Lithium battery SOH estimation and RUL prediction method based on combination of improved Archimedes algorithm and BiGRU

Abstract

The invention discloses an improved Archimedes algorithm and BiGRU combined lithium battery SOH estimation and RUL prediction method, and relates to the technical field of new energy automobiles and power grid energy storage. Extracting characteristic parameters containing battery aging state information as the input of a prediction model through a battery modeling and data driving method; correlation analysis is carried out on the extracted characteristic parameters, characteristic quantities with high correlation degree are reserved, and data preprocessing is carried out; inputting the screened characteristic quantity into a BiGRU model to carry out SOH estimation and RUL prediction; and optimizing the BiGRU model by using an improved Archimedes algorithm. According to the lithium battery SOH estimation and RUL prediction method combining the improved Archimedes algorithm and the BiGRU, a lithium battery SOH prediction and RUL estimation model based on an improved Archimedes algorithm optimized Bi GRU neural network is provided in the aspect of model solving, so that the model can better fit training data, the learning ability of the model is improved, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of new energy vehicles and power grid energy storage technology, and in particular to a lithium battery SOH estimation and RUL prediction method combining an improved Archimedean algorithm with a BiGRU. Background Art

[0002] With the development of economy, global energy consumption is increasing, but traditional energy such as coal, oil and natural gas are non-renewable, and the environmental problems caused by their use are becoming more and more serious, which seriously hinders the sustainable development of human society. Electricity is the representative of new energy, and batteries, as the carrier of electric energy, can help alleviate the problems of energy shortage and environmental pollution by improving their application technology. Therefore, it has become an inevitable trend for new energy to replace traditional mineral energy. The transformation of battery energy storage has also greatly increased the scale of energy storage lithium batteries, but even the best lithium batteries will face failure and aging problems.

[0003] As the use of lithium batteries increases, people are paying attention to potential safety hazards during their use. Monitoring and evaluating the battery status can further guide the operation and maintenance of the battery, which can prevent safety accidents caused by overcharging and voltage increase, resulting in internal short circuits, and is of great significance to ensuring the safe, stable and reliable operation of the battery management system. Summary of the invention

[0004] The purpose of the present invention is to provide a lithium battery SOH estimation and RUL prediction method combining an improved Archimedes algorithm with BiGRU. Aiming at the problem of incomplete screening of key influencing factors of the health status of lithium batteries, the working principle and aging mechanism of lithium batteries are studied. The SOH and RUL of lithium batteries cannot be directly obtained during actual use. The health status and remaining service life indicators are quantified through battery modeling and data-driven methods, and multiple characteristic parameters such as capacity increment peak, differential voltage inflection point, constant current charging time, etc. containing battery aging status information are extracted as inputs of the prediction model, and a SOH estimation and RUL prediction model based on the improved Archimedes algorithm and BiGRU neural network is established.

[0005] To achieve the above object, the present invention provides a lithium battery SOH estimation and RUL prediction method combining an improved Archimedes algorithm with BiGRU, comprising the following steps:

[0006] Step S1: studying the working principle and aging mechanism of lithium batteries, and extracting characteristic parameters containing battery aging status information as input of the prediction model through battery modeling and data-driven methods;

[0007] Step S2: performing correlation analysis on the extracted feature parameters, retaining the feature quantities with high correlation, and performing data preprocessing to improve the availability of the data;

[0008] Step S3: Input the selected feature quantities into the BiGRU model for SOH estimation and RUL prediction;

[0009] Step S4: Use the improved Archimedean algorithm to optimize the BiGRU model in order to minimize the loss function and improve the accuracy of the prediction results.

[0010] Preferably, the specific steps of step S1 are:

[0011] Step S11, in view of the problem of incomplete screening of key factors affecting the health status of lithium batteries, the main factors leading to battery aging and capacity decay mechanism are summarized: the electrochemical reactions occurring during the operation of lithium batteries include not only redox reactions of lithium ion insertion and deinsertion, but also side reactions, which cause internal aging and capacity decay of lithium batteries as the number of battery cycles increases;

[0012] Step S12, extracting characteristic parameters for estimating the SOH of the lithium battery from the lithium battery experimental data, taking the voltage, current, temperature parameter curves, the capacity increment curve, and the differential voltage curve as the analysis objects;

[0013] Step S13, the capacity increment analysis method studies the IC curve to understand the changes in the internal state of the battery during aging, and converts the slowly changing charging voltage platform into an IC peak by drawing the IC curve;

[0014] Step S14, using the data of the measured voltage not rising to the charging cut-off voltage to draw a voltage-capacity curve, and using the ampere-hour integration method to obtain the battery capacity during charging;

[0015] Step S15, the parameters in the data set are discrete data, and the integral operation in the ampere-hour integration method is converted into a superposition operation;

[0016] Step S16, the IC curve describes the relationship between dQ / dV-V, and the voltage-capacity curve is fitted and derived using the python software functions np.polyfit and np.poly1d.

[0017] Preferably, in step S14, the battery capacity during charging is obtained by the ampere-hour integration method, and the expression is as follows:

[0018]

[0019] Wherein, T represents the time when discharge ends, and i represents the discharge current at time t.

[0020] Preferably, in step S15, the integral operation in the ampere-hour integration method is converted into a superposition operation, and the expression is as follows:

[0021]

[0022] Wherein, I(t) represents the current value at the tth time point, T(t) represents the time at time t, T(t-1) represents the time at time t-1, and I(0) represents the current value at the initial time.

[0023] Preferably, the specific steps of step S2 are as follows:

[0024] Step S21, using the Z-score standardization method to process features;

[0025] Step S22, testing the correlation between continuous variables by Pearson correlation coefficient, extracting and retaining the feature quantities with high correlation;

[0026] Step S23: using a box plot method to remove abnormal values, removing outliers, and treating the removed data locations as missing values;

[0027] Step S24: Use the KNN interpolation algorithm to measure and identify adjacent points, and use adjacent observations to estimate missing values.

[0028] Preferably, in step S21, the Z-score normalization method is used to process the features, and the expression is as follows:

[0029]

[0030] In the formula, z represents the standardized value of Z-score. represents the original data value to be standardized, μ represents the mean of the data, and σ represents the standard deviation of the data;

[0031] In step S22, the correlation between continuous variables is tested by the Pearson correlation coefficient, and the expression is as follows:

[0032]

[0033] In the formula, represents the mean of variable x, x i represents the i-th observation in the sample data set x, represents the mean of variable y, y i represents the i-th observation in the sample data set y, and n represents the total number of samples;

[0034] In step S23, a box plot method is used to remove outliers, and the expression is as follows:

[0035] IQR = Q3-Q1;

[0036] UpLimit=Q3+IQR×1.5;

[0037] LowLimit = Q1-IQR×1.5;

[0038] x>UpLimit&x<LowLimit;

[0039] In the formula, IQR means, Q3 is the 75% quantile, Q1 is the 25% quantile, x is the data in the data set, UpLimit and LowLimit are the upper and lower limits of the box plot, and when the upper and lower limits are exceeded, it is considered as outlier data;

[0040] In step S24, adjacent observations are used to estimate missing values, and the expression is as follows:

[0041]

[0042] Where L2(x1,x2) represents the Euclidean distance between vectors x1 and x2; 1n represents the nth component in vector x1, x 2n Represents the nth component in vector x2, where N represents the number of dimensions of the vector.

[0043] Preferably, the specific steps of step S3 are as follows:

[0044] Step S31, calculating the GRU neural network update gate, reset gate and new candidate state;

[0045] Step S32, BiGRU combines the output results of the input sequences in two directions to generate a final output result.

[0046] Preferably, the calculation of the GRU unit in step S31 is represented by the following formula:

[0047] Update Gate:

[0048] z t =σ(W Z ·[h t-1 ,x t ]);

[0049] In the formula, σ represents the activation function, W Z represents the weight parameter of the update gate, x t and h t-1 Represent the input and hidden states at time step t respectively;

[0050] Reset the gate:

[0051] r t =σ(W r ·[h t-1 ,x t ]);

[0052] Where W r Represents the weight parameter of the reset gate;

[0053] New candidate status:

[0054]

[0055] Where: W h represents the weight parameter of the candidate hidden state, ⊙ represents the corresponding position multiplication of the matrix;

[0056] Update hidden state:

[0057]

[0058] The final output result in step S32 is expressed as follows:

[0059]

[0060] In the formula, and are the weights of the forward and backward hidden layers, respectively, both are constant values, and their sum is 1.

[0061] Preferably, the specific steps of step S4 include:

[0062] Step S41, population initialization, setting the initial parameters of the optimization algorithm, using the Sin chaos model to generate the initial population, and improving the population traversal capability;

[0063] Step S42, calculating the transfer operator TF and the density factor d;

[0064] Step S43: If TF≤0.5, it is the initial exploration stage of the population, and the acceleration of the object is obtained;

[0065] Step S44, if TF>0.5, it is the initial development stage of the population, and the acceleration of the object is obtained;

[0066] Step S45: Select the current individual and the global optimal individual for arithmetic crossover to generate a new offspring individual closer to the current optimal solution, thereby accelerating the group to move closer to the global optimal area;

[0067] Step S46: adopting the Golden Levy guidance mechanism so that when searching in an unknown range, a larger range can be reached, thereby making the algorithm jump out of the local optimum;

[0068] Step S47, introducing the relationship between the sine function and the unit circle, so that the population can traverse all points on the sine function;

[0069] Step S48: Add a greedy mechanism after the guiding mechanism to decide whether to update the target position by comparing the individual fitness before and after the individual position is updated, so as to retain the individuals with better fitness;

[0070] Step S49, determine whether the maximum number of iterations has been reached or the optimal solution has been found. If the optimal solution has been found, output the decision variable position. If the above conditions are not met, return to step S42 until the optimal solution is found.

[0071] Step S410: Using multiple characteristic parameters including the capacity increment peak, differential voltage inflection point, constant current charging time and other characteristic parameters containing battery aging status information as inputs of the prediction model, a SOH estimation and RUL prediction model based on the improved BiGRU is established.

[0072] Preferably, the Sin chaotic one-dimensional mapping expression in step S41 is as follows:

[0073]

[0074] Where, X n+1 represents the result of the function at time step n+1, X n Represents the value at the current time n, and N represents the number of iterations;

[0075] In step S42, the expression of the transfer operator TF is as follows:

[0076]

[0077] In the formula, t represents the current time, t max Indicates the maximum value of time;

[0078] The expression of density factor d is as follows:

[0079]

[0080] In step S43, during the initial exploration phase of the population, the acceleration update formula of the object is as follows:

[0081]

[0082] Where, den mr Indicates the density value at the current moment, vol mr Indicates the volume value at the current moment, acc mr Indicates the acceleration value at the current moment. represents the density value at time t+1;

[0083] In step S44, in the initial population development stage, the acceleration update formula of the object is as follows:

[0084]

[0085] Where, den best represents the optimal density, vol best represents the optimal volume, acc bestrepresents the optimal acceleration;

[0086] In step S45, the arithmetic crossover operator expression is as follows:

[0087]

[0088] In the formula, λ represents the weight parameter, represents the position of object i at time t+1, X best Indicates the best position;

[0089] In step S46, the calculation formula of the step length s is:

[0090] s=μ / |ν| 1 / β ;

[0091] In the formula, μ is a constant, v represents the speed, and β is an exponential constant;

[0092] In step S47, the individual position update formula is:

[0093]

[0094] Where, X t represents the position of a vector, represents the optimal position vector, represents the position of the current vector, θ1 and θ2 represent the golden section coefficient τ, dis represents the distance between the individual and the optimal position, R1 and R2 represent random values, and s represents the step size;

[0095] In step S48, the greedy mechanism expression is as follows:

[0096]

[0097] In the formula, fit(.) represents the fitness function.

[0098] Therefore, the present invention adopts the lithium battery SOH estimation and RUL prediction method combining the above-mentioned improved Archimedean algorithm with BiGRU, which has the following beneficial effects:

[0099] (1) This method quantifies the health status and remaining service life indicators through battery modeling and data-driven methods, extracts multiple characteristic parameters such as capacity increment peak, differential voltage inflection point, constant current charging time, etc. containing battery aging status information as input of the prediction model, and uses Z-score, box plot and KNN interpolation algorithm to process the characteristic data to improve data availability.

[0100] (2) In terms of model solving, this method proposes a lithium battery SOH prediction and RUL estimation model based on the improved Archimedean algorithm to optimize the BiGRU neural network, so that the model can better fit the training data, increase the model's learning ability, and improve the prediction accuracy.

[0101] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 A flowchart of an embodiment of a lithium battery SOH estimation and RUL prediction method combining an improved Archimedean algorithm with BiGRU according to the present invention;

[0103] Figure 2 This is a flow chart of an improved Archimedes algorithm of an embodiment of a lithium battery SOH estimation and RUL prediction method combining an improved Archimedes algorithm with BiGRU according to the present invention. DETAILED DESCRIPTION

[0104] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0105] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0106] Example

[0107] See also Figure 1-2 , a lithium battery SOH estimation and RUL prediction method combining an improved Archimedes algorithm and BiGRU is provided, characterized in that it comprises the following steps:

[0108] Step S1: Study the working principle and aging mechanism of lithium batteries, and extract characteristic parameters containing battery aging status information as input of the prediction model through battery modeling and data-driven methods. The specific steps are as follows:

[0109] Step S11, in view of the problem of incomplete screening of key factors affecting the health status of lithium batteries, the main factors leading to battery aging and capacity decay mechanism are summarized: the electrochemical reactions occurring during the operation of lithium batteries include not only redox reactions of lithium ion insertion and deinsertion, but also side reactions, which cause irreversible aging and capacity decay inside the lithium battery as the number of battery cycles increases;

[0110] Step S12, extracting characteristic parameters for estimating the SOH of the lithium battery from the lithium battery experimental data, taking the voltage, current, temperature parameter curves, the capacity increment curve, and the differential voltage curve as the analysis objects;

[0111] Step S13, the capacity increment analysis method studies the IC curve to understand the changes in the internal state of the battery during aging, and converts the slowly changing charging voltage platform into an IC peak by drawing the IC curve;

[0112] Step S14: When the battery is charged for a period of time and the rising rate of the charging measurement voltage decreases, a voltage-capacity curve is drawn using the data that the measured voltage has not risen to the charging cut-off voltage, and the battery capacity during charging is obtained by the ampere-hour integration method, and the expression is as follows:

[0113]

[0114] Wherein, T represents the time when discharge ends, and i represents the discharge current at time t.

[0115] Step S15, the parameters in the data set are discrete data, and the integral operation in the ampere-hour integration method is converted into a superposition operation; the expression is as follows:

[0116]

[0117] Wherein, I(t) represents the current value at the tth time point, T(t) represents the time at time t, T(t-1) represents the time at time t-1, and I(0) represents the current value at the initial time.

[0118] Step S16, the IC curve describes the relationship between dQ / dV-V, and the voltage-capacity curve is fitted and derived using the python software functions np.polyfit and np.poly1d.

[0119] Step S2: perform correlation analysis on the extracted feature parameters, retain the feature quantities with high correlation, and perform data preprocessing to improve the availability of the data. The specific steps are as follows:

[0120] Step S21, using the Z-score standardization method to process features; in step S21, the Z-score standardization method is used to process features, and the expression is as follows:

[0121]

[0122] In the formula, z represents the standardized value of Z-score. represents the original data value to be standardized, μ represents the mean of the data, and σ represents the standard deviation of the data;

[0123] Step S22: The correlation between continuous variables is tested by the Pearson correlation coefficient, with a value range of [-1, 1]. The feature quantities with high correlation are extracted and retained. The Pearson correlation coefficient between samples is obtained by estimating the covariance and standard deviation of the samples. The expression is as follows:

[0124]

[0125] In the formula, represents the mean of variable x, x i represents the i-th observation in the sample data set x, represents the mean of variable y, y i represents the i-th observation in the sample data set y, and n represents the total number of samples;

[0126] Step S23: Use the box plot method to remove outliers. Remove the outliers and regard the removed data positions as missing values. The upper and lower limits of the box plot are calculated as follows:

[0127] IQR = Q3-Q1;

[0128] UpLimit=Q3+IQR×1.5;

[0129] LowLimit = Q1-IQR×1.5;

[0130] x>UpLimit&x<LowLimit;

[0131] In the formula, IQR means, Q3 is the 75% quantile, Q1 is the 25% quantile, x is the data in the data set, UpLimit and LowLimit are the upper and lower limits of the box plot, and when the upper and lower limits are exceeded, it is considered as outlier data;

[0132] Step S24, select the KNN interpolation algorithm to measure and identify adjacent points, and use adjacent observations to estimate missing values. In the feature analysis in the previous section, the feature data set has been Z-score standardized, so the KNN interpolation algorithm with Euclidean distance as the measurement method can be directly used. Assuming that the dimension of the multidimensional space is N, x1 and x2 are two sample points, then the Euclidean distance L2 between x1 and x2 is, and the expression of the KNN interpolation algorithm is as follows:

[0133]

[0134] Where L2(x1,x2) represents the Euclidean distance between vectors x1 and x2; 1n represents the nth component in vector x1, x 2n Represents the nth component in vector x2, where N represents the number of dimensions of the vector.

[0135] Step S3: Input the selected feature quantity into the BiGRU model for SOH estimation and RUL prediction. The GRU neural network memorizes valuable information based on the output state of the previous node to learn the time dependency in the sequence. It is a recurrent neural network with state memory and multi-layer chain structure. The GRU neural network contains only two gate control structures: update gate z t and reset gate r t , these gates control the information flow within the GRU and help capture relevant temporal patterns. The specific steps are as follows:

[0136] Step S31, calculate the GRU neural network update gate, reset gate and new candidate state, and express the calculation of the GRU unit by the following formula:

[0137] Update Gate:

[0138] z t =σ(W Z ·[h t-1 ,x t ]);

[0139] In the formula, σ represents the activation function, W Z represents the weight parameter of the update gate, x t and h t-1 Represent the input and hidden states at time step t respectively;

[0140] Reset the gate:

[0141] r t =σ(W r ·[h t-1 ,x t ]);

[0142] Where W r Represents the weight parameter of the reset gate;

[0143] New candidate status:

[0144]

[0145] Where: W h represents the weight parameter of the candidate hidden state, ⊙ represents the corresponding position multiplication of the matrix;

[0146] Update hidden state:

[0147]

[0148] The final output result in step S32

[0149] Step S32, BiGRU combines the forward and backward information of the gated recurrent unit (GRU) neural network and connects the corresponding states together to obtain the final output, which increases the learning ability of the model and improves the prediction accuracy. BiGRU uses two GRUs to calculate the hidden vectors in the GRU units of forward propagation and backward propagation. and BiGRU combines the output results of the input sequences in two directions to produce the final output result y t , the expression is as follows:

[0150]

[0151] In the formula, and are the weights of the forward and backward hidden layers, respectively, both are constant values, and their sum is 1.

[0152] Step S4: Use the improved Archimedean algorithm to optimize the BiGRU model in order to minimize the loss function and improve the accuracy of the prediction results. The specific steps are as follows:

[0153] BiGRU requires an optimization algorithm to adjust its parameters in order to minimize the loss function and improve prediction accuracy. The BiGRU model contains a large number of learnable parameters, such as weight matrices, bias terms, and parameters of the gating mechanism, which determine how the model processes inputs and generates outputs. The optimization algorithm calculates the gradient of the loss function with respect to each parameter through backpropagation, and uses optimization methods such as gradient descent to adjust the parameters so that the model can better fit the training data.

[0154] The basic principle of Archimedes algorithm is to search for the optimal solution by simulating a spiral. The basic steps are as follows:

[0155] Population initialization:

[0156] O i =lb i +rand*(ub i -lb i ),i=1,2,...,N;

[0157] In the formula, O i represents the initial position of the ith particle or solution, lb i The lower bound of the i-th parameter, ub i represents the upper bound of the i-th parameter, and rand represents a random number between 0 and 1, which is used to randomly initialize the solution within the specified upper and lower bounds;

[0158] Initialize the volume (vol) and density (den) of each i-th object:

[0159]

[0160] Initialize the acceleration of the i-th object:

[0161] acc i =lb i +rand*(ub i -lb i );

[0162] Where ub and lb are the upper and lower boundaries of the variable.

[0163] Update volume, density:

[0164]

[0165] In the formula, rand represents a random number between 0 and 1;

[0166] Step S42, calculating the transfer operator TF and the density factor d; in the step S42, the expression of the transfer operator TF is as follows:

[0167]

[0168] In the formula, t represents the current time, t max Indicates the maximum value of time;

[0169] The expression of density factor d is as follows:

[0170]

[0171] Step S43: If TF≤0.5, it is the initial exploration stage of the population, and collisions occur between objects. A random material is selected and the acceleration of the object at iteration t+1 is updated:

[0172]

[0173] Where, den mr Indicates the density at the current moment, vol mr Indicates the volume at the current moment, acc mr Represents the acceleration parameter at the current moment, represents the density at time t+1;

[0174] Step S44: If TF>0.5, it is the initial development stage of the population, and the acceleration of the object at iteration t+1 is updated:

[0175]

[0176] Where, den best represents the optimal density, vol best represents the optimal volume, acc best represents the optimal acceleration;

[0177] Normalization:

[0178]

[0179] In the formula, It represents the normalized acceleration of the ith object at time t+1, u represents the normalized upper bound, which is usually 1, and l represents the normalized lower bound, which is usually 0.

[0180] Object position update:

[0181] If TF≤0.5, the object position is updated at the i-th t+1 iteration using:

[0182]

[0183] In the formula, represents the position vector of the i-th individual at time t, C1 is the control coefficient, rand represents a random number between 0 and 1, acc is the acceleration factor, and d is the distance metric;

[0184] If TF>0.5, the object position is updated at the i-th t+1 iteration using:

[0185]

[0186] In the formula, F is the control factor, C2 is the control coefficient, d is the distance measure, and T is the correction factor;

[0187] in,

[0188]

[0189] In the formula, P represents the probability factor, which determines the value of F.

[0190] In the original Archimedean algorithm, the population is initialized by random distribution, which leads to a certain degree of blindness in the early search of individuals, thus making the algorithm converge slowly; in the global development stage, the Archimedean algorithm only relies on random individuals to lead the population to find the optimal solution in the optimal area, which will lead to low accuracy of the algorithm; in the local development stage, when the optimal individual falls into the local extreme value space, the population will also fall into the local optimum, causing the algorithm to stagnate. Therefore, the Sin chaos model, arithmetic crossover operator and golden Levy guidance mechanism are used to improve it.

[0191] Step S41, population initialization, setting the initial parameters of the optimization algorithm, using the Sin chaos model to generate the initial population, and improving the population traversal capability.

[0192] The Sin chaos model is a chaotic model with good ergodicity and randomness and infinite mapping folding times. Its basic principle is to generate a chaotic sequence between the chaotic variable space [0,1] through a mapping relationship, and then transform it into the individual optimization variable space. The one-dimensional mapping expression of Sin chaos is as follows:

[0193] The expression of Sin chaotic one-dimensional mapping is as follows:

[0194]

[0195] Where, X n+1 represents the value of step n+1, X n represents the value of the nth step, and N represents the total number of steps in the sequence;

[0196] Arithmetic crossover operator: Select the current individual and the global optimal individual for arithmetic crossover, and generate new offspring individuals closer to the current optimal solution, thereby accelerating the group to move closer to the global optimal area. At the same time, the arithmetic crossover operator gives the current individual the ability to learn from excellent individuals, enhances the population's ability to share information, and thus increases population diversity. The expression of the arithmetic crossover operator is as follows:

[0197]

[0198] In the formula, λ represents the weight factor, represents the solution of the ith individual at step t+1, X best represents the global optimal solution among all current individuals;

[0199] Golden Levy Guidance Mechanism: Levy step is a random walk that obeys Levy distribution. Its alternating characteristics of short and long distance steps enable it to reach a larger range when searching in an unknown range, thus helping the algorithm to escape from the local optimum. The calculation formula of step length s is:

[0200] s=μ / |ν| 1 / β ;

[0201] In the formula, μ is a constant, v represents the speed, and β is an exponential constant;

[0202] Where μ and ν follow a normal distribution:

[0203]

[0204] In the formula, σ μ Usually it is taken as 1, Γ(β) is the Gamma function.

[0205] Step S47, introducing the relationship between the sine function and the unit circle, so that the population can traverse all points on the sine function, the individual position update formula is:

[0206]

[0207] Where, X t represents the position of a vector, represents the optimal position vector, Represents the position of the current vector, dis represents the distance between the individual and the optimal position, θ1 and θ2 represent the golden section coefficients τ, the purpose of which is to narrow the search space so that the algorithm can fully search the area that can produce excellent solutions in each iteration, thereby accelerating the convergence of the algorithm; R1 represents a random number, and R1∈[0,2π], R1 and Levy step size s jointly determine the search radius; R2 represents a random number, and R2∈[0,π], R2 determines the direction of the individual position update.

[0208] The specific parameter expressions in the formula are as follows:

[0209]

[0210] Step S48: Add a greedy mechanism after the guiding mechanism, and decide whether to update the target position by comparing the individual fitness before and after the individual position is updated, so as to retain the individuals with better fitness; in step S48, the greedy mechanism is expressed as follows:

[0211]

[0212] In the formula, fit(.) represents the fitness function.

[0213] Step S49, determine whether the maximum number of iterations has been reached or the optimal solution has been found. If the optimal solution has been found, output the decision variable position. If the above conditions are not met, return to step S42 until the optimal solution is found.

[0214] Step S410: Using multiple characteristic parameters including the capacity increment peak, differential voltage inflection point, constant current charging time and other characteristic parameters containing battery aging status information as inputs of the prediction model, a SOH estimation and RUL prediction model based on the improved BiGRU is established.

[0215] A SOH prediction model based on improved BiGRU is established. In order to ensure the effectiveness and accuracy of SOH prediction, the first 35%-40% of the data are selected as training data to ensure that the model can learn the intrinsic relationship between degradation characteristics.

[0216] According to NASA's lithium battery aging test, when the battery capacity reaches 70% of the rated capacity, that is, the battery SOH is 0.7, the battery life is considered to have ended. According to the mapping relationship between battery SOH and RUL, the battery SOH data is used to predict RUL, and a RUL prediction method based on the improved Archimedean algorithm to optimize the GRU fusion model is established.

[0217] Therefore, the present invention adopts the lithium battery SOH estimation and RUL prediction method combining the above-mentioned improved Archimedes algorithm and BiGRU, and proposes a lithium battery SOH prediction and RUL estimation model based on the improved Archimedes algorithm to optimize the BiGRU neural network in terms of model solving, so that the model can better fit the training data, increase the learning ability of the model, and improve the prediction accuracy.

[0218] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. Improved Archimedes algorithm combined with BiGRU lithium battery SOH estimation and RUL prediction method, characterized in that: The following steps are involved: Step S1: studying the working principle and aging mechanism of lithium batteries, and extracting characteristic parameters containing battery aging status information as input of the prediction model through battery modeling and data-driven methods; Step S2: performing correlation analysis on the extracted feature parameters, retaining the feature quantities with high correlation, and performing data preprocessing to improve the availability of the data; Step S3: Input the selected feature quantities into the BiGRU model for SOH estimation and RUL prediction; Step S4: Use the improved Archimedean algorithm to optimize the BiGRU model in order to minimize the loss function and improve the accuracy of the prediction results.

2. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 1 is characterized in that: The specific steps of step S1 are: Step S11, in view of the problem of incomplete screening of key factors affecting the health status of lithium batteries, the main factors leading to battery aging and capacity decay mechanism are summarized: the electrochemical reactions occurring during the operation of lithium batteries include not only redox reactions of lithium ion insertion and deinsertion, but also side reactions, which cause internal aging and capacity decay of lithium batteries as the number of battery cycles increases; Step S12, extracting characteristic parameters for estimating the SOH of the lithium battery from the lithium battery experimental data, taking the voltage, current, temperature parameter curves, the capacity increment curve, and the differential voltage curve as the analysis objects; Step S13, the capacity increment analysis method studies the IC curve to understand the changes in the internal state of the battery during aging, and converts the slowly changing charging voltage platform into an IC peak by drawing the IC curve; Step S14, using the data of the measured voltage not rising to the charging cut-off voltage to draw a voltage-capacity curve, and using the ampere-hour integration method to obtain the battery capacity during charging; Step S15, the parameters in the data set are discrete data, and the integral operation in the ampere-hour integration method is converted into a superposition operation; Step S16, the IC curve describes the relationship between dQ / dV-V, and the voltage-capacity curve is fitted and derived using the python software functions np.polyfit and np.poly1d.

3. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 2 is characterized in that: In step S14, the battery capacity during charging is obtained by the ampere-hour integration method, and the expression is as follows: Wherein, T represents the time when discharge ends, and i represents the discharge current at time t.

4. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 3 is characterized in that: In step S15, the integral operation in the ampere-hour integration method is converted into a superposition operation, and the expression is as follows: Wherein, I(t) represents the current value at the tth time point, T(t) represents the time at time t, T(t-1) represents the time at time t-1, and I(0) represents the current value at the initial time.

5. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 4 is characterized in that: The specific steps of step S2 are as follows: Step S21, using the Z-score standardization method to process features; Step S22, testing the correlation between continuous variables by Pearson correlation coefficient, extracting and retaining the feature quantities with high correlation; Step S23: using a box plot method to remove abnormal values, removing outliers, and treating the removed data locations as missing values; Step S24: Use the KNN interpolation algorithm to measure and identify adjacent points, and use adjacent observations to estimate missing values.

6. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 5 is characterized in that: In step S21, the Z-score normalization method is used to process the features, and the expression is as follows: In the formula, z represents the standardized value of Z-score. represents the original data value to be standardized, μ represents the mean of the data, and σ represents the standard deviation of the data; In step S22, the correlation between continuous variables is tested by the Pearson correlation coefficient, and the expression is as follows: In the formula, represents the mean of variable x, x i represents the i-th observation in the sample data set x, represents the mean of variable y, y i represents the i-th observation in the sample data set y, and n represents the total number of samples; In step S23, a box plot method is used to remove outliers, and the expression is as follows: IQR = Q3-Q1; UpLimit=Q3+IQR×1.5; LowLimit = Q1-IQR×1.5; x>UpLimit&x<LowLimit; In the formula, IQR stands for interquartile range, which is the difference between Q3 and Q1 in the data set, Q3 is the 75% quantile, Q1 is the 25% quantile, x is the data in the data set, UpLimit and LowLimit are the upper and lower limits of the box plot, and when the upper and lower limits are exceeded, it is considered as outlier data; In step S24, adjacent observations are used to estimate missing values, and the expression is as follows: Where L2(x1,x2) represents the Euclidean distance between vectors x1 and x2; 1n represents the nth component in vector x1, x 2n Represents the nth component in vector x2, where N represents the number of dimensions of the vector.

7. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 6 is characterized in that: The specific steps of step S3 are as follows: Step S31, calculating the GRU neural network update gate, reset gate and new candidate state; Step S32, BiGRU combines the output results of the input sequences in two directions to generate a final output result.

8. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 7 is characterized in that: In step S31, the calculation of the GRU unit is represented by the following formula: Update Gate: z t =σ(W Z ·[h t-1 ,x t ]); In the formula, σ represents the activation function, W Z represents the weight parameter of the update gate, x t and h t-1 Represent the input and hidden states at time step t respectively; Reset the gate: r t =σ(W r ·[h t-1 ,x t ]); Where W r Represents the weight parameter of the reset gate; New candidate status: Where: W h represents the weight parameter of the candidate hidden state, ⊙ represents the corresponding position multiplication of the matrix; Update hidden state: The final output result in step S32 is expressed as follows: In the formula, and are the weights of the forward and backward hidden layers, respectively, both are constant values, and their sum is 1.

9. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 7, characterized in that: The specific steps of step S4 include: Step S41, population initialization, setting the initial parameters of the optimization algorithm, using the Sin chaos model to generate the initial population, and improving the population traversal capability; Step S42, calculating the transfer operator TF and the density factor d; Step S43: If TF≤0.5, it is the initial exploration stage of the population, and the acceleration of the object is obtained; Step S44, if TF>0.5, it is the initial development stage of the population, and the acceleration of the object is obtained; Step S45: Select the current individual and the global optimal individual for arithmetic crossover to generate a new offspring individual closer to the current optimal solution, thereby accelerating the group to move closer to the global optimal area; Step S46: adopting the Golden Levy guidance mechanism so that when searching in an unknown range, a larger range can be reached, thereby making the algorithm jump out of the local optimum; Step S47, introducing the relationship between the sine function and the unit circle, so that the population can traverse all points on the sine function; Step S48: Add a greedy mechanism after the guiding mechanism to decide whether to update the target position by comparing the individual fitness before and after the individual position is updated, so as to retain the individuals with better fitness; Step S49, determine whether the maximum number of iterations has been reached or the optimal solution has been found. If the optimal solution has been found, output the decision variable position. If the above conditions are not met, return to step S42 until the optimal solution is found. Step S410: Using multiple characteristic parameters including the capacity increment peak, differential voltage inflection point, constant current charging time and other characteristic parameters containing battery aging status information as inputs of the prediction model, a SOH estimation and RUL prediction model based on the improved BiGRU is established.

10. The lithium battery SOH estimation and RUL prediction method combining the improved Archimedean algorithm and BiGRU according to claim 7, characterized in that: The Sin chaotic one-dimensional mapping expression in step S41 is as follows: Where, X n+1 represents the result of the function at time step n+1, X n Represents the value at the current time n, and N represents the number of iterations; In step S42, the expression of the transfer operator TF is as follows: In the formula, t represents the current time, t max Indicates the maximum value of time; The expression of density factor d is as follows: In step S43, during the initial exploration phase of the population, the acceleration update formula of the object is as follows: Where, den mr Indicates the density value at the current moment, vol mr Indicates the volume value at the current moment, acc mr Indicates the acceleration value at the current moment. represents the density value at time t+1; In step S44, in the initial population development stage, the acceleration update formula of the object is as follows: Where, den best represents the optimal density, vol best represents the optimal volume, acc best represents the optimal acceleration; In step S45, the arithmetic crossover operator expression is as follows: In the formula, λ represents the weight parameter, represents the position of object i at time t+1, X best Indicates the best position; In step S46, the calculation formula of the step length s is: s=m / n 1 / β ; In the formula, μ is a constant, v represents the speed, and β is an exponential constant; In step S47, the individual position update formula is: Where, X t represents the position of a vector, represents the optimal position vector, represents the position of the current vector, θ1 and θ2 represent the angle coefficients, dis represents the distance between the individual and the optimal position, R1 and R2 represent random values, and s represents the step size; In step S48, the greedy mechanism expression is as follows: In the formula, fit(.) represents the fitness function.

Images