A solid-state battery soc high-precision estimation method based on electric-power signal

Abstract

The application discloses a solid-state battery SOC estimation method based on electric-power double signal adaptive fusion. Voltage, current and surface mechanical stress signals are synchronously collected, a second-order RC equivalent circuit model is established, and a recursive least square method with a forgetting factor is used to identify parameters on line; a high-order polynomial stress model is established based on a stress-SOC nonlinear relationship, a stress offset and a temperature influence factor are introduced to compensate for the deviation caused by working conditions, sensor zero drift, pre-tightening force and temperature; voltage and stress observation are uniformly included in an adaptive unscented Kalman filtering framework, noise covariance is dynamically adjusted according to observation sensitivity, and double signal adaptive fusion and high-precision SOC estimation are realized.

Description

Technical Field

[0001] This invention relates to the field of battery management system technology, and in particular to a high-precision method for estimating the state of charge (SOC) of a solid-state battery based on electrical signals. Background Technology

[0002] With the rapid development of electric vehicles, increasingly higher demands are being placed on the energy density and safety of on-board energy storage systems. Lithium-ion batteries have become the mainstream choice due to their high energy density, low self-discharge rate, and long cycle life. However, traditional liquid lithium-ion batteries are limited by the electrochemical window, thermal stability, and flammability of the electrolyte, and their performance is gradually approaching its theoretical limit. Solid-state batteries (SSBs) replace liquid electrolytes with solid electrolytes, eliminating the risks of leakage and combustion, and have higher energy density and safety, making them an important direction for next-generation power batteries. However, during the charging and discharging process of SSBs, the electrode volume changes caused by lithium-ion insertion and extraction are more pronounced under the rigid solid-solid interface constraint, and the accumulation and release of mechanical stress may lead to interface degradation and even safety risks. State of Charge (SOC) is the most critical state variable of the Battery Management System (BMS), and its estimation accuracy directly affects range prediction, energy dispatch, and safety protection. However, SOC cannot be directly measured and can only be estimated indirectly through external signals such as current and voltage.

[0003] Currently, SOC estimation methods based on electrical signals are mainly divided into direct calculation methods, model-based methods, and data-driven methods. Model-based methods describe the voltage-current dynamic characteristics by establishing an equivalent circuit model and combine it with Kalman filtering to achieve online estimation, among which the second-order RC model is the most widely used. However, existing electrical signal-based methods essentially rely only on the single observation dimension of voltage, and their estimation accuracy is highly dependent on the accuracy of the equivalent circuit model parameters, which change with temperature, SOC, and aging. At the same time, when the voltage sensor fails or the measurement is disturbed, the system lacks an independent source of state information, resulting in insufficient robustness and fault tolerance redundancy.

[0004] On the other hand, the mechanical stress signal generated by the battery during charging and discharging provides an independent observation dimension for SOC estimation, where the measurement mechanism and electrical signal are mutually independent. In solid-state batteries, the rigid solid-solid interface constraint makes the change in electrode volume produce a more significant mechanical response. The stress signal originates directly from the volume change of the electrode material, does not depend on the parameters of the equivalent circuit model, and can provide effective redundancy information when the voltage channel is affected by parameter drift or measurement disturbance. However, existing methods based on stress signals also have limitations: the stress-SOC relationship exhibits nonlinear characteristics in different SOC ranges, with a mechanical plateau period in the medium SOC range, resulting in a significant decrease in sensitivity; stress measurement is easily affected by factors such as temperature changes, preload decay, and sensor zero drift; existing research lacks an effective online compensation mechanism to address the systematic bias and temperature effect of the stress model. Existing joint electrical-mechanical estimation methods use a two-step method based on EKF estimation followed by weighted fusion, employing fixed weights and failing to compensate for stress model bias and temperature effect online, thus failing to adapt to the dynamic changes in sensitivity of the two signals under different operating conditions.

[0005] Therefore, there is an urgent need for a SOC estimation method that can effectively fuse electrical and force signals with independent measurement mechanisms, dynamically weight each signal according to its reliability under different operating conditions, and adaptively compensate for stress model deviations and temperature effects, so as to achieve high-precision estimation of solid-state battery SOC under all operating conditions. Summary of the Invention

[0006] To overcome the problems of insufficient accuracy and poor robustness in existing methods for estimating the SOC of solid-state batteries based on a single electrical signal or a single stress signal, the present invention aims to provide a high-precision SOC estimation method for solid-state batteries based on electro-stress signals. Through multi-source signal collaborative fusion and adaptive weighting mechanism, it achieves high-precision and high-robust SOC estimation under wide operating conditions.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a high-precision estimation method for the state of charge (SOC) of a solid-state battery based on electro-force signals, comprising the following steps: Step S1: Simultaneously collect voltage, current, and surface mechanical stress signals of solid-state batteries at different temperatures on the experimental platform; Step S2: Based on the collected voltage and current signals, establish a second-order RC equivalent circuit model, and use the recursive least squares method with forgetting factor to identify the model parameters in real time online, and establish the relationship between OCV and SOC. Step S3: Establish a high-order polynomial model of stress-SOC based on the collected surface mechanical stress signal, introduce temperature influence factor and stress offset, compensate for stress deviation caused by temperature change, dynamic working conditions, sensor zero drift and preload change, and calculate the local sensitivity of stress to SOC. Step S4: Based on the second-order RC equivalent circuit model and the stress-SOC high-order polynomial model, construct the voltage and stress dual-observation equations, and integrate the voltage and stress dual-observation equations into the adaptive unscented Kalman filter framework; adaptively adjust the observation noise covariance according to the local sensitivity of stress to SOC, dynamically allocate the dual-channel correction weights, realize the fusion of electric and force signals, and obtain the SOC estimation result.

[0008] Furthermore, the experimental platform in step S1 includes: an Arbin single-cell battery testing system for executing test protocols and acquiring voltage and current data in real time; a mechanical constraint device that applies rigid constraints to the battery using a sandwich structure; a pressure sensor for acquiring mechanical stress signals on the battery surface; a constant temperature chamber for providing a constant ambient temperature; and a control host computer for controlling the execution of experimental protocols, monitoring safety boundaries, and storing data.

[0009] Furthermore, the second-order RC equivalent circuit model established in step S2 includes an ohmic internal resistance R0, a first RC network (composed of R1 and C1 in parallel, used to simulate electrochemical polarization), and a second RC network (composed of R2 and C2 in parallel, used to simulate concentration polarization), with the battery charging direction as the positive current direction, and its expression is as follows: ; in, This indicates the measured battery terminal voltage. Indicates open-circuit voltage. Parallel branches are used to simulate the electrochemical polarization of a battery, caused by the electrochemical reaction kinetics at the electrode surface. This reflects the voltage hysteresis phenomenon caused by the limitation of the electrochemical reaction rate during the charging and discharging process. This represents a relatively fast dynamic response, and the voltage it generates; The parallel branch is used to simulate the concentration polarization of the battery, caused by the diffusion process of the active material inside the battery. It reflects the voltage hysteresis phenomenon caused by the diffusion limitation of lithium ions in the electrodes and electrolyte during the charging and discharging process, representing a slower dynamic response. The voltage it generates is... .

[0010] Furthermore, in step S2, recursive least squares with a forgetting factor (FFRLS) is used to identify the model parameters in real time online. The process is as follows: the second-order RC equivalent circuit model is subjected to Laplace transform and discretized, transforming it into a difference equation in the form of ARMA (autoregressive moving average), and the small signal is defined. Establish the standard linear regression form , where the regression vector Includes current and historical terminal voltage and current information, parameter vector The system includes combinations of circuit parameters to be identified; online identification is performed using a recursive least squares method with a forgetting factor, and the recursive formula is as follows: ; in, For the gain vector, Let covariance matrix be the variance matrix. This is a forgetting factor used to reduce the weight of historical data to enhance the tracking ability of time-varying parameters; based on the identified parameter vector... The inverse kinematics yields all the physical parameters of the second-order RC equivalent circuit. .

[0011] Furthermore, the high-order polynomial stress model established in step S3, based on constant current discharge experimental data, uses SOC as the independent variable and surface mechanical stress as the dependent variable, and fits the stress-SOC relationship using a 12th-order polynomial. Its expression is: ; in For battery surface stress, For the first The coefficients of the order polynomial are determined by the least squares method.

[0012] Furthermore, the pressure offset state quantity introduced in step S3 is used to compensate for systematic deviations caused by dynamic operating condition deviations, sensor zero drift, and changes in preload. It is modeled as a random walk process. ; in This represents noise during the migration process. Correspondingly, the pressure observation equation is expressed as: ; in It is a 12th-order polynomial stress model. This is a pressure offset state quantity. This is to reduce noise in the pressure measurement. The pressure offset state quantity is then included as part of the state variables in subsequent step S4, and is estimated online and updated in real time by an adaptive unscented Kalman filter.

[0013] Furthermore, in step S3, the temperature influence factor is derived based on multi-temperature experimental data, and the stress model is temperature-corrected. Specifically, using 25℃ as the reference temperature, and based on stress-SOC experimental data at three temperature gradients of 15℃, 25℃, and 35℃, the offset coefficient of the stress curve relative to the reference temperature at each temperature is calculated. A second-order polynomial fitting is then performed with the ambient temperature T as the independent variable to obtain a continuous expression for the temperature influence factor. ; Where the temperature influence factor at 25°C is set to 1, and a, b, and c are fitting parameters; after introducing temperature correction, the stress observation model is modified as follows: ; in This is a 12th-order polynomial stress model calibrated at 25℃. This refers to the temperature influence factor corresponding to the current ambient temperature. This is a pressure offset state quantity.

[0014] Furthermore, the state vector in step S4 contains four variables to be estimated: .

[0015] in In a charged state, and The polarization voltage is for a second-order RC model. The pressure offset is used to compensate for the systematic deviation between the stress model and the actual pressure online, and is modeled as a random walk process.

[0016] The state equation is: ; in This represents process noise. Time constant. , The equivalent circuit parameters are provided in real time by online identification values, enabling online updates of model parameters.

[0017] Furthermore, the observation vector in step S4 contains two directly measurable physical quantities: the corresponding dual-observation equation is: ; in and The noise is measured for voltage and pressure, respectively. The two observation equations couple the electrical and force signals by sharing the state variable SOC, constraining the same SOC state from different physical dimensions.

[0018] Furthermore, the observation noise matrix of the adaptive unscented Kalman filter (AUKF) in step S4... Designed as a diagonal matrix that dynamically adjusts with sensitivity: ; in Based on basic noise, To calibrate noise, when the sensitivity of a certain channel decreases, the corresponding R increases, and the weight of that channel in the Kalman gain automatically decreases, thus achieving continuous dynamic adjustment of the correction weights of the two observation channels.

[0019] Furthermore, the adaptive unscented Kalman filter (AUKF) recursive process in step S4 includes the following steps: S4.1. Based on the state estimate and covariance of the previous time step, generate 2n+1 Sigma points (n=4) through Cholesky decomposition. S4.2. Perform nonlinear propagation of each Sigma point through the state equation to calculate the predicted state mean and covariance; S4.3. Regenerate Sigma points based on the predicted state, and calculate the predicted observation mean for each Sigma point using both voltage and pressure dual observation equations; S4.4. Update the observation noise matrix R(k) according to the current predicted stress sensitivity at the SOC using an adaptive rule; S4.5. Calculate the observation autocovariance and state-observation cross-covariance, calculate the Kalman gain and update the state estimate and covariance, and use the measured voltage and pressure values ​​to correct the state prediction value to obtain the state estimate value including SOC and pressure offset.

[0020] Furthermore, the method also includes a step of verifying three dynamic operating conditions—FUDS, UDDS, and NEDC—under different preload and temperature conditions. Using the ampere-hour integral method reference value as a benchmark, the root mean square error (RMSE) and mean absolute error (MAE) are calculated to quantify the accuracy estimate. The technical concept of this invention is as follows: By decoupling the physical mechanisms of the electrical response and mechanical expansion behavior of solid-state batteries, this invention integrates voltage and stress observations into a unified adaptive unscented Kalman filter framework. To address the systematic bias in the stress model, pressure offset is used as an AUKF state variable for online real-time compensation. To correct the drift effect of temperature on the stress reference, a temperature influence factor model is established. To address the inconsistency in the reliability of voltage and stress across different SOC ranges, an adaptive observation noise matrix based on physical sensitivity is designed. The dual-channel correction weights are dynamically adjusted according to the local rate of change of stress with respect to SOC, achieving the fusion of electrical and force signals in a probabilistically optimal sense. This enables high-precision and robust estimation of solid-state battery SOC under all operating conditions.

[0021] The beneficial effects of this invention are mainly reflected in: 1. This invention overcomes the inherent limitations of single signals under different operating conditions by adaptively fusing electrical signals and force signals that are independent of electromechanical measurement mechanisms, significantly improving the accuracy and robustness of solid-state battery SOC estimation under all operating conditions. Tests on three dynamic operating conditions (FUDS, UDDS, NEDC) under different preload forces (500N, 800N) and temperatures (15℃, 25℃, 35℃) show that the root mean square error of SOC estimation is less than 1.22%, which is more than 50% lower on average compared to single-signal methods based solely on voltage and stress.

[0022] 2. This invention effectively overcomes the interference of preload variation, sensor drift and temperature fluctuation on the stress reference by introducing pressure offset state quantity as online estimation of AUKF state variable and polynomial correction of temperature influence factor, so that the method maintains stable high-precision estimation capability in a wide preload range and a wide temperature range.

[0023] 3. The adaptive fusion mechanism based on physical sensitivity proposed in this invention can dynamically allocate correction weights according to the reliability of each observation channel under the current operating state, realizing probabilistic optimal fusion without manual setting, and improving the method's adaptability to different operating conditions and battery aging states. Attached Figure Description

[0024] Figure 1 This is a diagram of the dual-polarization equivalent circuit of the solid-state battery of the present invention; Figure 2 This invention provides the SOC estimation error for FUDS conditions under different initial preload forces of 500N and 800N. Figure 3 This is the SOC estimation error of the present invention under NEDS operating conditions at 15℃ and 35℃; Figure 4 This is a framework diagram of the SOC estimation technique of the present invention; Figure 5 This is a schematic diagram of the entire SOC estimation process of the present invention.

[0025] In the diagram, U ocv This represents the battery open-circuit voltage, which corresponds to the battery's state of charge (SOC); R oR1 is the ohmic internal resistance, used to characterize the instantaneous ohmic voltage drop generated by electrode materials, electrolyte, and current collector. The R1C1 branch describes the faster electrochemical polarization process inside the battery, where R1 represents the charge transfer resistance, C1 represents the corresponding polarization capacitance, and its polarization voltage is denoted as U1. The R2C2 branch describes the slower concentration polarization and lithium-ion diffusion process, where R2 and C2 represent the corresponding polarization resistance and polarization capacitance, respectively, and their polarization voltage is denoted as U2. The two RC branches characterize the dynamic response at both fast and slow time scales, enabling the model to accurately reflect the terminal voltage changes of the solid-state battery under dynamic charge-discharge conditions. Using the second-order RC equivalent circuit model, the measured current and the online identified R1 can be used to... o The parameters R1, C1, R2, and C2 are used to predict the battery terminal voltage, and U1 and U2 are introduced as internal polarization states into the subsequent AUKF state-space model to establish a dynamic observation relationship between the voltage signal and the SOC. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and not for limiting the scope of protection of the present invention.

[0027] The technical solution adopted by this invention to solve its technical problem is: a high-precision estimation method for the state of charge (SOC) of a solid-state battery based on an electro-force signal, comprising the following steps: Step S1: Simultaneously collect voltage, current, and surface mechanical stress signals of solid-state batteries at different temperatures on the experimental platform; Step S2: Based on the collected voltage and current signals, establish a second-order RC equivalent circuit model, and use the recursive least squares method with forgetting factor to identify the model parameters in real time online, and establish the relationship between OCV and SOC. Step S3: Establish a high-order polynomial model of stress-SOC based on the collected surface mechanical stress signal, introduce temperature influence factor and stress offset, compensate for stress deviation caused by temperature change, dynamic working conditions, sensor zero drift and preload change, and calculate the local sensitivity of stress to SOC. Step S4: Based on the second-order RC equivalent circuit model and the stress-SOC high-order polynomial model, construct the voltage and stress dual observation equations, and integrate the voltage and stress dual observation equations into the adaptive unscented Kalman filter framework; adaptively adjust the observation noise covariance according to the local sensitivity of stress to SOC, dynamically allocate the dual-channel correction weights, realize the fusion of electric and force signals, and obtain the SOC estimation results.

[0028] The expression for the second-order RC equivalent circuit model established in step S2 is as follows: ; in, This indicates the measured battery terminal voltage. Indicates open-circuit voltage. It represents the ohmic internal resistance of the battery, simulates the ohmic loss during the charge transfer process inside the battery, and reflects the instantaneous voltage drop caused by electrode materials, electrolyte and current collector during the charging and discharging process; The parallel branch is used to simulate the electrochemical polarization of a battery, caused by the electrochemical reaction kinetics at the electrode surface. It reflects the voltage lag phenomenon caused by the rate limitation of the electrochemical reaction during charging and discharging, representing a faster dynamic response. The voltage it generates is... ; The parallel branch is used to simulate the concentration polarization of the battery, caused by the diffusion process of the active material inside the battery. It reflects the voltage hysteresis phenomenon caused by the diffusion limitation of lithium ions in the electrodes and electrolyte during the charging and discharging process, representing a slower dynamic response. The voltage it generates is... .

[0029] The recursive least squares parameter identification process with forgetting factor in step S2 is as follows: The discretized battery model, and the Laplace transform of the equivalent circuit model, yields the following results: ; Taking the current as the input to the battery system, the transfer function is obtained as follows: ; Furthermore, the zero-order hold method is used to discretize the continuous transfer function. The zero-order hold method assumes a sampling period of... The internal input signal remains constant, allowing for precise sampling of the continuous system. Let... , Given the discrete poles of the two RC circuits, the Z-domain transfer function is obtained as follows: ; in To represent the unit delay operator, a new operator is introduced to simplify calculations. ~ The analytical relationship between the five discrete parameters and the circuit's physical parameters is as follows: ; Furthermore, the Z-domain transfer function is transformed into a time-domain difference equation form that can be directly used for parameter identification. The small signal is defined. That is, the polarization and ohmic components after subtracting the open-circuit voltage from the terminal voltage, we can obtain: ; Among them, regression variables The order of magnitude is This avoids the impact of large DC components on the accuracy of the identification values. Because the SOC changes slowly, three consecutive sampling points... It is approximately true.

[0030] The time-difference equation is written in standard linear regression form. , where the regression vector and parameter vector They are respectively: ; Online identification was performed using recursive least squares with a forgetting factor (FFRLS). Forgetting factor The introduction of this feature gradually reduces the weight of historical data in the algorithm, enhancing its ability to track time-varying parameters. The recursive process is as follows: ; in, For the gain vector, The covariance matrix is ​​the direct output of the predicted terminal voltage at each step of the RLS algorithm. This can be used to evaluate the model's identification accuracy. Based on the identified parameter vector... It is possible to solve for all the physical parameters of the equivalent circuit: ; The expression for the higher-order polynomial stress model established in step S3 is as follows: ; Differentiating the stress model expression yields the stress sensitivity function to SOC: ; in, For battery surface stress, For the first The coefficients of the polynomial are determined using the least squares method. The polynomial order is taken as n=12.

[0031] The stress offset introduced in step S3 The offset changes slowly over time and has no clear dynamic law, so it is modeled as a random walk process: ; in, This represents noise during the migration process. Correspondingly, the stress observation equation is expressed as: ; in, It is a 12th-order polynomial stress model. This is the stress offset. To reduce stress measurement noise, the SOC is corrected by comparing the actual measured stress with the stress estimated by the model, ensuring the accuracy and reliability of SOC estimation under various operating conditions.

[0032] The expression for the temperature factor in step S3 is as follows: ; in, The temperature influence factor is T, where T is the battery temperature, and the coefficients a, b, and c have values ​​of a = 1.296 × 10⁻⁴, b = 8.689 × 10⁻³, and c = 0.7018, respectively.

[0033] Furthermore, after introducing temperature correction, the stress observation model is revised as follows: ; in, This is a 12th-order polynomial stress model calibrated at 25℃. This refers to the temperature influence factor corresponding to the current ambient temperature. Let S be the stress offset, and S be the corrected stress. This is noise from stress measurement.

[0034] The state vector of AUKF in step S4 is as follows: ; in, In a charged state, and The polarization voltage is for a second-order RC model. The pressure offset is used to compensate for the systematic deviation between the stress model and the actual pressure online, and is modeled as a random walk process.

[0035] The state equation for the vector parameters is: ; in, This represents process noise. Time constant. , The equivalent circuit parameters are provided in real time by online identification values, enabling online updates of model parameters.

[0036] The observation vector contains two directly measurable physical quantities, and the corresponding two-observation equation is: ; in, and These are the noise levels from voltage and pressure measurements, respectively.

[0037] The formula for calculating the observation noise matrix is: ; in, Based on basic noise, To calibrate noise, when the sensitivity of a certain channel decreases, the corresponding R increases, and the weight of that channel in the Kalman gain automatically decreases, thus achieving continuous dynamic adjustment of the correction weights of the two observation channels.

[0038] The AUKF recursive process in step S4 is as follows: S4.1. Generate Sigma points. Based on state estimation. and covariance Nine Sigma points were generated through Cholesky decomposition: ; in, This represents the posterior state estimate from the previous time step, with a dimension of 4×1; Let represent the posterior covariance matrix of the previous time step, with a dimension of 4×4; This represents the i-th Sigma sampling point, with a dimension of 4×1.

[0039] S4.2. State Prediction. Each Sigma point is nonlinearly propagated through the vector parameter state equation in claim 7 to calculate the predicted state mean and covariance: ; in, This is the state transition function. To control the input, The process noise covariance matrix These are the weights used when calculating the weighted average. These are the weights used when calculating the weighted covariance. It is the predicted state mean. To predict state covariance.

[0040] S4.3. Regenerate Sigma points. Based on and Regenerate Sigma points This is to improve the approximation accuracy of nonlinear systems.

[0041] S4.4. Two-Observation Prediction. The predicted observation mean is calculated by simultaneously applying the two-observation equation in claim 7 to each Sigma point: ; in, It is a two-observation function. Corresponding voltage observation equation, Corresponding stress observation equation.

[0042] S4.5. Adaptive R(k) Update. Based on the sensitivity at the current predicted SOC, calculate the adaptive observation noise matrix according to the observation noise matrix calculation formula in claim 7.

[0043] S4.6. Covariance Calculation. Calculate the observed autocovariance. State-observation cross-covariance The formula is as follows: ; Among them, the observation autocovariance matrix The state-observation cross-covariance matrix has a dimension of 2×2. The dimension is 4×2.

[0044] S4.7. Measurement Update. Calculate the Kalman gain and update the state estimate and covariance formulas as follows: ; in, It is a 4×2 Kalman gain matrix. For the actual measurement vector, This is the updated state vector.

[0045] Furthermore, the SOC components in the state vector are expanded as follows: ; in, and These are the gains of the voltage channel and pressure channel for SOC correction, respectively, automatically calculated from the covariance matrix. When the voltage channel sensitivity is high... Dominant; when pressure channels are more reliable The contribution automatically increases. This mechanism achieves adaptive fusion of voltage and pressure signals in a probabilistically optimal sense, without the need for manual weight setting.

[0046] The above embodiments are used to explain and illustrate the present invention, but not to limit the present invention. Any modifications and changes made to the present invention within the spirit and scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A high-precision method for estimating the state of charge (SOC) of a solid-state battery based on electro-force signals, characterized in that, Includes the following steps: S1. Simultaneously collect voltage, current and surface mechanical stress signals of solid-state batteries at different temperatures on the experimental platform; S2. Based on the collected voltage and current signals, establish a second-order RC equivalent circuit model. Use the recursive least squares method with forgetting factor to identify the parameters of the second-order RC equivalent circuit model in real time and establish the relationship between OCV and SOC to obtain the real-time electrical parameters. S3. Based on the collected surface mechanical stress signals, establish a stress-SOC high-order polynomial model, introduce temperature influence factors and stress offset, compensate for the stress deviation caused by temperature changes, dynamic working conditions, sensor zero drift and preload changes in the stress-SOC high-order polynomial model, and construct the local sensitivity function relationship of stress to SOC. S4. Based on the second-order RC equivalent circuit model and the stress-SOC high-order polynomial model, construct the voltage and stress dual-observation equations and an adaptive unscented Kalman filter framework to obtain the voltage and stress dual-observation channels. Update the adaptive unscented Kalman filter framework according to the real-time electrical parameters, and estimate the stress offset online during the filtering recursion process. Substitute it into the local sensitivity function relationship of stress to SOC to calculate the local sensitivity value of SOC. Adjust the observation noise covariance of the adaptive unscented Kalman filter framework according to the local sensitivity value of SOC. Use the observation noise covariance to dynamically allocate the correction weights of the voltage and stress dual-observation channels to achieve power signal fusion and obtain the SOC estimation result.

2. The method for high-precision estimation of solid-state battery SOC based on power signals as described in claim 1, characterized in that, The expression for the second-order RC equivalent circuit model established in step S2 is as follows: ； in, This indicates the measured battery terminal voltage. Indicates open-circuit voltage. Indicates the ohmic internal resistance of the battery; The parallel branch is used to simulate the electrochemical polarization of the battery. C1 represents the corresponding polarization capacitance, and its polarization voltage is U1. R2 and C2 represent the corresponding polarization resistance and polarization capacitance, respectively, and their polarization voltage is denoted as U2. In the formula, t represents a continuous time variable, d / dt represents the derivative with respect to time t, and dU1 / dt and dU2 / dt represent the rates of change of the two polarization voltages with time, respectively.

3. The method for high-precision estimation of solid-state battery SOC based on power signals as described in claim 1, characterized in that, The recursive least squares parameter identification process with forgetting factor in step S2 is as follows: The discretized battery model, and the Laplace transform of the equivalent circuit model, yields the following results: ； Taking the current as the input to the battery system, the transfer function is obtained as follows: ； The continuous transfer function is discretized using the zero-order hold method, which assumes a sampling period. The internal input signal remains unchanged, and the continuous system is accurately sampled, so that... , Let be the discrete poles of the two RC circuits; where τ1 and τ2 are the time constants of the R1C1 and R2C2 branches, respectively, s is the complex frequency variable in the Laplace domain, and Δt is the discretized sampling period. The Z-domain transfer function is as follows: ； in Let p1 and p2 represent the unit delay operator; p1 and p2 correspond to the discrete poles of branches R1C1 and R2C2 respectively under the sampling period Δt, used to characterize the discrete attenuation characteristics of the corresponding polarization dynamics. To simplify the calculation, five discrete parameters are introduced. The analytical relationship between the discrete parameters and the circuit physical parameters is as follows: ； The Z-domain transfer function is transformed into a time-domain difference equation form that can be directly used for parameter identification, and a small signal is defined. That is, the polarization and ohmic components after subtracting the open-circuit voltage from the terminal voltage, we can obtain: ； Among them, regression variables The order of magnitude is This avoids the impact of large DC components on the accuracy of the identification values. Because the SOC changes slowly, three consecutive sampling points... It is approximately true, where k is the index of the discrete sampling time, and U ocv (k) represents the open-circuit voltage at the k-th sampling time; unless otherwise specified, k in the subsequent formulas in this paper represents the index of the discrete sampling time. The time-difference equation is written in standard linear regression form. The regression vector and parameter vector are as follows; where the parameter vector θ is composed of the five discrete parameters a1, a2, b0, b1, and b2 mentioned above, and the superscript T indicates the vector transpose: ； In the formula, k represents the index of the discrete sampling time; Online identification is performed using recursive least squares with a forgetting factor. The algorithm gradually reduces the weight of historical data to enhance its ability to track time-varying parameters. The recursive process is as follows: ； in, For the gain vector, The covariance matrix is ​​the direct output of the predicted terminal voltage at each step of the RLS algorithm. It can be used to evaluate the model identification accuracy, where k is the index of the current discrete sampling time, θ(k) is the parameter estimation vector at time k, φ(k) is the regression vector, and λ∈(0,1] is the forgetting factor; based on the identified parameter vector The inverse solution can be used to obtain all the physical parameters of the equivalent circuit. The inverse solution formula is as follows: 。 4. The high-precision SOC estimation method for solid-state batteries based on power signals as described in claim 1, characterized in that, The expression for the higher-order polynomial stress model established in step S3 is as follows: ； Differentiating the stress model expression yields the stress sensitivity function to SOC: ； in, For battery surface stress, For the first The coefficients of the polynomial are determined by the least squares method. The order of the polynomial is n=12. Here, h(·) represents the polynomial mapping function from SOC to battery surface stress, d represents the differential operator, dh / dSOC represents the local sensitivity of the stress model to SOC, i is the polynomial order index, and n is the highest order of the polynomial.

5. The high-precision SOC estimation method for solid-state batteries based on power signals as described in claim 1, characterized in that, The stress offset introduced in step S3 The stress offset changes slowly over time and has no clear dynamic law. Modeled as a random walk process: ； in, To account for noise during the migration process, the corresponding stress observation equation is expressed as: ； in, It is a 12th-order polynomial stress model. This is the stress offset. This is noise from stress measurement.

6. The high-precision SOC estimation method for solid-state batteries based on power signals as described in claim 1, characterized in that, The expression for the temperature factor in step S3 is as follows: ； in, Here, T represents the battery temperature, and the coefficients a, b, and c are determined by second-order polynomial least squares fitting of stress-SOC experimental data under multiple temperature gradients. After introducing temperature correction, the stress observation model is revised as follows: ； in, This is a high-order polynomial stress model calibrated at the reference temperature. This refers to the temperature influence factor corresponding to the current ambient temperature. Let S be the stress offset, and S be the corrected stress. To reduce stress measurement noise, the reference temperature is the reference temperature used during stress model calibration.

7. The method for high-precision estimation of solid-state battery SOC based on power signals as described in claim 1, characterized in that, The state vector of AUKF in step S4 is as follows: ； in, In a charged state, and The polarization voltage is for a second-order RC model. The pressure offset is used to compensate for the systematic deviation between the stress model and the actual pressure online. It is modeled as a random walk process. In the formula, x(k) represents the state vector at the k-th sampling time, the superscript T indicates the vector transpose, and k is the index of the discrete sampling time. The state equation for the vector parameters is: ； in, For process noise, time constant , The equivalent circuit parameters are provided in real time by online identification values, enabling online updates of model parameters; The observation vector contains two directly measurable physical quantities, and the corresponding two-observation equation is: ； in, and The noise levels for voltage and pressure measurements are respectively. The formula for calculating the observation noise matrix is: ； in, Based on basic noise, To calibrate noise, when the sensitivity of a channel decreases, the corresponding R increases, and the weight of that channel in the Kalman gain automatically decreases, achieving continuous dynamic adjustment of the correction weights for the two observation channels; R u,base and R s,base R represents the variance of the basic observation noise for the voltage channel and the stress channel, respectively. u,pen and R s,pen These are the sensitivity penalty coefficients for the two channels, ε. u and ε s A positive number is set to prevent the denominator from approaching zero. dOCV / dSOC and dh / dSOC represent the voltage-SOC relationship and the local sensitivity of the stress-SOC model at the current SOC, respectively. diag(·) represents the construction of a diagonal matrix from the elements in parentheses. R(k) represents the observation noise covariance matrix at the k-th sampling time.

8. The method for high-precision estimation of solid-state battery SOC based on power signals as described in claim 1, characterized in that, The AUKF recursive process in step S4 is as follows: S4.

1. Generate Sigma points based on state estimation and covariance Nine Sigma points were generated through Cholesky decomposition: ； in, This represents the posterior state estimate from the previous time step, with a dimension of 4×1; Let represent the posterior covariance matrix of the previous time step, with a dimension of 4×4; Let represent the i-th Sigma sampling point, with a dimension of 4×1; where n is the dimension of the state vector, n=4 in this invention; λ is the scale parameter of the unscented transformation, used to adjust the dispersion of the Sigma point relative to the state mean; i is the index of the Sigma point; the subscript i of the square root term of the matrix indicates taking its i-th column; χ0 is the center Sigma point, χ i With χ i+n These are the positive and negative Sigma points along the i-th direction, thus generating a total of 2n+1=9 Sigma points; S4.

2. State prediction: Each Sigma point is propagated nonlinearly through the vector parameter state equation in claim 7 to calculate the predicted state mean and covariance. ； in, This is the state transition function. To control the input, The process noise covariance matrix These are the weights used when calculating the weighted average. These are the weights used when calculating the weighted covariance. It is the predicted state mean. To predict state covariance; S4.

3. Regenerate Sigma points, based on and Regenerate Sigma points To improve the approximation accuracy of nonlinear systems; S4.

4. Two-observation prediction: Simultaneously apply the two-observation equation in claim 7 to each Sigma point to calculate the predicted observation mean. ； in, It is a two-observation function. Corresponding voltage observation equation, Corresponding stress observation equation; S4.

5. Adaptive R(k) update: Based on the sensitivity at the current predicted SOC, calculate the adaptive observation noise matrix according to the observation noise matrix calculation formula in claim 7; S4.

6. Covariance Calculation: Calculate the observed autocovariance. State-observation cross-covariance The formula is as follows: ； Among them, the observation autocovariance matrix The state-observation cross-covariance matrix has a dimension of 2×2. The dimension is 4×2; S4.

7. Measurement Update: Calculate the Kalman gain and update the state estimate and covariance formula as follows: ； in, It is a 4×2 Kalman gain matrix. For the actual measurement vector, The updated state vector, Expanding the SOC components in the state vector: ； in, and These represent the gains of the voltage channel and pressure channel for SOC correction, respectively, automatically calculated from the covariance matrix. When the voltage channel sensitivity is high... Dominant; when pressure channels are more reliable Its contribution increases automatically.

Images