A state of charge and state of power joint prediction method for state of health of a battery

Abstract

This invention provides a joint prediction method for state of charge (SOC) and state of power (SOP) of a battery, comprising: collecting voltage, current, temperature, and capacity data of a lithium-ion battery under different operating conditions and temperatures; constructing multi-temperature SOC-open-circuit voltage relationship curves; extracting health features characterizing the degree of battery aging; establishing battery health status characterization quantities; establishing a second-order resistive-capacitive equivalent circuit model of the lithium-ion battery to achieve battery SOC prediction; co-optimizing the FFRLS forgetting factor and MIAUKF key parameters; comprehensively considering temperature constraints, voltage constraints, and SOC constraints in SOP prediction, and coupling the health status characterization quantities into the power prediction process to construct a two-layer network model to achieve high-precision prediction of battery SOP. This invention can effectively improve the accuracy, convergence, and robustness of SOC prediction under complex operating conditions, noise variations, and temperature fluctuations.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion power battery state estimation and power prediction technology, and in particular to a method for jointly predicting the state of charge and state of power for battery health. Background Technology

[0002] As a core component of new energy vehicles and energy storage systems, the state monitoring results of lithium-ion power batteries directly affect vehicle safety, energy management, and power allocation strategies. State of Charge (SOC) characterizes the battery's current remaining capacity, while State of Operation (SOP) characterizes the battery's maximum available charge / discharge power within a given time domain; both are key state variables in battery management systems. Existing research shows a clear coupling relationship between SOC and SOP, with the accuracy of SOC estimation directly impacting the accuracy of SOP boundary calculations.

[0003] Existing SOC estimation methods mostly employ equivalent circuit models combined with Kalman filtering algorithms. While these methods offer some mechanistic interpretability, the model parameters and noise statistics are prone to change under complex dynamic operating conditions, temperature variations, and battery aging, leading to decreased estimation accuracy. Current SOP estimation methods typically consider only single factors such as voltage constraints or SOC constraints, failing to adequately account for the impact of temperature variations and battery aging on maximum usable power. Consequently, they struggle to accurately reflect the battery's true power capability throughout its entire lifespan.

[0004] Furthermore, existing joint estimation methods primarily focus on SOC-SOH joint estimation, while research on explicitly coupling health status information into the SOC-SOP joint prediction link is relatively insufficient. Especially in aging battery scenarios, if the capacity parameters, internal resistance parameters, or SOC-OCV relationships established at a single temperature of a new battery are still used for SOC and SOP calculations, it will lead to biases in remaining capacity estimation and overly optimistic available power boundaries, affecting the battery management system's judgment on power control and safety warnings.

[0005] Therefore, it is necessary to propose a joint SOC-SOP prediction method that considers battery health status, introduces health status characteristics into the entire process of SOC estimation and SOP prediction, and combines online parameter identification, temperature compensation filtering and deep time series prediction model to improve prediction accuracy, robustness and engineering applicability under complex operating conditions and different life stages. Summary of the Invention

[0006] 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 (SOC) and state of power (SOP) of a battery, which can jointly predict the SOC and SOP of a lithium battery and has important practical significance for improving the efficiency and safety of lithium batteries.

[0007] The method includes the following steps:

[0008] Step 1: Extract the state of charge-open circuit voltage curves at different temperatures under different operating conditions. The state of charge-open circuit voltage curves are SOC-OCV curves: By testing the battery under various operating conditions and different temperature environments, the corresponding voltage, current and temperature data are obtained, and SOC-OCV curves under different operating conditions and temperature conditions are constructed based on the voltage, current and temperature data.

[0009] Step 2: The Forgetting Factor Recursive Least Squares (FFRLS) method is used for parameter identification. The battery model is transformed into a mathematical form that can be identified by the least squares method. The parameters of the battery equivalent circuit model are identified using FFRLS, and the weights of the old and new data are dynamically adjusted by the forgetting factor to track the dynamic changes of the system.

[0010] Step 3: The State of Charge (SOC) is predicted using the Multi-Innovation Adaptive Unscented Kalman Filter (MIAUKF) model that considers temperature compensation. Based on the traditional unscented Kalman filter, the model combines multi-innovation theory and the method of adaptively adjusting the observation noise covariance matrix, and considers the influence of temperature on battery characteristics for temperature compensation, thereby predicting the SOC of the battery.

[0011] Step 4: The improved Polar Lights Optimizer (IPLO) algorithm is used to optimize the forgetting factor of FFRLS and the parameters of the MIAUKF model. By incorporating chaotic mapping into the Polar Lights Optimizer algorithm, the global search capability of the algorithm is enhanced. The improved IPLO algorithm is used to jointly optimize the forgetting factor of FFRLS and the relevant parameters in the MIAUKF model, thereby improving the accuracy of the parameters.

[0012] Step 5: Add three constraints to the State of Power (SOP) prediction: temperature constraint, voltage constraint, and State of Charge (SOC) constraint. When evaluating and predicting SOP, temperature, voltage, and SOC are used as constraints, and the impact of the three on the battery power state is comprehensively considered.

[0013] Step 6: A two-layer convolutional neural network-gated recurrent unit-multi-head attention mechanism-NBEATS network (CNN-GRU-MHA-NBEATS) model is used to predict the power state SOP. A two-layer CNN-GRU-MHA-N-BEATS model is constructed. The convolutional neural network (CNN) is used to extract local features of the data, the time series information is processed by the gated recurrent unit (GRU), the multi-head attention mechanism (MHA) is used to enhance the model's ability to capture the relationship between different features, and the NBEATS network is used to complete trend modeling and multi-step prediction.

[0014] Beneficial effects: This invention proposes a joint prediction framework that explicitly couples battery health status characteristics to the entire process of SOC and SOP prediction. By introducing information on capacity decay, internal resistance growth, and feature evolution, it enables adaptive correction of SOC and SOP as the battery ages, which can more realistically reflect the changes in remaining capacity and available power boundaries at different life stages.

[0015] This invention proposes a joint prediction framework that explicitly couples battery health status characteristics to the entire process of SOC and SOP prediction. By introducing information on capacity decay, internal resistance growth, and feature evolution, it enables adaptive correction of SOC and SOP as the battery ages, which can more realistically reflect the changes in remaining capacity and available power boundaries at different life stages.

[0016] This invention constructs a SOC prediction link that integrates a second-order RC model, online FFRLS parameter identification, temperature-compensated MIAUKF state estimation, and IPLO parameter co-optimization. This enables joint updates of model parameters, filter parameters, and temperature compensation terms, effectively improving the accuracy, convergence, and robustness of SOC prediction under complex operating conditions, noise variations, and temperature fluctuations. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a framework diagram of the method of the present invention. Detailed Implementation

[0019] like Figure 1 As shown, this invention provides a method for jointly predicting the state of charge and state of power for battery health, including the following steps:

[0020] Step 1: Extract the state of charge-open circuit voltage curves at different temperatures under different operating conditions. The state of charge-open circuit voltage curves are SOC-OCV curves: By testing the battery under various operating conditions and different temperature environments, the corresponding voltage, current and temperature data are obtained, and SOC-OCV curves under different operating conditions and temperature conditions are constructed based on the voltage, current and temperature data.

[0021] Step 2: The Forgetting Factor Recursive Least Squares (FFRLS) method is used for parameter identification. The battery model is transformed into a mathematical form that can be identified by the least squares method. The parameters of the battery equivalent circuit model are identified using FFRLS, and the weights of the old and new data are dynamically adjusted by the forgetting factor to track the dynamic changes of the system.

[0022] Step 3: The State of Charge (SOC) is predicted using the Multi-Innovation Adaptive Unscented Kalman Filter (MIAUKF) model that considers temperature compensation. Based on the traditional unscented Kalman filter, the model combines multi-innovation theory and the method of adaptively adjusting the observation noise covariance matrix, and considers the influence of temperature on battery characteristics for temperature compensation, thereby predicting the SOC of the battery.

[0023] Step 4: The improved Polar Lights Optimizer (IPLO) algorithm is used to optimize the forgetting factor of FFRLS and the parameters of the MIAUKF model. By incorporating chaotic mapping into the Polar Lights Optimizer algorithm, the global search capability of the algorithm is enhanced. The improved IPLO algorithm is used to jointly optimize the forgetting factor of FFRLS and the relevant parameters in the MIAUKF model, thereby improving the accuracy of the parameters.

[0024] Step 5: Add three constraints to the State of Power (SOP) prediction: temperature constraint, voltage constraint, and State of Charge (SOC) constraint. When evaluating and predicting SOP, temperature, voltage, and SOC are used as constraints, and the impact of the three on the battery power state is comprehensively considered.

[0025] Step 6: A two-layer convolutional neural network-gated recurrent unit-multi-head attention mechanism-N-BEATS network (CNN-GRU-MHA-N-BEATS) model is used to predict the power state SOP. The two-layer CNN-GRU-MHA-N-BEATS model is constructed. The convolutional neural network (CNN) is used to extract local features of the data, the time series information is processed by the gated recurrent unit (GRU), the multi-head attention mechanism (MHA) is used to enhance the model's ability to capture the relationship between different features, and the NBEATS network is used to complete trend modeling and multi-step prediction.

[0026] In step 1, a health state characterization quantity is obtained. This quantity characterizes the degree of battery degradation at the current stage of its lifespan and serves as a common input for subsequent SOC and SOP predictions. This allows for the coupled correction of the health state to the state estimation and power prediction results, expressed as:

[0027] ,

[0028] Among them, h k Let SOH be the health status representation vector at the k-th sampling time. k Let ΔR be the battery health state at the k-th sampling time. k E represents the incremental characteristic of internal resistance. cv,k As an isobaric energy characteristic, ΔV k The characteristic of voltage difference at equal time intervals, ΔT k Characterized by temperature differences at equal time intervals;

[0029] The state of SOH in a healthy state is defined in terms of capacity decay, and the expression is:

[0030] ,

[0031] Among them, Q n,k Q0 represents the current maximum available capacity of the battery at the k-th sampling time, and Q0 represents the initial rated capacity of the battery.

[0032] In step 2, the lithium-ion battery is modeled using a second-order RC equivalent circuit model to describe its ohmic internal resistance characteristics and two types of polarization dynamics. In the second-order RC equivalent circuit model, the first RC branch characterizes the electrochemical polarization process, and the second RC branch characterizes the concentration polarization process. The continuous state equation and observation equation of the second-order RC equivalent circuit model are as follows:

[0033] ,

[0034] ,

[0035] ,

[0036] Where U1 is the voltage of the first polarization branch and U2 is the voltage of the second polarization branch. The first derivative of the voltage U1 in the first polarization branch. U is the first derivative of the voltage U2 in the second polarization branch, R0 is the ohmic internal resistance, R1 and R2 are the internal resistances of the first and second polarizations respectively, C1 and C2 are the capacitances of the first and second polarizations respectively, I is the battery current, and U t U is the terminal voltage. oc Let the open-circuit voltage be the voltage. After discretizing the continuous model, we get:

[0037] ,

[0038] ,

[0039] ,

[0040] Among them, U 1,k and U 2,k These are the first and second polarization voltages at the k-th sampling time, respectively; U 1,k+1 and U 2,k+1 These are the first and second polarization voltages at the (k+1)th sampling time, respectively; R 1,k and R 2,k These are the first polarization internal resistance and the second polarization internal resistance at the k-th sampling time, respectively; C 1,k and C 2,k These are the first and second polarization capacitors at the k-th sampling time, respectively; R 0,k Let I be the ohmic internal resistance at the k-th sampling time; k U is the battery current at the k-th sampling time; oc,k U is the open-circuit voltage at the k-th sampling time; t,k Let be the terminal voltage at the k-th sampling time; e is a natural constant; the second-order RC equivalent circuit model can provide a mechanistic model basis for online parameter identification and SOC state filtering.

[0041] In step 2, the parameters of the second-order RC equivalent circuit model are identified online using the recursive least squares method (FFRLS) with a forgetting factor, so that the ohmic internal resistance, polarization internal resistance, and polarization capacitance can be updated in real time according to changes in operating conditions; the FFRLS parameter update formula is:

[0042] ,

[0043] ,

[0044] ,

[0045] in, This is the parameter estimation vector at the k-th sampling time. This is the parameter estimation vector at the (k+1)th sampling time; K k+1 The gain matrix at the (k+1)th sampling time; y k+1 This is the observation output at the (k+1)th sampling time. P is the regression vector at the (k+1)th sampling time. k Let P be the error covariance matrix at the k-th sampling time. k+1 Let be the error covariance matrix at the (k+1)th sampling time. is the forgetting factor; I is the identity matrix; the superscript T indicates matrix transpose; the superscript -1 indicates matrix inversion; the forgetting factor is used to increase the weight of newly sampled data in parameter updates, reduce the lagging effect of historical data on current identification results, and thus enhance the model's ability to track dynamic conditions and parameter drift.

[0046] In step 3, the SOC prediction is described using a discrete state equation based on the ampere-hour integration mechanism and combined with the current health state to correct the capacity, so as to reflect the dynamic change of the remaining battery capacity over time; the discretized expression of the SOC state equation is:

[0047] ,

[0048] Among them, SOC k and SOC k+1 These represent the state of charge at the k-th sampling time and the state of charge at the (k+1)-th sampling time, respectively. The charging / discharging efficiency is given by Δt, which is the sampling period. This is process noise;

[0049] To improve the accuracy of open-circuit voltage mapping under different temperature conditions, a temperature compensation term is further introduced in the SOC prediction process, resulting in the open-circuit voltage expression considering temperature compensation:

[0050] ,

[0051] in, Considering the temperature-compensated open-circuit voltage at the k-th sampling time, Let T be the mapping function of SOC and open-circuit voltage at the reference temperature. k Let T be the battery temperature at the k-th sampling time. ref For reference temperature, This is the temperature compensation coefficient; through the combined effect of capacity correction and temperature compensation, the accuracy of SOC prediction under different aging stages and different temperature conditions is improved.

[0052] In step 3, the SOC prediction is implemented using the Multi-Innovation Adaptive Unscented Kalman Filter (MIAUKF) model, which considers temperature compensation. The multi-innovation mechanism enhances the utilization of observation information, and the adaptive covariance update mechanism improves the filter's adaptability to changes in noise statistics. The state update formula for MIAUKF is:

[0053] ,

[0054] ,

[0055] in, This is the prior state estimate at the k-th sampling time; This is the posterior state estimate after combining observation correction at the k-th sampling time; This is the multi-information vector at the k-th sampling time. Let be the prior error covariance matrix at the k-th sampling time. S is the posterior error covariance matrix at the k-th sampling time; k Let be the information covariance matrix at the k-th sampling time; (symbol) Indicates the conditional estimation relationship;

[0056] The multi-innovation vector is defined as:

[0057] ,

[0058] Where, ν k is the information at the k-th sampling time; p is the multiple information length, i.e., the number of information elements involved in the current state correction;

[0059] The expression for the adaptive update of the process noise and measurement noise covariance is:

[0060] ,

[0061] ,

[0062] Among them, Q k R is the process noise covariance matrix at the k-th sampling time. k Let be the measurement noise covariance matrix at the k-th sampling time. This is the adaptive adjustment coefficient for the process noise covariance; To measure the adaptive adjustment coefficient of noise covariance; C k This is the observation matrix at the k-th sampling time. The multi-innovation vector is the vector at the k-th sampling time. By combining the multi-innovation vector with the adaptive covariance, the stability and robustness of SOC prediction under noisy and uncertain conditions are improved.

[0063] In step 4, the forgetting factor of FFRLS and the parameters of the MIAUKF model are jointly optimized using the chaotic mapping-improved aurora optimization algorithm IPLO to simultaneously consider parameter identification accuracy, SOC prediction accuracy, and algorithm convergence performance; the parameter vector to be optimized is defined as:

[0064] ,

[0065] Where Θ is the parameter vector to be optimized; λ is the forgetting factor in FFRLS; and α, β, and κ are the unscented transformation parameters in MIAUKF.

[0066] The population is initialized using a Logistic chaotic mapping:

[0067] ,

[0068] ,

[0069] in, Let be the chaotic variable of the i-th individual in the j-th dimension at the t-th iteration; represents the chaotic variable for the next iteration; μ is the control parameter for the Logistic chaotic mapping. Let LB be the initial position of the i-th individual in the j-th dimension parameter; j UB is the lower bound of the j-th dimension parameter; j This is the upper bound of the j-th dimension parameter; the superscript 0 indicates the initialization time;

[0070] The formula for updating the optimized aurora position is:

[0071] ,

[0072] ,

[0073] in, Let be the position of the i-th individual in the j-th dimension parameter at the t-th iteration; The position at the (t+1)th iteration; r1 and r2 are random numbers in the interval [0,1]; W1 and W2 are adaptive weights; v(t) is the rotational motion term at the tth iteration; A o For aurora walking; t is the current iteration number; T is the maximum iteration number;

[0074] The fitness function is defined as:

[0075] ,

[0076] Where J(Θ) is the fitness function value corresponding to the parameter vector Θ; , , These are the weighting coefficients for each error index; This represents the root mean square error of the terminal voltage. The root mean square error of SOC; The mean absolute error of SOC; the optimal parameters satisfy:

[0077] ,

[0078] Where, Θ * The optimal parameter vector is defined to minimize the fitness function J(Θ); argmin represents the independent variable that minimizes the objective function; through the IPLO collaborative optimization, the unified optimization of FFRLS and MIAUKF parameters is achieved, thereby improving the overall performance of the SOC prediction branch.

[0079] In step 5, the SOP prediction employs a combination of temperature, voltage, and SOC constraints to determine the maximum allowable charge and discharge currents of the battery within the target duration, and then calculates the battery power boundary. The maximum allowable discharge current and the maximum allowable charge current are as follows:

[0080] ,

[0081] ,

[0082] in, This represents the maximum allowable discharge current at the k-th sampling time. This represents the maximum allowable charging current at the k-th sampling time. and These are the maximum allowable discharge current and the maximum allowable charging current under voltage constraints, respectively. and These are the maximum allowable discharge current and the maximum allowable charging current under SOC constraints, respectively. and These are the maximum allowable discharge current and the maximum allowable charging current under temperature constraints, respectively; min{} represents taking the minimum value among multiple constraint currents.

[0083] The peak current under SOC constraint is:

[0084] ,

[0085] ,

[0086] Among them, SOC minThe lower limit of allowable SOC; SOC max The upper limit of allowed SOC; Δt p For target duration or power prediction time window;

[0087] The corresponding maximum available discharge power and charging power are as follows:

[0088] ,

[0089] ,

[0090] in, This represents the maximum available discharge power at the k-th sampling time. U represents the maximum available charging power at the k-th sampling time. t,k Let be the terminal voltage at the k-th sampling time. By solving the three constraints together, the problem of the power boundary deviating from reality under a single constraint can be avoided, thus improving the physical rationality of the SOP boundary calculation.

[0091] In step 6, the prediction of SOP is implemented using a two-layer CNN-GRU-MHA-NBEATS model, where CNN is used to extract local features, GRU is used to extract temporally dependent features, MHA is used to enhance the weights of key features, and NBEATS is used to achieve trend modeling and multi-step prediction; the input vector is defined as:

[0092] ,

[0093] Among them, Z k U is the feature vector from the k-th sampling time step to the two-layer CNN-GRU-MHA-NBEATS model; t Let be the terminal voltage at the k-th sampling time. This represents the maximum available discharge power boundary at the k-th sampling time. This represents the maximum available charging power boundary at the k-th sampling time.

[0094] The network output expression is:

[0095] ,

[0096] in, f is the predicted SOP value at the (k+τ)th sampling time; CNN-GRU-MHA-NBEATS () represents the nonlinear mapping function of the two-layer CNN-GRU-MHA-NBEATS network; Z k-L+1 Let L represent the input time window sequence consisting of the (k-L+1)th sampling time to the kth sampling time; L is the input time window length; τ is the prediction step size; and γ is the health state coupling weight vector. This is a correction term for the SOP prediction results based on health status;

[0097] By inputting the health status characterization quantity and the three-constraint power boundary into the deep prediction network, the SOP prediction result can simultaneously take into account battery aging characteristics, dynamic operating state and physical constraint boundary, thereby improving prediction accuracy and generalization ability.

[0098] 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.

[0099] In one embodiment, a lithium-ion power battery cell with a rated capacity of 50Ah was selected as the test object, and the experiment was conducted at a normal temperature of 25°C with a sampling period of 1 second. During the experiment, the battery's terminal voltage, current, surface temperature, and capacity data were collected to construct the original test dataset. The above parameters are merely examples and do not constitute a limitation on the scope of protection of this invention.

[0100] In step 1, based on the collected voltage, current, and temperature data, a correspondence between the state of charge (SOC) and the open-circuit voltage (OCV) is established, and features such as capacity decay, internal resistance growth, voltage difference, and temperature difference are extracted to construct a health state characterization vector. This step provides health state information for subsequent state estimation and power prediction.

[0101] In step 2, a second-order resistor-capacitor RC equivalent circuit model is established, and online parameter identification is performed using the recursive least squares method (FFRLS) with a forgetting factor. Optionally, the initial value of the forgetting factor is set to 0.98. Through online identification, the ohmic internal resistance, polarization internal resistance, and polarization capacitance parameters updated over time can be obtained, providing a mechanistic model basis for SOC prediction.

[0102] In step 3, a multi-inspiration adaptive unscented Kalman filter (MIAUKF) considering temperature compensation is used to predict the state of health (SOC). Optionally, the unscented transform parameters α are set to 0.01, β to 2, κ to 0, and the multi-inspiration length p to 3. The health status representation obtained in step 1 and the online identification parameters obtained in step 2 are input into the filter to obtain the SOC estimation result.

[0103] In step 4, the Aurora Optimization Algorithm Improved by Chaotic Mapping (IPLO) is used to jointly optimize the forgetting factor in FFRLS and the key parameters in MIAUKF. Optionally, the population size is set to 20 and the maximum number of iterations is set to 50. This step yields a better combination of parameters, improving the accuracy and stability of SOC prediction.

[0104] In step 5, before SOP prediction, the maximum allowable charging current and maximum allowable discharging current of the battery are calculated jointly based on voltage constraints, temperature constraints, and SOC constraints. Optionally, the lower limit of SOC is set to 10%, and the upper limit of SOC is set to 90%. Through this step, the maximum available charging power boundary and the maximum available discharging power boundary can be obtained.

[0105] In step 6, a prediction network consisting of a convolutional neural network (CNN), a gated recurrent unit (GRU), a multi-head attention mechanism (MHA), and an interpretable time-series neural basis unfolding analysis network (NBEATS) is constructed to predict the SOP. Optionally, the input time window length is set to 60, and the prediction step size is set to 1. This step outputs the SOP prediction result in the target time domain.

[0106] In the above embodiment, step 1 obtains the health state representation vector, step 2 obtains the online updated equivalent circuit parameters, step 3 obtains the SOC estimation result, step 5 obtains the charge and discharge power boundary, and step 6 obtains the SOP prediction result, thereby forming a joint prediction process of health state representation, parameter identification, SOC estimation and SOP prediction.

[0107] To illustrate the technical effectiveness of the method of this invention, it can be compared with methods that do not incorporate health status characterization metrics. Comparison metrics may include the root mean square error of SOC, the mean absolute error, and the SOP prediction error. The comparison results show that the method of this invention can obtain more stable SOC estimation results and more reasonable SOP prediction results.

[0108] Table 1. Comparison of exemplary results between the method of the present invention and different comparative methods.

[0109]

[0110] This invention provides a method for jointly predicting the state of charge and state of power of a battery. Many methods and approaches exist for implementing this technical solution; the above description is merely a preferred embodiment. 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 combined state-of-charge and state-of-power prediction oriented to state-of-health, characterized in that, Includes the following steps: Step 1: Extract the state of charge-open circuit voltage curves at different temperatures under different operating conditions. The state of charge-open circuit voltage curves are SOC-OCV curves: By testing the battery under various operating conditions and different temperature environments, the corresponding voltage, current and temperature data are obtained, and SOC-OCV curves under different operating conditions and temperature conditions are constructed based on the voltage, current and temperature data. Step 2: The recursive least squares method with forgetting factor (FFRLS) is used to identify parameters, transforming the battery model into a mathematical form that can be identified using least squares. FFRLS is used to identify the parameters of the battery equivalent circuit model, and the weights of the old and new data are dynamically adjusted by the forgetting factor to track the dynamic changes of the system. Step 3: Use the multi-innovation adaptive unscented Kalman filter (MIAUKF) model that considers temperature compensation to predict the state of charge (SOC). Step 4: The improved Aurora Optimization Algorithm (IPLO) is used to optimize the forgetting factor of FFRLS and the parameters of the MIAUKF model. The global search capability of the algorithm is enhanced by incorporating chaotic mapping into the Aurora Optimization Algorithm. The improved IPLO algorithm is used to jointly optimize the forgetting factor of FFRLS and the relevant parameters in the MIAUKF model. Step 5: Add three constraints to the power state SOP prediction: temperature constraint, voltage constraint, and state of charge (SOC) constraint. When evaluating and predicting SOP, temperature, voltage, and SOC are used as constraints. Step 6: A two-layer convolutional neural network-gated recurrent unit-multiple-head attention mechanism-NBEATS network CNN-GRU-MHA-NBEATS model is used to predict the power state SOP. A two-layer CNN-GRU-MHA-N-BEATS model is constructed. The convolutional neural network (CNN) is used to extract local features of the data, the gated recurrent unit (GRU) is used to process time series information, the multi-head attention mechanism (MHA) is used to enhance the model's ability to capture the relationship between different features, and the NBEATS network is used to complete trend modeling and multi-step prediction.

2. The method of claim 1, wherein, In step 1, a health state characterization quantity is obtained. This quantity characterizes the degree of battery degradation at the current stage of its lifespan and serves as a common input for subsequent SOC and SOP predictions. This allows for the coupled correction of the health state to the state estimation and power prediction results, expressed as: ， wherein h k is the health status representation vector at the kth sampling time, SOH k is the battery health status at the kth sampling time, ΔR k is the internal resistance increment feature, E cv,k is the isobaric energy feature, ΔV k is the equal time interval voltage difference feature, ΔT k is the equal time interval temperature difference feature; The state of SOH in a healthy state is defined in terms of capacity decay, and the expression is: ， where Q n,k is the current maximum available capacity of the battery at the kth sampling time, and Q0is the initial rated capacity of the battery.

3. The method of claim 2, wherein, In step 2, the lithium-ion battery is modeled using a second-order RC equivalent circuit model to describe its ohmic internal resistance characteristics and two types of polarization dynamics. In the second-order RC equivalent circuit model, the first RC branch characterizes the electrochemical polarization process, and the second RC branch characterizes the concentration polarization process. The continuous state equation and observation equation of the second-order RC equivalent circuit model are as follows: ， ， ， wherein U1 is a first polarization branch voltage, U2 is a second polarization branch voltage, is a first derivative of the first polarization branch voltage U1, is a first derivative of the second polarization branch voltage U2, R0 is an ohmic internal resistance, R1 and R2 are a first polarization internal resistance and a second polarization internal resistance, respectively, C1 and C2 are a first polarization capacitance and a second polarization capacitance, respectively, I is a battery current, U t is an open-circuit voltage; after discretization of the continuous model, we obtain: oc is an open-circuit voltage; after discretization of the continuous model, we obtain: ， ， ， wherein, U 1,k and U 2,k are the first and second polarization voltages at the kth sampling time, respectively; U 1,k+1 and U 2,k+1 are the first and second polarization voltages at the (k+1)th sampling time, respectively; R 1,k and R 2,k are the first and second polarization resistances at the kth sampling time, respectively; C 1,k and C 2,k are the first and second polarization capacitances at the kth sampling time, respectively; R 0,k is the ohmic resistance at the kth sampling time; I k is the battery current at the kth sampling time; U oc,k is the open-circuit voltage at the kth sampling time; U t,k is the terminal voltage at the kth sampling time; and e is the natural constant.

4. The method of claim 3, wherein, In step 2, the parameters of the second-order RC equivalent circuit model are identified online using the recursive least squares method (FFRLS) with a forgetting factor, so that the ohmic internal resistance, polarization internal resistance, and polarization capacitance can be updated in real time according to changes in operating conditions; the FFRLS parameter update formula is: ， ， ， wherein is the parameter estimation vector at the kth sampling instant; is the parameter estimation vector at the k+1th sampling instant; K k+1 is the gain matrix at the k+1th sampling instant; y k+1 is the observation output at the k+1th sampling instant; is the regression vector at the k+1th sampling instant; P k is the error covariance matrix at the kth sampling instant; P k+1 is the error covariance matrix at the k+1th sampling instant; is the forgetting factor; I is the identity matrix; the superscript T denotes matrix transposition; the superscript -1 denotes matrix inversion.

5. The method of claim 4, wherein, In step 3, the SOC prediction is described using a discrete state equation based on the ampere-hour integration mechanism and combined with the current health state to correct the capacity, so as to reflect the dynamic change of the remaining battery capacity over time; the discretized expression of the SOC state equation is: ， wherein SOC k and SOC k+1 are the state of charge at the kth sampling instant and the state of charge at the k+1th sampling instant, respectively, is the charging / discharging efficiency, Δt is the sampling period, is the process noise; A temperature compensation term is further introduced into the SOC prediction process, resulting in the open-circuit voltage expression considering temperature compensation: ， wherein, is the open circuit voltage compensated for temperature at the kth sampling instant, is the mapping function of SOC and open circuit voltage at reference temperature, T k is the battery temperature at the kth sampling instant, T ref is the reference temperature, is the temperature compensation coefficient.

6. The method of claim 5, wherein, In step 3, the SOC prediction is implemented using the Multi-Inspiration Adaptive Unscented Kalman Filter (MIAUKF) model, which considers temperature compensation. The state update formula of the MIAUKF is as follows: ， ， wherein, is a prior state estimate value at the kth sampling time; is a posterior state estimate value at the kth sampling time combined with observation correction; is a multi-innovation vector at the kth sampling time; is a prior error covariance matrix at the kth sampling time; is a posterior error covariance matrix at the kth sampling time;S k is an innovation covariance matrix at the kth sampling time;symbol denotes a conditional estimation relationship; The multi-innovation vector is defined as: ， wherein v k is the innovation at the kth sampling instant; p is the multi-innovation length; The expression for the adaptive update of the process noise and measurement noise covariance is: ， ， wherein Q k is a process noise covariance matrix at the kth sampling time; R k is a measurement noise covariance matrix at the kth sampling time; is a process noise covariance adaptive adjustment coefficient; is a measurement noise covariance adaptive adjustment coefficient; C k is an observation matrix at the kth sampling time; is a multiple innovation vector at the kth sampling time.

7. The method of claim 6, wherein, In step 4, the forgetting factor of FFRLS and the parameters of the MIAUKF model are jointly optimized using the chaotic mapping-improved aurora optimization algorithm IPLO. The parameter vector to be optimized is defined as: ， Where Θ is the parameter vector to be optimized; λ is the forgetting factor in FFRLS; and α, β, and κ are the unscented transformation parameters in MIAUKF. The population is initialized using a Logistic chaotic mapping: ， ， wherein, is the chaotic variable of the ith individual in the jth dimension at the tth iteration; is the chaotic variable of the next iteration; μ is the control parameter of the Logistic chaotic map; is the initial position of the ith individual in the jth dimension parameter; LB j is the lower bound of the jth dimension parameter; UB j is the upper bound of the jth dimension parameter; the superscript 0 indicates the initialization time; The formula for updating the optimized aurora location is: ， ， wherein, is the position of the ith individual in the jth dimension parameter at the tth iteration; is the position at the t+1th iteration; r1 and r2 are random numbers in the interval [0,1]; W1 and W2 are adaptive weights; v(t) is the turning movement term at the tth iteration; A o is the aurora walk term; t is the current iteration number; T is the maximum iteration number; The fitness function is defined as: ， Where J(Θ) is the fitness function value corresponding to the parameter vector Θ; , , These are the weighting coefficients for each error index; This represents the root mean square error of the terminal voltage. The root mean square error of SOC; The mean absolute error of SOC; the optimal parameters satisfy: ， where Θ * is the optimal parameter vector corresponding to the minimum of the fitness function J(Θ); arg min denotes taking the argument that minimizes the objective function.

8. The method according to claim 7, characterized in that, In step 5, the SOP prediction employs a combination of temperature, voltage, and SOC constraints to determine the maximum allowable charge and discharge currents of the battery within the target duration, and then calculates the battery power boundary. The maximum allowable discharge current and the maximum allowable charge current are as follows: ， ， in, This represents the maximum allowable discharge current at the k-th sampling time. This represents the maximum allowable charging current at the k-th sampling time. and These are the maximum allowable discharge current and the maximum allowable charging current under voltage constraints, respectively. and These are the maximum allowable discharge current and the maximum allowable charging current under SOC constraints, respectively. and These are the maximum allowable discharge current and the maximum allowable charging current under temperature constraints, respectively; min{} represents taking the minimum value among multiple constraint currents; The peak current under SOC constraint is: ， ， wherein SOC min is the lower limit of the allowed SOC; SOC max is the upper limit of the allowed SOC; At p is the target duration or power prediction time window; The corresponding maximum available discharge power and charging power are as follows: ， ， in, This represents the maximum available discharge power at the k-th sampling time. U represents the maximum available charging power at the k-th sampling time. t,k Let be the terminal voltage at the k-th sampling time.

9. The method according to claim 8, characterized in that, In step 6, the prediction of SOP is implemented using a two-layer CNN-GRU-MHA-NBEATS model, where CNN is used to extract local features, GRU is used to extract temporally dependent features, MHA is used to enhance the weights of key features, and NBEATS is used to achieve trend modeling and multi-step prediction; the input vector is defined as: ， Among them, Z k U is the feature vector from the k-th sampling time step to the two-layer CNN-GRU-MHA-NBEATS model; t Let be the terminal voltage at the k-th sampling time. This represents the maximum available discharge power boundary at the k-th sampling time. This represents the maximum available charging power boundary at the k-th sampling time. The network output expression is: ， in, f is the predicted SOP value at the (k+τ)th sampling time; CNN-GRU-MHA-NBEATS () represents the nonlinear mapping function of the two-layer CNN-GRU-MHA-NBEATS network; Z k-L+1 Let L represent the input time window sequence consisting of the (k-L+1)th sampling time to the kth sampling time; L is the input time window length; τ is the prediction step size; and γ is the health state coupling weight vector. This is a correction term for the SOP prediction result based on health status.

10. 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 9.

Images