A multi-scale SOC / SOH estimation method based on improved sine-cosine algorithm

Abstract

The present invention discloses a multi-scale SOC / SOH estimation method based on an improved sine-cosine algorithm, and belongs to the technical field of lithium battery SOC / SOH estimation. The technical solution first establishes a second-order RC equivalent circuit model of the battery; uses a double extended Kalman filter to estimate the state and parameters of the battery at multiple scales; uses an improved sinusoidal residual algorithm MSCA to optimize the system state error covariance Q and the measurement noise covariance R; and finally combines the double Kalman filter and the improved sine-cosine algorithm to perform SOC / SOH estimation. The technology provided by this patent can effectively improve the accuracy of SOC estimation, and the multi-scale online update method can effectively reduce the calculation cost. Accurate and real-time monitoring of the SOC / SOH of power lithium batteries can ensure that the battery is promptly prepared for corresponding maintenance or replacement, effectively discover and avoid unsafe behaviors of the battery, and provide protection for the stability of the power battery.

Description

Technical Field

[0001] The invention relates to the technical field of lithium battery SOC / SOH estimation, and in particular to a multi-scale SOC / SOH estimation method based on an improved sine-cosine algorithm. Background Art

[0002] As the country attaches great importance to clean energy, electric vehicles (EV) have become a research hotspot in recent years. Lithium-ion batteries have been widely used in electric vehicles, and accurate estimation of the state of charge (SOC) and state of health (SOH) has an important impact on the service life and durability of electric vehicles.

[0003] Definition of SOC: The ratio of the remaining capacity of a battery after it has been used for a period of time or has been left unused to its fully charged state. It can be expressed by a formula ranging from 0 to 1. When SOC = 0, it means that the battery is fully discharged, and when SOC = 1, it means that the battery is fully charged.

[0004]

[0005] Where Q Remain is the remaining capacity of the battery, Q Rated is the rated capacity of the battery, Q Discharge The significance of SOC estimation is that accurate estimation of SOC can avoid overcharging or over-discharging of the battery.

[0006] There are three existing methods: ampere-hour integration method, black box model, and nonlinear filtering algorithm. The Ah integration method is the most commonly used SOC estimation method. The black box model can theoretically achieve the same estimation effect as the true value, such as the rarely used neural network method. Because they have high requirements on data, and the data cannot obtain all working states. Among the nonlinear filtering algorithms, the extended Kalman filter (EKF) has been widely used, but when it is applied to SOC estimation, the state equation needs to be linearized, which reduces the accuracy of the model.

[0007] The sine-cosine algorithm was proposed by Mirjalili S in 2016. The main difference between this algorithm and conventional bionic algorithms is that it no longer adopts bionic ideas to simulate the behavior of organisms in nature to solve optimization problems, but uses sine and cosine functions to construct a specific update strategy to iterate the solution set and finally obtain a relatively optimal solution. The SCA algorithm and other swarm intelligence optimization algorithms cannot guarantee that the solution set found is the optimal solution in the search for the optimal solution, but the probability of approaching the optimal solution increases as the scale of the solution set and the number of iterations gradually increase. The iterative strategy of SCA is divided into two parts: global search and local development. Global search is achieved by using larger random perturbations on the solution set to explore unknown areas in the solution space. The current solution space is fully explored by using weaker random perturbations on the solution set. Summary of the invention

[0008] In order to improve the accuracy of charge / health state (SOC / SOH) estimation, the present invention proposes a multi-scale SOC / SOH estimation method based on an improved sine-cosine algorithm, which is characterized by comprising the following steps:

[0009] (1) Establish a second-order RC equivalent circuit model of the battery;

[0010] (2) Using double extended Kalman filtering to estimate the battery state and parameters at multiple scales;

[0011] (3) Using the improved sine-cosine algorithm MSCA to optimize the system state error covariance Q and measurement noise covariance R;

[0012] (4) Combine dual Kalman filtering and improved sine-cosine algorithm for SOC / SOH estimation.

[0013] Furthermore, the second-order RC equivalent circuit model of the battery in step 1 is established and the model parameters are identified. The circuit equation of the continuous system can be written as:

[0014]

[0015]

[0016] U=U OCV +U1+U2+R0I

[0017] Among them, R0 represents internal resistance, R1 and R2 represent polarization resistance, C1 and C2 represent polarization capacitance, U represents voltage, and U OCV Indicates open circuit voltage;

[0018] After discretizing the continuous system, the battery system can be expressed as follows:

[0019]

[0020]

[0021] U(k)=g(x k , S k , I k )=U OCV (SOC)-U1(k)-U2(k)-R0I

[0022] x k =[U 1，k , U 2，k , SOC k ] T

[0023] Where k represents the time step, Δt is the time between k and k+1, and x represents the state of the battery; when x and S are top or bottom labels, the value represents the state and parameter, is the noise matrix of the state filter, is the noise matrix of parameter filtering.

[0024] Furthermore, the dual Kalman filter in step 2 is composed of parameter filtering and state filtering. The parameter filtering is used for battery model parameter estimation, and the state filtering is used for SOC estimation. The parameter filtering and the state filtering are on different time scales. The parameter filtering state S includes two parts: ohmic internal resistance and capacity, that is: S k+1 =[R0,q].

[0025] Furthermore, the improved sine-cosine algorithm MSCA in step 3 is specifically as follows:

[0026] (3.1) Establish the sine-cosine algorithm model:

[0027]

[0028]

[0029] In the formula, t is the number of iterations, is the component of the position of individual i in the tth iteration in the jth dimension, r1, r2, r3 are random parameters, r1 is affected by the number of iterations, r2~U[0,2π], r3~U(0,∞], P j (t) is the component of the best candidate solution of the candidate solution set at the tth iteration in the jth dimension;

[0030] (3.2) A random function r4~U[0,1] is introduced to determine whether it is a sine update or a cosine update, eliminating the correlation between the iteration step size and direction;

[0031]

[0032] (3.3) Candidate solution Introducing inertia weight:

[0033]

[0034] Where ω0 is the inertia weight at the initial stage of iteration; ω e is the inertia weight at the end of iteration; t is the current iteration number; T is the maximum iteration number;

[0035] (3.4) Nonlinear improvement strategy of r1:

[0036]

[0037] The aim is to construct an r1 that takes into account both global search and local search and has a moderate convergence speed in the later iteration.

[0038] Furthermore, the steps of combining dual Kalman filtering and improved sine-cosine algorithm in step 4 are as follows:

[0039] (4.1) Algorithm initialization

[0040] State filter initialization:

[0041] Parameter filter initialization:

[0042] (4.2) Time update equation of parameter filtering:

[0043] (4.3) Measurement update equation for parameter filtering:

[0044]

[0045]

[0046]

[0047] (4.4) Optimize Qx and Rx according to MSCA in step 3;

[0048] (4.5) State filter time update equation:

[0049]

[0050] (4.6) State filter measurement update equation:

[0051]

[0052]

[0053]

[0054] Return to step (4.4) until MSCA reaches the maximum number of iterations;

[0055] in: Q x , Q S is the initial value of the algorithm, are the initial values ​​of the estimated parameters, is the initial value of the battery status, Q x , Q S are the initial values ​​of the noise matrix, and the variables with “^” represent estimated values.

[0056] Furthermore, ω0 is set to 0.9; ω e Set to 0.4; T is set to 50; m is set to 2 or 3.

[0057] Furthermore, the SOC / SOH estimation in step 4: SOH is estimated by ohmic internal resistance:

[0058]

[0059] Where R END is the ohmic internal resistance at the end of life, R is the estimated value of the state, and R NEW is the ohmic internal resistance of the battery when it leaves the factory, R END =2R NEW .

[0060] Furthermore, the SOC / SOH estimation in step 4: SOH is estimated by capacity:

[0061]

[0062] Where q is the current capacity and q0 is the initial capacity.

[0063] Beneficial effects of the present invention:

[0064] The technology provided by this patent can effectively improve the accuracy of SOC estimation, and the multi-scale online update method can effectively reduce the computing cost. In the field of electric vehicle BMS, battery SOC can provide basic data and judgment basis for battery pack balancing standards. At the same time, accurate and real-time monitoring of power lithium battery SOC can also ensure that the battery is promptly maintained or replaced, which can effectively discover and avoid unsafe battery behavior and provide protection for the stability of power batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is the second-order RC equivalent circuit model of lithium battery.

[0066] Figure 2 is the decreasing curve of ω1.

[0067] Figure 3 The decreasing curve of r1 when m=2 and m=3.

[0068] Figure 4 The fluctuation path curve of r1cos(r2) when m=2.

[0069] Figure 5 The fluctuation path curve of r1cos(r2) when m=3.

[0070] Figure 6 This is the MSCA-DEKF flow chart.

[0071] Figure 7 The simulation results of the standard SCA, two improved SCAs, EKF and DEKF for SOC estimation without initial error are shown in Figure 2, referring to the ampere-hour integration method.

[0072] Figure 8 The simulation results of the standard SCA, two modified SCAs, EKF, and DEKF are used for SOC estimation with reference to the ampere-hour integration method with initial SOC error.

[0073] Fig. 9 The simulation results of the standard SCA, two improved SCAs, DEKF for SOC estimation with initial R0 and capacity error are shown in Figure 2, and the ampere-hour integration method is used for reference.

[0074] Fig.10 The EKF is used for SOC estimation with initial R0 and capacity error, referring to the simulation results of the ampere-hour integration method.

[0075] Fig.11 It is the iterative convergence diagram of the algorithm.

[0076] Fig.12 This is a diagram of the parameter estimation simulation results.

[0077] Fig.13 This is the SOH estimation simulation result diagram. DETAILED DESCRIPTION

[0078] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0079] The second-order RC model is adopted in the present invention, and the model parameters are obtained through pulse discharge test and open circuit voltage test (OCV). DEKF is divided into state filtering and parameter filtering. The nonlinear decrement of the transformation parameter r1 in the sine-cosine algorithm (SCA) is proposed. The improved SCA (MSCA) is applied to the optimization of the covariance noise matrix in the state filtering. Parameter filtering is used to estimate the ohmic internal resistance and capacity online, and the time scale of the online parameter update is adjusted to 60 time steps to reduce the calculation cost. SOH can also be obtained through the ohmic internal resistance and capacity. The simulation results show that this method improves the accuracy of SOC estimation and corrects the initial errors of SOC and Ro in the first parameter estimation. The specific steps are as follows:

[0080] Step 1: Establish a second-order RC equivalent circuit model of the battery;

[0081] This model uses circuit elements such as resistors, capacitors, and constant voltage sources to form a circuit network to simulate the dynamic characteristics of the battery. This model is a one-time parameter model, which usually contains fewer parameters and is easy to derive state space equations. Therefore, it is widely used in system-level simulation analysis and real-time control. Equivalent circuit models based on parallel resistor and capacitor RC networks have been widely studied. For example: first-order RC model, second-order RC model, and third-order RC model. Using the second-order RC model, such as Figure 1 As shown: R0 represents the internal resistance, which reflects the sharp change of the terminal voltage. R1 and R2 represent the polarization resistance. C1 and C2 represent the polarization capacitance. U represents the voltage, U represents the OCV represents the open circuit voltage. The circuit equation for the continuous system can be written as:

[0082]

[0083]

[0084] U=U OCV +U1+U2+R0I

[0085] After discretizing the continuous system, the battery system can be expressed as follows:

[0086]

[0087]

[0088] U(k)=g(x k , S k , I k )=U OCV (SOC)-U1(k)-U2(k)-R0I

[0089] x k =[U 1，k , U 2，k , SOC k ] T

[0090] k represents the time step, Δt is the time between k and k+1, and x represents the state of the battery. When x and S are top or bottom labels, the values ​​represent the state and parameters, for example, is the noise matrix of the state filter, is the noise matrix of parameter filtering.

[0091] It is particularly important to note that there is a stable functional relationship between SOC and OCV, which can be expressed as follows:

[0092] f(SOC)=a1(SOC) 7 +a2(SOC) 6+a3(SOC) 5 +a4(SOC) 4 +a5(SOC) 3 +a6(SOC) 2 +a7(SOC)+a8

[0093] Step 2: Use double extended Kalman filtering to estimate the battery state and parameters at multiple scales;

[0094] The dual Kalman filter consists of parameter filtering and state filtering. The parameter filtering is used for battery model parameter estimation, and the state filtering is used for SOC estimation. The parameter filtering and state filtering are on different time scales. The parameter filtering state S includes two parts: ohmic internal resistance and capacity, namely: S k+1 =[R0,q].

[0095] Step 3: Use the improved sinusoidal residual algorithm MSCA to optimize the system state error covariance Q and measurement noise covariance R;

[0096] The SCA algorithm and other intelligent optimization algorithms cannot guarantee that the solution set obtained is the optimal solution when searching for the optimal solution, but the scale of the solution set and the number of iterations are gradually increased, and the more likely it is to approach the optimal solution. The iterative strategy of SCA is divided into two parts: global search and local development. Global search is achieved by using larger random perturbations on the solution set to explore unknown areas in the solution space, and weaker random perturbations are used on the solution set to fully explore the current solution space.

[0097] SCA updates the solution set by applying random perturbations as follows.

[0098]

[0099]

[0100] In the formula, t is the number of iterations, is the component of the position of individual i in the tth iteration in the jth dimension, r1, r2, r3 are random parameters, r1 is affected by the number of iterations, r2~U[0,2π], r3~U(0,∞], P j (t) is the component of the best candidate solution of the candidate solution set at the tth iteration in the jth dimension.

[0101] By introducing the random function r4~U[0,1] to eliminate the correlation between the iteration step and direction, the following iterative equation is obtained:

[0102]

[0103] The equation is the process of updating the solution in the SCA algorithm based on the current optimal solution and the candidate solutions in the solution set. The design purpose of the parameters is as follows: r1 is the limit of the range that the current solution can expand, and r2 is the extreme value of the updated solution approaching or moving away from the current optimal solution within the allowed range. r3, as a random number, is the degree of influence of the current optimal solution on the candidate solution. r4, as a random number, is used to determine whether it is a sine update or a cosine update, eliminating the correlation between the iteration step and direction.

[0104] In the intelligent algorithm, the inertia weight ω1 represents the ability of the candidate solution to inherit the information of the candidate solution of the previous iteration. In SCA, the candidate solution The inertia weight is a constant of 1. However, in previous experience, when the inertia weight in the intelligent algorithm is 1, it will affect the exploration and development capabilities of the algorithm. Referring to the optimization concept of the particle swarm algorithm, the following inertia weight is introduced.

[0105]

[0106] ω0 is the inertia weight at the initial stage of iteration, which is set to 0.9; ω e is the inertia weight at the end of the iteration, set to 0.4; t is the current number of iterations; T is the maximum number of iterations, set to 50. The curve of ω1 decreasing with the number of iterations is as follows: Figure 2 As shown in the figure, it can be seen that ω1 decreases nonlinearly with the increase of t. The algorithm can maintain strong random oscillation and search capabilities in the early stage. The reduction of inertia weight in the later stage strengthens the development ability of the algorithm and improves the accuracy of the algorithm.

[0107] Based on the improvement strategy, a nonlinear improvement strategy for r1 is proposed. The principle is the same, which aims to construct an r1 that takes into account both global search and local search and has a moderate convergence speed in the later iteration.

[0108]

[0109] In the formula, m is 2 or 3.

[0110] When r1 is m=2 or m=3, the descending curve is as follows Figure 3 As shown; when m = 2, the fluctuation path of r1cos(r2) is as follows Figure 4 As shown: When m = 3, the fluctuation path of r1cos(r2) is as follows: Figure 5 shown.

[0111] In order to obtain an accurate noise covariance matrix Q x and R x , using MSCA to calculate the noise covariance Q x and R x Optimized.

[0112] Step 4: Combine dual Kalman filtering and improved sine-cosine algorithm for SOC / SOH estimation;

[0113] DEKF is used for parameter adaptation and state estimation. The optimal result is obtained through the following algorithm flow. The optimization effect is affected by the fitness function and the measured value Y i With the estimated value x i The absolute error between them is used as the fitness function and can be expressed as:

[0114]

[0115] K is the current estimated time point.

[0116] like Figure 6 As shown, the MSCA-DEKF process for state estimation is as follows:

[0117] (4.1) Algorithm initialization

[0118] State filter initialization:

[0119] Parameter filter initialization:

[0120] (4.2) Time update equation of parameter filtering:

[0121] (4.3) Measurement update equation for parameter filtering:

[0122]

[0123]

[0124]

[0125] (4.4) Optimize Qx and Rx according to the MSCA algorithm in step 3;

[0126] (4.5) State filter time update equation

[0127]

[0128] (4.6) State filter measurement update equation

[0129]

[0130]

[0131]

[0132] Return to (4.4) until the maximum number of MSCA iterations is reached;

[0133] in: Q x , Q S is the initial value of the algorithm, are the initial values ​​of the estimated parameters, is the initial value of the battery status, Q x , Q S is the initial value of the noise matrix, and the variables with “^” represent estimated values.

[0134]

[0135]

[0136] In parameter filtering, ohmic resistance and capacity are estimated, both of which are related to SOH. Therefore, based on the accurate estimation of ohmic resistance and capacity, the relationship between ohmic resistance, capacity and SOH is established, thereby achieving the estimation of SOC and SOH.

[0137] SOH estimation method 1:

[0138]

[0139] R END is the ohmic internal resistance at the end of life, R is the estimated value of the state, and R NEW is the ohmic internal resistance of the battery when it leaves the factory, R END =2R NEW .

[0140] SOH estimation method 2:

[0141] Estimating SOH by Capacity

[0142]

[0143] q is the current capacity and q0 is the initial capacity.

[0144] The verification of the algorithm of the present invention is given below.

[0145] Generally speaking, the algorithm parameters directly affect the performance of the algorithm. The parameter S is optimized through the parameter filter. x0 includes the SOC and the voltage of the two RC branches. S1 includes the ohmic internal resistance and capacity. The sampling time is also adjusted. The parameter S of the battery model changes very little, which makes the sampling time of the parameter filter longer than that of the state filter. Therefore, the sampling time of the parameter filter is set to 60 time steps, and the sampling time of the state filter is set to 0.1s. The parameter identification results of the battery test are shown in Table 1.

[0146] In order to demonstrate the effectiveness of the algorithm for parameter identification, the initial value of the algorithm is set as the error, and the following error setting strategy is adopted. The initial value with error is marked with Δ at the bottom.

[0147] Strategy (1): All initial values ​​are set to the best available values.

[0148] Strategy (2): The initial SOC is set to 1, the initial value of the ohmic internal resistance is enlarged to 0.037 (amplification), and the initial value of the capacity is set to 24 Ah (actual capacity 80%).

[0149] Strategy (3): The initial SOC is set to 0.8, and the initial values ​​of the ohmic internal resistance and capacity are set to actual values.

[0150] x0=[0 0 1] T

[0151]

[0152]

[0153] x0=[0 0 1] T

[0154]

[0155]

[0156] x0=[0 0 0.8] T

[0157]

[0158]

[0159] Table 1 Battery parameters

[0160]

[0161]

[0162] Initialization of MSCA:

[0163] The maximum number of iterations is 50, the search dimension is 20, and the parameter search range is: the upper limit is 100 and the lower limit is 0.00000000001.

[0164] After setting the parameters, the estimation results show that for strategies (1), (2), and (3), the algorithm shows different accuracy.

[0165] like Figure 7As shown, in strategy (1), we can see that the standard SCA-DEKF, MSCA-DEKF: m=3 or m=2 outperform EKF and DEKF.

[0166] like Figure 8 As shown, in strategy (2), it can be seen that the estimation accuracy of SCA-DEKF, MSCA-DEKF: m = 3 or m = 2 is greatly improved compared with DEKF and EKF. At the same time, the SOC error is corrected in the first parameter estimation.

[0167] like Fig. 9 , Fig.10 As shown, in strategy (3), it can be seen that SCA-DEKF, MSCA-DEKF, m = 3 or m = 2 also have great improvements compared with DEKF and EKF, and the initial errors of Ro and capacity can be quickly restored to the correct values.

[0168] like Fig.11 As shown in Figure 2, the number of iterations required to achieve convergence is different under different strategies.

[0169] In strategy (1), MSCA-DEKF: m = 3 only needs 6 iterations to converge, which is better than the other iterations with correct initial parameters.

[0170] In strategy (2), MSCA-DEKF: m = 2 only needs 11 iterations to converge, which is better than MSCA-DEKF: m = 3 and standard SCA-DEKF.

[0171] In strategy (3), MSCA-DEKF: m = 3 only needs 14 iterations to converge, which is better than MSCA-DEKF: m = 2 and standard SCA-DEKF.

[0172] The experimental cells are from a new batch. Therefore, the theoretical SOH value based on capacity is considered to be 100%. According to the equation, SOH can be divided into SOH based on Ro and SOH based on capacity. The simulation results are shown in Fig.12 , Fig.13 shown.

[0173] from Fig.12 As can be seen from (a), (b), and (c), the Ro estimate has high reliability and converges quickly from the initial error to the correct value. Therefore, the SOH estimate based on Ro also has high reliability, such as Fig.13 As shown in (a), (b) and (c).

[0174] from Fig.12 It can be seen from (d), (e), and (f) that the capacity estimation has a large jitter, so the capacity-based SOH estimation also has a large jitter, e.g. Fig.13 As shown in (d), (e) and (f).

[0175] Two nonlinear descent strategies of r1 are applied to SCA, named MSCA. The noise covariance matrix of the state filter is optimized by MSCA for SOC estimation. Parameter filtering is used for SOH estimation. The method consisting of state filtering and parameter filtering is named MSCA-DEKF. Parameter estimation and state estimation are based on different time scales. The parameters of the battery model do not change significantly in a short time, so the parameter update is slower than the state update, which reduces the amount of calculation. When the initial value matrix takes the correct value, the standard SCA-DEKF, MSCA-DEKF: m=3 or m=2 outperform EKF and DEKF. When the initial SOC is set to 0.8, the estimation accuracy of the standard SCA-DEKF, MSCA-DEKF: m=3 or m=2 is greatly improved compared with DEKF and EKF. At the same time, when the time scale of the parameter filtering is set to 60 time steps, the SOC error in the first parameter estimation is corrected. When the initial Ro and capacity are set to 0.0037×10Ω and 30.24×1.05Ah, EKF cannot perform calculations close to Ref. Standard SCA-DEKF, MSCA-DEKF: m=3 or m=2 also show significant improvements over DEKF and EKF. The initial errors of Ro and capacity can be quickly restored to the correct values. When the parameters are set to the initial SOC error, MSCA-DEKF: m=2 only needs 11 iterations to converge, which is better than MSCA-DEKF: m=3 and standard SCA-DEKF. When the initial errors of the parameters are set to Ro and capacity, MSCA-DEKF: m=3 only needs 14 iterations to converge, which is better than MSCA-DEKF: m=2 and standard SCA-DEKF. MSCA-DEKF: m=3 only needs 6 iterations to converge, which is better than other products with correct initial parameters. SOH estimation based on Ro, m=2 or 3 can quickly correct the initial error and has high reliability. Capacity-based SOH estimation can correct some initial errors but has certain fluctuations.

[0176] The filtering method for multi-scale SOC estimation and online parameter update based on improved sine-cosine algorithm optimization proposed in this scheme has practical application value.

[0177] Finally, it should be noted that the above examples are only used to illustrate the technical solution of the present invention rather than to limit it. After reading this application, technical personnel in the relevant field may make various modifications or changes to the present invention with reference to the above embodiments, which are all within the scope of protection required by the pending application of the present invention.

Claims

1. A multi-scale SOC / SOH estimation method based on an improved sine-cosine algorithm, characterized in that: The steps include: (1) Establish a second-order RC equivalent circuit model of the battery; (2) Using double extended Kalman filtering to estimate the battery state and parameters at multiple scales; (3) Using the improved sine-cosine algorithm MSCA to optimize the system state error covariance Q and measurement noise covariance R; Improved sine-cosine algorithm MSCA, the specific contents are as follows: (3.1) Establish the sine-cosine algorithm model: In the formula, t is the number of iterations, is the component of the position of individual i in the tth iteration in the jth dimension, r1, r2, r3 are random parameters, r1 is affected by the number of iterations, r2~U[0,2π], r3~U(0,∞], P j (t) is the component of the best candidate solution of the candidate solution set at the tth iteration in the jth dimension; (3.2) A random function r4~U[0,1] is introduced to determine whether it is a sine update or a cosine update, eliminating the correlation between the iteration step size and direction; (3.3) Candidate solution Introducing inertia weight: Where ω0 is the inertia weight at the initial stage of iteration; ω e is the inertia weight at the end of iteration; t is the current iteration number; T is the maximum iteration number; (3.4) Nonlinear improvement strategy of r1: The aim is to construct an r1 that takes into account both global search and local search and has a moderate convergence speed in the later iteration; (4) Combined dual extended Kalman filter and improved sine-cosine algorithm for SOC / SOH estimation.

2. The multi-scale SOC / SOH estimation method based on the improved sine-cosine algorithm according to claim 1 is characterized in that: The second-order RC equivalent circuit model of the battery in step (1) is established and the model parameters are identified. The circuit equation of the continuous system can be written as: U=U OCV +U1+U2+R0I Among them, R0 represents internal resistance, R1 and R2 represent polarization resistance, C1 and C2 represent polarization capacitance, U represents voltage, and U represents OCV Indicates open circuit voltage; After discretizing the continuous system, the battery system can be expressed as follows: U(k)=g(x k ,S k ,I k )=U OCV (SOC)-U1(k)-U2(k)-R0I x k =[U 1，k ,The 2，k ,SOC k ] T Where k represents the time step, Δt is the time between k and k+1, and x represents the state of the battery; when x and S are top or bottom labels, the value represents the state and parameter, is the noise matrix of the state filter, is the noise matrix of parameter filtering.

3. The multi-scale SOC / SOH estimation method based on the improved sine-cosine algorithm according to claim 2 is characterized in that: In step (2), the dual Kalman filter is composed of parameter filtering and state filtering. The parameter filtering is used for battery model parameter estimation, and the state filtering is used for SOC estimation. The parameter filtering and the state filtering are on different time scales. The parameter filtering state S includes two parts: ohmic internal resistance and capacity, that is, S k+1 =[R0,q].

4. The multi-scale SOC / SOH estimation method based on the improved sine-cosine algorithm according to claim 3 is characterized in that: The steps of combining the double extended Kalman filter and the improved sine-cosine algorithm in step (4) are as follows: (4.1) Algorithm initialization State filter initialization: Parameter filter initialization: (4.2) Time update equation of parameter filtering: (4.3) Measurement update equation for parameter filtering: (4.4) According to the MSCA in step (3), Q x and R x Optimize (4.5) State filter time update equation: (4.6) State filter measurement update equation: Return to step (4.4) until MSCA reaches the maximum number of iterations; in: is the initial value of the algorithm, are the initial values ​​of the estimated parameters, is the initial value of the battery status, are the initial values ​​of the noise matrix, and the variables with "^" represent estimated values.

5. The multi-scale SOC estimation method based on improved sine-cosine algorithm optimization according to claim 4 is characterized in that: ω0 is set to 0.9; ω e Set to 0.4; Set T to 50; m is 2 or 3.

6. The multi-scale SOC estimation method based on improved sine-cosine algorithm optimization according to claim 4 is characterized in that: SOC / SOH estimation in step (4): SOH estimation by ohmic internal resistance: Where: R END is the ohmic internal resistance at the end of life, R is the estimated value of the state, and R NEW is the ohmic internal resistance of the battery when it leaves the factory, R END =2R NEW .

7. The multi-scale SOC estimation method based on improved sine-cosine algorithm optimization according to claim 4 is characterized in that: SOC / SOH estimation in step (4): SOH estimation by capacity: Among them: q is the current capacity, q0 is the initial capacity.

Images