State estimation method, device and equipment for power battery of sweeper

Abstract

The invention relates to a method, a device and equipment for estimating the state of a power battery of a sweeper. The method comprises the following steps: performing impedance characteristic acquisition on a power battery of the sweeper under various working conditions, and performing random low-frequency modulation multi-sine excitation signal measurement to obtain battery working condition characteristic data; performing fitting operation through the equivalent circuit model to obtain a battery health factor model; performing charging process feature extraction analysis to obtain a battery charging curve reconstruction model; carrying out extended Kalman filtering operation to obtain a battery state-of-charge value; performing segment processing and Gaussian process regression calculation on voltage and current data in the charging process to obtain a battery health state value; and updating the polynomial mapping coefficient of the battery health factor model and the weight parameter of the battery charging curve reconstruction model to obtain a target health factor model and a target charging curve reconstruction model. According to the invention, the adaptability and accuracy of the state estimation method in the whole life cycle of the battery are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and in particular to a method, device and equipment for estimating the state of a power battery of a sweeper. Background Art

[0002] Traditional power battery status estimation methods are mainly designed for fixed working conditions such as electric vehicles, and are difficult to meet the special working conditions of sweepers, such as frequent start-stop and severe load fluctuations. In the actual operation of sweepers, due to the complex working environment and changing operating modes, the battery's charging and discharging characteristics show nonlinear and time-varying characteristics, which brings great challenges to the accurate estimation of battery status.

[0003] In actual applications, there are problems with the charging process of sweeper power batteries, such as incomplete charging and difficulty in measuring impedance characteristics. On the one hand, due to the randomness and continuity of the task, sweepers often cannot obtain complete charging process data, which makes it difficult to apply the traditional health status assessment method based on the standard charging curve; on the other hand, the complex working environment causes the internal impedance characteristics of the battery to show obvious nonlinearity and time-varying, and traditional impedance measurement methods are difficult to obtain accurate impedance characteristics. Summary of the invention

[0004] The main purpose of the present invention is to provide a method, device and equipment for estimating the state of a power battery of a sweeper. The present invention ensures the adaptability and accuracy of the state estimation method throughout the entire life cycle of the battery.

[0005] To achieve the above object, the present invention provides a method for estimating the state of a power battery of a sweeper, comprising the following steps: The impedance characteristics of the power battery of the sweeper are collected under various working conditions, and the battery working condition characteristic data is obtained by measuring the random low-frequency modulated multi-sine excitation signal; According to the battery operating condition characteristic data, a fitting operation is performed through an equivalent circuit model, and a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency is established to obtain a battery health factor model; Based on the charging data in the battery operating condition characteristic data, a charging process characteristic extraction analysis is performed to obtain a battery charging curve reconstruction model; Performing an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value; Based on the battery charging curve reconstruction model, the voltage and current data during the charging process are processed in sections and Gaussian process regression is performed to obtain a battery health status value; According to the battery state of charge value and the battery health state value, the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model are updated to obtain a target health factor model and a target charging curve reconstruction model.

[0006] The present invention also provides a device for estimating the state of a power battery of a sweeper, comprising: The acquisition module is used to collect impedance characteristics of the power battery of the sweeper under various working conditions, and obtain battery working condition characteristic data by measuring the random low-frequency modulated multi-sine excitation signal; Establishing a module for performing a fitting operation through an equivalent circuit model according to the battery operating condition characteristic data, and establishing a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency to obtain a battery health factor model; An analysis module, used to extract and analyze charging process characteristics based on the charging data in the battery operating condition characteristic data, and obtain a battery charging curve reconstruction model; A calculation module, used for performing an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value; A processing module, used to reconstruct a model based on the battery charging curve, perform segmented processing and Gaussian process regression calculation on the voltage and current data during the charging process, and obtain a battery health status value; An updating module is used to update the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model according to the battery state of charge value and the battery health state value to obtain a target health factor model and a target charging curve reconstruction model.

[0007] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0008] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0009] In summary, the technical solution provided by the present invention effectively solves the problem of accuracy of battery impedance measurement under sweeper working conditions by adopting random low-frequency modulated multi-sinusoidal excitation signals for impedance feature acquisition, and the impedance feature data obtained is more reliable; a health factor model is constructed based on the equivalent circuit model and the polynomial mapping relationship, which accurately describes the attenuation characteristics of the battery capacity and Coulomb efficiency, and the battery state evaluation is more comprehensive; the extreme learning machine combined with the genetic algorithm and hill climbing optimization realizes the reconstruction of the charging curve, effectively solves the problem of incomplete charging data, and improves the robustness of the state estimation; the extended Kalman filter algorithm is used to perform online estimation of the battery state of charge, which significantly reduces the estimation error and ensures the reliability of the results; the accurate evaluation of the battery health state is realized through Gaussian process regression, and the probability correction is performed in combination with Bayesian theory to improve the accuracy of the evaluation results; a dynamic update mechanism for model parameters is established to ensure the adaptability and accuracy of the state estimation method throughout the battery life cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 1 is a schematic diagram of the steps of a method for estimating the state of a power battery of a sweeping vehicle in one embodiment of the present invention; Figure 2 This is a structural block diagram of a device for estimating the state of a power battery of a sweeping vehicle in one embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0011] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0013] Reference Figure 1 This embodiment provides a method for estimating the state of a power battery of a sweeper, comprising the following steps: S1, collect impedance characteristics of the power battery of the sweeper under various working conditions, and obtain battery working condition characteristic data by measuring the multi-sine excitation signal with random low-frequency modulation; Among them, the voltage and current data of the power battery of the sweeper are sampled when it is running under low-speed cleaning conditions, medium-speed cleaning conditions, high-speed cleaning conditions, no-load driving conditions, loaded driving conditions and comprehensive working conditions to obtain the original working condition operation data. The start-stop frequency in the original working condition operation data is statistically calculated to analyze the load change characteristics of the battery under various working conditions. Through this analysis process, the working condition characteristic parameters are obtained, such as the load response characteristics and charge and discharge cycles of the battery under different working conditions. According to the working condition characteristic parameters, the excitation amplitude of the random low-frequency modulated multi-sine excitation signal is set, and the frequency range of the excitation signal is set to 0.01Hz to 1000Hz to obtain the target excitation signal. This frequency range covers the dynamic response frequency band of the battery under different working conditions, and can fully excite the internal resistance characteristics of the battery. The target excitation signal is applied to the power battery. Through the action of the excitation signal, the internal voltage and current of the battery are induced to respond, and the voltage response and current response data of the battery under various working conditions are collected. Through these response data, impedance measurement data is obtained, which reflects the electrochemical behavior of the battery under different frequency excitations, including important parameters such as the battery's resistance and polarization characteristics. Using these impedance measurement data, the different internal resistance components of the battery are separated and calculated, including the ohmic internal resistance component, the electrochemical polarization internal resistance component, and the concentration polarization internal resistance component. Each type of internal resistance reflects the physical and electrochemical characteristics of the battery under different working conditions. According to the impedance characteristic parameters, the internal resistance change trend of the battery under different working conditions is analyzed. The change in internal resistance is closely related to the battery's aging process, capacity decline, and changes in the chemical reactions inside the battery. By analyzing the change trend of the internal resistance, the impedance change characteristic data is obtained to help judge the battery's health status and future performance decline. The original operating condition data, operating condition characteristic parameters, impedance measurement data, and impedance change characteristic data are fused and processed to obtain the battery operating condition characteristic data.

[0014] S2, according to the battery operating condition characteristic data, the equivalent circuit model is used for fitting operation, and a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency is established to obtain a battery health factor model; Specifically, according to the battery operating condition characteristic data, the voltage response and current response of the battery equivalent circuit model are Laplace transformed to obtain the s-domain transfer function parameter set of the battery. Laplace transform can convert the complex response of the battery in the time domain to the frequency domain, so that the voltage and current responses of the battery can be expressed in a more concise and systematic way, and the dynamic characteristics of the battery under various working conditions can be extracted. Based on the s-domain transfer function parameter set, nonlinear least squares fitting is performed to obtain various key parameters in the battery equivalent circuit model, such as ohmic internal resistance parameters, electrochemical polarization RC network parameters, and concentration polarization RC network parameters. The equivalent circuit model of the battery consists of resistors, capacitors, and inductors, which describe the electrical characteristics and electrochemical reaction processes of the battery. At this stage, the goal of the fitting process is to determine the internal resistance parameter matrix of the battery, that is, the impedance characteristics of the battery at different frequencies and different working conditions. Through nonlinear least squares fitting, the model error is effectively reduced, making the obtained battery internal resistance parameters more accurate. The relationship between each parameter in the battery internal resistance parameter matrix and the number of battery cycles is fitted by least squares. The internal resistance of the battery will decay as the number of cycles increases. By fitting the relationship between the battery internal resistance parameters and the number of cycles, a set of internal resistance decay characteristic curves is obtained, which reflects the performance degradation process of the battery under different working conditions as the battery is used over time. By analyzing these internal resistance decay characteristic curves, the aging mechanism and performance decay law of the battery can be understood. Based on the set of internal resistance decay characteristic curves, a capacity decay polynomial model is constructed, and a set of capacity decay coefficients is calculated. Capacity decay is an important indicator of battery aging, which is closely related to the change of internal resistance. Through polynomial modeling, the decay law of battery capacity with time and number of uses is described, and the mathematical expression of capacity decay is obtained. At the same time, based on the set of internal resistance decay characteristic curves, a polynomial model of coulomb efficiency decay is constructed, and a set of coulomb efficiency coefficients is calculated. Coulomb efficiency is a key parameter in the battery charging and discharging process, and its decay is also a manifestation of the battery aging process. By establishing a coulomb efficiency decay model, the efficiency loss of the battery in long-term use can be accurately predicted. The capacity decay mapping function and the coulomb efficiency mapping function are weighted and combined to construct the initial model of the battery health factor. In order to improve the accuracy of the battery health factor model, the model is optimized. The weight coefficients of the initial model of the battery health factor are cross-validated by five folds. Cross-validation is a model validation method that effectively avoids overfitting and improves the generalization ability of the model. The validation error threshold ε is set to 0.01, and the weight coefficients of the model are adjusted by iterative optimization by minimizing the root mean square error (RMSE) to make the model's prediction results closer to the actual battery health status. The iterative optimization process reduces the model error by adjusting the weight coefficients to ensure that the model can provide accurate health factor predictions under a variety of operating conditions. After iterative optimization and weight adjustment, the optimal set of weight coefficients is obtained, and the parameters of the initial model of the battery health factor are updated based on the optimal coefficients to obtain the battery health factor model.

[0015] S3, based on the charging data in the battery operating condition characteristic data, performing charging process feature extraction and analysis to obtain a battery charging curve reconstruction model; It should be noted that the charging data is processed in segments. There are significant different stages of voltage and current changes, such as the charging start segment, constant current charging segment, constant voltage charging segment, and charging termination segment. By analyzing the characteristics of the voltage rise rate and current change rate in the charging data, the charging process is divided into different stages to obtain a segmented charging feature data matrix. The data of each stage represents different characteristics of the charging process. Based on the segmented charging feature data matrix, a genetic algorithm is used to extract features of the charging data of each stage. The genetic algorithm is an optimization method that can find the global optimal solution in a complex search space. The goal of the genetic algorithm is to extract the most representative voltage feature points and current feature points from the voltage and current data of each charging segment, which can fully reflect the key changes in the charging process. Through the genetic algorithm, the selection of feature points is optimized to ensure that the selected feature points can effectively represent the overall characteristics of the charging process and form a feature point set. The feature point set is constructed with a feature vector. The voltage feature vector and the current feature vector contain the voltage and current change information of each charging stage, respectively, representing the electrical characteristics of the battery during the charging process. The voltage feature vector and the current feature vector are combined into a training sample set to form an extreme learning machine input data matrix. Based on the constructed input data matrix, an extreme learning machine network structure with L hidden nodes is constructed. Extreme learning machine is a training method for neural networks, which is fast and efficient. In this way, the input data matrix is ​​mapped to an output space with nonlinear relationships to achieve accurate modeling of the charging curve. After the network structure is constructed, the input weights and bias parameters of the extreme learning machine are optimized using the hill climbing algorithm. The hill climbing algorithm is a local search algorithm that makes the network better fit the charging data by iteratively optimizing the weights and bias parameters, thereby improving the prediction ability of the model. Based on the initial network parameter set, the output weights of the extreme learning machine are analytically solved. According to the input and output relationship of the network, the charging curve mapping model is obtained. The model can predict the curve characteristics of the charging process based on the input voltage and current feature points. The error calculation is performed between the prediction results of the charging curve mapping model and the actual charging curve. By comparing the difference between the prediction results and the actual charging curve, the model evaluation index is obtained to measure the prediction accuracy of the model. Based on the model evaluation index, iterative optimization is performed to adjust the parameters of the network to obtain the optimal network parameter set. The optimization process improves the accuracy and generalization ability of the model under different charging conditions by minimizing the error. During the optimization process, the prediction ability of the model is continuously enhanced by continuously adjusting parameters such as weights and biases. When the error reaches a predetermined threshold, the model is considered to have reached the optimal state. According to the optimal network parameter set, the parameters of the charging curve mapping model are updated, and finally the battery charging curve reconstruction model is obtained.

[0016] S4, performing an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value; Specifically, the actual capacity value is obtained by performing parameter calculation based on the capacity decay mapping function in the battery health factor model. The capacity decay mapping function reflects the decay law of the battery capacity over time and the number of times used. By calculating the parameters of the mapping function, the actual capacity value of the battery is obtained. Similarly, the coulomb efficiency mapping function in the battery health factor model is used to calculate the parameters to obtain the coulomb efficiency value. Coulomb efficiency is related to the energy conversion efficiency during the battery charging and discharging process, and reflects the energy utilization efficiency of the battery during the cycle process. According to the actual capacity value and the coulomb efficiency value, the state space equation of the battery is constructed. The state space equation is used to describe the dynamic behavior of the system in the time domain, including the state variables and state transfer equations of the system. In the case of batteries, the state variables include the state of charge (SOC), internal resistance, capacity, etc., and the state transfer equation describes how these variables change over time. The state space equation of the battery is constructed by combining the effects of capacity decay and coulomb efficiency. At the same time, the nonlinear relationship between the battery terminal voltage and the state of charge is modeled. There is a complex nonlinear relationship between the terminal voltage and the state of charge of the battery, especially during the charging and discharging process, the voltage will show different curves as the state of charge changes. Through nonlinear modeling, the relationship between the battery terminal voltage and the state of charge is more accurately reflected, and the observation equation model is obtained. Based on the state equation model and the observation equation model, the extended Kalman filter operation is performed. In the extended Kalman filter, the state prediction value and the error covariance matrix are calculated. The state prediction value is the state of the next moment predicted based on the previous state information and the state equation of the system, while the error covariance matrix describes the uncertainty of the prediction value. In the prediction process, the prior state quantity is calculated based on the previous state quantity and the state transfer equation. The state quantity represents the optimal estimate of the system in the absence of new measurement information. The error covariance matrix provides quantitative information about the accuracy of the prediction value, reflecting the prediction error and uncertainty of the model. According to the prior state quantity and the observation equation model, combined with the actual battery measurement data, the filter gain coefficient is obtained by calculating the Kalman gain matrix. The Kalman gain matrix is ​​the core part of the Kalman filter, which determines the weighted ratio between the predicted value and the observed value. In the extended Kalman filter, the Kalman gain coefficient can effectively balance the difference between the predicted and actual measured values, thereby ensuring that the final filtering result can both reflect the dynamic changes of the system and effectively correct the measurement error. Based on the filter gain coefficient, the state prediction value is corrected to obtain the a posteriori state quantity. The a posteriori state quantity represents the final estimate of the battery state based on the existing prediction value and the actual observation data. By combining the a priori state quantity with the measurement data, the Kalman filter can accurately update and correct the battery's state of charge. After each filtering, the error covariance matrix is ​​updated according to the a posteriori state quantity. Through the iterative process of the extended Kalman filter, the battery's state of charge value is obtained.

[0017] S5, based on the battery charging curve reconstruction model, the voltage and current data during the charging process are processed in segments and Gaussian process regression is performed to obtain the battery health status value; Among them, the voltage curve and current curve of the battery are segmented according to the charging data output by the battery charging curve reconstruction model. The charging process is divided into multiple stages, such as the constant current stage, the constant voltage stage, etc., and the voltage and current change characteristics of each stage have significantly different laws. The voltage curve and current curve are segmented according to the different stages of the charging process to obtain more detailed stage data. The voltage and current data of the segmented data of the charging stage are integrated to calculate the charging time and charging capacity parameters of each stage. The charging capacity parameter reflects the energy stored in the battery in a certain stage, while the charging time indicates the charging time of the battery in this stage. Through these parameters, the energy accumulation of the battery during the charging process and the duration of the charging process are quantified. Based on the voltage sequence in the segmented data of the charging stage, the voltage difference between adjacent sampling points is calculated and divided by the sampling time interval to obtain the voltage change rate parameter. The voltage change rate can reflect the dynamic changes of the battery during the charging process. Too fast or too slow voltage changes mean that there is an abnormality in the battery, such as increased internal resistance or capacity attenuation. The charging capacity parameter, charging time parameter and voltage change rate parameter are combined to construct a training sample matrix as the input data of the Gaussian process regression model. The combination of these parameters forms a multidimensional feature space, which reflects the changing characteristics of the battery in various aspects during the charging process. By inputting these features into the Gaussian process regression model, a health status prediction model based on battery charging data is established. Gaussian process regression is a non-parametric regression method that effectively handles the uncertainty in battery health status prediction and can provide high-precision prediction results when dealing with complex nonlinear relationships. In the Gaussian process regression model, the parameters of the kernel function are optimized by the maximum likelihood estimation method. The kernel function is a key component of Gaussian process regression, which determines the measurement method of similarity between data. The kernel function parameters are optimized by the maximum likelihood estimation method, so that the model can better fit the voltage and current data during the battery charging process, and improve the model's prediction ability for the battery health status. After optimization, the optimal kernel function parameter set is used to calculate the prediction mean and prediction variance of the Gaussian process regression model. The predicted mean represents the model's estimated result of the battery's health status, while the predicted variance provides the uncertainty of the estimated value, reflecting the confidence level of the model under given input data. Perform error analysis on the predicted results. Through the analysis of the errors, the prediction accuracy of the model is understood and a basis for correction is provided. On this basis, the Bayesian theory is applied to make probability corrections to the predicted results. By combining prior knowledge and observed data, the Bayesian theory can effectively correct the predicted results and obtain the estimated health status. The corrected health status value is compared and analyzed with the preset health status threshold. This threshold is determined based on the battery's design standards, usage conditions, and historical data, and is used to determine whether the battery's current health status meets normal usage requirements.If the corrected health status value is lower than the preset threshold, it means that the battery has performance degradation, insufficient capacity or other faults; conversely, if the health status value is higher than the threshold, it means that the battery is in good working condition. In this way, the battery health status is monitored in real time to obtain the battery health status value.

[0018] S6, updating the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model according to the battery state of charge value and the battery health state value, to obtain a target health factor model and a target charging curve reconstruction model.

[0019] Specifically, according to the battery state of charge value and the battery health state value, the residual calculation of the capacity decay mapping function coefficient and the coulomb efficiency mapping function coefficient in the battery health factor model is performed. The difference between the capacity decay and coulomb efficiency predicted by the model and the actual measurement results is quantified to obtain a residual vector, which reflects the deviation between the model output and the actual measurement value. The coefficients of the capacity decay mapping function and the coulomb efficiency mapping function are recursively updated and calculated based on the residual vector. The update process uses the feedback mechanism of the residual to adjust the coefficients of the mapping function so that the model can more accurately fit the actual health status of the battery in the next iteration. The recursive update calculation method gradually adjusts the coefficients of the capacity decay and coulomb efficiency according to the size and direction of the residual vector, thereby optimizing the prediction ability of the health factor model and ensuring that it can adapt to the dynamic changes of the battery under different usage conditions. At the same time, for the extreme learning machine in the battery charging curve reconstruction model, a sensitivity analysis is performed to optimize the input weights and bias parameters. The extreme learning machine is an efficient feedforward neural network, and its training process depends on the optimization of input weights and bias parameters. Through sensitivity analysis, the influence of each parameter on the model output is evaluated, and the parameter sensitivity matrix is ​​obtained to reflect the sensitivity of the model output to different parameters. According to the sensitivity matrix, the input weights and bias parameters are optimized and adjusted so that the extreme learning machine can better adapt to the characteristic changes in the battery charging process, improve the prediction accuracy of the charging curve reconstruction model, and obtain the updated network parameters. According to the updated mapping coefficient set, the battery health factor model is reconstructed. By substituting the coefficients of the updated capacity attenuation and coulomb efficiency mapping function into the original health factor model, the battery health factor is recalculated to obtain the target health factor model. Through continuous iterative updates, the health factor model reflects the actual health status of the battery more and more accurately. Similarly, based on the updated network parameters, the parameters of the battery charging curve reconstruction model are updated to obtain the target charging curve reconstruction model.

[0020] In one example, the impedance characteristics of the power battery of a sweeper are collected under various working conditions, and the battery working condition characteristic data is obtained by measuring the random low-frequency modulated multi-sine excitation signal, including: The voltage and current data of the power battery of the sweeper are sampled when it is running under low-speed sweeping conditions, medium-speed sweeping conditions, high-speed sweeping conditions, no-load driving conditions, loaded driving conditions and comprehensive operating conditions to obtain the original operating data; Perform statistical calculations on the start-stop frequency in the original operating data, and analyze the load change characteristics to obtain the operating characteristic parameters; According to the characteristic parameters of the working condition, the excitation amplitude of the random low-frequency modulated multi-sine excitation signal is set, and the excitation frequency range is set to 0.01 Hz to 1000 Hz to obtain the target excitation signal; Apply the target excitation signal to the power battery, collect the voltage response and current response of the power battery under various working conditions, obtain impedance measurement data, and based on the impedance measurement data, separate and calculate the ohmic internal resistance component, the electrochemical polarization internal resistance component, and the concentration polarization internal resistance component to obtain impedance characteristic parameters; According to the impedance characteristic parameters, the internal resistance change trend of the power battery under various working conditions is analyzed to obtain the impedance change characteristic data, and the original working condition operation data, working condition characteristic parameters, impedance measurement data and impedance change characteristic data are fused to obtain the battery working condition characteristic data.

[0021] In this example, the power battery of the sweeper is sampled under different working conditions, including low-speed sweeping conditions, medium-speed sweeping conditions, high-speed sweeping conditions, no-load driving conditions, loaded driving conditions, and comprehensive working conditions. Under these different working conditions, the changes in battery voltage and current reflect the working state and performance of the battery. Collecting and analyzing these data provides a basis for subsequent battery state estimation. Under each working condition, the battery voltage is obtained through sampling. and current Data. These data are recorded in real time by the battery management system (BMS) or other collection equipment. The collected raw operating data includes the changes of the battery voltage and current over time under different operating conditions, expressed as and ,in Indicates time. In order to analyze the performance of the battery under different working conditions, these data are analyzed. Calculate the start-stop frequency. The start-stop frequency refers to the number of charge and discharge cycles of the battery under the working condition. In the battery management system, the frequency is counted by detecting changes in the battery charging and discharging process. For example, under low-speed cleaning conditions, the battery starts and stops frequently, resulting in a high start-stop frequency, while under high-speed cleaning conditions, the battery start-stop frequency is relatively low. By counting the start-stop frequency ( ), obtain the frequency of battery load changes under different working conditions. This feature can reflect the load characteristics of the battery under working conditions, which in turn affects the assessment of battery health. Analyze the load change characteristics. The load change characteristics refer to the load changes experienced by the battery during the working condition. By calculating the battery current ( ) reflects the change of load, and the load change rate ( ) is expressed by the following formula: ; in, and Respectively represent the current value when the load starts and ends, and are the time points when the load starts and ends. Through this analysis, the load changes under the working conditions are obtained. According to the characteristic parameters of the working conditions, the excitation amplitude of the random low-frequency modulated multi-sine excitation signal is set, and the excitation frequency range is set to 0.01Hz to 1000Hz to generate the target excitation signal. This signal can simulate the electrochemical reaction characteristics of the battery under different loads and working conditions, and measure the impedance response of the battery through multi-frequency excitation signals. The selection of the excitation signal is very critical. Excitation signals of different frequencies will cause different electrochemical reactions inside the battery, thereby reflecting the different impedance characteristics of the battery. Target excitation signal It is expressed as: ; in, Indicates The amplitude of the excitation signal, It is frequency, is the phase of the signal, is the total number of excitation signals. By setting excitation signals of different frequency ranges, the impedance characteristics of the battery are measured. After the target excitation signal is applied to the battery, the voltage response and current response data of the battery under various working conditions are collected. Through these data, the impedance measurement data of the battery is obtained. The impedance of the battery ( ) is calculated by the ratio of voltage to current: ; in, and The impedance of the battery is divided into multiple components, including the ohmic internal resistance ( )、Electrochemical polarization internal resistance( ) and concentration polarization internal resistance ( ). These components represent the impedance characteristics of different electrochemical reactions of the battery. The ohmic internal resistance is related to the conductivity of the battery, the electrochemical polarization internal resistance is related to the charge and discharge reaction rate of the battery, and the concentration polarization internal resistance is related to the chemical concentration gradient of the battery. In order to separate and calculate these internal resistance components, the impedance spectrum analysis method is used. The impedance model of the battery is described by the electrochemical equivalent circuit model. A common equivalent circuit model is as follows: ; in, is an imaginary unit, and are the time constants of electrochemical polarization and concentration polarization, respectively. By fitting the impedance data, these internal resistance components are separated. By analyzing the impedance characteristic parameters, the internal resistance change trend of the battery under different working conditions is obtained, which reflects the health status of the battery, because the increase in internal resistance indicates that the battery performance is declining. The internal resistance change characteristic data ( ) The degree of battery aging is evaluated through time series analysis and statistical methods. As the battery is used for a longer time, the internal resistance will gradually increase, resulting in a decrease in battery capacity and coulomb efficiency. The original operating data, operating characteristic parameters, impedance measurement data, and impedance change characteristic data are fused to obtain the battery's operating characteristic data. Through machine learning algorithms, principal component analysis, or other data fusion methods, these data are weighted to obtain comprehensive battery operating characteristic data.

[0022] In one example, according to the battery operating condition characteristic data, a fitting operation is performed through an equivalent circuit model, and a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency is established to obtain a battery health factor model, including: According to the battery operating condition characteristic data, Laplace transform is performed on the voltage response and current response of the battery equivalent circuit model to obtain the s-domain transfer function parameter set; Based on the s-domain transfer function parameter set, the battery internal resistance parameter matrix is ​​obtained by fitting the ohmic internal resistance parameters, electrochemical polarization RC network parameters, and concentration polarization RC network parameters of the equivalent circuit model through nonlinear least squares fitting. Perform least square fitting on the relationship between each parameter in the battery internal resistance parameter matrix and the number of battery cycles to obtain a set of internal resistance attenuation characteristic curves; According to the set of internal resistance attenuation characteristic curves, a capacity attenuation polynomial model is constructed, and a set of capacity attenuation coefficients is calculated. At the same time, a capacity attenuation mapping function is established; Based on the internal resistance attenuation characteristic curve set, a Coulomb efficiency attenuation polynomial model is constructed, and the Coulomb efficiency coefficient set is calculated. At the same time, a Coulomb efficiency mapping function is established; The capacity decay mapping function and the Coulomb efficiency mapping function are weightedly combined to construct an initial model of the battery health factor; The weight coefficients of the initial model of the battery health factor were cross-validated by five-fold, and the validation error threshold ε=0.01 was set. The optimal weight coefficient set was obtained by iterative optimization by minimizing the root mean square error RMSE. According to the optimal weight coefficient set, the parameters of the initial battery health factor model are updated to construct a battery health factor model.

[0023] In this example, based on the battery operating characteristic data, the voltage response and current response of the battery equivalent circuit model are Laplace transformed to obtain the s-domain transfer function parameter set. The battery equivalent circuit model consists of the ohmic internal resistance ( )、Electrochemical polarization internal resistance( ) and concentration polarization internal resistance ( ). In order to obtain the response characteristics of the battery at different frequencies, the Laplace transform is used to convert the voltage and current responses from the time domain to the s domain (complex frequency domain). The voltage and current signals of the battery in the time domain are converted into the transfer function form of the Laplace domain. For the battery equivalent circuit model, the transfer function is expressed as: ; in, and are the Laplace transform results of the battery voltage and current, is a complex frequency domain variable, and is the time constant of electrochemical polarization and concentration polarization. Through these transfer functions, the frequency domain response characteristics of the battery are obtained. Based on the s-domain transfer function parameter set, the internal resistance parameters of the battery are estimated by the nonlinear least squares fitting method, including ohmic internal resistance, polarization internal resistance and concentration polarization internal resistance. The goal of nonlinear least squares fitting is to minimize the error between the model output (i.e., the predicted value of the voltage and current response of the battery) and the experimental data. The voltage and current responses obtained by the experiment are set to and , then the least squares fitting objective is expressed as: ; in, is the number of data points, For the By minimizing the above error, we can obtain the parameter set of the battery equivalent circuit model, namely the ohmic internal resistance parameter , electrochemical polarization internal resistance parameters , concentration polarization internal resistance parameters and their time constants and These parameters constitute the internal resistance parameter matrix of the battery, which is expressed as: ; According to the relationship between each parameter in the internal resistance parameter matrix and the number of battery cycles, the attenuation characteristic curve of the internal resistance is fitted by the least square fitting method. Assuming that the number of battery cycles is , the internal resistance parameter will change after each charge and discharge process, and the internal resistance decay is expressed as a function related to the number of cycles. By fitting the relationship between the internal resistance parameter and the number of battery cycles, the internal resistance decay characteristic curve is obtained. For example, the relationship of capacity decay is approximately: ; in, is the initial ohmic internal resistance, is the attenuation coefficient, is the number of cycles. Similarly, attenuation models are established for electrochemical polarization internal resistance and concentration polarization internal resistance respectively. Based on the set of internal resistance attenuation characteristic curves, polynomial models of capacity attenuation and coulomb efficiency attenuation are constructed. Assume that the polynomial model of capacity attenuation is: ; in, Indicates The capacity of the battery at the time of the cycle, are the polynomial coefficients to be determined. These coefficients are calculated through least squares fitting, and the capacity decay mapping function is established. Similarly, the decay model of coulomb efficiency is expressed as: ; in, For the The coulombic efficiency of the battery at the time of the cycle, is the polynomial coefficient to be determined. Through similar steps, the mapping function of Coulomb efficiency decay is established. The capacity decay mapping function and the Coulomb efficiency mapping function are weightedly combined to obtain the initial model of battery health factor. The battery health factor is calculated by the weighted combination of capacity decay and Coulomb efficiency, and the formula is: ; in, and is a weighted coefficient, which indicates the contribution of capacity attenuation and Coulomb efficiency to the battery health factor. The weight coefficient of the initial model of the battery health factor is optimized through five-fold cross validation to evaluate the generalization ability of the model and avoid overfitting. The validation error threshold is set to , iterative optimization is performed by minimizing the root mean square error, and the formula is as follows: ; in, is the health factor predicted by the model, is the actual measured health factor, is the number of data points in the validation set. By optimizing RMSE, the optimal weight coefficient set is obtained. According to the optimal weight coefficient set, the parameters of the initial battery health factor model are updated, and finally an accurate battery health factor model is constructed.

[0024] In one example, based on the charging data in the battery operating condition characteristic data, a charging process feature extraction analysis is performed to obtain a battery charging curve reconstruction model, including: The charging data in the battery operating condition characteristic data is processed in sections, and the charging start section, constant current charging section, constant voltage charging section, and charging termination section are divided according to the voltage rise rate and the current change rate to obtain a sectioned charging characteristic data matrix; Based on the segmented charging characteristic data matrix, the voltage characteristic points and current characteristic points of each segment of charging data are extracted by genetic algorithm to obtain a characteristic point set; Construct feature vectors for the feature point set, combine the voltage feature vector and the current feature vector into a training sample set, and obtain the extreme learning machine input data matrix; According to the extreme learning machine input data matrix, an extreme learning machine network structure with L hidden nodes is constructed, and the input weights and bias parameters are optimized by the hill climbing algorithm to obtain the initial network parameter set; Based on the initial network parameter set, the output weight of the extreme learning machine is analytically solved to obtain the charging curve mapping model, and the error between the prediction result of the charging curve mapping model and the actual charging curve is calculated to obtain the model evaluation index; The model evaluation index is iteratively optimized to obtain an optimal network parameter set, and the parameters of the charging curve mapping model are updated according to the optimal network parameter set to obtain a battery charging curve reconstruction model.

[0025] In this example, the charging data in the battery operating condition characteristic data is processed in segments. The charging data includes the changes in the battery voltage and current over time during the charging process. The charging process is divided into the charging start segment, constant current charging segment, constant voltage charging segment, and charging termination segment. The judgment is made based on the voltage rise rate and current change rate. The voltage rise rate and current change rate are the key features of the charging process and are calculated using the following formula: ; ; in, and The time points The battery voltage and current at is the time interval. According to these change rates, the division of each charging stage is determined. For example, when the voltage rise rate is high, it is in the charging start stage or constant current charging stage; when the voltage tends to be constant and the current change is small, it enters the constant voltage charging stage; when the current drops sharply, it indicates that it enters the charging termination stage. After division, a charging feature data matrix is ​​obtained, each row represents the data of a charging stage, which contains information such as voltage and current. Based on the segmented charging feature data matrix, the voltage feature points and current feature points of each charging stage are extracted by genetic algorithm. Genetic algorithm is a heuristic search algorithm that simulates the process of natural selection and gradually optimizes parameters through operations such as selection, crossover, and mutation. Genetic algorithm is used to automatically select key feature points in the voltage and current curves to represent the feature information of each charging stage. The goal of feature point extraction is to identify the key moments of voltage and current changes during the charging process, such as the voltage changes from rapid rise to stability and the current drops from high value to low value. Set an appropriate objective function, such as maximizing the discrimination of feature points or minimizing errors, and use genetic algorithm to extract the most representative feature point set. Based on the extracted voltage and current feature point sets, feature vectors are constructed. The voltage feature points and current feature points of each charging stage are combined into a feature vector and form a complete training sample set. Assume that the voltage feature point is , the current characteristic point is , then the characteristic vector of each charging stage is expressed as: ; These feature vectors constitute the input data matrix of the extreme learning machine model. Assume that the input data matrix is , each column corresponds to a set of feature points, and each row is a training sample. Use these data to build an extreme learning machine model. The extreme learning machine is a feedforward neural network with a single hidden layer, and the number of nodes in the hidden layer is specified by the user. Set the extreme learning machine network structure and assume that the number of hidden layer nodes is , then the weight matrix between the input data and the hidden layer and the bias matrix Need to be optimized. The output of the extreme learning machine is: ; in, is the activation function (usually sigmoid or ReLU), is the input matrix, is the weight from the input layer to the hidden layer, is the bias. By calculating the output of the hidden layer , and get the final output weight matrix , the matrix and the output data The relationship between them is: ; The input weights and bias parameters are optimized using a hill climbing algorithm. The hill climbing algorithm is a greedy optimization algorithm that searches for better solutions through local search. Based on the initial weights and biases, the hill climbing algorithm minimizes the error between the predicted results and the actual charging data by adjusting the weights and biases. The optimization goal is to minimize the error of the model, such as the mean square error (MSE): ; in, is the charging curve output predicted by the model, is the actual charging curve data, is the number of samples. By minimizing the MSE, we get the initial network parameter set, that is, the optimized input weights and biases. Based on the initial network parameter set, we analytically solve the output weights of ELM and get the weight matrix of the output layer. After obtaining the complete ELM network model, the mapping function of the battery charging curve is obtained. The prediction result of the charging curve is compared with the actual charging curve to calculate the error and obtain the model evaluation index, which includes the root mean square error (RMSE) and the determination coefficient ( )wait: ; ; in, is the mean of the actual charging data. The accuracy of the model is determined by analyzing the model evaluation indicators, and it is optimized iteratively as needed. Based on the optimal set of network parameters, the parameters of the charging curve mapping model are updated to obtain the final reconstruction model of the battery charging curve. The model can predict the changes in the charging process by inputting the charging data of the battery.

[0026] In one example, an extended Kalman filter operation is performed on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value, including: Perform parameter calculation on the capacity decay mapping function in the battery health factor model to obtain the actual capacity value, and perform parameter calculation based on the coulomb efficiency mapping function in the battery health factor model to obtain the coulomb efficiency value; According to the actual capacity value and the coulomb efficiency value, the battery state space equation is constructed to obtain the state equation model, and the nonlinear relationship between the battery terminal voltage and the state of charge is modeled, and the observation equation is constructed to obtain the observation equation model; Based on the state equation model and the observation equation model, the state prediction value and the error covariance matrix are calculated to obtain the prior state quantity, and the Kalman gain matrix is ​​calculated based on the prior state quantity and the observation equation to obtain the filter gain coefficient; Based on the filter gain coefficient, the state prediction value is corrected and calculated to obtain the a posteriori state quantity, and according to the a posteriori state quantity, the battery state of charge value is obtained by updating the error covariance matrix.

[0027] In this example, the capacity decay mapping function and coulomb efficiency mapping function in the battery health factor model are the basis of battery health management, which can reflect the difference between the actual performance of the battery and the ideal state. The capacity decay mapping function is obtained by fitting historical data, which represents the capacity decay of the battery under different usage conditions. Assume that the capacity decay mapping function is: ; in, Indicates the battery time The capacity of is the initial capacity, is the capacity decay rate. By performing regression analysis on the battery usage data, the decay rate is obtained. According to the decay rate , calculate the actual capacity of the battery at any time. The Coulomb efficiency mapping function related to capacity decay is used to describe the efficiency change of the battery during the charging and discharging process. Assume that the Coulomb efficiency mapping function is: ; in, For battery in time The Coulomb efficiency at is the initial Coulomb efficiency, is the Coulomb efficiency decay rate. Similar to capacity decay, the parameter is obtained through experimental data. and , and then calculate the Coulombic efficiency. According to the actual capacity value And the Coulomb efficiency value , construct the state space equation of the battery. The state space equation is a mathematical model that describes the dynamic changes of the system and is used to predict the system state. In battery management, the system state includes the battery state of charge (SOC), battery voltage ( ) and current ( ) etc. The SOC of a battery is a key parameter that describes the remaining capacity of a battery. There is a complex nonlinear relationship between the SOC of a battery and the voltage and current of the battery, which is modeled by the state equation and the observation equation. The battery state space equation is expressed as: ; in, is a state vector, including the battery state of charge (SOC) and other possible state quantities, such as battery voltage ( )、Current( ); is the control input (e.g., charge and discharge current); and The state transfer matrix and control matrix describe the dynamic behavior of the system respectively; is the process noise, which is assumed to be zero-mean Gaussian noise. The nonlinear relationship between battery voltage and SOC is modeled by the observation equation. The observation equation is expressed as: ; in, is the observed quantity, usually the battery terminal voltage; It is a nonlinear function, which represents the relationship between battery voltage and SOC; is the observation noise, which is assumed to be zero-mean Gaussian noise. Based on the state space model and the observation equation, the Kalman filter can estimate the battery SOC through two steps: prediction and update. The Kalman filter calculates the state prediction value and the error covariance matrix to obtain the prior state quantity. Assume that the prior state quantity is , and its calculation formula is: ; in, is the estimated value at the previous moment, is the next moment state of the prediction. The error covariance matrix represents the uncertainty of the prediction value, which is expressed as: ; in, is the error covariance matrix at the previous moment, is the process noise covariance matrix, which represents the uncertainty of the system. The Kalman gain matrix is ​​calculated in combination with the observation equation to combine the prior state and the observation value in an optimal way. The Kalman gain matrix calculation formula is: ; in, is the observation matrix, which represents the relationship between states and observations, is the observation noise covariance matrix, which represents the uncertainty of the observation. Kalman gain It is used to perform weighted averaging of the predicted state and the actual observation to obtain the optimal state estimate. Based on the Kalman gain, the state prediction value is corrected and calculated to obtain the posterior state quantity. The update formula of the posterior state quantity is: ; in, is the actual observed battery voltage, It is the battery voltage prediction value obtained based on the predicted state. Through this update step, the Kalman filter algorithm effectively corrects the state estimation and obtains the accuracy of the state estimation at subsequent moments through the updated error covariance matrix. According to the posterior state quantity, the battery state of charge value is obtained through the error covariance matrix update.

[0028] In one example, based on the battery charging curve reconstruction model, the voltage and current data during the charging process are processed segmentally and Gaussian process regression is performed to obtain the battery health status value, including: According to the charging data output by the battery charging curve reconstruction model, the voltage curve and the current curve are segmented according to the charging stage to obtain the charging stage segmented data; Integrate the voltage and current data of the segmented data of the charging stage, and calculate the charging time of each stage to obtain the charging capacity parameter and charging time parameter; According to the voltage sequence in the segmented data of the charging stage, the voltage difference between adjacent sampling points is calculated and divided by the sampling time interval to obtain the voltage change rate parameter; The charging capacity parameter, charging time parameter and voltage change rate parameter are combined to construct a training sample matrix and obtain Gaussian process regression input data; The Gaussian process regression input data is input into the Gaussian process regression model, and the kernel function parameters are optimized by the maximum likelihood estimation method to obtain the optimal kernel function parameter set. Based on the optimal kernel function parameter set, the prediction mean and prediction variance of the Gaussian process regression model are calculated to obtain the prediction result of the battery health status; The prediction results are analyzed for errors, and the prediction results are probability corrected using Bayesian theory to obtain a corrected health status value. The corrected health status value is then compared and analyzed with the preset health status threshold to obtain the battery health status value.

[0029] In this example, the battery charging process is divided into several different stages, such as the charging start stage, constant current charging stage, constant voltage charging stage, and charging termination stage. According to the voltage and current curves in the charging data, these curves are segmented to obtain segmented data for different charging stages. The voltage rise rate and current change rate are used as criteria for determining the division of each stage. For example, if the voltage rise rate exceeds a certain set threshold, it is determined to be a constant current stage; when the voltage reaches the set constant voltage value, it enters the constant voltage charging stage. According to these judgment rules, the battery charging curve is divided into multiple charging stages. Integrate the voltage and current data of each stage. Calculate the charging time and charging capacity of each charging stage. The charging capacity is estimated by integrating the current over time. Assuming that in the first The charging current in this stage is , charging time is , then the charging capacity of this stage It is expressed as: ; in, and are the start and end time of the stage respectively. By integrating the current, the charging capacity of each stage is obtained, and thus the total capacity of the charging process is calculated. By calculating the difference between the voltages of adjacent sampling points and dividing it by the sampling time interval, the voltage change rate of each charging stage is calculated. If the voltage sequence is , then the voltage change rate It is expressed as: ; in, and The battery at the time and The voltage value, is the sampling time interval. By calculating the voltage change rate, the charging characteristics of the battery at different stages are evaluated. The charging capacity parameters, charging time parameters and voltage change rate parameters are combined to construct a training sample matrix. These parameters jointly describe the performance of the battery during the charging process and serve as input data for the Gaussian process regression model. Assume that the charging capacity, charging time and voltage change rate form a feature vector: ; in, Indicates The characteristic vectors of the charging stage. By combining the characteristic vectors of all stages into a training sample set, the input data matrix of Gaussian process regression is obtained. The Gaussian process regression model is a non-parametric regression method that can make predictions based on training data without explicitly assuming the function form. In Gaussian process regression, it is assumed that the output variable (battery health status) is generated by the input variable (charging process characteristics) through a Gaussian process and has a Gaussian distribution. The kernel function parameters are optimized by the maximum likelihood estimation method to find the optimal kernel function parameter set. Assume that the kernel function of Gaussian process regression is: ; in, is the signal amplitude, is the length scale, and are the two eigenvectors in the input data matrix. The kernel function parameters are optimized by the maximum likelihood estimation method. and , and obtain the optimal kernel function, thereby improving the prediction accuracy of the regression model. After obtaining the optimal kernel function parameters, the Gaussian process regression model is used for prediction, and the mean and variance of the prediction are calculated. The mean of the prediction result represents the estimated value of the battery health state, while the variance represents the uncertainty of the prediction. The mean of the prediction and variance Calculated by the following formula: ; ; in, is the output data of the training sample (battery health status), is the covariance matrix of the training samples, is the noise variance, is the covariance between the new sample and the training sample. Through these formulas, the predicted results of the battery health status and the corresponding uncertainty are obtained. The prediction results are analyzed for errors, and the prediction results are corrected by probability through Bayesian theory. The Bayesian method updates the prediction results based on prior knowledge and observed data to obtain the corrected health status value. Assume that the prior distribution is The prediction result is , then the posterior distribution is: ; in, is the likelihood function, indicating that given a health state When the observation data The probability of is a prior distribution. Through Bayesian updating, the corrected health status value is obtained. The health status of the battery is determined by comparing the corrected health status value with the preset health status threshold. If the corrected health status value is lower than a certain set threshold, it means that there is a problem with the battery and maintenance or replacement is required.

[0030] In one example, according to the battery state of charge value and the battery health state value, the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model are updated to obtain the target health factor model and the target charging curve reconstruction model, including: According to the battery state of charge value and the battery health state value, the capacity attenuation mapping function coefficient and the coulomb efficiency mapping function coefficient of the battery health factor model are residually calculated to obtain a residual vector; Based on the residual vector, recursively update and calculate the coefficients of the capacity attenuation mapping function and the Coulomb efficiency mapping function to obtain an updated mapping coefficient set; Perform sensitivity analysis on the input weights and bias parameters of the extreme learning machine in the battery charging curve reconstruction model to obtain a parameter sensitivity matrix, and optimize and adjust the input weights and bias parameters based on the parameter sensitivity matrix to obtain updated network parameters; According to the updated mapping coefficient set, the battery health factor model is reconstructed to obtain the target health factor model, and based on the updated network parameters, the battery charging curve reconstruction model is updated to obtain the target charging curve reconstruction model.

[0031] In this example, given the state of charge and health state of the battery, the residual vector is calculated based on the capacity decay mapping function and coulomb efficiency mapping function in the battery health factor model. The capacity decay mapping function and coulomb efficiency mapping function represent the change law of battery capacity decay and coulomb efficiency, respectively, which are represented by polynomials or other mathematical functions. Assume that the capacity decay mapping function is: ; in, Indicates the relationship between battery capacity and state of charge. is the coefficient of the mapping function, and SOC is the state of charge value. Similarly, the Coulomb efficiency mapping function is expressed as: ; in, It represents the relationship between battery Coulombic efficiency and state of charge. are the coefficients of the mapping function, and SOC is the state of charge value. In order to optimize these mapping functions, a residual vector is calculated, which represents the difference between the actual measured battery health state value and the health state value calculated based on the model. The residual vector It is expressed as: ; in, is the actual measured battery health status value, It is the battery health status value calculated based on the model. By calculating the residual vector, the model error is evaluated and the basis for subsequent parameter updates is provided. Based on the residual vector, the coefficients of the capacity decay mapping function and the coulomb efficiency mapping function are recursively updated. The coefficients are updated using gradient descent or other optimization algorithms. Assume that in the In the iterations, the coefficients of the mapping function are and , the coefficients are updated using the following formula: ; ; in, is the learning rate, and They are the gradients of the loss function to the coefficients of the capacity attenuation mapping function and the Coulomb efficiency mapping function. It is expressed as: ; in, is the sample size, It is The residual of each sample. By recursively updating these coefficients, the model gradually reduces the error in each iteration and improves the accuracy of the battery health factor model. At the same time, a sensitivity analysis is performed on the input weights and bias parameters of the extreme learning machine in the battery charging curve reconstruction model. The extreme learning machine is a machine learning model based on a single hidden layer neural network, which has the advantages of fast training and high accuracy. In the extreme learning machine, the input weights and bias parameters directly affect the performance of the model, and a sensitivity analysis is performed. Assume that the network structure of the extreme learning machine is: ; in, is the output of the model, is the output weight, is the activation function, is the output of the hidden layer node, For input data, is the bias parameter. By calculating the impact of each input weight and bias parameter on the model output, the parameter sensitivity matrix is ​​obtained, which indicates the sensitivity of each parameter to the model prediction result. The calculation is achieved through the following formula: ; in, Indicates The input weights Through sensitivity analysis, we can identify the parameters that have a greater impact on the model output, so as to optimize these parameters in a targeted manner. Based on the parameter sensitivity matrix, we can optimize the input weights and bias parameters. The optimization goal is to minimize the error between the model's predicted results and the actual results. We use the gradient descent method to update the weights and bias parameters as follows: ; ; in, and are the gradients of the loss function with respect to weights and biases, respectively. The battery health factor model is reconstructed according to the updated input weights and bias parameters to obtain the target health factor model. At the same time, based on the updated network parameters, the battery charging curve reconstruction model is updated to obtain the target charging curve reconstruction model. Through iterative optimization, the battery health factor model and charging curve reconstruction model are continuously improved so that they can provide accurate battery status estimation under different working environments.

[0032] Reference Figure 2 This embodiment provides a device for estimating the state of a power battery of a sweeping vehicle, comprising: The acquisition module 1 is used to collect impedance characteristics of the power battery of the sweeper under various working conditions, and obtain battery working condition characteristic data by measuring the multi-sine excitation signal with random low-frequency modulation; Establishing module 2, which is used to perform fitting calculations through an equivalent circuit model according to the battery operating condition characteristic data, and establish a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency to obtain a battery health factor model; Analysis module 3, used to extract and analyze the charging process characteristics based on the charging data in the battery operating condition characteristic data, and obtain a battery charging curve reconstruction model; Operation module 4 is used to perform extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain the battery state of charge value; Processing module 5, used to rebuild the model based on the battery charging curve, perform segmented processing and Gaussian process regression calculation on the voltage and current data during the charging process, and obtain the battery health status value; The updating module 6 is used to update the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model according to the battery state of charge value and the battery health state value, so as to obtain a target health factor model and a target charging curve reconstruction model.

[0033] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0034] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0035] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0036] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0037] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0038] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0039] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for estimating the state of a power battery of a sweeping vehicle, characterized in that: The following steps are involved: The impedance characteristics of the power battery of the sweeper are collected under various working conditions, and the battery working condition characteristic data is obtained by measuring the random low-frequency modulated multi-sine excitation signal; According to the battery operating condition characteristic data, a fitting operation is performed through an equivalent circuit model, and a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency is established to obtain a battery health factor model; Based on the charging data in the battery operating condition characteristic data, a charging process characteristic extraction analysis is performed to obtain a battery charging curve reconstruction model; Performing an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value; Based on the battery charging curve reconstruction model, the voltage and current data during the charging process are processed in sections and Gaussian process regression is performed to obtain a battery health status value; According to the battery state of charge value and the battery health state value, the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model are updated to obtain a target health factor model and a target charging curve reconstruction model.

2. The method for estimating the state of the power battery of a sweeping vehicle according to claim 1, characterized in that: The impedance characteristics of the power battery of the sweeper are collected under various working conditions, and the battery working condition characteristic data is obtained by randomly modulating a multi-sine excitation signal at a low frequency, including: The voltage and current data of the power battery of the sweeper are sampled when it is running under low-speed sweeping conditions, medium-speed sweeping conditions, high-speed sweeping conditions, no-load driving conditions, loaded driving conditions and comprehensive operating conditions to obtain the original operating data; Statistically calculating the start-stop frequency in the original operating condition data, and analyzing the load change characteristics to obtain operating condition characteristic parameters; According to the working condition characteristic parameters, the excitation amplitude of the random low-frequency modulated multi-sine excitation signal is set, and the excitation frequency range is set to 0.01 Hz to 1000 Hz to obtain a target excitation signal; Applying the target excitation signal to the power battery, collecting the voltage response and current response of the power battery under various operating conditions to obtain impedance measurement data, and based on the impedance measurement data, separately calculating the ohmic internal resistance component, the electrochemical polarization internal resistance component, and the concentration polarization internal resistance component to obtain impedance characteristic parameters; According to the impedance characteristic parameters, the internal resistance change trend of the power battery under various operating conditions is analyzed to obtain impedance change characteristic data, and the original operating condition operation data, the operating condition characteristic parameters, the impedance measurement data and the impedance change characteristic data are fused to obtain the battery operating condition characteristic data.

3. The method for estimating the state of the power battery of a sweeping vehicle according to claim 2, characterized in that: According to the battery operating condition characteristic data, a fitting operation is performed through an equivalent circuit model, and a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency is established to obtain a battery health factor model, including: According to the battery operating condition characteristic data, Laplace transform is performed on the voltage response and current response of the battery equivalent circuit model to obtain an s-domain transfer function parameter set; Based on the s-domain transfer function parameter set, the ohmic internal resistance parameters, electrochemical polarization RC network parameters, and concentration polarization RC network parameters of the equivalent circuit model are fitted by nonlinear least squares to obtain a battery internal resistance parameter matrix; Performing least square fitting on the relationship between each parameter in the battery internal resistance parameter matrix and the number of battery cycles to obtain a set of internal resistance attenuation characteristic curves; According to the internal resistance attenuation characteristic curve set, a capacity attenuation polynomial model is constructed, and a capacity attenuation coefficient set is calculated, and at the same time, a capacity attenuation mapping function is established; Based on the internal resistance attenuation characteristic curve set, a Coulomb efficiency attenuation polynomial model is constructed, and a Coulomb efficiency coefficient set is calculated and obtained, and at the same time, a Coulomb efficiency mapping function is established; Performing a weighted combination of the capacity decay mapping function and the coulomb efficiency mapping function to construct an initial model of a battery health factor; The weight coefficients of the initial model of the battery health factor are cross-validated by five folds, the validation error threshold ε=0.01 is set, and iterative optimization is performed by minimizing the root mean square error RMSE to obtain the optimal weight coefficient set; According to the optimal weight coefficient set, the parameters of the initial battery health factor model are updated to construct a battery health factor model.

4. The method for estimating the state of the power battery of a sweeper according to claim 3, characterized in that: The method of extracting and analyzing the charging process characteristics based on the charging data in the battery operating condition characteristic data to obtain a battery charging curve reconstruction model includes: The charging data in the battery operating condition characteristic data is processed in sections, and the charging start section, the constant current charging section, the constant voltage charging section, and the charging termination section are divided according to the voltage rise rate and the current change rate to obtain a sectioned charging characteristic data matrix; Based on the segmented charging characteristic data matrix, extracting the voltage characteristic points and the current characteristic points of each segment of charging data by a genetic algorithm to obtain a characteristic point set; Constructing a feature vector for the feature point set, combining the voltage feature vector and the current feature vector into a training sample set, and obtaining an extreme learning machine input data matrix; According to the extreme learning machine input data matrix, an extreme learning machine network structure with L hidden layer nodes is constructed, and the input weights and bias parameters are optimized by a hill climbing algorithm to obtain an initial network parameter set; Based on the initial network parameter set, the output weight of the extreme learning machine is analytically solved to obtain a charging curve mapping model, and the prediction result of the charging curve mapping model is compared with the actual charging curve to obtain a model evaluation index; The model evaluation index is iteratively optimized to obtain an optimal network parameter set, and the parameters of the charging curve mapping model are updated according to the optimal network parameter set to obtain a battery charging curve reconstruction model.

5. The method for estimating the state of the power battery of a sweeping vehicle according to claim 4, characterized in that: The performing of an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value includes: Performing parameter calculation on the capacity decay mapping function in the battery health factor model to obtain an actual capacity value, and performing parameter calculation based on the coulomb efficiency mapping function in the battery health factor model to obtain a coulomb efficiency value; According to the actual capacity value and the coulomb efficiency value, a battery state space equation is constructed to obtain a state equation model, and a nonlinear relationship between the battery terminal voltage and the state of charge is modeled to construct an observation equation to obtain an observation equation model; Based on the state equation model and the observation equation model, the state prediction value and the error covariance matrix are calculated to obtain a priori state quantity, and the Kalman gain matrix is ​​calculated according to the priori state quantity in combination with the observation equation to obtain a filter gain coefficient; Based on the filter gain coefficient, the state prediction value is corrected and calculated to obtain a posterior state quantity, and according to the posterior state quantity, the battery state of charge value is obtained by updating the error covariance matrix.

6. The method for estimating the state of the power battery of a sweeping vehicle according to claim 5, characterized in that: The battery charging curve reconstruction model is based on the battery charging curve, and the voltage and current data in the charging process are processed in sections and Gaussian process regression is performed to obtain the battery health status value, including: According to the charging data output by the battery charging curve reconstruction model, the voltage curve and the current curve are segmented according to the charging stage to obtain charging stage segmented data; Integrating the voltage and current data of the segmented data of the charging stage, and calculating the charging time of each stage to obtain a charging capacity parameter and a charging time parameter; According to the voltage sequence in the charging stage segmented data, the voltage difference between adjacent sampling points is calculated and divided by the sampling time interval to obtain a voltage change rate parameter; Performing feature combination on the charging capacity parameter, the charging time parameter and the voltage change rate parameter, constructing a training sample matrix, and obtaining Gaussian process regression input data; Inputting the Gaussian process regression input data into the Gaussian process regression model, optimizing the kernel function parameters by the maximum likelihood estimation method to obtain an optimal kernel function parameter set, and calculating the predicted mean and predicted variance of the Gaussian process regression model based on the optimal kernel function parameter set to obtain a predicted result of the battery health state; The prediction result is subjected to error analysis, and the prediction result is subjected to probability correction through Bayesian theory to obtain a corrected health status value, and the battery health status value is obtained by comparing and analyzing the corrected health status value with a preset health status threshold.

7. The method for estimating the state of the power battery of a sweeping vehicle according to claim 6, characterized in that: The updating of the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model according to the battery state of charge value and the battery health state value to obtain a target health factor model and a target charging curve reconstruction model includes: According to the battery state of charge value and the battery health state value, performing residual calculation on the capacity attenuation mapping function coefficient and the coulomb efficiency mapping function coefficient of the battery health factor model to obtain a residual vector; Based on the residual vector, recursively update and calculate the coefficients of the capacity attenuation mapping function and the coulomb efficiency mapping function to obtain an updated mapping coefficient set; Performing sensitivity analysis on the input weights and bias parameters of the extreme learning machine in the battery charging curve reconstruction model to obtain a parameter sensitivity matrix, and optimizing and adjusting the input weights and bias parameters based on the parameter sensitivity matrix to obtain updated network parameters; According to the updated mapping coefficient set, the battery health factor model is reconstructed to obtain a target health factor model, and based on the updated network parameters, the battery charging curve reconstruction model is parameter updated to obtain a target charging curve reconstruction model.

8. A device for estimating the state of a power battery of a sweeping vehicle, characterized in that: For implementing the steps of the method according to any one of claims 1 to 7, the device comprises: The acquisition module is used to collect impedance characteristics of the power battery of the sweeper under various working conditions, and obtain battery working condition characteristic data by measuring the random low-frequency modulated multi-sine excitation signal; Establishing a module for performing a fitting operation through an equivalent circuit model according to the battery operating condition characteristic data, and establishing a polynomial mapping relationship between the health factor and the capacity and coulomb efficiency to obtain a battery health factor model; An analysis module, used to extract and analyze charging process characteristics based on the charging data in the battery operating condition characteristic data, and obtain a battery charging curve reconstruction model; A calculation module, used for performing an extended Kalman filter operation on the polynomial mapping relationship in the battery health factor model to obtain a battery state of charge value; A processing module, used to reconstruct a model based on the battery charging curve, perform segmented processing and Gaussian process regression calculation on the voltage and current data during the charging process, and obtain a battery health status value; An updating module is used to update the polynomial mapping coefficients of the battery health factor model and the weight parameters of the battery charging curve reconstruction model according to the battery state of charge value and the battery health state value to obtain a target health factor model and a target charging curve reconstruction model.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images