Battery health state monitoring method based on entropy feature technology

Abstract

The invention discloses a battery health state monitoring method based on an entropy feature technology. The battery health state monitoring method based on the entropy feature technology comprises the following steps: data acquisition: acquiring a voltage signal generated in a battery recycling process; extracting and screening information entropy features, and extracting and screening the information entropy features in the voltage signals; and training and predicting information entropy features, inputting the information entropy features into a cpo-wlssvm-adaboost model for training, and outputting a health state estimation value of the battery. According to the method, after the information entropy features in the voltage signals are extracted, the information entropy features are input into the cpo-wlssvm-adaboost model for training and prediction, and accurate estimation of the health condition of the battery is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health status monitoring, and in particular to a battery health status monitoring method based on entropy feature technology. Background Art

[0002] Lithium-ion batteries have the advantages of high energy density and simple management. They are widely used in energy storage, electric vehicles and other scenarios. However, many problems and challenges have emerged in the continuous use and development of lithium batteries. The hottest topic is the capacity degradation of lithium batteries during repeated use. Changes in battery chemical materials and the influence of the external physical environment may lead to capacity degradation. The result of capacity degradation will not only lead to poor battery performance, but also cause safety hazards during battery use. Therefore, the health status monitoring technology of lithium batteries has become a top priority. Monitoring the health status (SOH) and remaining life (RUL) of batteries through battery management systems can effectively improve the use of lithium batteries and avoid possible safety problems.

[0003] In recent years, researchers have made in-depth progress in this area. Due to the technical difficulties in measuring the internal parameters of lithium batteries, the estimation of SOH usually uses external parameter data that is easy to measure, such as the battery's current, voltage, temperature, etc. The collected external data is combined with the existing two major research methods: model-based and data-driven to complete the accurate estimation of SOH. With the emergence and rapid development of artificial intelligence, data-driven methods have gradually become mainstream. Compared with model-based methods, data-driven methods do not need to consider the complex reaction principles and aging mechanisms inside the battery. They only need to use deep learning of data sets to achieve ideal prediction effects, and are gradually favored by researchers. However, the integrity, relevance and high quality of the data all pose challenges to the accurate estimation of battery SOH. Therefore, how to accurately estimate the battery health status through external voltage data has become an important issue that needs to be solved urgently. Summary of the invention

[0004] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description and other drawings of the description.

[0005] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a battery health status monitoring method based on entropy feature technology.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a battery health status monitoring method based on entropy feature technology, comprising the following steps: Data collection, collecting voltage signals generated during battery cycling; Extraction and screening of information entropy features: Extraction and screening of information entropy features in voltage signals; Training and prediction of information entropy features, input the information entropy features into the cpo-wlssvm-adaboost model for training, and output the estimated value of the battery's health status.

[0007] The present invention achieves accurate estimation of the battery health state by extracting the information entropy features in the voltage signal and then using the cpo-wlssvm-adaboost model for training and prediction.

[0008] In some embodiments, in the extraction and screening of information entropy features, the voltage signal ranges from 3.5 V to 4.19 V. The present invention divides the charging voltage segment of 3.5 V to 4.19 V into 0.1 V segments, a total of 7 voltage segments, and extracts information entropy features for each voltage segment.

[0009] In some embodiments, in the extraction and screening of information entropy features, the voltage signal ranges from 3.6v to 3.7v. The preferred charging voltage segment of the present invention is the information entropy feature in the range of 3.6v to 3.7v.

[0010] In some embodiments, the information entropy features include power spectrum entropy features and spread entropy features. Power spectrum entropy represents a measure of uncertainty in the battery during the cycle. As the battery life decreases, the uncertainty will increase, and the power spectrum entropy in the voltage data will also increase accordingly; spread entropy overcomes some of the defects of approximate entropy, sample entropy, and permutation entropy, and has the advantages of fast calculation speed and less impact by mutation signals. It has been well applied in signal monitoring and abnormal judgment. Spread entropy represents the irregularity and complexity of the battery during the cycle. As the battery life decreases, the irregularity and complexity will increase, and the spread entropy in the voltage data will also increase.

[0011] In some embodiments, the steps of extracting the power spectrum entropy feature are: The discrete voltage signal is transformed into a power spectrum using Fourier transform; According to Parseval's theorem, the sum of the energies of each power spectrum is calculated; The power spectrum entropy feature PSDE is calculated according to the proportion of the power spectrum and the size of the power spectrum energy. The formula is: ,in, Indicates the input voltage segment, Indicates that Fourier transform is performed on the voltage segment. represents the obtained power spectrum, It represents the percentage of the kth power spectrum in the entire spectrum, and also represents the probability of the kth frequency component appearing. It represents the maximum power spectrum entropy that can be generated by the input voltage segment and is used for normalization.

[0012] In some embodiments, the walk entropy is a method for evaluating the irregularity and complexity of a nonlinear time series, and is calculated for a time series of length N. The steps of spreading entropy are: Use linear or nonlinear mapping functions to transform nonlinear time series Map to ; Using linear transformation Each Assign to An integer, where represents the jth element of the classification signal obtained after linear transformation, round is the rounding function, and c is the number of categories; Calculate the embedding vector , creating an embedding vector with embedding dimension m and time delay d , the formula is: ; use Each sequence Map to In for The scatter pattern of each embedding vector Each corresponds to a scatter pattern , whose elements , is the embedding vector The first element of is the second element, and so on. because for There are m elements in , and each element has c possible values, so the number of scatter patterns is equal to ; for Scattering Pattern , the probability of each scattering mode is as follows: , in, Represents the time series assigned The dispersion pattern for The number of Represents the time series assigned The dispersion pattern divided by the total number of embedding vectors; Finally, according to the definition of Shannon entropy, the value of dispersion entropy (DE) is calculated as follows: , in, Represents the maximum scatter entropy that can be produced and is used for normalization.

[0013] In some embodiments, the training and prediction of information entropy features include the following steps: The crested porcupine optimizer (CPO) algorithm is used to optimize the regularization parameters and initial values ​​of the kernel width of the Wlssvm model to find the optimal parameters; The Wlssvm model is used as a weak predictor and synthesized into a strong predictor through the Adaboost ensemble algorithm; Input information entropy features for training and prediction.

[0014] In some embodiments, the crested porcupine optimizer (CPO) algorithm comprises: In the initialization phase, the optimization problem consists of multiple dimensions (d), each of which needs to be minimized or maximized through optimization techniques. The CPO algorithm optimizes these problems by randomly initializing their dimensions in the search space of each problem. The formula is: ,in, represents the number of individuals, is the first candidate solution in the search space, and are the lower and upper bounds of the search range, respectively. is a vector randomly initialized between 0 and 1; The cyclic population reduction technique accelerates the convergence speed by letting some CPs leave the population during the optimization process, and then rejoins the population to improve diversity and avoid falling into the local minimum. The mathematical model is: , in, is the variable that determines the number of loops. is the current function evaluation number, is the maximum number of function evaluations, represents the remainder or modulus operator, and the minimum number of individuals in the newly generated population is Therefore, the population size cannot be less than ; The exploration phase is divided into the first defense strategy and the second defense strategy. In the first defense strategy, when the CP becomes aware of a predator it begins to raise and flap its quills, giving the impression of a larger prey, using the formula: ,in, is the position of the i-th CP in the t-th iteration, is the current best CP position, and is a random number, pass Calculated, is the position of another CP chosen randomly; In the second defense strategy, the CP uses vocal methods to make noise and threaten the predator. As the predator approaches, the porcupine's voice becomes louder and louder. The formula is: , in, is a random vector between O and 1, is a random number, and are the positions of two randomly selected CPs; The development phase is divided into the third defense strategy and the fourth defense strategy. In the third defense strategy, CP secretes a foul odor that spreads around it to prevent predators from approaching, as given by: , in, is a fitness function, is a random vector, is a time-dependent factor, and are the locations of three randomly selected CPs; In the fourth defense strategy, CP uses physical attack with the formula: , in, is a control parameter, and is a random number, is a random vector, is a factor based on fitness.

[0015] In some embodiments, the construction algorithm of the WLSSVM-Adaboost strong predictor is as follows: Input training set , penalty factor , core width , initial value , minimum , search step length ; Initialization, initial weights of training samples: ; Iterative process : (a) In the current sample weight distribution Use LSSVM to classify the training samples and generate weak classifiers (b) Calculation Error rate: (c) if , then in Reduce the step size by one And return to step (a); (d) Calculate the weak classifier Weight: ; (e) Update sample weights and normalize: , in, is the weight update parameter, is the normalization factor; (f) output strong classifier: , in, The function is the set of variable points x corresponding to the maximum value of the function in the brackets. For the training set The predicted value vector in .

[0016] In some embodiments, after obtaining the final SOH estimate, the trained model is verified and evaluated using a test set, and the steps are as follows: 60% of the extracted information entropy features are used as the test set to input SVM,LSTM, and the original model LSSVM-Adaboost is trained to obtain the estimated value; The root mean square error RMSE, r 2 The coefficient evaluates the effect of the algorithm model, and the calculation formula is: , , in, and Respectively represent The actual SOH and estimated SOH after cycles, Indicates the average value of SOH.

[0017] By adopting the above technical solution, the beneficial effects of the present invention are: 1. The present invention extracts the information entropy features from the voltage signal and inputs them into the cpo-wlssvm-adaboost model for training and prediction, thereby achieving accurate estimation of the battery health status.

[0018] 2. The present invention solves the problem of how to estimate the health status of the battery through external voltage data in battery health monitoring through entropy feature technology, thereby achieving accurate health estimation of the battery.

[0019] 3. The present invention only extracts the required power spectrum entropy and spread entropy from the charging voltage segment of 3.6V-3.7V, and then uses the cpo-wlssvm-adaboost model for training and prediction, thereby achieving accurate estimation of battery health status, which can be applied to the battery management system of lithium batteries.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure.

[0021] Undoubtedly, these and other objects of the present invention will become more apparent after the following detailed description of the preferred embodiment described with reference to various figures and drawings.

[0022] In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, one or several preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation of the present invention.

[0024] In the drawings, the same reference numerals are used for the same components and the drawings are schematic and not necessarily drawn to scale.

[0025] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only one or several embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on such drawings without paying creative work.

[0026] Figure 1 It is an overall flow chart of a battery health status monitoring method based on entropy feature technology of the present invention; Figure 2 is a flow chart of a battery health status monitoring method based on entropy feature technology in some embodiments of the present invention; Figure 3 This is a flow chart of the operation of the CPO-WLSSVM-ADABOOST model in some embodiments of the present invention. DETAILED DESCRIPTION

[0027] 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 specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but are not used to limit the present invention.

[0028] In the present invention, unless otherwise clearly specified and limited, the first feature "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0029] Reference Figure 1 , Figure 1 The figure is an overall flow chart of a battery health status monitoring method based on entropy feature technology of the present invention.

[0030] According to some embodiments of the present invention, the present invention provides a battery health status monitoring method based on entropy feature technology, comprising the following steps: Data collection, collecting voltage signals generated during battery cycling; Extraction and screening of information entropy features: Extraction and screening of information entropy features in voltage signals; Training and prediction of information entropy features, input the information entropy features into the cpo-wlssvm-adaboost model for training, and output the estimated value of the battery's health status.

[0031] The present invention achieves accurate estimation of the battery health state by extracting the information entropy features in the voltage signal and then using the cpo-wlssvm-adaboost model for training and prediction.

[0032] According to some embodiments of the present invention, optionally, in the extraction and screening of information entropy features, the voltage signal ranges from 3.5v to 4.19v. The present invention divides the charging voltage segment of 3.5v to 4.19v into 0.1v segments, a total of 7 voltage segments, and extracts information entropy features for each voltage segment.

[0033] According to some embodiments of the present invention, optionally, in the extraction and screening of information entropy features, the voltage signal ranges from 3.6v to 3.7v. The present invention preferably selects information entropy features in the charging voltage range of 3.6v to 3.7v.

[0034] According to some embodiments of the present invention, optionally, the information entropy feature includes a power spectrum entropy feature and a spread entropy feature. The power spectrum entropy represents a measure of the uncertainty of the battery in the cycle. As the battery life decreases, the uncertainty will increase, and the power spectrum entropy in the voltage data will also increase; the spread entropy overcomes some of the defects of approximate entropy, sample entropy and permutation entropy, and has the advantages of fast calculation speed and less influence by mutation signals. It has been well applied in signal monitoring and abnormal judgment. The spread entropy represents the irregularity and complexity of the battery in the cycle. As the battery life decreases, the irregularity and complexity will increase, and the spread entropy in the voltage data will also increase.

[0035] According to some embodiments of the present invention, optionally, the step of extracting the power spectrum entropy feature is: The discrete voltage signal is transformed into a power spectrum using Fourier transform; According to Parseval's theorem, the sum of the energies of each power spectrum is calculated; The power spectrum entropy feature PSDE is calculated according to the proportion of the power spectrum and the size of the power spectrum energy. The formula is: ,in, Indicates the input voltage segment, Indicates that Fourier transform is performed on the voltage segment. represents the obtained power spectrum, It represents the percentage of the kth power spectrum in the entire spectrum, and also represents the probability of the kth frequency component appearing. It represents the maximum power spectrum entropy that can be generated by the input voltage segment and is used for normalization.

[0036] According to some embodiments of the present invention, optionally, the walk entropy is a method for evaluating the irregularity and complexity of a nonlinear time series, and calculates a time series of length N. The steps of spreading entropy are: Use linear or nonlinear mapping functions to transform nonlinear time series Map to ; Using linear transformation Each Assign to An integer, where represents the jth element of the classification signal obtained after linear transformation, round is the rounding function, and c is the number of categories; Calculate the embedding vector , creating an embedding vector with embedding dimension m and time delay d , the formula is: ; use Each sequence Map to In for The scatter pattern of each embedding vector Each corresponds to a scatter pattern , whose elements , is the embedding vector The first element of is the second element, and so on. because for There are m elements in , and each element has c possible values, so the number of scatter patterns is equal to ; for Scattering Pattern , the probability of each scattering mode is as follows: , in, Represents the time series assigned The dispersion pattern for The number of Represents the time series assigned The dispersion pattern divided by the total number of embedding vectors; Finally, according to the definition of Shannon entropy, the value of dispersion entropy (DE) is calculated as follows: , in, Indicates the maximum scatter entropy that can be produced and is used for normalization.

[0037] According to some embodiments of the present invention, optionally, the training and prediction of information entropy features include the following steps: The crested porcupine optimizer (CPO) algorithm is used to optimize the regularization parameters and initial values ​​of the kernel width of the Wlssvm model to find the optimal parameters; The Wlssvm model is used as a weak predictor and synthesized into a strong predictor through the Adaboost ensemble algorithm; Input information entropy features for training and prediction.

[0038] According to some embodiments of the present invention, optionally, the crested porcupine optimizer (CPO) algorithm comprises: In the initialization phase, the optimization problem consists of multiple dimensions (d), each of which needs to be minimized or maximized through optimization techniques. The CPO algorithm optimizes these problems by randomly initializing their dimensions in the search space of each problem. The formula is: , in, represents the number of individuals, is the first candidate solution in the search space, and are the lower and upper bounds of the search range, respectively. is a vector randomly initialized between 0 and 1; The cyclic population reduction technique accelerates the convergence speed by letting some CPs leave the population during the optimization process, and then rejoins the population to improve diversity and avoid falling into the local minimum. The mathematical model is: , in, is the variable that determines the number of loops. is the current function evaluation number, is the maximum number of function evaluations, represents the remainder or modulus operator, and the minimum number of individuals in the newly generated population is Therefore, the population size cannot be less than ; The exploration phase is divided into the first defense strategy and the second defense strategy. In the first defense strategy, when the CP becomes aware of a predator it begins to raise and flap its quills, giving the impression of a larger prey, using the formula: ,in, is the position of the i-th CP in the t-th iteration, is the current best CP position, and is a random number, pass Calculated, is the position of another CP chosen randomly; In the second defense strategy, the CP uses vocal methods to make noise and threaten the predator. As the predator approaches, the porcupine's voice becomes louder and louder. The formula is: , in, is a random vector between O and 1, is a random number, and are the positions of two randomly selected CPs; The development phase is divided into the third defense strategy and the fourth defense strategy. In the third defense strategy, CP secretes a foul odor that spreads around it to prevent predators from approaching, as given by: , in, is a fitness function, is a random vector, is a time-dependent factor, and are the locations of three randomly selected CPs; In the fourth defense strategy, CP uses physical attack with the formula: , in, is a control parameter, and is a random number, is a random vector, is a factor based on fitness.

[0039] According to some embodiments of the present invention, optionally, the construction algorithm of the WLSSVM-Adaboost strong predictor is as follows: Input training set , penalty factor , core width , initial value , minimum , search step length ; Initialization, initial weights of training samples: ; Iterative process : (a) In the current sample weight distribution Use LSSVM to classify the training samples and generate weak classifiers (b) Calculation Error rate: (c) if , then in Reduce the step size by one And return to step (a); (d) Calculate the weak classifier Weight: ; (e) Update sample weights and normalize: , in, is the weight update parameter, is the normalization factor; (f) output strong classifier: , in, The function is the set of variable points x corresponding to the maximum value of the function in the brackets. For the training set The predicted value vector in .

[0040] According to some embodiments of the present invention, optionally, after obtaining the final SOH estimation value, the trained model is verified and evaluated using a test set, the steps are: 60% of the extracted information entropy features are used as the test set to input SVM,LSTM, and the original model LSSVM-Adaboost is trained to obtain the estimated value; The root mean square error RMSE, r 2 The coefficient evaluates the effect of the algorithm model, and the calculation formula is: , , in, and Respectively represent The actual SOH and estimated SOH after cycles, Indicates the average value of SOH.

[0041] Example 1 Reference Figure 1 , Figure 3 This embodiment provides a battery health status monitoring method based on entropy feature technology, including the following steps: Data collection, collecting voltage signals generated during battery cycling; Extraction and screening of information entropy features: Extraction and screening of information entropy features in voltage signals; Training and prediction of information entropy features, input the information entropy features into the cpo-wlssvm-adaboost model for training, and output the estimated value of the battery's health status.

[0042] In the extraction and screening of information entropy features, the range of voltage signals is 3.5v-4.19v. Specifically, each 0.1v is a section, namely 3.5v-3.6v, 3.6v-3.7v, 3.7v-3.8v, 3.8v-3.9v, 3.9v-4.0v, 4.0v-4.1v, and 4.1v-4.19v.

[0043] Furthermore, the information entropy features include power spectrum entropy features and spread entropy features. The power spectrum entropy represents a measure of the uncertainty of the battery in the cycle. As the battery life decreases, the uncertainty will increase, and the power spectrum entropy in the voltage data will also increase. The spread entropy represents the irregularity and complexity of the battery in the cycle. As the battery life decreases, the irregularity and complexity will increase, and the spread entropy in the voltage data will also increase. Specifically, The steps for extracting power spectrum entropy features are: The discrete voltage signal is transformed into a power spectrum using Fourier transform; According to Parseval's theorem, the sum of the energies of each power spectrum is calculated; The power spectrum entropy feature PSDE is calculated according to the proportion of the power spectrum and the size of the power spectrum energy. The formula is: , in, Indicates the input voltage segment, Indicates that Fourier transform is performed on the voltage segment. represents the obtained power spectrum, It represents the percentage of the kth power spectrum in the entire spectrum, and also represents the probability of the kth frequency component appearing. It represents the maximum power spectrum entropy that can be generated by the input voltage segment and is used for normalization.

[0044] Scattering entropy overcomes some of the defects of approximate entropy, sample entropy and permutation entropy, and has the advantages of fast calculation speed and less influence by mutation signals. It has been well applied in signal monitoring and abnormal judgment. Scattering entropy is a method for evaluating the irregularity and complexity of nonlinear time series. It calculates a time series of length N. The steps of spreading entropy are: Use linear or nonlinear mapping functions to transform nonlinear time series Map to ; Using linear transformation Each Assign to An integer, where represents the jth element of the classification signal obtained after linear transformation, round is the rounding function, and c is the number of categories; Calculate the embedding vector , creating an embedding vector with embedding dimension m and time delay d , the formula is: ; use Each sequence Map to In for The scatter pattern of each embedding vector Each corresponds to a scatter pattern , whose elements , is the embedding vector The first element of is the second element, and so on. because for There are m elements in , and each element has c possible values, so the number of scatter patterns is equal to ; for Scattering Pattern , the probability of each scattering mode is as follows: , in, Represents the time series assigned The dispersion pattern for The number of Represents the time series assigned The dispersion pattern divided by the total number of embedding vectors; Finally, according to the definition of Shannon entropy, the value of dispersion entropy (DE) is calculated as follows: , in, Indicates the maximum scatter entropy that can be produced and is used for normalization.

[0045] Furthermore, the training and prediction of information entropy features include the following steps: The crested porcupine optimizer (CPO) algorithm is used to optimize the regularization parameters and initial values ​​of the kernel width of the Wlssvm model to find the optimal parameters; The Wlssvm model is used as a weak predictor and synthesized into a strong predictor through the Adaboost ensemble algorithm; Input information entropy features for training and prediction; Specifically, the Crested Porcupine Optimizer (CPO) is a new meta-heuristic algorithm with the advantages of defense strategy simulation, dynamic interaction and update, group dynamic adjustment strategy, and high improvement rate. The Crested Porcupine Optimizer (CPO) algorithm includes: In the initialization phase, the optimization problem consists of multiple dimensions (d), each of which needs to be minimized or maximized through optimization techniques. The CPO algorithm optimizes these problems by randomly initializing their dimensions in the search space of each problem. The formula is: ,in, represents the number of individuals, is the first candidate solution in the search space, and are the lower and upper bounds of the search range, respectively. is a vector randomly initialized between 0 and 1; The cyclic population reduction technique accelerates the convergence speed by letting some CPs leave the population during the optimization process, and then rejoins the population to improve diversity and avoid falling into the local minimum. The mathematical model is: , in, is the variable that determines the number of loops. is the current function evaluation number, is the maximum number of function evaluations, represents the remainder or modulus operator, and the minimum number of individuals in the newly generated population is Therefore, the population size cannot be less than ; The exploration phase is divided into the first defense strategy and the second defense strategy. In the first defense strategy, when the CP becomes aware of a predator it begins to raise and flap its quills, giving the impression of a larger prey, using the formula: ,in, is the position of the i-th CP in the t-th iteration, is the current best CP position, and is a random number, pass Calculated, is the position of another CP chosen randomly; In the second defense strategy, the CP uses vocal methods to make noise and threaten the predator. As the predator approaches, the porcupine's voice becomes louder and louder. The formula is: ,in, is a random vector between O and 1, is a random number, and are the positions of two randomly selected CPs; The development phase is divided into the third defense strategy and the fourth defense strategy. In the third defense strategy, CP secretes a foul odor that spreads around it to prevent predators from approaching, as given by: , in, is a fitness function, is a random vector, is a time-dependent factor, and are the locations of three randomly selected CPs; In the fourth defense strategy, CP uses physical attack with the formula: , in, is a control parameter, and is a random number, is a random vector, is a factor based on fitness.

[0046] The time complexity analysis of CPO shows that its performance depends mainly on the dimension of the problem and the maximum number of function evaluations, which is consistent with other metaheuristic algorithms. Overall, the time complexity of CPO in big O notation is .

[0047] WLSSVM increases the weight coefficient to improve the robustness of the model, and retains the calculation speed on the basis of LSSVM. It is widely used in nonlinear function estimation. Adaboost, as a variation of boosting, synthesizes weak predictors with general prediction ability through some method to finally form a strong predictor with stronger prediction ability. The construction algorithm of WLSSVM-Adaboost strong predictor is as follows: Input training set , penalty factor , core width , initial value , minimum , search step length ; Initialization, initial weights of training samples ; Iterative process : (a) In the current sample weight distribution Use LSSVM to classify the training samples and generate weak classifiers (b) Calculation Error rate: (c) if , then in Reduce the step size by one And return to step (a); (d) Calculate the weak classifier Weight: ; (e) Update sample weights and normalize: , in, is the weight update parameter, is the normalization factor; (f) output strong classifier: , in, The function is the set of variable points x corresponding to the maximum value of the function in the brackets. For the training set The predicted value vector in .

[0048] Furthermore, after using the information entropy feature to input cpo-wlssvm-adaboost for training and prediction to obtain the final SOH estimate, the trained model is verified and evaluated using the test set. The steps are as follows: 60% of the extracted information entropy features are used as the test set to input SVM,LSTM, and the original model LSSVM-Adaboost is trained to obtain the estimated value; The root mean square error RMSE, r 2 The coefficient evaluates the effect of the algorithm model, and the calculation formula is: , , in, and Respectively represent The actual SOH and estimated SOH after cycles, Indicates the average value of SOH.

[0049] Example 2 Reference Figure 2 This embodiment provides a battery health status monitoring method based on entropy feature technology. The difference between this embodiment and embodiment 1 is that: In the extraction and screening of information entropy features, the voltage signal range is 3.6v-3.7v. After using the information entropy features to input cpo-wlssvm-adaboost for training and prediction to obtain the final SOH estimation value, the trained model is verified and evaluated using the test set. Specifically, by comparing the estimation obtained by cpo-wlssvm-adaboost training with the estimation obtained by SVM, LSTM, and the original model LSSVM-Adaboost, cpo-wlssvm-adaboost can improve the estimation accuracy and reduce the root mean square error RMSE to 0.8%, r 2 The coefficient can be increased to about 0.99. Therefore, the cpo-wlssvm-adaboost of the present invention can significantly improve the prediction accuracy and reduce the prediction error.

[0050] It should be understood that the embodiments disclosed in the present invention are not limited to the specific processing steps or materials disclosed herein, but should be extended to equivalent substitutions of such features understood by ordinary technicians in the relevant field. It should also be understood that the terms used herein are only used for the purpose of describing specific embodiments and are not meant to be limiting.

[0051] The "embodiment" mentioned in the specification means that the specific features or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0052] In addition, the described features or characteristics may be combined in one or more embodiments in any other suitable manner. In the above description, some specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of the embodiments of the present invention. However, those skilled in the relevant art will understand that the present invention can be implemented without one or more of the above specific details or can also be implemented using other methods, components, materials, etc.

Claims

1. A battery health status monitoring method based on entropy feature technology, characterized in that: The following steps are involved: Data collection, collecting voltage signals generated during battery cycling; Extraction and screening of information entropy features: Extraction and screening of information entropy features in voltage signals; Training and prediction of information entropy features, input the information entropy features into the cpo-wlssvm-adaboost model for training, and output the estimated value of the battery's health status.

2. The method for monitoring the battery health status based on entropy characteristic technology according to claim 1, characterized in that: In the extraction and screening of information entropy features, the voltage signal ranges from 3.5v to 4.19v.

3. The method for monitoring the battery health status based on entropy characteristic technology according to claim 2 is characterized in that: In the extraction and screening of information entropy features, the voltage signal ranges from 3.6v to 3.7v.

4. The method for monitoring the battery health status based on entropy characteristic technology according to claim 1, characterized in that: Information entropy features include power spectrum entropy features and dispersion entropy features.

5. The method for monitoring the battery health status based on entropy characteristic technology according to claim 4 is characterized in that: The steps for extracting power spectrum entropy features are: The discrete voltage signal is transformed into a power spectrum using Fourier transform; According to Parseval's theorem, the sum of the energies of each power spectrum is calculated; The power spectrum entropy feature PSDE is calculated according to the proportion of the power spectrum and the size of the power spectrum energy. The formula is: ,in, Indicates the input voltage segment, Indicates that Fourier transform is performed on the voltage segment. represents the obtained power spectrum, It represents the percentage of the kth power spectrum in the entire spectrum, and also represents the probability of the kth frequency component appearing. It represents the maximum power spectrum entropy that can be generated by the input voltage segment and is used for normalization.

6. The method for monitoring the battery health status based on entropy characteristic technology according to claim 4 is characterized in that: The walk entropy is a method used to evaluate the irregularity and complexity of nonlinear time series. It calculates a time series of length N. The steps of spreading entropy are: Use linear or nonlinear mapping functions to transform nonlinear time series Map to ; Using linear transformation Each Assign to An integer, where represents the jth element of the classification signal obtained after linear transformation, round is the rounding function, and c is the number of categories; Calculate the embedding vector , creating an embedding vector with embedding dimension m and time delay d , the formula is: ； use Each sequence Map to In for The scatter pattern of each embedding vector Each corresponds to a scatter pattern , whose elements , is the embedding vector The first element of is the second element, and so on. because for There are m elements in , and each element has c possible values, so the number of scatter patterns is equal to ; for Scattering Pattern , the probability of each scattering mode is as follows: ， in, Represents the time series assigned The dispersion pattern for The number of Represents the time series assigned The dispersion pattern divided by the total number of embedding vectors; Finally, according to the definition of Shannon entropy, the value of dispersion entropy (DE) is calculated as follows: ， in, Represents the maximum scatter entropy that can be produced and is used for normalization.

7. The method for monitoring battery health status based on entropy feature technology according to claim 1, characterized in that: The training and prediction of information entropy features include the following steps: The crested porcupine optimizer (CPO) algorithm is used to optimize the regularization parameters and initial values ​​of the kernel width of the Wlssvm model to find the optimal parameters; The Wlssvm model is used as a weak predictor and synthesized into a strong predictor through the Adaboost ensemble algorithm; Input information entropy features for training and prediction.

8. The method for monitoring the battery health status based on entropy characteristic technology according to claim 7, characterized in that: The Crested Porcupine Optimizer (CPO) algorithm includes: In the initialization phase, the optimization problem consists of multiple dimensions (d), each of which needs to be minimized or maximized through optimization techniques. The CPO algorithm optimizes these problems by randomly initializing their dimensions in the search space of each problem. The formula is: ， in, represents the number of individuals, is the first candidate solution in the search space, and are the lower and upper bounds of the search range, respectively. is a vector randomly initialized between 0 and 1; The cyclic population reduction technique accelerates the convergence speed by letting some CPs leave the population during the optimization process, and then rejoins the population to improve diversity and avoid falling into the local minimum. The mathematical model is: ， in, is the variable that determines the number of loops. is the current function evaluation number, is the maximum number of function evaluations, represents the remainder or modulus operator, and the minimum number of individuals in the newly generated population is Therefore, the population size cannot be less than ; The exploration phase is divided into the first defense strategy and the second defense strategy. In the first defense strategy, when the CP becomes aware of a predator it begins to raise and flap its quills, giving the impression of a larger prey, using the formula: ,in, is the position of the i-th CP in the t-th iteration, is the current best CP position, and is a random number, pass Calculated, is the position of another CP chosen randomly; In the second defense strategy, the CP uses vocal methods to make noise and threaten the predator. As the predator approaches, the porcupine's voice becomes louder and louder. The formula is: ， in, is a random vector between O and 1, is a random number, and are the positions of two randomly selected CPs; The development phase is divided into the third defense strategy and the fourth defense strategy. In the third defense strategy, CP secretes a foul odor that spreads around it to prevent predators from approaching, as given by: ， in, is a fitness function, is a random vector, is a time-dependent factor, and are the locations of three randomly selected CPs; In the fourth defense strategy, CP uses physical attack with the formula: ， in, is a control parameter, and is a random number, is a random vector, is a factor based on fitness.

9. The method for monitoring the battery health status based on entropy characteristic technology according to claim 7, characterized in that: The construction algorithm of the WLSSVM-Adaboost strong predictor is as follows: Input training set , penalty factor , core width , initial value , minimum , search step length ; Initialization, initial weights of training samples: ; Iterative process : (a) In the current sample weight distribution Use LSSVM to classify the training samples and generate weak classifiers (b) Calculation Error rate: (c) if , then in Reduce the step size by one And return to step (a); (d) Calculate the weak classifier Weight: ; (e) Update sample weights and normalize: ， in, is the weight update parameter, is the normalization factor; (f) output strong classifier: ， in, The function is the set of variable points x corresponding to the maximum value of the function in the brackets. For the training set The predicted value vector in .

10. The method for monitoring battery health status based on entropy feature technology according to claim 9, characterized in that: After obtaining the final SOH estimate, use the test set to verify and evaluate the trained model. The steps are: 60% of the extracted information entropy features were used as test sets to be input into SVM, LSTM, and the original model LSSVM-Adaboost for training to obtain the estimated value; The root mean square error RMSE, r 2 The coefficient evaluates the effect of the algorithm model, and the calculation formula is: ， ， in, and Respectively represent The actual SOH and estimated SOH after cycles, Indicates the average value of SOH.

Images