Method for jointly estimating SOC of electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm

Abstract

The invention discloses a lithium ion battery parameter identification and state of charge (SOC) estimation method based on a particle swarm optimization and extended Kalman filter (PSO-EKF) algorithm, and relates to the technical field of lithium ion battery management systems. The method is used for solving the problem that a traditional SOC estimation method is insufficient in precision, adaptability and efficiency. According to the method, a bird flock foraging behavior is simulated through a PSO algorithm, a parameter-population scale is optimized by taking a convergence rate and diversity as targets, and rapid and accurate search of battery model parameters is realized; meanwhile, the SOC is accurately estimated by using the nonlinear processing capability of the EKF algorithm. And the two methods supplement each other, so that the precision and efficiency of parameter identification and SOC estimation are remarkably improved. The method provides powerful support for optimization and upgrading of a lithium ion battery management system, and has wide application prospects and remarkable economic and social benefits.

Description

Technical Field

[0001] The present invention relates to a state estimation technology for an electrochemical energy storage system, and more specifically, to a method for jointly estimating the state of charge (SOC) of an electrochemical energy storage system based on particle swarm optimization-extended Kalman filter (PSO-EKF) parameter identification and an extended Kalman filter (EKF) algorithm. Background Art

[0002] Battery SOC estimation is important for optimizing the battery's charge and discharge processes, extending battery life, and preventing overcharging or over-discharging. It is widely used in battery management systems, electric vehicles, renewable energy storage systems, and other fields that rely on electrochemical energy storage devices.

[0003] The existing technology has the following deficiencies:

[0004] Due to the complexity of battery electrochemical characteristics and its nonlinear characteristics under different working conditions, traditional SOC estimation methods are highly dependent on model parameters. When the battery's electrochemical model parameters (such as battery internal resistance, capacity, etc.) change, the traditional EKF algorithm may have estimation deviations and it is difficult to provide high-precision estimation results in a highly dynamic environment.

[0005] In view of the above problems, the present invention proposes a solution. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm, XX to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm includes the following steps:

[0009] A Thevenin lithium-ion battery model was established, and the lithium-ion battery was identified online using a PSO-EKF joint method. The identified parameters were used to estimate the SOC using the EKF algorithm.

[0010] Establish a population optimization module. It is divided into three parts: establishing convergence speed indicators, establishing diversity indicators, and establishing a population optimization model.

[0011] In a preferred embodiment, establishing a Thevenin lithium-ion battery model comprises the following steps:

[0012] likeFigure 2 As shown in Figure 1, the lithium-ion battery model is a first-order RC equivalent circuit model.

[0013] The model consists of an ideal open circuit voltage source U ocv , an ohmic internal resistance R0, and a parallel RC circuit. Among them, U oc represents the battery's open-circuit voltage, and R0 is the battery's internal resistance. The RC circuit, consisting of resistor Rp and capacitor Cp, simulates the battery's electrochemical polarization. The battery's terminal voltage is represented by U, and IL represents the input current. The electrical characteristic equation for this model in the time domain is as follows:

[0014]

[0015] In a preferred embodiment, the lithium-ion battery is identified online based on the PSO-EKF joint method, comprising the following steps:

[0016] The EKF algorithm predicts the state value at time k:

[0017]

[0018] where f(x k ,u k ) is the nonlinear state equation; g(x k ,u k ) is the nonlinear output equation; w k is the system noise, with a mean of 0; the covariance is Q k ,v k is the system observation noise, with a mean of 0; the covariance R k , v k And w k Not related to each other.

[0019] In order to deal with nonlinear discrete systems, we use the state function f(x k ,u k ) and the measurement function g(x k ,u k At time k, f(x k ,u k ), g(x k ,u k ) The proposed distribution obtained by the first-order local linear approximation of the Taylor series has a large truncation error, which leads to a decrease in filtering accuracy, but the computational complexity is relatively small:

[0020]

[0021] definition is the state transition matrix, is the observation matrix, rewritten as:

[0022]

[0023] Further limiting, the complete recursive process of the extended Kalman filter algorithm is as follows:

[0024]

[0025] Further qualifying, the a priori estimates of the state variables:

[0026] x k / k-1 =f(x k ,u k-1 )

[0027] Further limiting, the error covariance matrix of the prediction k time is:

[0028]

[0029] Further limit, calculate the gain matrix K k for:

[0030]

[0031] Further limit, according to the observed value U k The updated state estimate is:

[0032] x k =x k / k-1 +K k [y k -C k x k / k-1 ]

[0033] Further limiting, the updated error covariance matrix is:

[0034] P k =(ik k C k )P k / k-1

[0035] Further limiting, the generated error is:

[0036] e k =v exp -v mdl

[0037] The use of the extended Kalman filter algorithm to identify the parameters of the battery equivalent circuit model online requires the formation of a state equation with the model parameters as state variables. p ,OCV,R0,R p , C p, the output is the state space equation of the battery under U0 continuous time. In practical applications, since the current and voltage data are collected at discrete time points, the collection results are a series of scattered data, so it is necessary to discretize the continuous state equation.

[0038]

[0039] Let t0 = KT s , t=(k+1)T s , the discretized space state equation is:

[0040]

[0041] get:

[0042] Joint OCV k+1 =OCV k , R 0,k+1 =R 0,k ,R p,k+1 =R p,k and C p,k+1 =C p,k Get the state model and observation model:

[0043]

[0044] For the variable I included in the EKF algorithm p ,OCV,R0,R p , C p According to the state model, C p Compared with the other four parameters, I p ,OCV,R0,R p The corresponding noise (Q w1 ,Q w2 ,Q w3 ,Q w4 ) is a group of objects to be identified, and C p The corresponding noise Q w5 Perform another set of identification using the particle swarm optimization (PSO) algorithm, setting the population size to 20 and the parameters to be identified as [D*F], where D is the dimension and F is the population size. For the first set, D = 4 and F = 20; for the second set, D = 1 and F = 20.

[0045] PSO is initialized as a group of random particles. During the position update process, it performs global and local searches based on this. Guided by the acceleration factor, particles can search along the direction of the historical optimal position and the group optimal position, and eventually the entire population converges to the global optimal solution. The update process of particle speed and position in PSO is:

[0046]

[0047] to I p ,OCV,R0,R p The observation model is processed to obtain the objective function of the first set of optimization objects:

[0048] F=min((R p,k +Q w,1 )(I p,k +Q w,2 )+(R 0,k +Q w,3 )I L,k +(OCV k +Q w,4 )-U0) 2

[0049] The second set of objective functions:

[0050]

[0051] Where I0 is the current flowing through the battery, R p , I p The value obtained by optimizing the first set of objective functions.

[0052] In a preferred embodiment, the identified parameters are used to estimate the SOC based on the EKF algorithm, which includes the following steps:

[0053] The state space equation based on the state variables of the Thevenin model:

[0054]

[0055] Uk=U p,k +R 0,k I L,k +OCV(SOCk)+v k

[0056] The state transfer matrix after linearization using Taylor series is:

[0057] The linearized observation matrix is:

[0058]

[0059] In an optimal way, in order to reduce the impact of population size on the PSO-EKF model, we establish a population optimization model based on convergence speed and diversity, which is divided into three parts: establishing a convergence speed index, establishing a diversity index and establishing a population optimization model.

[0060] The particle position and velocity update formulas are as follows:

[0061] x i,d (t+1)=x i,d (t)+x i,d (t-1) (4)

[0062] v i,d (t+1)=ωv i,d (t)+c1r1(p i,d (t)-x i,d (t))+c2r2(g d (t)-x i,d (t)) (5)

[0063] Where t is the evolutionary generation, d∈D is the dth dimension of the search space, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers uniformly distributed between [0, 1]; i,d (t) is the position information of the particle at the tth iteration, v i,d (t) is the velocity information of the particle at the tth iteration, p i,d (t) is the historical optimal position of the particle at the tth iteration, g d (t) is the global optimal position at the tth iteration. In this case, D = [4, 1].

[0064] Convergence speed index:

[0065] The difference between the average values of the minimum Euclidean distances between the Pareto frontiers generated by three consecutive iterations of the algorithm is called the convergence rate of the algorithm. The calculation formula is:

[0066]

[0067] Where n i+1 is the number of non-dominated solutions generated by the algorithm in the i+1th iteration, B i-1 , B i , B i+1 is the Pareto frontier generated by three consecutive iterations of the algorithm, d i Frontier B i+1 The i-th solution to the frontier B i The minimum Euclidean distance, Frontier B i+1 With Frontier B i The average of the smallest Euclidean distances between them.

[0068] This metric requires the Pareto front generated by three consecutive iterations to calculate. If S > 0, it indicates that the algorithm converged relatively quickly in the current iteration and the algorithm's population size was small. Conversely, it indicates that the algorithm converged relatively slowly in the current iteration and the algorithm's population size was large.

[0069] Establishing diversity metrics:

[0070] The distribution of non-dominated solutions in the Pareto front generated by the algorithm is called the diversity of the algorithm. The calculation formula is:

[0071]

[0072] Where, d i is the Euclidean distance between the i-th non-dominated solution obtained by the algorithm and its nearest non-dominated solution, d i The diversity index reflects the distribution of non-dominated solutions in the Pareto front generated by the algorithm. The more evenly the particles are distributed in the solution space, the smaller the SP value is, and the better the diversity of the algorithm is.

[0073] The mathematical description of the change in algorithm diversity is:

[0074] C=SP i -SP i-1

[0075] Where C is the indicator of algorithm population diversity change, SP i is the population diversity value of the Pareto front generated by the algorithm's iteration i.

[0076] If C>0, it means that the population diversity of the current iteration algorithm is relatively poor and the population size of the algorithm is small; otherwise, it means that the population diversity of the algorithm is relatively good and the population size is large.

[0077] Establishing a population optimization model

[0078] Based on the description of the relationship between population size and optimization performance, this case designed a population size fitness function, which is described as:

[0079] S>0&C>0

[0080] S≤0&C<0

[0081] S>0&C<0(1)

[0082] S≤0&C>0(2)

[0083] The algorithm's population size is determined by the population size fitness function. When the algorithm satisfies equation (1) five times in a row, the algorithm stops calculating the fitness function. Since the algorithm is evolving in a positive direction, the situation described in equation (2) theoretically does not exist. This method not only provides a basis for adjusting the algorithm's population size, but also minimizes unnecessary computational steps.

[0084] The population increase mechanism is:

[0085] N(t+1)=N(t)+k1×N S (t)(3)

[0086] Where t represents the tth iteration; k1 is a random number between the interval [0, 1]; N(t) represents the population size of the algorithm at the tth iteration; N S (t) is the number of non-dominated solutions generated in the tth iteration of the algorithm.

[0087] The population reduction mechanism is:

[0088] N(t+1)=N(t)-k2×(N(t)-N S (t))

[0089] Where k2 is a random number between the interval [0, 1].

[0090] Input: Randomly initialize a population of N particles and initialize the particle position x i (0), speed v i (0), inertia weight ω, learning factors c1 and c2. Set the initial position of each particle to the current historical optimal position p i (0), select the global optimal position g(0) from the initial external archive;

[0091] 1. Update the particle's position x i (t) and velocity v i (t);

[0092] 2. Evaluate the fitness value of each particle and perform non-dominated sorting;

[0093] 3. Update external archives;

[0094] 4. Update individual optimal position p i (t), global optimal position g(t);

[0095] 5. Calculate CS and CSP values;

[0096] 6. Determine whether the current iteration algorithm should use a population increase or decrease mechanism;

[0097] 7. Automatically adjust the population size of the algorithm.

[0098] The present invention is based on the method of estimating the SOC of electrochemical energy storage system by combining PSO-EKF parameter identification and EKF algorithm:

[0099] The present invention aims to further improve the accuracy and robustness of battery SOC estimation by combining precise model parameter identification with nonlinear state estimation, overcome the defect of traditional SOC estimation methods in the prior art that the accuracy decreases when the battery dynamic characteristics change, and meet the needs of modern battery management systems in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0100] Figure 1 This is an algorithm flow chart based on the forgetting factor recursive least squares method and extended Kalman filter.

[0101] Figure 2 It is the second-order RC equivalent circuit model of lithium-ion battery.

[0102] Figure 3 It is the flow chart of the improved Kalman filter algorithm.

[0103] Figure 4 This is a schematic diagram of the SOC estimation results.

[0104] Figure 5 This is a schematic diagram of SOC estimation error. DETAILED DESCRIPTION

[0105] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0106] A lithium-ion battery SOC estimation system algorithm based on extended Kalman filter of particle swarm optimization, such as Figure 1 As shown, the following steps are included:

[0107] S1: Build a Thevenin lithium-ion battery model.

[0108] Furthermore, if Figure 2 As shown in Figure 1, the lithium-ion battery model is a first-order RC equivalent circuit model.

[0109]

[0110] Specifically, the model includes an ideal open-circuit voltage source U ocv , an ohmic internal resistance R0, and a parallel RC circuit. Among them, U ocv Represents the open circuit voltage of the battery, and R0 is the internal resistance of the battery. The RC loop consists of resistor R p and capacitor C p The terminal voltage of the battery is expressed as U, IL Represents the input current. Based on the circuit principle and Figure 2 The electrical characteristic equation of the model in the time domain can be obtained:

[0111] Furthermore, the above formula can be expressed in the frequency domain as follows:

[0112]

[0113] Furthermore, the above formula can be written into the transfer function expression as follows:

[0114]

[0115] Furthermore, let τ1=R1C1

[0116] The solution is:

[0117]

[0118] S2: Online parameter identification of lithium-ion batteries based on PSO-EKF joint.

[0119] Specifically, the recursive process of the improved extended Kalman filter algorithm is as follows:

[0120] Furthermore, the state value at time k is predicted to be:

[0121]

[0122] where f(x k ,u k ) is the nonlinear state equation; g(x k ,u k ) is the nonlinear output equation; w k is the system noise, with a mean of 0; the covariance is Q k ,v k is the system observation noise, with a mean of 0; the covariance R k , v k And w k Not related to each other.

[0123] In order to deal with nonlinear discrete systems, we use the state function f(x k ,u k ) and the measurement function g(x k ,u k At time k, f(x k ,u k ), g(x k ,u k ) The proposed distribution obtained by the first-order local linear approximation of the Taylor series has a large truncation error, which leads to a decrease in filtering accuracy, but the computational complexity is relatively small:

[0124]

[0125] definition is the state transition matrix, is the observation matrix, rewritten as:

[0126]

[0127] The complete recursive process of the extended Kalman filter algorithm is as follows:

[0128]

[0129] Get a priori estimates of the state variables:

[0130] x k / k-1 =f(x k ,u k-1 )

[0131] Furthermore, the error covariance matrix of the prediction k time is:

[0132]

[0133] Furthermore, the gain matrix K is calculated k for:

[0134]

[0135] Furthermore, according to the observation value U k The updated state estimate is:

[0136] x k =x k / k-1 +K k [y k -C k x k / k-1 ]

[0137] Furthermore, the updated error covariance matrix is:

[0138] P k =(ik k C k )P k / k-1

[0139] Furthermore, the generated error is:

[0140] e k =v exp -v mdl

[0141] The use of the extended Kalman filter algorithm to identify the parameters of the battery equivalent circuit model online requires the formation of a state equation with the model parameters as state variables.p ,OCV,R0,R p , C p , the output is the state space equation of the battery under U0 continuous time. In practical applications, since the current and voltage data are collected at discrete time points, the collection results are a series of scattered data, so it is necessary to discretize the continuous state equation.

[0142]

[0143] The solution of the continuous state equation is:

[0144]

[0145] Let t0 = KT s , t=(k+1)T s , the discretized space state equation is:

[0146]

[0147] get:

[0148]

[0149] Joint OCV k+1 =OCV k , R 0,k+1 =R 0,k ,R p,k+1 =R p,k and C p,k+1 =C p,k Get the state model and observation model:

[0150]

[0151] The Jacobian matrix of the state equation is:

[0152]

[0153] in

[0154]

[0155] The linearized observation matrix is:

[0156]

[0157] For the variable I included in the EKF algorithm p ,OCV,R0,R p , C p According to the state model, C p Compared with the other four parameters, I p,OCV,R0,R p The corresponding noise (Q w1 ,Q w2 ,Q w3 ,Q w4 ) is a group of objects to be identified, and C p The corresponding noise Q w5 Perform another set of identification using the particle swarm optimization (PSO) algorithm, setting the population size to 20 and the parameters to be identified as [D*F], where D is the dimension and F is the population size. For the first set, D = 4 and F = 20; for the second set, D = 1 and F = 20.

[0158] Furthermore, the position of the first particle f and its flying speed are:

[0159] x f,d =[x f,1 ,x f,2 ,…,x f,D ]

[0160] v f,d =[v f,1 ,v f,2 ,…,v f,D ]

[0161] The historical optimal position of the particle and the optimal position of the population are:

[0162] p best =[p f,1 ,p f,2 ,…,p f,D ]

[0163] g best =[g f,1 ,g f,2 ,…,g f,D ]

[0164] PSO is initialized as a group of random particles. During the position update process, it performs global and local searches based on this. Guided by the acceleration factor, particles can search along the direction of the historical optimal position and the group optimal position, and eventually the entire population converges to the global optimal solution. The update process of particle speed and position in PSO is:

[0165]

[0166] to I p ,OCV,R0,R p The observation model is processed to obtain the objective function of the first set of optimization objects:

[0167] F=min((R p,k +Qw,1 )(I p,k +Q w,2 )+(R 0,k +Q w,3 )I L,k +(OCV k +Q w,4 )-U0) 2

[0168] The solution is:

[0169]

[0170] The second set of objective functions:

[0171]

[0172] Where I0 is the current flowing through the battery, R p , I p The value obtained by optimizing the first set of objective functions.

[0173] S3: Lithium-ion battery SOC estimation based on EKF algorithm

[0174] Based on the Thevenin model electrical equation:

[0175]

[0176] State space equations for state variables:

[0177]

[0178] Uk=U p,k +R 0,k I L,k +OCV(SOCk)+v k

[0179] The state transfer matrix after linearization using Taylor series is:

[0180]

[0181] The linearized observation matrix is:

[0182]

[0183] Applying the extended Kalman filter algorithm to estimate battery SOC essentially uses the ampere-hour integration method to calculate SOC. The error between the actual battery terminal voltage measurement and the model's prediction is used to correct the ampere-hour integration result. Large errors in this error generate a large correction term, accelerating the correction of the SOC estimate. This method overcomes the drawbacks of the ampere-hour integration method, which requires precise initial values and is prone to cumulative errors. It converges quickly even when the initial error is large, achieving good estimation results even under dynamic conditions with drastic current fluctuations.

[0184] To verify the effectiveness of the present invention in estimating SOC, a 280Ah lithium iron phosphate battery manufactured by Ruipu Lanjun Company was used as the research object, and a simulated energy storage operating condition experiment was conducted. As shown in the figure, compared with the conventional extended Kalman filter algorithm, the lithium-ion battery SOC estimation system proposed in this invention, which uses particle swarm optimization parameter identification and extended Kalman filtering, has higher battery SOC estimation accuracy, with an error of no more than 2%.

[0185] The establishment of a population optimization model based on convergence speed and diversity is divided into three parts: establishing a convergence speed index, establishing a diversity index and establishing a population optimization model.

[0186] MOPSO is a heuristic algorithm based on information sharing between particles. Each particle in the population represents a potential solution to the problem to be optimized. The algorithm iteratively searches for the optimal solution by following the behavior of the optimal particle in the solution space. The particle position and velocity update formula is as follows:

[0187] x i,d (t+1)=x i,d (t)+x i,d (t-1) (4)

[0188] v i,d (t+1)=ωv i,d (t)+c1r1(p i,d (t)-x i,d (t))+c2r2(g d (t)-x i,d (t)) (5)

[0189] Where t is the evolutionary generation, d∈D is the dth dimension of the search space, ω is the inertia weight, c1 and c2 are learning factors, r1 and r2 are random numbers uniformly distributed between [0, 1]; i,d (t) is the position information of the particle at the tth iteration, v i,d (t) is the velocity information of the particle at the tth iteration, p i,d (t) is the historical optimal position of the particle at the tth iteration, g d(t) is the global optimal position at the tth iteration. In this case, D = [4, 1].

[0190] Establish convergence speed indicator:

[0191] The convergence rate index is one of the important indicators to describe the optimization performance of the algorithm. The difference between the average values of the minimum Euclidean distances between the Pareto fronts generated by three consecutive iterations of the algorithm is called the convergence rate of the algorithm. The calculation formula is:

[0192]

[0193] Where n i+1 is the number of non-dominated solutions generated by the algorithm in the i+1th iteration, B i-1 , B i , B i+1 is the Pareto frontier generated by three consecutive iterations of the algorithm, d i Frontier B i+1 The i-th solution to the frontier B i The minimum Euclidean distance, Frontier B i+1 With Frontier B i The average of the smallest Euclidean distances between them.

[0194] This metric requires the Pareto front generated by three consecutive iterations to calculate. If S > 0, it indicates that the algorithm converged relatively quickly in the current iteration and the algorithm's population size was small. Conversely, it indicates that the algorithm converged relatively slowly in the current iteration and the algorithm's population size was large.

[0195] Establishing diversity metrics:

[0196] Diversity index is an important component in describing the optimization performance of the algorithm.

[0197] The distribution of solutions is called the diversity of the algorithm. The calculation formula is:

[0198]

[0199] Where, d i is the Euclidean distance between the i-th non-dominated solution obtained by the algorithm and its nearest non-dominated solution, d i The diversity index reflects the distribution of non-dominated solutions in the Pareto front generated by the algorithm. The more evenly the particles are distributed in the solution space, the smaller the SP value is, and the better the diversity of the algorithm is.

[0200] The mathematical description of the change in algorithm diversity is:

[0201] C=SP i -SP i-1 (9)

[0202] Where C is the indicator of algorithm population diversity change, SP i is the population diversity value of the Pareto front generated by the algorithm's iteration i.

[0203] If C>0, it means that the population diversity of the current iteration algorithm is relatively poor and the population size of the algorithm is small; otherwise, it means that the population diversity of the algorithm is relatively good and the population size is large.

[0204] Establish a population optimization model:

[0205] Based on the description of the relationship between population size and optimization performance, this case designed a population size fitness function, which is described as:

[0206] S>0&C>0(10)

[0207] S≤0&C<0(11)

[0208] S>0&C<0(12)

[0209] S≤0&C>0(13)

[0210] The description of the population fitness function is shown in Table 1.

[0211] Table 1 Analysis of population fitness function

[0212]

[0213] The algorithm's population size is determined by the population size fitness function. When the algorithm satisfies Equation (12) five times in a row, the algorithm stops calculating the fitness function. Since the algorithm is evolving in a positive direction, the situation of Equation (13) does not exist in theory. This method not only provides a basis for adjusting the algorithm's population size, but also minimizes unnecessary calculation steps.

[0214] The population increase mechanism is:

[0215] N(t+1)=N(t)+k1×N S (t)(14)

[0216] Where t represents the tth iteration; k1 is a random number between the interval [0, 1]; N(t) represents the population size of the algorithm at the tth iteration; N S (t) is the number of non-dominated solutions generated in the tth iteration of the algorithm.

[0217] The particles added by the population increase mechanism are the particles corresponding to the non-dominated solutions in the target space. When the algorithm executes the population increase mechanism, some particles are randomly selected from the non-dominated particles and their position information is assigned to the newly added particles, thereby improving the effectiveness of the newly added particles.

[0218] The population reduction mechanism is:

[0219] N(t+1)=N(t)-k2×(N(t)-N S (t))(15)

[0220] Where k2 is a random number between the interval [0, 1].

[0221] The particles reduced by the population reduction mechanism are the particles corresponding to the dominated solutions in the target space. When the algorithm executes the population reduction mechanism, some dominated particles are deleted, thereby accelerating the convergence of the algorithm and improving its performance to a certain extent.

[0222] Input: Randomly initialize a population of N particles and initialize the particle position x i (0), speed v i (0), inertia weight ω, learning factors c1 and c2. Set the initial position of each particle to the current historical optimal position p i (0), select the global optimal position g(0) from the initial external archive;

[0223] 1. Update the particle position x according to equations (4) and (5) i (v) and speed v i (t);

[0224] 2. Evaluate the fitness value of each particle and perform non-dominated sorting;

[0225] 3. Update external archives;

[0226] 4. Update individual optimal position p i (t), global optimal position g(t);

[0227] 5. Calculate CS and CSP values according to equations (6), (7), (8) and (9);

[0228] 6. According to equations (10), (11), (12) and (13), determine whether the current iteration algorithm should use the population increase or decrease mechanism;

[0229] 7. Automatically adjust the population size of the algorithm according to formula (14) or formula (15).

[0230] The present invention discloses a method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm. The method comprises a parameter identification module and an SOC estimation module, and the modules are signal-connected.

[0231] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0232] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0233] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0234] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0235] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0236] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm, characterized by: Including parameter identification module and SOC estimation module; The parameter identification module uses the particle swarm algorithm to improve the extended Kalman filter algorithm to perform real-time identification of battery model parameters; the SOC estimation module is used to use the extended Kalman filter algorithm, combined with the battery model parameter identification results, to estimate the battery SOC value in real time.

2. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 1 is characterized by: The parameter identification module includes a first input unit, a PSO-EKF algorithm unit and a first output unit; the first input unit is used to receive the input signal of the battery; the PSO-EKF algorithm unit is used to identify the battery model parameters in real time based on the input signal and the PSO-EKF algorithm; the first output unit is used to output the identified battery model parameters.

3. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 1 is characterized by: The SOC estimation module includes a second input unit, an extended Kalman filter algorithm unit, and a second output unit. The second input unit is used to receive the identified battery model parameters; the extended Kalman filter algorithm unit is used to estimate the SOC value of the battery in real time based on the battery model and the identified battery model parameters; and the second output unit is used to output the estimated SOC value.

4. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 2, characterized in that: The PSO-EKF algorithm unit includes a PSO-EKF algorithm setting unit and a parameter updating unit. The PSO-EKF algorithm unit is used to improve the extended Kalman filter algorithm and update parameters using the particle swarm algorithm; the parameter updating unit is used to update the battery model parameters using the PSO-EKF parameters and input signals.

5. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 3 is characterized by: The extended Kalman filter algorithm unit also includes a state prediction unit, an observation update unit and an error covariance update unit; the state prediction unit is used to predict the state of the battery; the observation update unit is used to update the state of the battery according to the observation value; the error covariance update unit is used to update the error covariance matrix.

6. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 2 is characterized in that : Through the parameter identification module, the Thevenin lithium-ion battery model is established. The electrical characteristic equation of this model in the time domain is as follows: The present lithium-ion battery model includes an ideal open-circuit voltage source U ocv , an ohmic internal resistance R0, and a parallel RC circuit; where U oc Represents the open circuit voltage of the battery, R0 is the internal resistance of the battery; the RC loop consists of the resistor R p and capacitor C p The terminal voltage of the battery is expressed as U, I L Represents the input current.

7. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 3 is characterized in that : The parameter identification module uses the PSO algorithm to improve the EKF algorithm; The EKF algorithm predicts the state value at time k: where f(x k ,u k ) is the nonlinear state equation; g(x k ,u k ) is the nonlinear output equation; w k is the system noise, with a mean of 0; the covariance is Q k ,v k is the system observation noise, with a mean of 0; the covariance R k , v k And w k Unrelated; PSO algorithm improves EKF: the variables included in the EKF algorithm p ,OCV,R0,R p , C p Identify and p ,OCV,R0,R p The corresponding noise (Q w1 ,Q w2 ,Q w3 ,Q w4 ) is a group of objects to be identified, and C p The corresponding noise Q w5 Make another group; use particle swarm algorithm for identification, set the population size to F, and the parameter to be identified to be [D*F]; the first group D=4, F=20, the second group D=1, F=20; Furthermore, the position of the first particle f and its flying speed are: x f,d =[x f,1 ,x f,2 ,…,x f,D ] v f,d =[v f,1 ,v f,2 ,…,v f,D ] The historical optimal position of the particle and the optimal position of the population are: p best =[p f,1 ,p f,2 ,…,p f,D ] g best =[g f,1 ,g f,2 ,…,g f,D ] to I p ,OCV,R0,R p The observation model is processed to obtain the objective function of the first set of optimization objects: F=min((R p,k +Q w,1 )(I p,k +Q w,2 )+(R 0,k +Q w,3 )I L,k +(OCV k +Q w,4 )-U0) 2 The second set of objective functions:

8. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 1 is characterized in that :Electrochemical energy storage system SOC can be used in energy storage battery management system and electric vehicle system.

9. The method for estimating the SOC of an electrochemical energy storage system based on PSO-EKF parameter identification and EKF algorithm according to claim 7, characterized in that: A population optimization model is established based on convergence speed and diversity, which is divided into three parts: establishing convergence speed indicators, establishing diversity indicators and establishing a population optimization model.

Images