping (Tent) Chaotic randomness, characterized by its chaotic nature, exhibits good regularity and unpredictability, enabling algorithms to generate more widely and evenly distributed random numbers within the search space. This significantly enhances the algorithm's traversal capability within the search space. In the snake algorithm, this facilitates a more comprehensive exploration of the possible solution space. When the algorithm approaches the optimal solution, the characteristics of chaotic random numbers allow for finer adjustments to the search step size and direction, enabling the algorithm to converge to the optimal solution faster and more stably. The range in the original algorithm is random numbers The specific formula for the improved parameters is as follows:

[0102] ,

[0103] when At that time, adopt Scaling the random numbers spreads their distribution, allowing the algorithm to optimize in a wider search space and avoid getting trapped in local optima too early; when At that time, adopt Transforming the random numbers allows the algorithm to finely adjust the search step size and direction as it approaches the optimal solution.

[0104] In step 2, an input adaptive twin physical information diffusion model (APIDM) is constructed to simultaneously predict the battery's state of equilibrium (SOH) and state of charge (SOC). Specifically:

[0105] The Input Adaptive Twin Physical Information Diffusion Model (APIDM) consists of two parallel-connected APIDMs: APIDM1 and APIDM2. The APIDM model is an improvement on the PIDM model. Compared with the data-driven approach of current mainstream deep learning models, the PIDM model is built based on the internal physical processes and mechanisms of the battery, and can analyze the influence of the lithium ion diffusion coefficient in the electrode material on the battery capacity decay. Moreover, deep learning models are highly dependent on training data, while the PIDM model is based on deterministic physical laws, has a certain degree of anti-interference ability, and the prediction results are more stable.

[0106] First, the AM-FM oscillation components after JMD decomposition and jump components The first input to the adaptive twin physical information diffusion sub-model APIDM1 ,Right now Then, the model undergoes a "forward noise addition-backward noise reduction" process: the input to APIDM1... Gradually add Gaussian noise, as shown in the following formula:

[0107] ,

[0108] Where t is the diffusion time step; This is the cumulative noise figure; The sample contains Gaussian noise. It follows a Gaussian distribution; I represents the identity matrix. , representing the noise figure ( (This represents the noise variance, which increases with t).

[0109] Although the PIDM model incorporates physical residual loss, its forward noise addition process cannot adaptively adjust based on the physical residual. It grows linearly; therefore, for By adding an exponential decay factor, the noise intensity is reduced for sample data with large physical residuals, and vice versa. The improved formula is as follows:

[0110] ,

[0111] in, This represents the adjustment factor, which controls the degree of influence of the physical residual on the noise intensity. Indicates input data The physical residual, the formula is: , The attenuation coefficient is... Indicates the number of battery cycles. That is, brand new, unused batteries; then the improved ones The formula is: ;

[0112] Next, the reverse denoising process is performed, and the specific formula is as follows:

[0113] ,

[0114] ,

[0115] in, The first input parameters are the APIDM1 parameters of the adaptive twin physical information diffusion sub-model. The noise in the model prediction;

[0116] The loss function of the first input adaptive twin physical information diffusion sub-model APIDM1 is generated, and the model is trained and the output is predicted. Unlike deep learning models that use the error between the predicted value and the true value as the loss function, the loss function of the APIDM model incorporates more of the physical information of the battery and the constraints of the initial conditions. The specific construction is as follows:

[0117] ,

[0118] in, Indicates the weight of each time step; Denotes the regularization coefficient; e is the expectation, i.e., for t, and Take the average of all combinations;

[0119] The model is iterated based on the constructed APIDM1 loss function, and the loss function is calculated simultaneously until the loss function converges, at which point the predicted value is output. The predicted SOH value is obtained by summing the results.

[0120] The predicted output SOH of APIDM1 is also used as one of the inputs to APIDM2 to construct the second input adaptive twin physical information diffusion sub-model APIDM2. ,in For time steps, , , Let k represent the voltage, current, and SOC value, respectively. Similarly, the formulas for forward noise addition, reverse noise reduction, and loss function are as follows:

[0121] ,

[0122] ,

[0123] ,

[0124] in, , ; These are the model parameters for APIDM2; Similarly, the regularization coefficient is used; the iteration continues according to the mathematical formula of the second input adaptive twin physical information diffusion sub-model APIDM2, while calculating the loss function until the loss function converges and the SOC estimate is output.

[0125] In step 3, using the SOC output in step 2 as a constraint, and combining parameters such as battery internal resistance and voltage, the continuous peak charging current and continuous peak discharging current required for predicting the battery's SOP are estimated. Specifically, before predicting the SOP, the peak charging and discharging currents are first calculated. The maximum and minimum values ​​of SOC are added to the peak currents as constraints. The charging current should be constrained by the maximum SOC, and similarly, the discharging current should be constrained by the minimum SOC, thereby preventing overcharging and over-discharging of the battery. The formulas for calculating the minimum charging current and maximum discharging current of SOP with SOC as a constraint are as follows:

[0126] ,

[0127] in, , These represent the minimum charging current and the maximum discharging current, respectively. Indicates the charging / discharging duration; This represents the state of charge at time t; Indicates the coulombic efficiency of the battery; This refers to the battery capacity.

[0128] In step 4, the SOP is estimated using the Adaptive Extended Kalman Particle Filter (AEPKF) model, as detailed below:

[0129] The state-space equations and observation equations of the AEPPKF are constructed as follows:

[0130] ,

[0131] ,

[0132] in, This represents the state vector at time t; and This represents the voltage of the internal polarization reaction of the battery at time t; Indicates the sampling time interval; and It is a time constant; and Indicates the battery's internal resistance; Represents the battery current at time t; , , All of these are system process noises, which follow a Gaussian distribution; Represents the observed value at time t;

[0133] Let be the open-circuit voltage at time t; This indicates the ohmic internal resistance of the battery; Indicates observation noise;

[0134] Next, the state vector Updates, resampling, and state estimation:

[0135] (1) Status update

[0136] ,

[0137] in, This represents the i-th updated state vector; Indicates the filter gain; This indicates the deviation between the observed and predicted values; This represents the observation noise covariance at time t; Indicates the forgetting factor; This represents the residual (the deviation between the updated observation and the predicted value). Represents the observation matrix; This represents the covariance matrix before the update.

[0138] (2) Weight calculation and normalization

[0139] ,

[0140] ,

[0141] in, Let represent the weights of the i-th updated state vector; Indicates the observed likelihood; Indicates the state transition probability; This represents the probability density of importance.

[0142] (3) State estimation

[0143] ,

[0144] ,

[0145] in, Let represent the SOC state update value at time t; then the final SOP expression for the battery under SOC constraints can be estimated as follows:

[0146] ,

[0147] in, This represents the sustained peak charging power of the battery under SOC-limited conditions at time k+L; This represents the sustained peak discharge power of the battery under SOC-limited conditions at time k+L; This represents the minimum continuous charging current under the SOC upper limit constraint; This represents the maximum continuous discharge current under the SOC lower limit constraint; State of charge (SOC) at time t t Corresponding open-circuit voltage; SOC t t represents the state of charge at time t; k represents the start time of discrete prediction; L represents the prediction step size; k+L represents the end time of prediction.

[0148] In one embodiment, a lithium-ion power battery cell with a rated capacity of 50Ah was selected as the test object. Cyclic aging and dynamic operating condition tests were conducted at room temperature (25℃), with a sampling period of 1 second. During the experiment, terminal voltage, current, surface temperature, capacity, number of cycles, and corresponding SOH, SOC, and SOP calibration data were collected and divided into training, validation, and test sets in chronological order. The above-mentioned test object, sampling period, and ambient temperature are merely examples and do not constitute a limitation on the scope of protection of this invention.

[0149] In step 1, the self-learning skip mode decomposition method (JMD) is used to optimize the input parameters of the SOH time series obtained from the test. The adaptive domain transition-tent chaotic cooperative snake optimization algorithm (TSO) is then employed to jointly optimize the number of modes M, the bandwidth penalty parameter α, and the skip regularization parameter β. Optionally, the population size of TSO is set to 30, the maximum number of iterations is set to 80, the search range of M is set to 3 to 8, the search range of α is set to 100 to 3000, and the search range of β is set to 0.0001 to 0.1. Through this step, the optimal JMD parameter combination, several AM-FM oscillation components, and skip components used to characterize the abrupt decline of SOH can be obtained, thereby separating the steady decline trend, local fluctuations, and abrupt information in the original SOH sequence.

[0150] In step 2, the AM-FM oscillation component and jump component obtained in step 1 are used as the input to the first input adaptive twin physical information diffusion sub-model APIDM1. Forward noise addition, reverse noise reduction, and physical residual constraint training are performed on each component, and the prediction results of each component are superimposed and reconstructed to obtain the SOH prediction result. Optionally, the diffusion time step is set to 100, and the physical residual regularization coefficient is set to 0.01. Through this step, the SOH prediction sequence at the target prediction time can be obtained, enabling subsequent SOC estimation to synchronously perceive the battery aging state.

[0151] In step 2, the battery terminal voltage, current, temperature, and the SOH prediction result output by APIDM1 are further combined to form the input variables of the second input adaptive twin physical information diffusion sub-model APIDM2, and the same forward noise addition, backward noise reduction, and physical consistency loss function are used for training. Through this step, the SOC estimation result at the target prediction time can be obtained, and the SOC prediction result is constrained by both real-time operating condition data and SOH degradation information, reducing the cumulative error caused by relying solely on voltage and current data.

[0152] In step 3, using the SOC estimation result output in step 2 as a constraint, and combining it with battery capacity, coulombic efficiency, open-circuit voltage, ohmic internal resistance, and prediction time window, the sustained peak charging current that satisfies the upper limit constraint of SOC and the sustained peak discharging current that satisfies the lower limit constraint of SOC are calculated respectively. Optionally, the lower limit of SOC is set to 10%, the upper limit of SOC is set to 90%, and the prediction time window is set to 10s. Through this step, charging current boundaries and discharging current boundaries that do not cause overcharging or over-discharging risks can be obtained, providing safety constraints for subsequent SOP estimation.

[0153] In step 4, the SOC estimation result obtained in step 2 and the battery terminal voltage are used as inputs to the adaptive extended Kalman particle filter (AEPKF) model. The continuous peak charge / discharge current boundary obtained in step 3 is then used for particle prediction, weight update, normalized resampling, and state estimation. Optionally, the particle number is set to 100, and the forgetting factor is set to 0.95. Through this step, the continuous peak charging power and continuous peak discharging power under SOC constraints can be obtained, i.e., the battery SOP estimation result.

[0154] In the above embodiment, step 1 obtains the SOH decomposition component and the jump component, step 2 obtains the SOH prediction result and the SOC estimation result, step 3 obtains the continuous peak charge and discharge current that satisfies the SOC safety boundary, and step 4 obtains the SOP estimation result under the SOC constraint condition, thereby forming a progressive joint prediction process that combines SOH decomposition prediction, SOC co-estimation and SOP safety boundary calculation.

[0155] The results of the prediction performance comparison are shown in Table 1 below.

[0156] Table 1. Comparison of exemplary results between the method of the present invention and different comparative methods.

[0157]

[0158] As shown in Table 1, under the same test data and prediction step size, the prediction errors of SOH, SOC, and SOP of the method of this invention are all smaller than those of the comparative methods, and the overall R² is the highest. Compared with the stepwise prediction method of LSTM-SOH+EKF-SOC+traditional SOP, the RMSE of SOH prediction of the method of this invention is reduced from 2.31% to 1.08%, the RMSE of SOC prediction is reduced from 2.76% to 1.29%, and the RMSE of SOP prediction is reduced from 8.91% to 4.36%, while the overall R² is increased from 0.914 to 0.981. This indicates that the present invention, through JMD-TSO decomposition optimization, APIDM twin physical information diffusion prediction, and AEPKF joint estimation, can achieve higher prediction accuracy and better stability under battery aging and dynamic operating conditions.

[0159] This invention provides a method for jointly predicting the state of charge and state of power of a battery by combining the battery's health status. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment of the invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. All components not explicitly stated in this embodiment can be implemented using existing technologies.

Claims

1. A method for jointly predicting the state of charge and state of power of a battery by combining the battery's state of health, characterized in that, Includes the following steps: Step 1: Based on the parameter self-learning jump mode decomposition method JMD, the healthy state SOH sequence is decomposed. The adaptive domain transition-tent chaos cooperative snake optimization algorithm TSO is used to optimize the core parameters of JMD, and the SOH signal is decomposed into multiple amplitude frequency-modulation AM-FM oscillation components and jump components. Step 2: Construct an Input Adaptive Twin Physical Information Diffusion Model (APIDM): a first input adaptive twin physical information diffusion sub-model APIDM1 and a second input adaptive twin physical information diffusion sub-model APIDM2. The SOH oscillation component and jump component obtained from Step 1 are used as inputs to the first input adaptive twin physical information diffusion sub-model APIDM1 to predict the SOH of the battery. Combining the actual current, temperature and voltage data of the battery, the predicted SOH result is used as the synchronous input to the second input adaptive twin physical information diffusion sub-model APIDM2 to output the predicted SOC result. Step 3: Using the SOC output in Step 2 as a constraint, and combining the battery internal resistance and voltage, estimate the continuous peak charge and discharge current required for predicting the battery power state SOP. Step 4: Using voltage and SOC as inputs, the SOP is estimated using the Adaptive Extended Kalman Particle Filter (AEPKF) model.

2. The method of claim 1, wherein, In step 1, the parameter-based self-learning skip mode decomposition method JMD decomposes the SOH sequence, including: JMD decomposition is performed on the SOH sequence, which decomposes the sequence into AM-FM oscillation components. Jump components and center frequency The updated formula is: ， ， ， in, The Fourier transform of the Kth oscillation component in the (i+1)th iteration; For the Fourier transform of the SOH sequence; For the Fourier transform of the jump component in the i-th iteration; α is the Fourier transform of the Lagrange multipliers in the i-th iteration; α is the parameter for the bandwidth of the equilibrium oscillation component; For frequency; Let K be the center frequency of the Kth oscillation component in the i-th iteration; This represents the jump component after the (i+1)th iteration update; This represents the center frequency of the k-th oscillation component in the (i+1)-th iteration; The penalty parameter is D; the first derivative matrix is ​​T; the transpose is I; the identity matrix is ​​d; and the differential symbol is d. For the Lagrange multipliers related to the auxiliary variables in the i-th iteration, the update formula is: ; As an auxiliary variable; Let be the Lagrange multiplier in the i-th iteration. Repeat the iteration until convergence to obtain the oscillatory and jumping components of the decomposed SOH sequence.

3. The method according to claim 2, characterized in that, In step 1, the optimization of the core parameters of JMD using the Adaptive Domain Transition-Tent Chaotic Cooperative Snake Optimization Algorithm (TSO) includes: Initialize the core parameters to be optimized in JMD, including the number of modalities M, bandwidth penalty parameter α, and jump regularization parameter β, and construct the initialization parameter vector. The TSO algorithm iteratively optimizes the parameters. The parameter update formula is determined by the parameters Q and Temp, and the update formula is as follows: ， ， in, and This represents the current parameter vector and random parameter vector values ​​of JMD; This represents the updated value of the j-th dimension parameter of the i-th individual in the (t+1)-th iteration; This indicates that the j-th dimension parameter value of an individual is randomly selected during the t-th iteration; It is a random number; and The corresponding fitness values ​​are respectively and ; Represents the optimal parameter vector values; The corresponding fitness value is , and These are the lower and upper bounds for the algorithm's search, respectively. and is a constant; exp is the natural exponential function; Q and Temp represent the algorithm's search power and update strategy adjustment, respectively, with the following formulas: ， ， Where b and B represent the maximum number of iterations and the current number of iterations, respectively; Represents a constant; Generating random numbers using tent mapping (Tent) : ， Where rand represents a random number.

4. The method of claim 3, wherein, Step 2 includes: The Input Adaptive Twin Physical Information Diffusion Model (APIDM) consists of two APIDMs connected in parallel: the first Input Adaptive Twin Physical Information Diffusion Submodel APIDM1 and the second Input Adaptive Twin Physical Information Diffusion Submodel APIDM2. First, the AM-FM oscillation components after JMD decomposition and jump components The first input to the adaptive twin physical information diffusion sub-model APIDM1 , Then, a forward noise addition-backward noise reduction process is performed on the model: the input of the first input is the adaptive twin physical information diffusion sub-model APIDM1. Gradually add Gaussian noise, as shown in the following formula: ， in, This indicates that, given an initial sample x0, the first... Noisy samples at each diffusion time step The conditional probability distribution; For diffusion time steps; This is the cumulative noise figure; For the first Noisy samples at each diffusion time step; It follows a Gaussian distribution; , For noise variance; To Adding an exponential decay factor, the formula is improved as: ， in, This indicates the first time under the condition of initial input sample x0. The noise variance parameter of the diffusion book is adaptively adjusted in a simplified step; Indicates the regulating factor; Indicates input data The physical residual, the formula is: , The attenuation coefficient is... This indicates the number of battery cycles; therefore, the improved... The formula is: ; Next, the reverse denoising process is performed, and the specific formula is as follows: ， ， in, Indicates that in the given first Noisy samples in each diffusion step Conditions, No. -1 diffusion non-sample The conditional probability distribution; This indicates that the first input adaptive twin physical information diffusion submodel APIDM1 is in the... Each diffusion time step is based on noisy samples. The generated mean prediction function; The noise in the model prediction; Loss function for generating the first input adaptive twin physical information diffusion submodel APIDM1 Train the model and predict its output: ， in, Indicates the weight of each time step; Represents the regularization coefficient; For the expectation, that is, for , and Take the average of all combinations; The model parameters represent the first input adaptive twin physical information diffusion sub-model APIDM1; This represents the SOH physical residual corresponding to the input data; The model is iterated based on the loss function of the constructed first input adaptive twin physical information diffusion sub-model APIDM1, while simultaneously calculating the loss function until the loss function converges, at which point the predicted value is output. The predicted SOH value is obtained by summing the results. The predicted output SOH of the first input adaptive twin physical information diffusion sub-model APIDM1 is simultaneously used as the input of the second input adaptive twin physical information diffusion sub-model APIDM2 to construct the input variables of the second input adaptive twin physical information diffusion sub-model APIDM2. ,in For time steps, , , Representing the voltage, current, and SOC values ​​at time k, respectively, the formulas for the forward denoising, reverse denoising, and loss function of the second input adaptive twin physical information diffusion sub-model APIDM2 are as follows: ， ， ， in, , ; These are the model parameters for APIDM2; The regularization coefficient is used. This represents the input variables of the adaptive twin physical information diffusion submodel APIDM2 given the second input. Noisy sample added at the τth diffusion time step under the condition The conditional probability distribution; This indicates adding noisy samples at the given τ-th diffusion time step. Sample at the (τ-1)th diffusion time step under the condition The conditional probability distribution; This represents the loss function of the second input adaptive twin physical information diffusion submodel APIDM2; Let represent the noisy sample at the τ-th diffusion time step; This represents the cumulative noise figure of the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step; This represents the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step, combined with the health state SOH condition and based on the noisy samples. The generated mean prediction function; This represents the noise predicted by the second input adaptive twin physical information diffusion sub-model APIDM2 at the τth diffusion time step, combined with the SOH condition of the health state. This indicates the input variables under the condition of SOH (Healthy State). The corresponding State of Charge (SOC) physical residual is calculated; the iteration continues according to the mathematical formula of the second input adaptive twin physical information diffusion sub-model APIDM2, while calculating the loss function until the loss function converges, and the SOC estimate is output.

5. The method of claim 4, wherein, Step 3 includes: First, calculate the peak charging and discharging currents before predicting the SOP (Start of Production). Then, apply the maximum and minimum SOC (State of Charge) values ​​to the peak currents as constraints. The charging current should be constrained by the maximum SOC, and the discharging current by the minimum SOC. The formulas for calculating the minimum charging current and maximum discharging current at SOP with SOC as the constraint are as follows: ， in, , These represent the minimum charging current and the maximum discharging current, respectively. Indicates the duration of charging or discharging; This represents the state of charge at time t; Indicates the coulombic efficiency of the battery; Battery capacity; Indicates the upper limit of the allowed state of charge; This indicates the lower limit of the permissible state of charge.

6. The method of claim 5, wherein, Step 4 includes: The state-space equations and observation equations of the AEPPKF are constructed using the following formulas: ， ， in, This represents the state vector at time t, where e is the natural constant; This represents the voltage of the first polarized branch at time t; This represents the voltage of the second polarization branch at time t; Indicates the sampling time interval; and It is a time constant; and Indicates the battery's internal resistance; Represents the battery current at time t; , , The system process noise follows a Gaussian distribution; Represents the observed value at time t; Let be the open-circuit voltage at time t; This indicates the ohmic internal resistance of the battery; This indicates observation noise.

7. The method according to claim 6, characterized in that, Step 4 also includes: Perform state vector Update, resampling, and state estimation; The state update formula is: ， in, This represents the i-th updated state vector; Indicates the filter gain; This indicates the deviation between the observed and predicted values; This represents the observation noise covariance at time t; Indicates the forgetting factor; Represents the residual; Represents the observation matrix; This represents the covariance matrix before the update.

8. The method according to claim 7, characterized in that, Step 4 also includes: Weight calculation and normalization: ， ， in, Let represent the unnormalized weight of the i-th particle at time k; This represents the normalized weight of the i-th particle at time k; Indicates the observed likelihood; Indicates the state transition probability; This represents the probability density of importance. The state estimation formula is: ， ， in, Represents the SOC state update value at time t; N represents the total number of particles; The final battery SOP expression under SOC constraints is estimated to be: ， in, This represents the sustained peak charging power of the battery under SOC-limited conditions at time k+L; This represents the sustained peak discharge power of the battery under SOC-limited conditions at time k+L; This represents the minimum continuous charging current under the SOC upper limit constraint; This represents the maximum continuous discharge current under the SOC lower limit constraint; State of charge (SOC) at time t t Corresponding open-circuit voltage; SOC t t represents the state of charge at time t; k represents the start time of discrete prediction; L represents the prediction step size; k+L represents the end time of prediction.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing program code that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, It stores a computer program or instructions that, when run on a computer, perform the steps of the method as described in any one of claims 1 to 8.

Images