Data-driven ebike end-of-life battery state of health estimation method and system

Abstract

The application discloses a data-driven electric bicycle retired battery health state estimation method and system, relates to the electric bicycle battery state monitoring field, and comprises the following steps: obtaining battery charging voltage data, and calculating an IC curve; extracting features related to battery life from the IC curve, and verifying the effectiveness of the extracted features by using an improved fuzzy C-means clustering algorithm; determining the structure of a BP neural network, and accelerating the convergence speed of the BP neural network by using an NSGA-II genetic algorithm; and taking the formed NSGA-II-BP neural network as the basis, taking the aging characteristics of the retired battery as the input of the NSGA-II-BP neural network, and taking the battery monomer capacity as the output, the state estimation model is built. The application combines the BP neural network and the NSGA-II optimization genetic algorithm and applies them to battery health state estimation. The application has great significance for reducing the development cost of electric bicycles and popularizing retired batteries.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electric bicycle battery state monitoring, in particular to a data-driven electric bicycle retired battery health state estimation method and system. BACKGROUND

[0002] Lithium-ion power batteries are widely used in electric bicycles due to their high energy density, low self-discharge rate, long life, etc. However, lithium-ion batteries are used and decayed immediately. A large number of lithium-ion power batteries are retired from electric vehicles due to capacity attenuation. Although the retired batteries cannot normally meet the use of electric bicycles, they can still be used in many other energy and power requirements lower power systems, such as wind energy storage, smart grid, etc. However, in the recycling process of retired batteries, the problem of continuous aging of batteries still exists. The performance of the battery in all aspects is deteriorating, mainly manifested as capacity attenuation and internal resistance increase. The imbalance of parameters in the battery pack leads to different charging and discharging depths of each single battery, which further leads to different capacity attenuation and heat generation.

[0003] The state of health (SoH) is one of the most important parameters for evaluating the health condition of lithium-ion power batteries. Aging and improper operation of the battery can significantly affect the safe and stable operation of power equipment. When the performance of the battery decreases to a certain extent, problems such as battery leakage, insulation damage, and local short circuit can cause safety accidents.

[0004] The existing power battery SoH estimation methods are mainly divided into three categories. The first category is direct measurement method, the second category is model-based SoH algorithm, and the third category is data-driven method. This method uses machine learning algorithm to train a large number of offline battery data to establish an aging model, and then estimates the health state of the remaining battery based on the model. This method is simple and reliable. With the support of a large number of training samples, an accurate estimation model can be obtained, and the problem of large number of retired batteries and large difference can be solved, further improving the estimation efficiency and accuracy. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the problem to be solved by the present application is how to estimate the health state of the retired battery in real time to ensure the safe operation of the energy storage system.

[0007] To solve the above technical problems, the application provides the following technical scheme: a data-driven electric bicycle retired battery state of health estimation method, which comprises the following steps: obtaining electric bicycle battery charging voltage data through maximum capacity test experiment, and obtaining an incremental capacity IC curve based on numerical derivation method, and performing filtering and noise reduction processing on the obtained IC curve; based on the key points in the IC curve, part of the charging curve is intercepted, and features related to battery life are extracted from the curve, and then data standardization is performed, and an improved fuzzy C-means clustering algorithm is used to sort the retired batteries to verify the effectiveness of the extracted features; a BP neural network is constructed to estimate the state of health of the power battery, the structure of the BP neural network is determined, the convergence speed of the BP neural network is accelerated by using NSGA-II genetic algorithm, and the best initial weight suitable for the BP neural network is found out to optimize the performance of the BP neural network; based on the formed NSGA-II-BP neural network, the extracted aging features of the retired battery are taken as the input of the NSGA-II-BP neural network, and the corresponding battery monomer capacity is taken as the output, so as to complete the construction of the battery state of health estimation model; the NSGA-II-BP neural network SoH estimation model of different types of batteries is constructed, and the accuracy of the model is tested, and the effectiveness of the constructed SoH estimation model under different working conditions is verified based on the test data of the retired batteries under different working conditions.

[0008] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the battery charging voltage data is a curve with charging voltage as height and charging time as length; the capacity IC curve is a curve with the derivative of battery capacity to battery voltage as height and charging voltage as length, and the calculation formula is,

[0009]

[0010] Wherein, N represents the sampling interval, Q represents the battery capacity, and V represents the battery voltage.

[0011] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the extracted features related to battery life include key voltage value, gradient, energy and fluctuation rate; the improved fuzzy C-means clustering algorithm comprises subtractive clustering, finding the best initial value of FCM, sorting the retired batteries by using the extracted features related to battery life through the fuzzy C-means clustering algorithm, and verifying the effectiveness of the extracted aging features.

[0012] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the structure of the BP neural network comprises transfer function, number of hidden layers and number of nodes, and then the weights and thresholds are initialized; the hidden layer Hf and output layer O p The output of the output layer O

[0013]

[0014] where w ij and w jp are the weights of the hidden layer and the output layer, respectively, a j and b p are the thresholds of the hidden layer and the output layer, respectively, l H and m0 are the number of neurons of the hidden layer and the output layer, respectively, x i is the i-th aging feature of the input; the default transfer function of the hidden layer and the output layer is tansing, and the expression is,

[0015]

[0016] The error e p between the expected output and the actual output is calculated, and the specific formula is as follows:

[0017] e p = Y p - O p p = 1, 2, …, m0

[0018] where Y p is the expected output, and O p is the actual output; the weights are corrected according to the calculated error, and the weights are adjusted in time to optimize the BP neural network, and the formula is,

[0019]

[0020] where η is the learning rate; the threshold is corrected according to the calculated error, and the threshold is adjusted in time, and the formula is,

[0021]

[0022] The loop is iterated until e p satisfies the requirement of the set expected error value.

[0023] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the accelerating convergence speed of the BP neural network comprises: starting from the fast non-dominated genetic algorithm NSGA-II, globally searching the weight value and the threshold value, and determining the optimal initial weight value suitable for the BP neural network; the BP neural network starts the training process with the optimal initial value provided by the fast non-dominated genetic algorithm, and seeks the optimal solution; the starting from the fast non-dominated genetic algorithm NSGA-II comprises: determining the topological structure of the neural network, and initializing the threshold value and the weight value; the globally searching the weight value and the threshold value comprises: coding the individuals in the population, and the individual coding comprises the weight value from the input layer to the hidden layer, the weight value from the hidden layer to the output layer, and the threshold value of the hidden layer and the output layer; the objective function of the NSGA-II is defined as the square error between the expected value obtained by inputting the training sample during the training of the BP network and the actual output value y , and the formula is, i

[0024]

[0025] wherein N represents the logarithm of the network training sample data, and s represents the node number of the network output; the optimization purpose of the objective function is to obtain a minimum value, the purpose of the fitness function is to obtain a maximum value, and the fitness function F is defined as the reciprocal of the objective function plus a small positive number ξ m , and the formula is,

[0026] F=(J m +ξ m ) -1 .

[0027] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the globally searching the weight value and the threshold value further comprises: selecting, crossing, mutating, non-dominant sorting, calculating the crowdedness, and performing the elite reservation operation on the population, so as to determine the optimal initial weight value suitable for the BP neural network; the selection operation comprises: adopting the Monte Carlo method, that is, selecting according to the fitness size, and the probability p i of the individual being selected is,

[0028]

[0029] wherein n represents the population size, f i represents the fitness of the individual i, and the fitness is proportional to the probability of the individual being selected; the crossing comprises: adopting the real number crossing between two individuals, and the individuals and after the crossing operation are,

[0030] ​

[0031] wherein, a is a random number between [0, 1]; the mutation operation includes, using a random number in any range, replacing any gene in the individual code to complete mutation to obtain The mutation operation formula is represented as,

[0032]

[0033] wherein, x min , x max are respectively the minimum value and the maximum value of the gene in the individual, and β is a random number between [0, 1]; the non-dominated sorting operation includes, constructing the battery characteristic population size as n, setting i = 1, comparing the size relationship of the fitness function between the individual x i and the individual x j to obtain the dominated and non-dominated relationship, if there is no individual x j better than x i , marking x i as a non-dominated individual; letting i = i + 1, turning to the comparison of the individual x i+1 and the individual x j , until all the non-dominated individuals are found; marking the obtained non-dominated individual set as the first level non-dominated layer of the population, ignoring the marked non-dominated individuals, continuing the optimization comparison to obtain the second level non-dominated layer, ignoring the marked non-dominated individuals, until the entire population is divided into non-dominated layers and dominated layers; the calculation of the crowding degree includes, initializing the crowding distance of any individual in the same layer as 0, i.e. n d = 0, arranging the individuals in the same layer in ascending order according to the fitness function value, letting the crowding degrees of the two individuals at the boundary be infinite, obtaining the crowding degree of the individual in the middle of the sorting, and the formula is represented as,

[0034] n d = n d + (F(i+1)-F(i-1))

[0035] The elite reservation operation includes: merging the parent population and the offspring population to form a new population, ordering the new population in a non-dominated order from low to high, sequentially putting the whole layer population into the new parent, when the individuals of the kth level are all put into the new parent set, the number of individuals in the set is less than n, and when the individuals of the k+1th level are all put into the new parent set, the number of individuals in the set is greater than n, then the individuals of the k+1th level are all arranged in descending order according to the crowding distance, then all the individuals of the level greater than k+1 are eliminated, and the size of the parent population is kept as n; the decoded weight and threshold value are substituted into the fitness, and it is judged whether the constraint condition is met, if yes, the optimal weight and threshold value are returned, the battery state of health model is built, and if not, the genetic optimization is continuously cycled.

[0036] As a preferred scheme of the data-driven electric bicycle retired battery state of health estimation method, wherein: the different working conditions are different working temperatures of the battery and different charge-discharge rates of the battery.

[0037] Another object of the present application is to provide a data-driven electric bicycle retired battery state of health estimation system, which can estimate the state of health of the retired battery in real time.

[0038] To solve the above technical problems, the present application provides the following technical solutions: a data-driven electric bicycle retired battery state of health estimation method system, comprising: a data preprocessing module, a feature extraction module, an NSGA-II-BP neural network construction module and a SoH estimation model evaluation module; the data preprocessing module is used for preprocessing the IC curve, obtaining the electric bicycle battery charging voltage data through the maximum capacity test experiment, and obtaining the incremental capacity IC curve by using the numerical derivative method, and filtering and denoising the obtained IC curve; the feature extraction module is used for extracting the features related to the battery life, based on the key points in the IC curve, part of the charging curve is intercepted, and the features related to the battery life are extracted from the curve, and then the data is standardized, and the improved fuzzy C-means clustering algorithm is used to sort the retired batteries to verify the effectiveness of the extracted features; the NSGA-II-BP neural network construction module is used for constructing the NSGA-II-BP neural network model, constructing the BP neural network to estimate the state of health of the power battery, determining the structure of the BP neural network, using the NSGA-II genetic algorithm to speed up the convergence speed of the BP neural network, and finding the best initial weight suitable for the BP neural network to optimize the performance of the BP neural network; based on the formed NSGA-II-BP neural network, the extracted retired battery aging features are used as the input of the NSGA-II-BP neural network, and the corresponding battery monomer capacity is used as the output, and the battery state of health estimation model is built; the SoH estimation model evaluation module is used for evaluating the SoH model, constructing the NSGA-II-BP neural network SoH estimation model of different types of batteries, and testing the accuracy of the model, based on the retired battery test data under different working conditions, verifying the effectiveness of the constructed SoH estimation model under different working conditions.

[0039] A computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the data-driven electric bicycle retired battery state of health estimation method as described above when executing the computer program.

[0040] A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the data-driven electric bicycle retired battery state of health estimation method as described above.

[0041] The application has the beneficial effects that: the application is based on ICA intercepting part of the charging curve and extracting battery aging characteristics therefrom, combines a BP neural network with an NSGA-II optimized genetic algorithm to apply to electric bicycle battery health state estimation, and establishes a data-driven electric bicycle retired battery health state estimation method, which can accurately estimate the electric bicycle retired battery health state under different working conditions and different battery types, and has high accuracy. This method can guarantee the safe operation of power equipment, and has great significance for reducing the development cost of electric bicycles and the popularization and reuse of retired batteries. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings. Among them:

[0043] Figure 1 The flowchart of the data-driven electric bicycle retired battery health state estimation method in Example 1.

[0044] Figure 2 The BP neural network structure diagram of the data-driven electric bicycle retired battery health state estimation method in Example 1.

[0045] Figure 3 The SoH estimation scheme verification flowchart of the data-driven electric bicycle retired battery health state estimation method in Example 1. DETAILED DESCRIPTION

[0046] In order to make the above-mentioned purposes, features and advantages of the application more apparent and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification.

[0047] In the following description, many specific details are set forth in order to provide a thorough understanding of the application, but the application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the application, therefore the application is not limited by the specific embodiments disclosed below.

[0048] Example 1

[0049] Reference Figures 1-3 , the first embodiment of the application provides a data-driven electric bicycle retired battery health state estimation method, which includes, as shown in Figure 1 :

[0050] Step 1: Obtain the electric bicycle battery charging voltage data through the maximum capacity test experiment, and use the numerical derivative method to obtain the incremental capacity IC curve, and filter and denoise the obtained IC curve.

[0051] The battery charging voltage data is a curve with charging voltage as the height and charging time as the length. The capacity IC curve is a curve with the derivative of the battery capacity to the battery voltage as the height and the charging voltage as the length. The selected data is a ternary lithium LiNCM battery with a capacity decay of less than 80%. Eight retired LiNCM battery modules are selected, and finally 110 battery monomers are obtained. The calculation formula is,

[0052]

[0053] Where N represents the sampling interval, Q represents the battery capacity, and V represents the battery voltage.

[0054] The IC curve is filtered to obtain a smooth IC curve. The filtering method used is Kalman filtering, which is based on the linear system state equation to achieve the optimal estimation of the current state according to the observation data. Since the observation data may contain system noise, etc.

[0055] Step 2: Based on the key points in the IC curve, part of the charging curve is cut off, and features related to battery life are extracted from the curve. Then, data standardization is performed, and an improved fuzzy C-means clustering algorithm is used to sort the retired batteries to verify the effectiveness of the extracted features.

[0056] The key points of the filtered IC curve are used to cut off the appropriate part of the charging curve. Then, features related to battery life are extracted from the curve. Four key features include:

[0057] Key voltage value (V): The peak position of the IC curve, or the central voltage value extracted from the part of the charging curve.

[0058] Gradient (G): The gradient of the curve, which is the slope value of the curve, which is roughly calculated using the following formula:

[0059]

[0060] Where t1 and t2 represent the start and end times of the charging curve, respectively. Since the ΔV of different battery monomers is set to the same value, G can be simplified to the time difference t2-t1.

[0061] Energy (E): The energy of the curve is the time integral of the voltage, which is represented by the following formula:

[0062]

[0063] Variance of oscillation (VO): The variance of oscillation of the line mainly explains the deviation from the average value of voltage, which can be expressed by the following formula,

[0064]

[0065] wherein μ and f represent the average value of voltage and the sampling frequency, respectively.

[0066] After obtaining each characteristic parameter, data normalization processing is performed, so that each index is on the same order of magnitude for comprehensive comparison. Data standardization can be expressed as:

[0067]

[0068] y is the characteristic value, and the value range after data standardization is 0-1.

[0069] The improved fuzzy C-means clustering algorithm includes using subtraction clustering (SUB) to find the best initial value of FCM and avoid converging to a local optimum. The extracted features related to battery life are used by the fuzzy C-means clustering algorithm to sort the retired batteries, and the effectiveness of the extracted aging features is verified.

[0070] Step 3: Construct a BP neural network to estimate the state of power battery health, determine the structure of the BP neural network, use the NSGA-II genetic algorithm to speed up the convergence speed of the BP neural network, and find the best initial weight suitable for the BP neural network to optimize the performance of the BP neural network.

[0071] As shown in Figure 2 , the structure of the BP neural network includes the transfer function, the number of hidden layers, and the number of nodes, and then the weights and thresholds are initialized.

[0072] Calculate the output of the hidden layer H f and the output layer O p , which is expressed by the formula,

[0073]

[0074] wherein w ij and w jp represent the weights of the hidden layer and the output layer, a j and b p represent the thresholds of the hidden layer and the output layer, l H and m0 represent the number of neurons in the hidden layer and the output layer, respectively, and x i represents the i-th aging feature of the input.

[0075] The default transfer function of the hidden layer and the output layer is the 'tansing' expression,

[0076]

[0077] Calculate the error e between the expected output and the actual output. p The specific formula is as follows:

[0078] e p =Y p -O p p = 1, 2, ..., m0

[0079] Among them, Y p Represented as the expected output, O p This represents the actual output.

[0080] The weights are corrected, and based on the calculation error, the weights are promptly calibrated to optimize the BP neural network. The formula is expressed as follows:

[0081]

[0082] Where η represents the learning rate.

[0083] The threshold is corrected by calibrating it promptly based on the calculation error. The formula is expressed as follows:

[0084]

[0085] Iterate until e p It meets the requirements of the set expected error value.

[0086] To accelerate the generation of the BP neural network, it is necessary to optimize it using an optimization algorithm. This invention proposes the NSGA-II-BP neural network, whose training process starts with a fast non-dominated genetic algorithm to globally search for weights and thresholds to find the best initial weights that fit the BP neural network. Then, the BP neural network starts the training process with the optimal initial values ​​provided by the fast non-dominated genetic algorithm to seek the optimal solution. The specific process is as follows.

[0087] Determine the topology of the neural network and initialize the threshold and weights.

[0088] Individuals in the population are encoded, and the individual encoding includes the weights from the input layer to the hidden layer, from the hidden layer to the output layer, and the thresholds of the hidden layer and the output layer.

[0089] The objective function of NSGA-II is defined as the expected value obtained by inputting training samples during the training of the BP network. Compared with the actual output value y i The squared error between (k) is expressed by the formula:

[0090]

[0091] Wherein, N represents the logarithm of network training sample data pairs, and s represents the number of nodes of network output.

[0092] The objective function optimization purpose is to obtain a minimum value, and the fitness function purpose is to obtain a maximum value, and the fitness function F is defined as the reciprocal of the objective function plus a small positive number ξ m , which is represented as,

[0093] F = (J m + ξ m ) -1 .

[0094] The population performs selection, crossover, mutation, non-dominated sorting, crowdedness calculation, and elite preservation operations, realizes the "survival of the fittest" of the population, and determines the best initial weight value adapted to the BP neural network.

[0095] The selection operation includes using the Monte Carlo method, that is, selecting according to the fitness size, and the probability p i of the individual being selected is represented as,

[0096]

[0097] Wherein, n represents the population size, and f i represents the fitness of individual i, and the fitness is proportional to the probability of the individual being selected.

[0098] The crossover includes using real number crossover between two individuals, and the individuals and after the crossover operation are represented as,

[0099]

[0100] Wherein, α represents a random number between [0, 1].

[0101] The mutation operation includes using a random number in any range to replace any gene in the individual code to complete mutation to obtain The mutation operation formula is represented as,

[0102]

[0103] Wherein, x min , x max respectively represent the minimum value and the maximum value of the gene in the individual, and β represents a random number between [0, 1].

[0104] The non-dominated sorting operation includes constructing a battery feature population with a size of n, setting i = 1, and for all j = 1,..., n and j ≠ i, comparing the individual x i and the individual xj The size of the fitness function between the individuals is compared to obtain the dominance and non-dominance relationship; if there is no individual x j better than x i , the individual x i is marked as a non-dominant individual; let i = i + 1, and go to the comparison of the individual x i+1 and the individual x j , until all the non-dominant individuals are found.

[0105] The obtained non-dominant individual set is marked as the first non-dominant layer of the population, the marked non-dominant individuals are ignored, and the optimization comparison is continued to obtain the second non-dominant layer, the marked non-dominant individuals are ignored, and the whole population is divided into non-dominant layers and dominant layers.

[0106] The calculation of the crowding degree includes initializing the crowding distance of any individual in the same layer as 0, i.e. n d = 0, arranging the individuals in the same layer in ascending order according to the fitness function value, setting the crowding degrees of the two individuals at the boundary as infinity, and obtaining the crowding degree of the individual in the middle of the arrangement, which is expressed by the formula,

[0107] n d = n d + (F(i+1)-F(i-1))

[0108] The elite reservation operation includes merging the parent population and the child population to form a new population, arranging the new population in order from low to high according to the non-dominant sorting level, and sequentially putting the whole layer population into the new parent population. When the individuals of the k-th level are all put into the new parent set, the number of individuals in the set is less than n, and when the individuals of the k+1-th level are all put into the new parent set, the number of individuals in the set is greater than n, the crowding degree of all individuals of the k+1-th level is calculated and arranged in descending order, and then all individuals with a level greater than k+1 are eliminated, and the size of the parent population is maintained as n.

[0109] The decoded weight and threshold value code is substituted into the fitness, and it is judged whether it meets the constraint condition. If it meets, the optimal weight and threshold value are returned, and the battery health state model is built. If it does not meet, the genetic optimization is continued.

[0110] Step 4: Based on the formed NSGA-II-BP neural network, the extracted aging characteristics of the retired battery are taken as the input of the NSGA-II-BP neural network, and the corresponding battery monomer capacity is taken as the output, and the battery health state estimation model is built.

[0111] Step 5: NSGA-II-BP neural network SoH estimation model of LiNCM battery is constructed, and the model accuracy is tested, and the effectiveness of the constructed SoH estimation model under different working conditions is verified based on the test data of the retired battery under different working conditions.

[0112] Different working conditions are different working temperatures of the battery and different charge and discharge rates of the battery. Based on the formed NSGA-II-BP neural network, the extracted aging characteristics of the retired battery are taken as the input of the NSGA-II-BP neural network, and the corresponding battery monomer capacity is taken as the output, so as to complete the construction of the battery health state estimation model. Figure 3 The test flow of the SoH estimation model is drawn to verify the effectiveness and universality of the proposed SoH estimation scheme under different environments. Firstly, for the retired LiNCM battery under the conventional environment (temperature 25℃, charge and discharge rate 1 / 3C), the aging characteristics are extracted, the NSGA-II-BP neural network SoH estimation model of the LiNCM battery is constructed, and the model accuracy is tested. In the test, 90 of 110 LiNCM retired batteries are selected as training, and 20 are selected as test, the test environment is 1 / 3C, 1 / 5C, 1 / 2C charge and discharge rate environment under 25℃, 1 / 3C charge and discharge rate under 10℃ and 40℃.

[0113] Step 6: The SoH estimation model is re-built for different types of LiFeO4 retired batteries, and the effectiveness and universality of the proposed SoH estimation scheme are verified.

[0114] The aging characteristics of the retired battery are taken as the input, and the selected total sample of the battery is 150 retired LiFePO4 batteries, of which 110 batteries are selected to train the SoH estimation model, and 40 batteries are used to test the model accuracy, and the experimental working condition is 25℃, 1 / 3C charge and discharge rate.

[0115] Embodiment 2

[0116] The second embodiment of the application is different from the first embodiment in that the data-driven electric bicycle retired battery health state estimation method further comprises, for the purpose of verifying the technical effects used in the method, the embodiment compares the test results by scientific demonstration means to verify the real effect of the method.

[0117] The LiNCM retired battery 110 is used for 110 sections, the LiFePO4 retired battery 150 is used for 150 sections, the data is tested using the NSGA-II-BP neural network, and the results in table 1 and table 2 are obtained. For the test results of the LiNCM retired battery, it can be seen that the capacity estimation error of the constructed SoH estimation model under different temperatures and different charge-discharge rates is less than 2%, the maximum error of capacity estimation under the conventional environment (25 DEG C, 1 / 3C) is 0.93%, and the error is less than 1%; for the test results of the LiFePO4 retired battery, it can be seen that the maximum error is less than 2%, the minimum error is only 0.29%, the average absolute error and the root mean square error are less than 1.50%, therefore, the proposed SoH estimation scheme is still applicable to the retired LiFePO4 battery, and has high accuracy and certain universality.

[0118] Table 1: Test results of LiNCM retired battery under different working conditions

[0119]

[0120] Table 2: Test results of LiFePO4 retired battery under different working conditions

[0121]

[0122] According to the experimental results of the above research materials, it can be seen that the present application can reduce background noise and influence, effectively extract the characteristics of the retired battery aging, and accurately estimate the health status of the retired battery of the electric bicycle under different working conditions and different battery types, and has high accuracy. This method can guarantee the safe operation of the power equipment, and has great significance for reducing the development cost of the electric bicycle and the popularization and reuse of the retired battery.

[0123] Although the present application is described above in conjunction with the drawings, the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative but not limiting, and those skilled in the art can make many modifications under the inspiration of the present application without departing from the purpose of the present application, and these all belong to the protection of the present application.

[0124] Example 3

[0125] The third embodiment of the present application, which is different from the first two embodiments, is a data-driven electric bicycle retired battery state of health estimation method system, characterized in that it comprises a data preprocessing module, a feature extraction module, an NSGA-II-BP neural network construction module and a SoH estimation model evaluation module; the data preprocessing module is used for preprocessing the IC curve, obtaining the electric bicycle battery charging voltage data through the maximum capacity test experiment, and obtaining the incremental capacity IC curve by using the numerical derivation method, and filtering and denoising the obtained IC curve; the feature extraction module is used for extracting the features related to the battery life, intercepting part of the charging curve based on the key points in the IC curve, and extracting the features related to the battery life from the curve, and then performing data standardization, and using the improved fuzzy C-means clustering algorithm to sort the retired batteries to verify the effectiveness of the extracted features; the NSGA-II-BP neural network construction module is used for constructing the NSGA-II-BP neural network model, constructing the BP neural network to estimate the state of health of the power battery, determining the structure of the BP neural network, using the NSGA-II genetic algorithm to speed up the convergence speed of the BP neural network, and finding the best initial weight suitable for the BP neural network to optimize the performance of the BP neural network; based on the formed NSGA-II-BP neural network, the extracted retired battery aging features are taken as the input of the NSGA-II-BP neural network, and the corresponding battery monomer capacity is taken as the output, and the battery state of health estimation model is built; the SoH estimation model evaluation module is used for evaluating the SoH model, constructing the NSGA-II-BP neural network SoH estimation model of different types of batteries, and testing the accuracy of the model, based on the retired battery test data under different working conditions, verifying the effectiveness of the constructed SoH estimation model under different working conditions.

[0126] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the prior art that essentially contribute or the parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes instructions for making a computer device (which can be a personal computer, a server or a network device) execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk and various program code storage media.

[0127] The logic and / or steps represented in the flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber (optical), and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In this context, a "computer-readable medium" can be any means that can store the program for use by or in connection with the instruction execution system, apparatus, or device.

[0128] The computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer.

[0129] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following techniques, which are well known in the art of making integrated circuits, can be used alone or in any combination to implement the application: a discrete logic circuit(s) having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

[0130] It should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, but not limit the present application, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced, without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of claims of the present application.

Claims

1. A data-driven method for estimating the health status of retired electric bicycle batteries, characterized by: include, The charging voltage data of electric bicycle batteries were obtained through maximum capacity test experiments, and the incremental capacity IC curve was obtained by numerical differentiation method. The obtained IC curve was then filtered and noise-reduced. Based on key points in the IC curve, a portion of the charging curve is extracted, and features related to battery life are extracted from the curve. Then, the data is standardized, and an improved fuzzy C-means clustering algorithm is used to sort retired batteries to verify the effectiveness of the extracted features. A BP neural network was constructed to estimate the health state of the power battery. The structure of the BP neural network was determined, the NSGA-II genetic algorithm was used to accelerate the convergence speed of the BP neural network, and the optimal initial weights that fit the BP neural network were found to optimize the performance of the BP neural network. Based on the established NSGA-II-BP neural network, the extracted aging features of retired batteries are used as the input of the NSGA-II-BP neural network, and the corresponding battery cell capacity is used as the output to complete the construction of the battery health status estimation model. We constructed NSGA-II-BP neural network SoH estimation models for different types of batteries and tested the accuracy of the models. Based on test data of retired batteries under different operating conditions, we verified the effectiveness of the constructed SoH estimation models under different operating conditions.

2. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 1, characterized in that: The battery charging voltage data is a curve with charging voltage as the height and charging time as the length; the capacity IC curve is a curve with the derivative of battery capacity with respect to battery voltage as the height and charging voltage as the length, and the calculation formula is expressed as follows: Where N represents the sampling interval, Q represents the battery charge, and V represents the battery voltage.

3. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 2, characterized in that: The extracted features related to battery life include key voltage values, gradients, energy, and volatility; The improved fuzzy C-means clustering algorithm includes: using subtractive clustering to find the optimal initial value of FCM; using the extracted battery life-related features to sort retired batteries using the fuzzy C-means clustering algorithm; and verifying the effectiveness of the extracted aging features.

4. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 3, characterized in that: The structure of the BP neural network includes a transfer function, the number of hidden layers, and the number of nodes, followed by initialization of weights and thresholds; Calculate hidden layer H f and output layer O p The output, expressed by the formula, is: Among them, w ij and w jp Let a represent the weights of the hidden layer and the output layer, respectively. j and b p Let l represent the threshold values ​​for the hidden layer and the output layer, respectively. H Let m0 and x represent the number of neurons in the hidden layer and the output layer, respectively. i This is represented by the i-th aging feature in the input; The default transfer function for the hidden and output layers is tansing, and the expression is: Calculate the error e between the expected output and the actual output. p The specific formula is as follows: e p =Y p -O p p=1,2,...,m0 Among them, Y p Represented as the expected output, O p This represents the actual output; The weights are corrected, and based on the calculation error, the weights are promptly calibrated to optimize the BP neural network. The formula is expressed as follows: Where η represents the learning rate; The threshold is corrected by calibrating it promptly based on the calculation error. The formula is expressed as follows: Iterate until e p It meets the requirements of setting the expected error value.

5. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 4, characterized in that: The method to accelerate the convergence speed of the BP neural network includes starting with the fast non-dominated genetic algorithm NSGA-II, globally searching for weights and thresholds to determine the optimal initial weights that fit the BP neural network; the BP neural network starts the training process with the optimal initial values ​​provided by the fast non-dominated genetic algorithm to seek the optimal solution; The process starting with the fast nondominated genetic algorithm NSGA-II includes determining the topology of the neural network and initializing the threshold and weights. The global search weights and thresholds include encoding individuals in the population, whereby the individual encoding includes weights from the input layer to the hidden layer, from the hidden layer to the output layer, and thresholds for the hidden layer and the output layer. The objective function of NSGA-II is defined as the expected value obtained by inputting training samples during the training of the BP network. Compared with the actual output value y i The squared error between (k) is expressed by the formula: Where N represents the number of training sample data pairs for the network, and s represents the number of output nodes of the network; The objective function aims to obtain a minimum value, while the fitness function aims to obtain a maximum value. The fitness function F is defined as the objective function plus a small positive number ξ. m The reciprocal of is represented as . F=(J m +ξ m ) -1 。 6. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 5, characterized in that: The global search weights and thresholds also include the following operations: population selection, crossover, mutation, non-dominated sorting, crowding calculation, elite retention, etc., to determine the optimal initial weights that are compatible with the BP neural network. The selection operation includes using the Monte Carlo method, that is, selecting according to fitness, where the probability p of an individual being selected is... i Represented as, Where n represents the group size, f i This represents the fitness of individual i, and fitness is directly proportional to the probability that an individual is selected. The crossing includes using real-number crossing between two individuals, where the individuals... and The individual generated after the crossover operation is represented as follows. Where α represents a random number between [0,1]; The mutation operation includes replacing any gene in the individual's coding system with a random number within an arbitrary range to complete the mutation. The mutation operation formula is expressed as follows: Where, x min x max Let represent the minimum and maximum values ​​of genes in an individual, respectively, and let β represent a random number between [0,1]. The non-dominated sorting operation includes constructing a battery feature population of size n, where i = 1, and for all j = 1, ..., n and j ≠ i, comparing individual x. i and individual x j The relationship between the fitness functions of individuals is used to determine dominance and non-dominance; if no individual x exists... j Better than x i Then mark x i For non-dominated individuals; let i = i + 1, then switch to individual x. i+1 and individual x j The comparison continues until all non-dominated individuals are found; The obtained set of non-dominated individuals is marked as the first level of non-dominated layer of the population. The marked non-dominated individuals are ignored. The optimization comparison is continued to obtain the second level of non-dominated layer. The marked non-dominated individuals are ignored until the entire population is divided into non-dominated layer and dominant layer. The calculation of crowding degree includes initializing the crowding distance of any individual in the same layer to 0, i.e., n d =0, arrange individuals in the same layer in ascending order of fitness function value, set the crowding degree of the two boundary individuals to infinity, and obtain the crowding degree of the individuals in the middle of the sort. The formula is expressed as, n d =n d +(F(i+1)-F(i-1)) The elite retention operation includes merging the parent and offspring populations to form a new population, sorting the new population in ascending order of non-dominant sorting level, and sequentially placing the entire population into the new parent generation. When all individuals of level k are placed into the new parent generation set, the number of individuals in the set is less than n, and when all individuals of level k+1 are placed into the new parent generation set, the number of individuals in the set is greater than n. Then, the crowding degree of all individuals of level k+1 is calculated and sorted in descending order. Finally, all individuals of level greater than k+1 are eliminated to maintain the parent population size of n. The decoded weights and thresholds are encoded and substituted into the fitness to determine whether they meet the constraints. If they do, the optimal weights and thresholds are returned to build the battery health status model. If they do not meet the constraints, the genetic optimization continues in a loop.

7. The data-driven method for estimating the health status of retired electric bicycle batteries as described in claim 6, characterized in that: The different operating conditions refer to different battery operating temperatures and different charge / discharge rates.

8. A system employing the data-driven method for estimating the health status of retired electric bicycle batteries as described in any one of claims 1 to 7, characterized in that: It includes a data preprocessing module, a feature extraction module, an NSGA-II-BP neural network construction module, and an SoH estimation model evaluation module; The data preprocessing module is used to preprocess the IC curve, obtain the electric bicycle battery charging voltage data through the maximum capacity test experiment, and use the numerical differentiation method to obtain the incremental capacity IC curve, and perform filtering and noise reduction processing on the obtained IC curve. The feature extraction module is used to extract features related to battery life. It extracts a portion of the charging curve based on key points in the IC curve and extracts features related to battery life from the curve. Then, it performs data standardization and uses an improved fuzzy C-means clustering algorithm to sort retired batteries to verify the effectiveness of the extracted features. The NSGA-II-BP neural network construction module is used to construct the NSGA-II-BP neural network model. The BP neural network is constructed to estimate the health state of the power battery. The structure of the BP neural network is determined, the NSGA-II genetic algorithm is used to accelerate the convergence speed of the BP neural network, and the optimal initial weights that are suitable for the BP neural network are found to optimize the performance of the BP neural network. Based on the formed NSGA-II-BP neural network, the extracted aging features of retired batteries are used as the input of the NSGA-II-BP neural network, and the corresponding battery cell capacity is used as the output to complete the construction of the battery health state estimation model. The SoH estimation model evaluation module is used to evaluate the SoH model, construct NSGA-II-BP neural network SoH estimation models for different types of batteries, test the accuracy of the models, and verify the effectiveness of the constructed SoH estimation model under different operating conditions based on retired battery test data under different operating conditions.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the data-driven method for estimating the health status of retired electric bicycle batteries according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the data-driven method for estimating the health status of retired electric bicycle batteries according to any one of claims 1 to 7.

Images