A solid-state battery soh estimation method of a wide-temperature self-adaptive boosting algorithm
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEAT UNIV OF SCI & TECH
- Filing Date
- 2026-06-02
- Publication Date
- 2026-08-07
AI Technical Summary
[0005]本发明的发明目的在于：针对上述存在的问题，提供一种宽温下自适应提升算法的固态电池SOH估计方法，期望改善现有的传统方法在宽温环境下精度衰减、自适应能力不足的问题
1、本发明提出一种宽温下自适应提升算法的固态电池SOH估计方法，通过构建LSSVM-Adaboost融合模型、设计改进的CFO算法、开发嵌入式温度传感器及实现模型自进化机制，有效解决了现有的传统方法在宽温域下的精度衰减与自适应能力不足问题，为固态电池的安全应用提供关键技术支撑。
Smart Images
Figure CN122525394A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of battery management technology, and specifically relates to a solid-state battery SOH estimation method based on an adaptive boosting algorithm under wide temperature range. Background Technology
[0002] With the large-scale application of solid-state batteries in electric vehicles, energy storage power stations, and smart grids, accurate assessment of their state of health (SOH) has become a core technological bottleneck for battery safety management and secondary utilization. Traditional SOH estimation methods generally suffer from insufficient adaptability in wide temperature environments (-20℃ to 60℃): data-driven models based on single algorithms such as neural networks and Kalman filters often rely on room temperature data for training, making it difficult to capture the nonlinear changes in battery aging mechanisms over a wide temperature range. For example, in low-temperature scenarios, the dramatic increase in electrolyte viscosity leads to a 3-5 fold increase in internal resistance, while traditional models, lacking a dynamic internal resistance compensation mechanism, result in SOH estimation errors exceeding 12%. In high-temperature scenarios, the capacity drop caused by SEI film thickening and accelerated lithium dendrite growth cannot be effectively identified by the model due to fixed parameter settings.
[0003] At the parameter optimization level, traditional methods such as grid search and random search are prone to getting trapped in local optima in complex parameter spaces and do not consider the dynamic impact of temperature on model performance. Particle swarm optimization (PSO), while possessing global search capabilities, has slow convergence speed and computational complexity as high as O(N³) in optimizing multidimensional parameters (such as LSSVM kernel function coefficients and Adaboost iteration count). Genetic algorithms (GA), on the other hand, disrupt the continuity of parameter combinations due to crossover and mutation operations, resulting in SOH estimation errors fluctuating by more than 8%. The recently proposed Cordyceps sinensis optimization algorithm (CFO) is superior to PSO and GA in terms of global search capabilities, but its original version suffers from drawbacks such as a single step size strategy and a rigid population mechanism: the fixed step size design leads to low exploration efficiency in the early stages and a tendency to get trapped in local optima in the later stages; the static population size is prone to "premature convergence" in complex parameter spaces, and it does not integrate diversity preservation strategies.
[0004] However, the crudeness of environmental compensation strategies further exacerbates the limitations of traditional methods. Existing solutions mostly employ linear interpolation or fixed threshold adjustments, which cannot adapt to the nonlinear changes in battery aging characteristics over a wide temperature range. For example, the temperature segmentation model switching strategy proposed in the paper "SOH Estimation of Lithium-ion Batteries Considering Temperature Influence" (2023) requires manual threshold setting, and there are estimation jump problems at the boundaries between temperature zones; the table lookup method widely used in commercial BMS systems suffers from insufficient data density, resulting in SOH estimation resolution of less than 10% in the temperature range of -20℃ to 60℃. Summary of the Invention
[0005] The purpose of this invention is to provide a solid-state battery SOH estimation method with an adaptive boosting algorithm under wide temperature range, addressing the aforementioned problems. This aims to improve the accuracy degradation and insufficient adaptive capability of existing traditional methods under wide temperature ranges.
[0006] The technical solution adopted in this invention is as follows: a solid-state battery SOH estimation method with an adaptive boosting algorithm under wide temperature range, the method comprising the following steps: Step S100: Construct an LSSVM-Adaboost fusion model, and improve the model's generalization ability by combining the non-linear feature mapping capability of LSSVM with the iterative enhancement mechanism of Adaboost. Step S200: Introduce the improved Cordyceps sinensis optimization algorithm CFO to globally optimize the hyperparameters of the fusion model in step S100, and improve the CFO algorithm by adopting a piecewise nonlinear step size decay strategy and a dynamic population adjustment mechanism. Step S300: A temperature sensor is installed on the solid-state battery to collect ambient temperature data in real time and dynamically adjust the parameters of the fusion model to achieve adaptive solid-state battery health state (SOH) estimation under a wide temperature environment. Step S400: Output the estimated state of health (SOH) of the solid-state battery and its confidence interval, and transmit it to the battery management system (BMS) via the CAN bus; Step S500: Generate a battery health assessment report based on the SOH estimation results. The battery health assessment report includes confidence intervals, error indicators, and model update suggestions.
[0007] It is important to note that LSSVM-Adaboost is an algorithm that combines Least Squares Support Vector Machine (LSSVM) with Adaboost ensemble learning. Adaptive State of Health (SOH) estimation of solid-state batteries under wide temperature environments is achieved by optimizing the LSSVM-Adaboost fusion model using an improved Cordyceps sinensis optimization algorithm (CFO). The CFO population is initialized using the Latin hypercube concept to ensure a uniform population distribution and expand the search range, thereby adjusting the algorithm's convergence speed.
[0008] By adopting the above method, through constructing an LSSVM-Adaboost fusion model, designing an improved CFO algorithm, a temperature sensor, and implementing a model self-evolution mechanism, the problem of accuracy decay and insufficient adaptive capability of existing traditional methods in a wide temperature range is effectively solved, providing key technical support for the safe application of solid-state batteries.
[0009] Furthermore, the hyperparameters of the LSSVM-Adaboost fusion model include the kernel function parameters of the LSSVM module. and regularization coefficient and the number of iterations of the Adaboost module. Weak classifier weights ; The objective of global hyperparameter optimization is to minimize the SOH estimation error, with the error range controlled within a certain range. Within; and during the optimization process, a cross-validation strategy is adopted, with the validation set accounting for 20% of the original data.
[0010] It is important to note that existing technologies do not consider the dynamic impact of temperature on model performance at the parameter optimization level, or have slow convergence speed and high computational complexity in multi-dimensional parameter optimization, resulting in large fluctuations in SOH estimation error. Therefore, this invention controls the SOH estimation error by setting hyperparameters. The cross-validation strategy is an existing technique. The validation set consists of data using solid-state batteries, with the original data being charge and discharge data of solid-state batteries. Feature factors and SOH values of the batteries are extracted from the validation set. The feature factors and SOH values are input into the prediction set, where the SOH values are both input and output values. The dataset is divided into an 80% training set and a 20% test set. The training set is trained using feature factors and SOH values, while the prediction set is used to predict the SOH values of the test set.
[0011] Furthermore, the LSSVM-Adaboost fusion model specifically includes the following: The LSSVM module uses the RBF kernel function to map low-dimensional features of battery voltage, current, and internal resistance to a high-dimensional space. Kernel parameters... The system automatically adjusts its performance using a cross-validation strategy, with initial values set in the range of 0.0001-10000. A regularization coefficient C is introduced to suppress overfitting, with initial values set in the range of 1-100. The Adaboost module concatenates multiple weak classifiers and assigns weights to each weak classifier. Initialize to 0.5; The LSSVM module and the Adaboost module are cascaded, with the output features of the LSSVM module used as the input features of the Adaboost module. A feedback mechanism is also introduced, which inputs the SOC estimate from the previous time step and the current voltage / current data into the fusion model.
[0012] Furthermore, dynamic population adjustment mechanisms include the following: Define a population entropy value H to assess diversity; ; in, For the first The frequency of a combination of parameters.
[0013] Furthermore, the confidence interval for the SOH estimation results is calculated as follows: Based on the model training error distribution, a Gaussian confidence model is constructed. Set the confidence interval range as follows: ; in, Output values for the LSSVM-Adaboost fusion model; When the confidence interval exceeds this range, the model is triggered to retrain until the confidence interval width falls within this range.
[0014] Furthermore, in step S300, the LSSVM-Adaboost fusion model transmits the ambient temperature data collected by the embedded temperature sensor in real time via the I2C protocol.
[0015] Furthermore, the method also includes a self-evolutionary mechanism for the LSSVM-Adaboost fusion model: Set an upper limit on the number of charge-discharge cycles for the solid-state battery. Once the upper limit is reached, add the new hyperparameters to the training set and re-execute the CFO optimization process to obtain a new parameter combination. At the same time, the new parameter combination needs to inherit 1 / 5 of the old parameter combination from the previous version, and accelerate the convergence of the parameter combination while maintaining its stability.
[0016] Furthermore, a solid-state battery SOH estimation device with an adaptive boosting algorithm under wide temperature range is provided, the device comprising the following modules; The model building module is responsible for the initialization and cascading of the LSSVM-Adaboost fusion model; Parameter optimization module: Integrates and improves the CFO algorithm to perform global optimization of hyperparameters; Results output module: Generates SOH estimates, confidence intervals, and health assessment reports; Self-evolution module: periodically triggers model retraining and parameter inheritance.
[0017] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention proposes a solid-state battery SOH estimation method with an adaptive boosting algorithm under a wide temperature range. By constructing an LSSVM-Adaboost fusion model, designing an improved CFO algorithm, developing an embedded temperature sensor, and implementing a model self-evolution mechanism, it effectively solves the problems of accuracy decay and insufficient adaptive capability of existing traditional methods under a wide temperature range, and provides key technical support for the safe application of solid-state batteries.
[0018] 2. This invention reduces the SOH estimation error over a wide temperature range by setting a threshold for the confidence interval width; it also shortens the parameter optimization time and reduces computational complexity through an improved CFO algorithm; and further achieves seamless adaptation over a wide temperature range through a differentiated temperature compensation strategy. Attached Figure Description
[0019] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a module architecture diagram of the present invention. Detailed Implementation
[0020] The present invention will now be described in detail with reference to the accompanying drawings.
[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0022] With the planned application of solid-state batteries in various fields, accurate assessment of their State of Health (SOH) has become a core technological bottleneck for battery safety management and secondary utilization. Traditional SOH estimation methods generally suffer from insufficient adaptability in wide temperature environments (-20℃ to 60℃); moreover, at the parameter optimization level, traditional methods are prone to getting trapped in local optima, without considering the dynamic impact of temperature on model performance.
[0023] Therefore, this invention proposes a solid-state battery SOH estimation method with an adaptive boosting algorithm under a wide temperature range. By constructing an LSSVM-Adaboost fusion model, designing an improved CFO algorithm, developing an embedded temperature sensor, and implementing a model self-evolution mechanism, this method effectively solves the problems of accuracy decay and insufficient adaptive capability of traditional methods under a wide temperature range, providing key technical support for the safe application of solid-state batteries.
[0024] Example 1 like Figures 1-2 As shown, one embodiment of the present invention is a solid-state battery SOH estimation method based on an adaptive boosting algorithm under wide temperature range, the method comprising the following steps: Step S100: Construct an LSSVM-Adaboost fusion model, and improve the model's generalization ability by combining the non-linear feature mapping capability of LSSVM with the iterative enhancement mechanism of Adaboost. The LSSVM-Adaboost fusion model specifically includes the following: The LSSVM module uses the RBF kernel function to map low-dimensional features of battery voltage, current, and internal resistance to a high-dimensional space. Kernel parameters... The system automatically adjusts its performance using a cross-validation strategy, with an initial value of 0.1. A regularization coefficient C is introduced to suppress overfitting, with an initial value of 1.0, which is dynamically increased to 1.5 times in low-temperature scenarios. The Adaboost module concatenates 50 weak classifiers, assigning weights to each weak classifier. Initialize to 0.5; The LSSVM module and the Adaboost module are cascaded, with the output features of the LSSVM module used as the input features of the Adaboost module. A feedback mechanism is also introduced, in which the SOC estimate from the previous time step is input into the model along with the current voltage / current data.
[0025] Therefore, the sample weights are adjusted based on the classification error of the previous round, the weights of misclassified samples are increased by 30%, and the total number of iterations T is dynamically determined by the improved CFO algorithm; and the error is controlled within ±3% when the LSSVM module and the Adaboost module are cascaded.
[0026] Furthermore, the hyperparameters of the LSSVM-Adaboost fusion model include the kernel function parameters of the LSSVM module. and regularization coefficient and the number of iterations of the Adaboost module. Weak classifier weights ; The objective of global hyperparameter optimization is to minimize the SOH estimation error, with the error range controlled within a certain range. Within; and during the optimization process, a cross-validation strategy is adopted, with the validation set accounting for 20% of the original data.
[0027] Step S200: Introduce the improved Cordyceps sinensis optimization algorithm CFO to globally optimize the hyperparameters of the fusion model in step S100, and improve the CFO algorithm by adopting a piecewise nonlinear step size decay strategy and a dynamic population adjustment mechanism. Dynamic population adjustment mechanisms include the following: Define a population entropy value H to assess diversity; ; in, For the first The frequency of a combination of parameters.
[0028] When H < 0.8 or the error does not improve after 5 consecutive generations, new individuals are generated through Levy flight with a perturbation amplitude of 1.2 times the current step size; the top 50% of individuals in fitness are retained, inefficient solutions are eliminated, and the population size will be dynamically reduced from the initial 50 to 20.
[0029] Step S300: A temperature sensor is installed on the solid-state battery to collect ambient temperature data in real time and dynamically adjust the parameters of the fusion model to achieve adaptive solid-state battery health state (SOH) estimation under a wide temperature environment. The LSSVM-Adaboost fusion model transmits ambient temperature data collected by embedded sensors in real time via the I2C protocol.
[0030] The strategy for adjusting differentiated parameters in a wide-temperature environment is as follows: Low temperature scenario: Increase the LSSVM regularization coefficient C to 1.5 times to suppress overfitting caused by the sharp increase in internal resistance; In high-temperature scenarios: Adaboost automatically increases the number of iterations by 30% and strengthens the weight coefficient α of aging feature samples to 0.7; For normal temperature scenarios: Keep the default parameters and only enable the basic noise filtering function.
[0031] Step S400: Output the estimated state of health (SOH) of the solid-state battery and its confidence interval, and transmit it to the battery management system (BMS) via the CAN bus; A Gaussian confidence model is constructed based on the model training error distribution. Step S500: Generate a battery health assessment report based on the SOH estimation results. The battery health assessment report includes confidence intervals, error metrics, and model update recommendations. The confidence interval range is set as follows: ; in, Output values for the LSSVM-Adaboost fusion model; when the confidence interval width exceeds 10%, trigger the model retraining process.
[0032] Step S600: The self-evolution mechanism of the LSSVM-Adaboost fusion model; Set an upper limit on the number of charge-discharge cycles for the solid-state battery. Once the upper limit is reached, add the new hyperparameters to the training set and re-execute the CFO optimization process to obtain a new parameter combination. At the same time, the new parameter combination needs to inherit 1 / 5 of the old parameter combination from the previous version, and accelerate the convergence of the parameter combination while maintaining its stability.
[0033] This embodiment utilizes the above methods to assess the health status of solid-state batteries in a wide temperature range. Furthermore, it considers the dynamic impact of temperature on model performance during parameter optimization, thereby reducing SOH estimation errors in a wide temperature range, shortening parameter optimization time, and reducing the complexity of parameter calculations. This embodiment further achieves seamless adaptation within a wide temperature range of -20℃ to 60℃ through a differentiated temperature compensation strategy. And by setting a self-evolution mechanism for the model, it ensures that the calculation accuracy of the model remains relatively high even after multiple cycles.
[0034] Example 2 Another embodiment of the present invention provides that the specific method and range for determining the hyperparameters of the LSSVM-Adaboost fusion model include the following: Kernel function parameters of the LSSVM module Automatic adjustment is achieved through five-fold cross-validation, with the initial range set to [0.01, 1.0], and the optimization objective is to minimize the validation set error. Number of iterations of the Adaboost module The value is dynamically determined by the improved CFO algorithm, with the initial range set to [20, 200], and the optimization objective is to balance accuracy and computational efficiency.
[0035] Within the aforementioned parameter range, the regularization parameters of the LSSVM module were determined through multiple experiments, with an initial range set to [0.1, 10], aiming to improve the model's generalization ability. Simultaneously, the learning rate of the Adaboost module was dynamically adjusted, with an initial range set to [0.001, 0.1], aiming to avoid overfitting while ensuring model convergence speed. Furthermore, the weight allocation strategy of the fusion model was optimized by introducing a temperature compensation factor to ensure optimal synergistic performance of each module across a wide temperature range. This method not only improves the model's robustness but also further enhances the accuracy and adaptability of solid-state battery SOH estimation.
[0036] The low temperature scenario is (-20℃~0℃), the normal temperature scenario is (0℃~40℃), and the high temperature scenario is (40℃~60℃).
[0037] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A solid-state battery SOH estimation method based on an adaptive boosting algorithm under wide temperature range, characterized in that, The method includes the following steps: Step S100: Construct an LSSVM-Adaboost fusion model, and improve the model's generalization ability by combining the non-linear feature mapping capability of LSSVM with the iterative enhancement mechanism of Adaboost. Step S200: Introduce the improved Cordyceps sinensis optimization algorithm CFO to globally optimize the hyperparameters of the fusion model in step S100, and improve the CFO algorithm by adopting a piecewise nonlinear step size decay strategy and a dynamic population adjustment mechanism. Step S300: A temperature sensor is installed on the solid-state battery to collect ambient temperature data in real time and dynamically adjust the hyperparameters of the fusion model to achieve adaptive solid-state battery health state (SOH) estimation under a wide temperature environment; Step S400: Output the estimated state of health (SOH) of the solid-state battery and its confidence interval, and transmit it to the battery management system (BMS) via the CAN bus; Step S500: Generate a battery health assessment report based on the SOH estimation results. The battery health assessment report includes confidence intervals, error indicators, and model update suggestions.
2. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 1, characterized in that, The hyperparameters of the LSSVM-Adaboost fusion model include the kernel function parameters of the LSSVM module. and regularization coefficient and the number of iterations of the Adaboost module. Weak classifier weights ; The objective of global hyperparameter optimization is to minimize the SOH estimation error, with the error range controlled within a certain range. Within; and during the optimization process, a cross-validation strategy is adopted, with the validation set accounting for 20% of the original data.
3. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 2, characterized in that, The LSSVM-Adaboost fusion model specifically includes the following: The LSSVM module uses the RBF kernel function to map low-dimensional features of battery voltage, current, and internal resistance to a high-dimensional space. Kernel parameters... The system automatically adjusts its performance using a cross-validation strategy, with initial values set in the range of 0.0001-10000. A regularization coefficient C is introduced to suppress overfitting, with initial values set in the range of 1-100. The Adaboost module concatenates multiple weak classifiers and assigns weights to each weak classifier. Initialize to 0.5; The LSSVM module and the Adaboost module are cascaded, and the output features of the LSSVM module are used as the input features of the Adaboost module. A feedback mechanism is introduced to input the previous SOC estimate and the current voltage / current data into the fusion model.
4. The solid-state battery SOH estimation method with a wide-temperature adaptive enhancement algorithm according to claim 1, the strategy for improving the CFO algorithm specifically includes two stages: In the population initialization phase, the optimal Latin hypercube concept is used to replace the original random initialization; In the larval parasitism process, an inverse proportional weighting factor is added, and the original formula is as follows: in, X pl CF,i Indicates the first i The parasitic stage of Cordyceps sinensis; X so CF,i This is its position in the exploration phase; r 3 is a random scalar obtained from a standard normal distribution. Because r The value of 3 is likely close to zero, therefore the displacement is small in most cases; however, in a few cases... r 3 will take a larger value, thus producing a significant displacement, pushing Cordyceps sinensis into a new exploration area, and helping to break out of the local optimum. The formula after adding the inverse proportional weighting factor is as follows: w It is an inverse proportional weighting factor, with a large value in the early stages and a gradually decreasing value in the later stages, which can achieve a large search range in the early stages and improve the local search capability in the later stages.
5. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 4, characterized in that, Dynamic population adjustment mechanisms include the following: Define a population entropy value H to assess diversity; ； in, For the first The frequency of a combination of parameters.
6. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 2, characterized in that, The confidence interval for the SOH estimation result is calculated as follows: Based on the model training error distribution, a Gaussian confidence model is constructed. Set the confidence interval range as follows: ； in, Output values for the LSSVM-Adaboost fusion model; When the confidence interval exceeds this range, the model is triggered to retrain until the confidence interval width falls within this range.
7. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 1, characterized in that, In step S300, the LSSVM-Adaboost fusion model transmits the ambient temperature data collected by the embedded sensor in real time via the I2C protocol.
8. The solid-state battery SOH estimation method based on a wide-temperature adaptive boosting algorithm according to claim 1, characterized in that, The method also includes a self-evolution mechanism for the LSSVM-Adaboost fusion model: Set an upper limit on the number of charge-discharge cycles for the solid-state battery. Once the upper limit is reached, add the new hyperparameters to the training set and re-execute the CFO optimization process to obtain a new parameter combination. At the same time, the new parameter combination needs to inherit 1 / 5 of the old parameter combination from the previous version, and accelerate the convergence of the parameter combination while maintaining its stability.
9. A solid-state battery SOH estimation device using a wide-temperature adaptive boosting algorithm, comprising the solid-state battery SOH estimation method using a wide-temperature adaptive boosting algorithm as described in any one of claims 1-8, characterized in that, The device includes the following modules; The model building module is responsible for the initialization and cascading of the LSSVM-Adaboost fusion model; Parameter optimization module: Integrates and improves the CFO algorithm to perform global optimization of hyperparameters; Results output module: Generates SOH estimates, confidence intervals, and health assessment reports; Self-evolution module: periodically triggers model retraining and parameter inheritance.