Battery health state estimation method and system based on partial constant current charging

Abstract

This invention provides a method and system for estimating battery health status based on partial constant current charging, relating to the field of lithium-ion battery technology. The method includes: acquiring charging data of the current battery during partial constant current charging; acquiring model parameters of the second-order equivalent circuit model of the current battery based on the charging data; inputting the model parameters into a pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery; and obtaining an estimated health status value of the current battery based on the maximum usable capacity and the battery's rated capacity. This invention can complete battery health status estimation using only partial constant current charging data, reducing the requirement for completeness of charging conditions and improving the feasibility of online applications. Furthermore, it only requires real-time identification of a small number of circuit model parameters and one forward propagation calculation, resulting in low computational complexity and suitability for resource-constrained battery management system deployments.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery technology, and specifically to a method and system for estimating battery health status based on local constant current charging. Background Technology

[0002] Lithium-ion batteries, due to their advantages such as high energy density, long cycle life, and low self-discharge rate, have been widely used in electric vehicles, energy storage systems, and portable electronic devices. However, with increasing cycle count and changing operating conditions, aging phenomena such as loss of active materials, increased polarization, and rising internal resistance gradually intensify, leading to a continuous decline in the battery's maximum usable capacity. This, in turn, affects the vehicle's range, power performance, and operational safety. Therefore, accurate estimation of the battery's State of Health (SOH) is of great importance.

[0003] Among existing battery health state estimation methods, those based on full charge and full discharge capacity testing offer high accuracy but require lengthy testing times, making them unsuitable for online applications. Methods based on complex electrochemical models, while possessing strong mechanistic explanation capabilities, suffer from complex model building, high parameter identification costs, and significant engineering deployment challenges. Purely mathematically driven methods, while offering fast inference speeds, often lack physical constraints and exhibit insufficient stability when applied across different operating conditions and aging stages. Furthermore, obtaining complete charge and discharge curves during actual electric vehicle operation is often difficult; in most cases, only data from partial constant-current charging phases can be collected. Relying on complete charging curves or full capacity tests for battery health state estimation fails to meet the online estimation requirements of real-world scenarios. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a battery health state estimation method based on local constant current charging, comprising: Obtain charging data of the battery during the current local constant current charging process; Based on charging data, obtain the model parameters of the second-order equivalent circuit model of the current battery; Input the model parameters into a pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery; Based on the maximum available capacity and the battery's rated capacity, an estimated value for the current battery's health status is obtained; The physical sensing neural network model was trained offline in the following way: based on cyclic charge and discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels were collected to establish a battery aging dataset; a second-order equivalent circuit model was established, and based on the battery aging dataset, the model parameter set of the battery at different aging levels was obtained; the physical sensing neural network model was trained with the model parameter set as input and the corresponding true value of maximum usable capacity as output.

[0005] Preferably, acquiring charging data of the current battery during partial constant current charging includes: Collect the time series of charging current, measured terminal voltage and battery temperature during a preset period of the current battery in the local constant current charging stage, and collect the stationary terminal voltage in the stationary state before charging starts. Based on the charging current and the battery's rated capacity, determine the constant current charging rate during the current battery's partial constant current charging process.

[0006] Preferably, based on charging data, the model parameters of the second-order equivalent circuit model of the current battery are obtained, including: The initial value of the open-circuit voltage is taken as the resting terminal voltage in the resting state before charging begins, and the initial polarization voltage is determined to be zero. Based on the initial value of the open-circuit voltage, the initial polarization voltage, and the current values ​​of the model parameters to be identified, the calculated terminal voltage is obtained using a pre-established expression for the calculated terminal voltage. The Levenberg-Marquardt algorithm is adopted, with the objective function being to minimize the sum of squared voltage errors between the measured and calculated terminal voltages. The model parameters to be identified are iteratively fitted to obtain updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, and the identified model parameters are output. The expression for calculating the terminal voltage is constructed based on a second-order equivalent circuit model under local constant current charging conditions. The formula for calculating the terminal voltage is as follows:

[0007] in, Let be the calculated terminal voltage at time t. The initial value of the open-circuit voltage. For constant current charging current, For ohm resistance, The resistance of the first polarization branch, The resistance of the second polarization branch, The time constant of the first polarization branch. The time constant of the second polarization branch; These are the model parameters to be identified.

[0008] Preferably, the Levenberg-Marquardt algorithm is used, with the objective function being to minimize the sum of squared voltage errors between the measured and calculated terminal voltages. The algorithm iteratively fits the model parameters to be identified, obtaining updated model parameters. These updated parameters are then used as the current values ​​for the model parameters to be identified in the next iteration, until the convergence condition is met. The identified model parameters are then output, including: Calculate the error vector and Jacobian matrix of the model parameters to be identified at the current values. Calculate the iteration step size and the gradient of the objective function based on the Jacobian matrix and the error vector. Then, iteratively fit the model parameters to be identified based on the iteration step size and the gradient of the objective function to obtain the updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, at which point the iteration terminates. The current values ​​of the model parameters to be identified at the time of termination are then determined as the identified model parameters.

[0009] Preferably, training a physical perception neural network model, using the model parameter set as input and the corresponding maximum available capacity true value as output, includes: The input features of the training samples are obtained from the model parameter set. The input features include temperature, ohmic resistance, first polarization resistance, first polarization time constant, second polarization resistance, and second polarization time constant. The true value of the maximum usable capacity corresponding to the training sample is also obtained. The input features are used as input to the physical sensing neural network to be trained. The maximum available capacity prediction value is calculated through forward propagation, and the capacity loss term is calculated. The capacity loss term is used to characterize the error between the maximum available capacity prediction value and the actual maximum available capacity value in the battery aging dataset. Using the model parameters of the training samples, the predicted terminal voltage is calculated based on the second-order equivalent circuit model, and a voltage consistency loss term is constructed based on the error between the predicted terminal voltage and the measured terminal voltage in the battery aging dataset. The model parameters of the training samples are corrected based on the voltage consistency loss term to obtain the corrected model parameters; Temperature and corrected model parameters are used as inputs to the physical sensing neural network to be trained. The maximum available capacity prediction is calculated through forward propagation, and a capacity loss term is constructed based on the error between the maximum available capacity prediction and the actual maximum available capacity in the battery aging dataset. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation until the capacity loss term converges, thus completing the training of the physical perception neural network model. The formula for calculating the voltage consistency loss term is as follows:

[0010] in, This is the voltage consistency loss term. The number of training samples. For the first The time series length of each training sample The first data collected centrally for battery aging The training sample at the th ... The measured terminal voltage at each moment. For the first in the battery aging dataset The training sample at the th ... Predicted terminal voltage at each moment; The formula for calculating the capacity loss term is as follows:

[0011] in, The number of training samples. For capacity loss, For the first The true maximum usable capacity of a training sample. For the first The maximum usable capacity prediction for each training sample.

[0012] Preferably, based on cyclic charge-discharge experiments at different temperatures and constant current charging rates, charging data at different aging stages and the corresponding true values ​​of the maximum usable capacity are collected to establish a battery aging dataset, including: Under different temperatures and constant current charging rates, multiple sets of cyclic charge-discharge experiments were performed on the battery. Each set of experiments included: constant current charging to the upper limit voltage, constant voltage charging to the cutoff current and resting, and constant current discharging to the lower limit voltage and resting. During the charging process of each set of cycles, the charging current, measured terminal voltage and battery temperature were collected in real time. After each set number of cycles, the discharge capacity is recorded after the discharge ends, and this is determined as the true value of the maximum usable capacity for this aging stage. The battery aging dataset is created by associating and storing the current, voltage, temperature, charging rate, and corresponding maximum usable capacity collected at different cycle stages.

[0013] Preferably, based on the maximum available capacity and the battery's rated capacity, an estimated value for the current battery health status is obtained, including: Based on the maximum available capacity and the battery's rated capacity, calculate the estimated current battery health status according to the battery health status formula; The formula for battery health status is as follows: ; in, The maximum usable capacity of the current battery, as output by the physical sensing neural network model. For the battery's rated capacity, This is an estimate of the current battery health status.

[0014] Based on the same inventive concept, the present invention also provides a battery health state estimation system based on partial constant current charging, the system comprising: The charging data acquisition module is used to acquire the charging data of the current battery during the local constant current charging process; The model parameter acquisition module is used to obtain the model parameters of the second-order equivalent circuit model of the current battery based on charging data. The battery maximum capacity determination module is used to input model parameters into a pre-trained physical sensing neural network model to obtain the current battery maximum usable capacity. The battery status estimation module is used to obtain an estimated value of the current battery health status based on the maximum available capacity and the battery's rated capacity. The physical sensing neural network model was trained offline in the following way: based on cyclic charge and discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels were collected to establish a battery aging dataset; a second-order equivalent circuit model was established, and based on the battery aging dataset, the model parameter set of the battery at different aging levels was obtained; the physical sensing neural network model was trained with the model parameter set as input and the corresponding true value of maximum usable capacity as output.

[0015] Preferably, the charging data acquisition module is specifically used for: Collect the time series of charging current, measured terminal voltage and battery temperature during a preset period of the current battery in the local constant current charging stage, and collect the stationary terminal voltage in the stationary state before charging starts. Based on the charging current and the battery's rated capacity, determine the constant current charging rate during the current battery's partial constant current charging process.

[0016] Preferably, the model parameter acquisition module is specifically used for: The initial value of the open-circuit voltage is taken as the resting terminal voltage in the resting state before charging begins, and the initial polarization voltage is determined to be zero. Based on the initial value of the open-circuit voltage, the initial polarization voltage, and the current values ​​of the model parameters to be identified, the calculated terminal voltage is obtained using a pre-established expression for the calculated terminal voltage. The Levenberg-Marquardt algorithm is adopted, with the objective function being to minimize the sum of squared voltage errors between the measured and calculated terminal voltages. The model parameters to be identified are iteratively fitted to obtain updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, and the identified model parameters are output. The expression for calculating the terminal voltage is constructed based on a second-order equivalent circuit model under local constant current charging conditions. The formula for calculating the terminal voltage is as follows:

[0017] in, Let be the calculated terminal voltage at time t. The initial value of the open-circuit voltage. For constant current charging current, For ohm resistance, The resistance of the first polarization branch, The resistance of the second polarization branch, The time constant of the first polarization branch. The time constant of the second polarization branch; These are the model parameters to be identified.

[0018] Preferably, the model parameter acquisition module is specifically used for: Calculate the error vector and Jacobian matrix of the model parameters to be identified at the current values. Calculate the iteration step size and the gradient of the objective function based on the Jacobian matrix and the error vector. Then, iteratively fit the model parameters to be identified based on the iteration step size and the gradient of the objective function to obtain the updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, at which point the iteration terminates. The current values ​​of the model parameters to be identified at the time of termination are then determined as the identified model parameters.

[0019] Preferably, the system also includes a physical perception neural network model building module, used for: The input features of the training samples are obtained from the model parameter set. The input features include temperature, ohmic resistance, first polarization resistance, first polarization time constant, second polarization resistance, and second polarization time constant. The true value of the maximum usable capacity corresponding to the training sample is also obtained. Using the model parameters of the training samples, the predicted terminal voltage is calculated based on the second-order equivalent circuit model, and a voltage consistency loss term is constructed based on the error between the predicted terminal voltage and the measured terminal voltage in the battery aging dataset. The model parameters of the training samples are corrected based on the voltage consistency loss term to obtain the corrected model parameters; Temperature and the corrected model parameters are used as inputs to the physical sensing neural network to be trained. The maximum available capacity prediction is calculated through forward propagation, and a capacity loss term is constructed based on the error between the maximum available capacity prediction and the actual maximum available capacity in the battery aging dataset. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation until the capacity loss term converges, thus completing the training of the physical perception neural network model. The formula for calculating the voltage consistency loss term is as follows:

[0020] in, This is the voltage consistency loss term. The number of training samples. For the first The time series length of each training sample The first data collected centrally for battery aging The training sample at the th ... The measured terminal voltage at each moment. For the first in the battery aging dataset The training sample at the th ... Predicted terminal voltage at each moment; The formula for calculating the capacity loss term is as follows:

[0021] in, The number of training samples. For capacity loss, For the first The true maximum usable capacity of a training sample. For the first The maximum usable capacity prediction for each training sample.

[0022] Preferably, the physical perception neural network model is specifically used for: Under different temperatures and constant current charging rates, multiple sets of cyclic charge-discharge experiments were performed on the battery. Each set of experiments included: constant current charging to the upper limit voltage, constant voltage charging to the cutoff current and resting, and constant current discharging to the lower limit voltage and resting. During the charging process of each set of cycles, the charging current, measured terminal voltage and battery temperature were collected in real time. After each set number of cycles, the discharge capacity is recorded after the discharge ends, and this is determined as the true value of the maximum usable capacity for this aging stage. The battery aging dataset is created by associating and storing the current, voltage, temperature, charging rate, and corresponding maximum usable capacity collected at different cycle stages.

[0023] Preferably, the battery state estimation module is specifically used for: Based on the maximum available capacity and the battery's rated capacity, calculate the estimated current battery health status according to the battery health status formula; The formula for battery health status is as follows: ; in, The maximum usable capacity of the current battery, as output by the physical sensing neural network model. For the battery's rated capacity, This is an estimate of the current battery health status.

[0024] Based on the same inventive concept, the present invention also provides an electronic device, comprising: at least one processor and a memory; wherein the memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a battery health state estimation method based on local constant current charging as described above is implemented.

[0025] Based on the same inventive concept, the present invention also provides a readable storage medium having an executable program stored thereon, which, when executed, implements a battery health state estimation method based on local constant current charging as described above.

[0026] Compared with the closest existing technology, the present invention has the following beneficial effects: This invention provides a battery health state estimation method based on partial constant current charging, comprising: acquiring charging data of the current battery during partial constant current charging; acquiring model parameters of the second-order equivalent circuit model of the current battery based on the charging data; inputting the model parameters into a pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery; and obtaining an estimated value of the current battery health state based on the maximum usable capacity and the rated capacity of the battery. The physical sensing neural network model is trained offline in the following manner: based on cyclic charge-discharge experiments at different temperatures and different constant current charging rates, charging data and corresponding true values ​​of the maximum usable capacity at different aging levels are collected to establish a battery aging dataset; a second-order equivalent circuit model is established, and based on the battery aging dataset, a set of model parameters for the battery at different aging levels is obtained; the physical sensing neural network model is trained with the model parameter set as input and the corresponding true value of the maximum usable capacity as output. This invention can estimate battery health status using local constant current charging data, significantly reducing the requirement for complete charging conditions. It can be directly embedded into the daily local charging scenarios of electric vehicles, improving the feasibility of online applications. In the offline training stage, second-order equivalent circuit model parameters are introduced as input features of the neural network, integrating physical laws into the data-driven model, enhancing the model's generalization ability and interpretability under different temperatures, charging rates, and aging stages. In the online estimation stage, only a small number of circuit model parameters need to be identified in real time, and a neural network forward propagation calculation is performed once. The computational complexity is low, making it suitable for deployment in resource-constrained battery management systems. Attached Figure Description

[0027] Figure 1 A schematic flowchart of the battery health state estimation method based on local constant current charging provided by the present invention; Figure 2 The overall flowchart of the battery health state estimation method based on local constant current charging provided by the present invention; Figure 3 A schematic diagram of the circuit structure of the second-order equivalent circuit model provided by the present invention; Figure 4 This is a schematic diagram of the process for identifying the parameters of the second-order equivalent circuit model of a battery under different aging conditions provided by the present invention. Figure 5 A schematic diagram of the physical sensing neural network model provided by this invention; Figure 6 A schematic diagram of the battery health status estimation process provided by the present invention; Figures 7(a)-7(d) are comparison diagrams of the measured terminal voltage curves and the fitting of the second-order equivalent circuit model to the four terminal voltage curves provided by the present invention at different aging stages. Figure 8This is a comparison chart of the battery health status estimate obtained using this invention and the actual value. Figure 9 A structural diagram of the battery health state estimation system based on local constant current charging provided by the present invention; Figure 10 A schematic diagram of the electronic device provided by the present invention. Detailed Implementation

[0028] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0029] Example 1: This invention provides a battery health state estimation method based on local constant current charging. Specifically, Figure 1 A flowchart illustrating the battery health state estimation method based on partial constant current charging provided in this embodiment of the invention is shown in the figure, including the following steps: S1: Obtain the charging data of the current battery during the local constant current charging process; S2: Based on charging data, obtain the model parameters of the second-order equivalent circuit model of the current battery; S3: Input the model parameters into the pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery; S4: Based on the maximum available capacity and the battery's rated capacity, obtain an estimate of the current battery's health status; The physical sensing neural network model was trained offline in the following way: based on cyclic charge and discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels were collected to establish a battery aging dataset; a second-order equivalent circuit model was established, and based on the battery aging dataset, the model parameter set of the battery at different aging levels was obtained; the physical sensing neural network model was trained with the model parameter set as input and the corresponding true value of maximum usable capacity as output.

[0030] like Figure 2 The diagram shown is the overall flowchart of the battery health state estimation method based on local constant current charging provided by this invention. The following is in conjunction with... Figure 2The overall process is summarized below. The battery health state estimation method based on local constant current charging provided by this invention includes two parts: an offline preparation stage and an online estimation stage, which are connected by a physical-aware neural network model. The offline preparation stage is used to construct the physical-aware neural network model (PRN), while the online estimation stage is used to estimate the current battery health state in real time during actual charging scenarios.

[0031] Offline preparation phase: The purpose of the offline preparation phase is to train a physical sensing neural network model that can map the parameters of the equivalent circuit model to the maximum usable capacity of the battery using historical data under different aging levels and operating conditions. This phase includes the following steps: Step A1: Based on cyclic charge-discharge experiments at different temperatures and constant current charging rates, collect charging data and corresponding maximum usable capacity values ​​for different aging levels to establish a battery aging dataset.

[0032] Specifically, under different temperatures and different constant current charging rates, multiple sets of cyclic charge-discharge experiments were performed on the battery. Each set of experiments included: constant current charging to the upper limit voltage, constant voltage charging to the cutoff current and resting, and constant current discharging to the lower limit voltage and resting. During the charging process of each set of cycles, the charging current, measured terminal voltage and battery temperature were collected in real time. After each set number of cycles, the discharge capacity is recorded after the discharge ends, and this is determined as the true value of the maximum usable capacity for this aging stage. The battery aging dataset is created by associating and storing the current, voltage, temperature, charging rate, and corresponding maximum usable capacity collected at different cycle stages.

[0033] In one specific implementation, multiple sets of cyclic charge-discharge experiments were conducted on the same type of lithium-ion battery under different temperatures (e.g., 25°C, 35°C, 45°C) and different constant current charging rates (e.g., 0.5C, 1C, 1.5C). Each set of experiments was performed according to the following procedure: first, the battery was charged with a constant current at a set rate until the terminal voltage reached the upper limit voltage; then, charging was continued in a constant voltage manner until the current dropped to the cutoff current; after a period of rest, the battery was discharged with a constant current at a set rate until the terminal voltage reached the lower limit voltage; and then rested again. The above charge-discharge cycle was repeated until the battery capacity decayed to below 80% of the rated capacity.

[0034] During each charging cycle, the time series of charging current, measured terminal voltage, and battery temperature are collected in real time. Understandably, the sampling frequency of the terminal voltage can be set according to actual needs; in this embodiment, the sampling interval is 1 second.

[0035] After each set number of cycles (e.g., every 50 cycles), the discharge capacity is recorded after the constant current discharge ends, and this is used as the true value of the maximum usable capacity for this aging stage.

[0036] Then, the current, voltage, temperature, charging rate and corresponding maximum usable capacity collected at different cycle stages are correlated and stored to establish a battery aging dataset.

[0037] Step A2: Establish a second-order equivalent circuit model and obtain the model parameter set of the battery under different aging levels based on the battery aging dataset.

[0038] like Figure 3 The diagram shows the circuit structure of the second-order equivalent circuit model. The second-order equivalent circuit model is a second-order RC equivalent circuit model, also known as a second-order ECM equivalent circuit model. The specific construction process is as follows: Open circuit voltage of the battery Determined to be a voltage source, the ohm resistor Connected in series to the positive terminal of the voltage source; The resistor of the first polarization branch With capacitor The first RC parallel circuit is connected in series after the ohmic resistor, and the first polarization voltage is connected in parallel to form a first RC parallel circuit. It satisfies the first differential equation and defines the first polarization time constant; The resistor of the second polarization branch With capacitor The first RC parallel circuit is connected in series with the second RC parallel circuit, forming a second RC parallel circuit. The second polarization voltage is connected in series with the first RC parallel circuit. It satisfies the second differential equation and defines the second polarization time constant; Based on Kirchhoff's voltage law, the terminal voltage equation is obtained; Solving the first and second differential equations under constant current charging conditions yields a first analytical expression for the first polarization voltage as a function of time and a second analytical expression for the second polarization voltage as a function of time. Substituting the first and second analytical expressions into the terminal voltage equation, we obtain the expression for the calculated terminal voltage as a function of time under constant current charging conditions. The formula for the first differential equation is as follows:

[0039] in, This is the first polarization voltage. The resistance of the first polarization branch, The capacitor of the first polarization branch; The charging current is constant current; The formula for the first polarization time constant is as follows:

[0040] in, The time constant of the first polarization branch. The resistance of the first polarization branch, The capacitor of the first polarization branch; The formula for the second differential equation is as follows:

[0041] in, This is the second polarization voltage. The resistance of the second polarization branch, The capacitor for the second polarization branch; The charging current is constant current; The formula for the second polarization time constant is as follows:

[0042] in, The time constant of the second polarization branch. The resistance of the second polarization branch, The capacitor for the second polarization branch; The formula for the terminal voltage equation is as follows:

[0043] in, This is the open-circuit voltage of the battery. For constant current charging current, For ohm resistance, This is the first polarization voltage. This is the second polarization voltage; The first analytical expression is as follows:

[0044] in, Let be the first polarization voltage at time t. For constant current charging current, The resistance of the first polarization branch, The time constant of the first polarization branch; The second analytical expression is as follows:

[0045] in, Let be the second polarization voltage at time t. For constant current charging current, The resistance of the second polarization branch, The time constant of the second polarization branch; The formula for calculating the terminal voltage during the offline preparation phase is as follows:

[0046] in, The calculated terminal voltage at time t during the offline preparation phase. The initial value of the open-circuit voltage. For constant current charging current, For ohm resistance, The resistance of the first polarization branch, The resistance of the second polarization branch, The time constant of the first polarization branch. The time constant of the second polarization branch; the parameter to be identified in the expression for calculating the terminal voltage is... Let the parameter vector to be identified be . .

[0047] For each set of charging data (corresponding to one aging stage) in the aging dataset established in step A1, the Levenberg-Marquardt algorithm is used. With the objective function of minimizing the sum of squared voltage errors between the measured and calculated terminal voltages, the algorithm iteratively fits the model parameters to be identified, obtaining updated model parameters. These updated model parameters are then used as the current values ​​for the model parameters to be identified in the next iteration, until the convergence condition is met, at which point the identified model parameters are output. The specific process is as follows: First, the terminal voltage in the static state before the start of charging in this set of charging data is taken as the initial value of the open circuit voltage, and the initial polarization voltage is determined to be zero.

[0048] For the first in the battery aging dataset The training sample (i.e., the nth training sample) (Group charging data), its sampling point sequence is as follows , The number of sampling points, the first The time is recorded as The corresponding measured terminal voltage is The calculated terminal voltage is The error vector between the measured terminal voltage and the calculated terminal voltage is then... The components are defined as follows:

[0049] The objective function is half of the sum of squared voltage errors (i.e., the sum of squared deviations).

[0050] Set the number of iterations And set the parameters to be identified. The initial value is then determined, and the following iterative process is executed: (1) Calculate the current parameter vector error vector And Jacobi matrix ; (2) Calculate the gradient of the objective function:

[0051] in, The gradient value, For the first The parameter vector of the next iteration , For the first The error vector of the next iteration. This is the Jacobian matrix corresponding to the error vector; (3) Selecting the damping factor Solve for the iteration step size :

[0052] in, The iteration step size, For the first The damping factor in the next iteration is a positive number. It is the identity matrix. Let Jacobi be the matrix corresponding to the error vector. The gradient value; (4) Update the parameter vector:

[0053] (5) Determine the convergence condition: If the norm of the iteration step size is less than the preset threshold for the step size, or the norm of the gradient is less than the preset threshold for the gradient, or the number of iterations reaches the preset maximum number of iterations, then terminate the iteration; otherwise, let It returns the steps for calculating the Jacobian matrix.

[0054] The convergence conditions are as follows:

[0055] or,

[0056] or, ; in, Let the norm of the iteration step size be . Set a threshold for the step size; Let be the norm of the gradient. Set a threshold for the gradient; For the number of iterations, This represents the maximum number of iterations.

[0057] The current value of the model parameter to be identified at the time of termination of iteration is determined as the identified model parameter.

[0058] The above parameter identification process is performed on each aging stage in the battery aging dataset to obtain the model parameters corresponding to each aging stage. These parameters are then associated with and stored in relation to the actual values ​​of temperature and maximum usable capacity for that stage, forming a model parameter set.

[0059] Step A3: Use the model parameter set obtained in step A2 as the input features of the training samples, and use the corresponding maximum available capacity true value as the output label of the training samples to train the physical perception neural network model.

[0060] Specifically, the input features include temperature, ohmic resistance, first polarization resistance, first polarization time constant, second polarization resistance, and second polarization time constant; Using the model parameters of the training samples, the predicted terminal voltage is calculated based on the second-order equivalent circuit model, and a voltage consistency loss term is constructed based on the error between the predicted terminal voltage and the measured terminal voltage in the battery aging dataset. The model parameters of the training samples are corrected based on the voltage consistency loss term to obtain the corrected model parameters; Temperature and corrected model parameters are used as inputs to the physical sensing neural network to be trained. The maximum available capacity prediction is calculated through forward propagation, and a capacity loss term is constructed based on the error between the maximum available capacity prediction and the actual maximum available capacity in the battery aging dataset. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation until the capacity loss term converges, thus completing the training of the physical perception neural network model.

[0061] In one specific implementation, for the first The model parameters to be corrected for each training sample for:

[0062] in, For the first The model parameters to be corrected for each training sample. For the first The ohmic resistance corresponding to each training sample For the first The first polarization resistance corresponding to each training sample For the first The first polarization event constant corresponding to each training sample. For the first The second polarization resistance corresponding to each training sample For the first The second polarization event constant corresponding to each training sample; Using the model parameters from the training samples (i.e., the model parameters to be corrected), the second-order equivalent circuit model is used to calculate the... The training sample at the th ... The predicted terminal voltage at each time point is:

[0063] in, The first battery aging dataset is calculated based on the parameters of the model to be corrected. The training sample at the th ... Predicted terminal voltage at each moment, For the first The initial open-circuit voltage values ​​corresponding to each training sample. For the first The constant current charging current corresponding to each training sample For the first The training sample of the th training sample Each sampling time.

[0064] The formula for calculating the voltage consistency loss term is as follows:

[0065] in, This is the voltage consistency loss term. The number of training samples. For the first The time series length of each training sample The first data collected centrally for battery aging The training sample at the th ... The measured terminal voltage at each moment. For the first in the battery aging dataset The training sample at the th ... Predicted terminal voltage at each moment; For the Using a training sample, the voltage error corresponding to that training sample is minimized to correct the parameters of the model to be corrected, resulting in the corrected model parameters:

[0066] in, For the first The model parameters were corrected based on the training samples.

[0067] The corrected model parameters are expressed as follows:

[0068] in, For the first Model parameters corrected from training samples For the first The corrected ohmic resistance of each training sample For the first The first polarization resistance corrected for each training sample For the first The first polarization event constant after correction of the training samples For the first The second polarization resistance corrected for each training sample For the first The second polarization event constant after correction of the training samples.

[0069] Temperature and the corrected model parameters are used together as input features to the physical perception neural network model. The input feature vector of each training sample is:

[0070] in, For the first The battery temperature corresponding to each training sample For the first Each training sample is input into the input feature vector of the physical perception neural network model.

[0071] The physical perception neural network model for the first The maximum available capacity prediction value for each training sample output:

[0072] in, For the first The predicted maximum usable capacity of a training sample. The nonlinear mapping function established for the physical perception neural network model. These are the network parameters of the physical sensing neural network model; The formula for calculating the capacity loss term is as follows:

[0073] in, The number of training samples. For capacity loss, For the first The true maximum usable capacity of a training sample. For the first Predicted maximum available capacity of training samples Update the network parameters of the physical perception neural network model by minimizing the capacity loss term:

[0074] in, These are the network parameters of the trained physical perception neural network model. A nonlinear mapping function established for a physical perception neural network model.

[0075] In this embodiment, the voltage consistency loss term is used to correct the model parameters of the second-order equivalent circuit model, enabling the corrected model parameters to more accurately characterize the dynamic characteristics of the battery's terminal voltage during local constant-current charging. The capacity loss term is used to train the network parameters of the physical sensing neural network model, allowing the physical sensing neural network model to output a predicted maximum usable capacity based on the corrected model parameters. Therefore, the voltage consistency loss term is not directly used to update the network parameters of the physical sensing neural network model, but rather serves as a physical consistency constraint term in the model parameter correction process. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation. This process is repeated until the capacity loss term converges, completing the training of the physical sensing neural network model. After training, the network establishes a mapping relationship from the equivalent circuit model parameters to the battery's maximum usable capacity.

[0076] Online estimation stage The online estimation stage is used to estimate the current battery health status in real time in actual charging scenarios using a trained physical sensing neural network model.

[0077] When an electric vehicle connects to a charging station and begins charging, charging data of the battery during the partial constant current charging process is obtained via S1. Specifically, after the constant current charging phase begins, the time series of charging current, measured terminal voltage, and battery temperature within a preset time period are collected, along with the stationary terminal voltage before charging begins. Simultaneously, based on the charging current and the battery's rated capacity, the constant current charging rate of the battery during the partial constant current charging process is determined.

[0078] As is understood, in this embodiment, "partial" refers to a portion of the data segment during the constant current charging phase, rather than the complete charging process. In actual operation of electric vehicles, it is often difficult to obtain a complete charge-discharge curve; in most cases, only data from a partial constant current charging phase can be collected. This embodiment utilizes partial data to estimate the battery's health status, overcoming the problem of obtaining a complete charge-discharge curve in existing technologies.

[0079] In some preferred embodiments, the duration of this data segment does not exceed 80% of the complete constant current charging time. This embodiment only needs to collect data from the first few seconds to several minutes of the constant current charging phase to complete the subsequent estimation of battery health status, without waiting for the complete charging process to end, which meets the needs of practical applications.

[0080] Then, based on the charging data, the model parameters of the current battery's second-order equivalent circuit model are obtained through S2. After obtaining the local constant current charging data, the second-order equivalent circuit model of the current battery is identified online using the same Levenberg-Marquardt algorithm as in step A2. Specifically, the terminal voltage in the resting state before charging begins is used as the initial value of the open-circuit voltage, and the initial polarization voltage is determined to be zero. Using a pre-established expression for calculating the terminal voltage, with the goal of minimizing the sum of squared voltage errors between the measured and calculated terminal voltages, the identified model parameters of the current battery are output through iterative fitting using the Levenberg-Marquardt algorithm. .

[0081] S3 inputs the model parameters into a pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery. Specifically, it inputs the identified model parameters of the current battery. Together with the currently collected temperature Together, they constitute the input feature vector: .

[0082] The input feature vector is fed into the physical sensing neural network model trained in step A3, and after one forward propagation calculation, the maximum usable capacity of the current battery is obtained. .

[0083] S4 is based on the current maximum available battery capacity obtained in step A3. Combined with the rated capacity of this battery model This yields an estimate of the current State of Health (SOH) of the battery.

[0084] The formula for battery health status is as follows: ; in, The maximum usable capacity of the current battery, as output by the physical sensing neural network model. For the battery's rated capacity, This is an estimate of the current battery health status.

[0085] This is an estimate of the current battery health status, which can be directly used to assess the battery's current capacity retention and serve as a reference for subsequent maintenance decisions.

[0086] It can also compare and determine whether the current battery health status meets the safety calibration standard based on the preset end-of-life standard and the calculated battery health status.

[0087] In this embodiment, the local constant current (CC) charging data refers to the partial charging data in constant current charging mode, where the continuous charging time does not exceed 80% of the full constant current charging time. Since this invention only requires partial data to complete parameter identification, it is applicable to various scenarios in practical applications where users end charging early or remove the charging gun after only partially charging, thus possessing strong engineering practical value.

[0088] It should be noted that the Levenberg-Marquardt algorithm in step A2 uses the exact same calculation formula and iterative logic as the online identification in S2. The only difference lies in the data source: the former comes from historical data in the offline aging dataset, while the latter comes from real-time data collected from the current battery. Both use the aforementioned terminal voltage expression for parameter fitting, ensuring consistency and comparability of the input features between the offline training phase and the online application phase.

[0089] Furthermore, those skilled in the art will understand that the specific parameter values ​​in the above steps (such as sampling interval, loop interval, maximum number of iterations, loss weight coefficient, etc.) can be adjusted according to the actual application scenario and battery type, and are not limited here.

[0090] The core idea of ​​this method is to establish a mapping relationship from physical parameters to capacity in the offline stage, while the online stage only requires extracting the physical parameters and performing neural network inference once to complete the estimation. This method makes full use of the rich data resources in the offline stage for model training, while ensuring low computational complexity and high real-time performance in the online stage, making it suitable for deployment in the battery management system of actual electric vehicles.

[0091] The key aspects and experimental results of this embodiment will be further explained below with reference to the accompanying drawings.

[0092] like Figure 4 The diagram shown illustrates the flowchart for parameter identification of the second-order equivalent circuit model of a battery under different aging conditions, as provided by this invention. The process specifically includes: (1) Read a set of local constant current charging data (corresponding to an aging stage) to obtain charging current, terminal voltage time series and temperature data; (2) Take the terminal voltage in the static state before charging starts in this set of data as the initial value of the open circuit voltage, and set the initial polarization voltage to zero; (3) Calculate the terminal voltage using the expression for calculating the terminal voltage, combined with the current value of the parameter to be identified; (4) Calculate the error vector and sum of squares between the measured terminal voltage and the calculated terminal voltage; (5) Determine whether the convergence condition is met: If it is met, output the current parameters as the identification result of the aging stage; if it is not met, calculate the Jacobian matrix and gradient, solve the iteration step size, update the parameter vector, and return to the step of calculating the Jacobian matrix to continue iterating until convergence.

[0093] The above process is repeated for each aging stage in the dataset, ultimately yielding a complete record of how the model parameters change with the degree of aging. This process embodies the core work of this method in the offline stage—transforming the raw charging data into a set of physically meaningful model parameters, providing high-quality input features for subsequent PRN network training.

[0094] like Figure 5 The diagram shown is a structural schematic of the physical sensing neural network model provided by this invention. The physical sensing neural network adopts a multi-layer fully connected structure. The input layer of this network contains 6 nodes, each corresponding to a temperature. Ohmic internal resistance First polarization resistor First polarization time constant Second polarization resistor Second polarization time constant The network consists of several hidden layers, each containing multiple neurons, which are passed layer by layer using a fully connected approach. The number of hidden layers and the number of nodes per layer can be optimized based on the training data scale. Their function is to extract deep features related to battery capacity decay from the input physical parameters. The network's output layer is a single node, used to output the predicted maximum usable capacity. .

[0095] This structure allows the network to learn both capacity mapping relationships and be constrained by the physical laws of voltage, thereby improving the model's generalization ability and estimation accuracy.

[0096] like Figure 6 The diagram shows a flowchart of the battery health state estimation process provided by this invention. In a real-world charging scenario, firstly, local constant-current charging data (time series of charging current, terminal voltage, and temperature) of the current battery is acquired, and the terminal voltage in the quiescent state before charging begins is collected as the initial value of the open-circuit voltage. Then, using the same method as offline identification, the model parameters of the current battery are identified online using the terminal voltage calculation formula and the Levenberg-Marquardt algorithm to obtain the model parameters in the current state. Finally, these parameters are combined with the current temperature to form an input feature vector, which is input into the trained physical sensing neural network. After one forward propagation, the maximum usable capacity of the current battery is obtained. Then, the current battery health status estimate is calculated based on the battery health status formula.

[0097] Throughout the online estimation process, parameter identification requires only a small amount of local constant current charging data, and neural network inference involves only one forward propagation calculation, resulting in low computational complexity and suitability for real-time deployment in resource-constrained battery management systems. Furthermore, the identification algorithm is completely consistent with the offline stage, ensuring the consistency of input features across time.

[0098] Figures 7(a)-7(d) show a comparison of the measured terminal voltage curves and the fitted terminal voltage curves of the second-order equivalent circuit model under four different aging stages provided by this invention. At different cycle numbers (e.g., the 50th, 100th, 200th, 500th, and 1000th cycles), the terminal voltage curves (fitted curves) calculated by substituting the identified model parameters into the terminal voltage calculation formula all show a high degree of overlap with the measured terminal voltage curves (measured curves). Experimental results show that, under different aging stages, the model parameters identified using the Levenberg-Marquardt algorithm can accurately describe the dynamic characteristics of the battery's terminal voltage, and the fitting accuracy meets the requirements of subsequent capacity estimation, verifying the effectiveness and reliability of the parameter identification method in this embodiment. The maintenance of this fitting accuracy also proves that the second-order equivalent circuit model's ability to represent the battery's dynamic behavior is stable under different aging stages, thus ensuring the quality of the input features for the physical sensing neural network.

[0099] like Figure 8 The figure shows a comparison between the estimated battery health status obtained using this invention and the actual value. Experimental data shows that there is good consistency between the estimated and actual values, and the battery health status estimation error remains at a low level, proving the effectiveness of the battery health status estimation method based on local constant current charging provided by this invention on different aging stages and different individual batteries.

[0100] Example 2: Based on the same inventive concept, this invention also provides a battery health state estimation system based on partial constant current charging, the structure of which is as follows: Figure 9 As shown, the system includes: The charging data acquisition module 901 is used to acquire the charging data of the current battery during the local constant current charging process; The model parameter acquisition module 902 is used to acquire the model parameters of the second-order equivalent circuit model of the current battery based on the charging data. The battery maximum capacity determination module 903 is used to input model parameters into a pre-trained physical sensing neural network model to obtain the current maximum usable capacity of the battery. The battery state estimation module 904 is used to obtain an estimated value of the current battery health state based on the maximum available capacity and the battery rated capacity. The physical sensing neural network model was trained offline in the following way: based on cyclic charge and discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels were collected to establish a battery aging dataset; a second-order equivalent circuit model was established, and based on the battery aging dataset, the model parameter set of the battery at different aging levels was obtained; the physical sensing neural network model was trained with the model parameter set as input and the corresponding true value of maximum usable capacity as output.

[0101] Preferably, the charging data acquisition module is specifically used for: Collect the time series of charging current, measured terminal voltage and battery temperature during a preset period of the current battery in the local constant current charging stage, and collect the stationary terminal voltage in the stationary state before charging starts. Based on the charging current and the battery's rated capacity, determine the constant current charging rate during the current battery's partial constant current charging process.

[0102] Preferably, the model parameter acquisition module is specifically used for: The initial value of the open-circuit voltage is taken as the resting terminal voltage in the resting state before charging begins, and the initial polarization voltage is determined to be zero. Based on the initial value of the open-circuit voltage, the initial polarization voltage, and the current values ​​of the model parameters to be identified, the calculated terminal voltage is obtained using a pre-established expression for the calculated terminal voltage. The Levenberg-Marquardt algorithm is adopted, with the objective function being to minimize the sum of squared voltage errors between the measured and calculated terminal voltages. The model parameters to be identified are iteratively fitted to obtain updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, and the identified model parameters are output. The expression for calculating the terminal voltage is constructed based on a second-order equivalent circuit model under local constant current charging conditions. The formula for calculating the terminal voltage is as follows:

[0103] in, Let be the calculated terminal voltage at time t. The initial value of the open-circuit voltage. For constant current charging current, For ohm resistance, The resistance of the first polarization branch, The resistance of the second polarization branch, The time constant of the first polarization branch. The time constant of the second polarization branch; These are the model parameters to be identified.

[0104] Preferably, the model parameter acquisition module is specifically used for: Calculate the error vector and Jacobian matrix of the model parameters to be identified at the current values. Calculate the iteration step size and the gradient of the objective function based on the Jacobian matrix and the error vector. Then, iteratively fit the model parameters to be identified based on the iteration step size and the gradient of the objective function to obtain the updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, at which point the iteration terminates. The current values ​​of the model parameters to be identified at the time of termination are then determined as the identified model parameters.

[0105] Preferably, the system also includes a physical perception neural network model building module, used for: The input features of the training samples are obtained from the model parameter set. The input features include temperature, ohmic resistance, first polarization resistance, first polarization time constant, second polarization resistance, and second polarization time constant. The true value of the maximum usable capacity corresponding to the training sample is also obtained. Using the model parameters of the training samples, the predicted terminal voltage is calculated based on the second-order equivalent circuit model, and a voltage consistency loss term is constructed based on the error between the predicted terminal voltage and the measured terminal voltage in the battery aging dataset. The model parameters of the training samples are corrected based on the voltage consistency loss term to obtain the corrected model parameters; Temperature and the corrected model parameters are used as inputs to the physical sensing neural network to be trained. The maximum available capacity prediction is calculated through forward propagation, and a capacity loss term is constructed based on the error between the maximum available capacity prediction and the actual maximum available capacity in the battery aging dataset. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation until the capacity loss term converges, thus completing the training of the physical perception neural network model. The formula for calculating the voltage consistency loss term is as follows:

[0106] in, This is the voltage consistency loss term. The number of training samples. For the first The time series length of each training sample The first data collected centrally for battery aging The training sample at the th ... The measured terminal voltage at each moment. For the first in the battery aging dataset The training sample at the th ... Predicted terminal voltage at each moment; The formula for calculating the capacity loss term is as follows:

[0107] in, The number of training samples. For capacity loss, For the first The true maximum usable capacity of a training sample. For the first The maximum usable capacity prediction for each training sample.

[0108] Preferably, the physical perception neural network model is specifically used for: Under different temperatures and constant current charging rates, multiple sets of cyclic charge-discharge experiments were performed on the battery. Each set of experiments included: constant current charging to the upper limit voltage, constant voltage charging to the cutoff current and resting, and constant current discharging to the lower limit voltage and resting. During the charging process of each set of cycles, the charging current, measured terminal voltage and battery temperature were collected in real time. After each set number of cycles, the discharge capacity is recorded after the discharge ends, and this is determined as the true value of the maximum usable capacity for this aging stage. The battery aging dataset is created by associating and storing the current, voltage, temperature, charging rate, and corresponding maximum usable capacity collected at different cycle stages.

[0109] Preferably, the battery state estimation module is specifically used for: Based on the maximum available capacity and the battery's rated capacity, calculate the estimated current battery health status according to the battery health status formula; The formula for battery health status is as follows: ; in, The maximum usable capacity of the current battery, as output by the physical sensing neural network model. For the battery's rated capacity, This is an estimate of the current battery health status.

[0110] Example 3: Based on the same inventive concept, such as Figure 10 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0111] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in a readable storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of the battery health state estimation method based on local constant current charging in the above embodiments.

[0112] Example 4: Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the readable storage medium here can include both the built-in storage medium within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the storage medium to implement the steps of the battery health state estimation method based on local constant current charging in the above embodiments.

[0113] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0115] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0116] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit its scope of protection. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading the present invention, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the claims pending approval.

Claims

1. A battery health state estimation method based on local constant current charging, characterized in that, include: Obtain charging data of the battery during the current local constant current charging process; Based on the charging data, obtain the model parameters of the second-order equivalent circuit model of the current battery. The model parameters are input into a pre-trained physical sensing neural network model to obtain the maximum usable capacity of the current battery. Based on the maximum available capacity and the battery's rated capacity, an estimated value of the current battery's health status is obtained; The physical sensing neural network model is trained offline in the following way: based on cyclic charge-discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels are collected to establish a battery aging dataset; a second-order equivalent circuit model is established, and based on the battery aging dataset, a set of model parameters for the battery at different aging levels is obtained; the physical sensing neural network model is trained with the set of model parameters as input and the corresponding true value of maximum usable capacity as output.

2. The method according to claim 1, characterized in that, The step of acquiring charging data of the current battery during the local constant current charging process includes: Collect the time series of charging current, measured terminal voltage and battery temperature during a preset period of the current battery in the local constant current charging stage, and collect the stationary terminal voltage in the stationary state before charging starts. Based on the charging current and the battery's rated capacity, the constant current charging rate during the current battery's partial constant current charging process is determined.

3. The method according to claim 1, characterized in that, The step of obtaining the model parameters of the second-order equivalent circuit model of the current battery based on the charging data includes: The resting terminal voltage in the resting state before charging begins is taken as the initial value of the open circuit voltage, and the initial polarization voltage is determined to be zero. Based on the initial value of the open-circuit voltage, the initial polarization voltage, and the current values ​​of the model parameters to be identified, the calculated terminal voltage is obtained using a pre-established expression for the calculated terminal voltage. The Levenberg-Marquardt algorithm is adopted, with the objective function being to minimize the sum of squared voltage errors between the measured terminal voltage and the calculated terminal voltage. The model parameters to be identified are iteratively fitted to obtain updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, and the identified model parameters are output. The expression for the calculated terminal voltage is constructed based on the second-order equivalent circuit model under local constant current charging conditions. The formula for calculating the terminal voltage is as follows: in, Let be the calculated terminal voltage at time t. The initial value of the open-circuit voltage. For constant current charging current, For ohmic resistance, The resistance of the first polarization branch, The resistance of the second polarization branch, The time constant of the first polarization branch. The time constant of the second polarization branch; These are the model parameters to be identified.

4. The method according to claim 3, characterized in that, The Levenberg-Marquardt algorithm is employed, with the objective function being to minimize the sum of squared voltage errors between the measured terminal voltage and the calculated terminal voltage. Iterative fitting is performed on the model parameters to be identified to obtain updated model parameters. These updated model parameters are then used as the current values ​​of the model parameters to be identified in the next iteration, until the convergence condition is met. The identified model parameters are then output, including: Calculate the error vector and Jacobian matrix of the model parameters to be identified at the current values, calculate the iteration step size and the gradient of the objective function based on the Jacobian matrix and the error vector, and iteratively fit the model parameters to be identified based on the iteration step size and the gradient of the objective function to obtain the updated model parameters. The updated model parameters are used as the current values ​​of the model parameters to be identified in the next iteration until the convergence condition is met, at which point the iteration terminates. The current values ​​of the model parameters to be identified at the time of termination of the iteration are then determined as the identified model parameters.

5. The method according to claim 1, characterized in that, The step of training the physical sensing neural network model, taking the model parameter set as input and the corresponding maximum available capacity true value as output, includes: The input features of the training samples are obtained from the model parameter set. The input features include temperature, ohmic resistance, first polarization resistance, first polarization time constant, second polarization resistance, and second polarization time constant. The true value of the maximum usable capacity corresponding to the training sample is also obtained. Using the model parameters of the training samples, the predicted terminal voltage is calculated according to the second-order equivalent circuit model, and a voltage consistency loss term is constructed based on the error between the predicted terminal voltage and the measured terminal voltage in the battery aging dataset. The model parameters of the training samples are corrected based on the voltage consistency loss term to obtain the corrected model parameters; The temperature and the corrected model parameters are used as inputs to the physical sensing neural network to be trained. The maximum available capacity prediction is calculated through forward propagation, and a capacity loss term is constructed based on the error between the maximum available capacity prediction and the actual maximum available capacity in the battery aging dataset. The gradient is calculated based on the capacity loss term, and the network parameters are updated through backpropagation until the capacity loss term converges, thus completing the training of the physical perception neural network model. The formula for calculating the voltage consistency loss term is as follows: in, This is the voltage consistency loss term. The number of training samples. For the first The time series length of each training sample The first data collected centrally for battery aging The training sample at the th ... The measured terminal voltage at each moment. For the first in the battery aging dataset The training sample at the th ... Predicted terminal voltage at each moment; The formula for calculating the capacity loss term is as follows: in, The number of training samples. For capacity loss, For the first The true maximum usable capacity of a training sample. For the first The maximum usable capacity prediction for each training sample.

6. The method according to claim 1, characterized in that, The cyclic charge-discharge experiments based on different temperatures and constant current charging rates collected charging data at different aging stages and the corresponding actual maximum usable capacity values ​​to establish a battery aging dataset, including: Under different temperatures and constant current charging rates, multiple sets of cyclic charge-discharge experiments were performed on the battery. Each set of experiments included: constant current charging to the upper limit voltage, constant voltage charging to the cutoff current and resting, and constant current discharging to the lower limit voltage and resting. During the charging process of each set of cycles, the charging current, measured terminal voltage and battery temperature were collected in real time. After each set number of cycles, the discharge capacity is recorded at the end of the discharge cycle and determined as the true value of the maximum usable capacity for this aging stage. The current, voltage, temperature, charging rate, and corresponding maximum usable capacity collected at different cycle stages are correlated and stored to establish a battery aging dataset.

7. The method according to claim 1, characterized in that, The process of obtaining an estimated current battery health status based on the maximum available capacity and the battery's rated capacity includes: Based on the maximum available capacity and the battery rated capacity, calculate the estimated health status of the current battery according to the battery health status formula; The formula for the battery health status is as follows: ； in, The maximum usable capacity of the current battery, as output by the physical sensing neural network model. For the battery's rated capacity, This is an estimate of the current battery health status.

8. A battery health state estimation system based on partial constant current charging, characterized in that, include: The charging data acquisition module is used to acquire the charging data of the current battery during the local constant current charging process; The model parameter acquisition module is used to acquire the model parameters of the second-order equivalent circuit model of the current battery based on the charging data. The battery maximum capacity determination module is used to input the model parameters into a pre-trained physical sensing neural network model to obtain the current maximum usable capacity of the battery. The battery status estimation module is used to obtain an estimated value of the current battery health status based on the maximum available capacity and the battery rated capacity. The physical sensing neural network model is trained offline in the following way: based on cyclic charge-discharge experiments at different temperatures and constant current charging rates, charging data and corresponding true values ​​of maximum usable capacity at different aging levels are collected to establish a battery aging dataset; a second-order equivalent circuit model is established, and based on the battery aging dataset, a set of model parameters for the battery at different aging levels is obtained; the physical sensing neural network model is trained with the set of model parameters as input and the corresponding true value of maximum usable capacity as output.

9. An electronic device, characterized in that, include: At least one processor and memory; The memory and processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the battery health state estimation method based on local constant current charging as described in any one of claims 1 to 7 is implemented.

10. A readable storage medium, characterized in that, It contains an executable program, which, when executed, implements the battery health state estimation method based on local constant current charging as described in any one of claims 1 to 7.

Images