Method for predicting residual life of battery

Abstract

The invention relates to a battery residual life prediction method. The method comprises the following steps: constructing a battery capacity aging model; setting the current discharge-charge cycle number as k, extracting initial state particles # imgabs1 # according to the initial prior probability density # imgabs0 #, setting the number N of the particles as 50, and i = 1, 2,..., N; carrying out importance sampling based on IALO-PF; calculating the average value of the particles to obtain state estimation; and predicting the remaining service life. According to the method, particle distribution is optimized by introducing the IALO algorithm, the resampling rate is reduced, and particle depletion is relieved. In addition, for the problems of convergence speed and local optimality of a standard ALO algorithm, the ALO algorithm is improved. And the prediction precision of the RUL is obviously improved.

Description

Technical Field

[0001] The invention belongs to the technical field of battery life prediction, and in particular relates to a method for predicting remaining battery life. Background Art

[0002] Due to the global energy crisis and environmental pollution, electric vehicles (EVs) have become a major trend in the automotive industry. Lithium-ion batteries are widely used in electric vehicles due to their high power density and long cycle life. As an indispensable component of electric vehicles, it is necessary to reasonably monitor the status of the battery through a battery management system (BMS). Therefore, it is crucial to achieve accurate prediction of the remaining useful life (RUL) and maintain the battery in time before an impending failure event. Lithium battery remaining useful life prediction methods are divided into three categories: model-based, data-driven, and hybrid.

[0003] Model-based methods are usually combined with adaptive filtering techniques, such as Kalman filtering, particle filtering (PF), etc. These methods are flexible in application and can handle process uncertainties. As a random statistical filtering algorithm, the PF method has been widely used in the field of RUL prediction. However, the standard PF method inevitably has the problem of particle diffusion, which causes the weights of most particles to become diffusible after several iterations. This has a negative impact on the accuracy and robustness of the remaining service life prediction. A feasible way is to embed the intelligent optimization algorithm into the standard PF framework to optimize the particle distribution. PSO, CS, WOA, GA and other optimization algorithms can only obtain the optimal solution after each iteration, but the ALO algorithm can ensure that the positions of all ant lions are in the optimal n positions after each iteration. Therefore, the present invention chooses to embed the ALO algorithm in the PF framework to improve the problem of particle degradation. However, the ALO algorithm still has certain limitations, such as easy to fall into local optimality, slow convergence in the later stage, and easy to have premature convergence. Therefore, the present invention aims to solve the particle degradation problem of PF in the process of predicting the remaining service life of lithium batteries, optimize the traditional ALO algorithm, and predict the remaining service life of lithium batteries by IALO improved PF. Summary of the invention

[0004] The present invention aims to solve the particle degradation problem of PF in the process of predicting the remaining service life of lithium batteries. The traditional ALO algorithm is optimized and the remaining service life of lithium batteries is predicted by PF improved by IALO. Therefore, the present invention proposes a method for predicting the remaining service life of batteries.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for predicting remaining battery life, the method comprising the following steps: Step 1: Build a battery capacity aging model; Step 2: Set the current number of discharge-charge cycles to k, based on the initial prior probability density , extract the initial state particles , the number of particles N is set to 50, and i=1, 2, …, N; Step 3: Importance sampling based on IALO-PF; Step 4: Calculate the average value of the particles to obtain the state estimate; Step 5: Predict the remaining useful life.

[0006] As a further technical solution of the present invention, in step 1, the step of constructing a battery capacity aging model includes: The RUL prediction accuracy based on the PF method depends on an accurate aging model. The battery capacity decay to 80% of the rated capacity is set as reaching the end of life. In the empirical model, the double exponential model has better global regression ability than other models and is widely used in data regression analysis. Therefore, the battery capacity aging model uses a double exponential model, and the double exponential model of battery capacity aging is expressed as: ; Where k is the number of discharge-charge cycles, represents the battery capacity at the kth cycle, are the model parameters to be identified in the aging experiment, where It is related to the internal resistance of the battery. Related to battery aging rate; Considering the complexity and real-time nature of the internal reactions of the aging process, the battery aging process is assumed to be a first-order Markov process and expressed as:

[0007] where g(·) and h(·) represent the state transition equation and measurement equation of the aging model, respectively; represents the state vector at period k, which is composed of model parameters to composition; is the battery capacity data at cycle k; and represents the process and measurement noise of the system.

[0008] As a further technical solution of the present invention, in step three, the step of importance sampling based on IALO-PF includes: Choose the prior PDF as the state PDF, i.e. ; IALO algorithm for iterative optimization , the steps are as follows: Step S1: IALO algorithm parameter initialization: set the maximum number of iterations , the number of ants and ant lions n, the dimension dim, the bounds of each dimension and , t is the current iteration; Step S2: Initialize ant lions and ants: Randomly initialize the positions of all ants and all antlion locations ; Step S3: Set the fitness function and record the elite ant lion: Set the fitness function f, calculate the fitness value of the ant lion and sort them from small to large, and then select the ant lion with the smallest fitness value as the elite ant lion individual, recorded as ; Step S4: Select antlions through roulette strategy: introduce judgment function Use the roulette strategy to select antlions , if the position A of the selected ant lion in the population is less than , this choice is considered appropriate; otherwise, the ant lion is reselected by roulette until the condition is met; by adding a judgment function to the roulette strategy, the fitness value of the selected sample is judged, thereby improving the global search ability and avoiding the algorithm from falling into the local optimum; Step S5: Random movement: The ants move randomly around the ant lion, and the minimum and maximum values ​​of the ant walking boundary are expressed as: ; The random walk process is expressed as: ; Step S6: Update the position of the ant: Normalize the position of the ant: , the ant's position will be updated to: ; Step S7: Introduce the Logistic chaos operator: The Logistic chaos mapping expression is: ;in are all in the range of [0, 1]. represents the initial solution, Indicates that by A set of new solutions mapped to the solution space; Set the position of ant i to , normalized Each dimension of can be normalized to Location : ; in Represents the minimum and maximum values ​​of the r-th dimension variable of all ants; Mapping the normalized variables to obtain a chaotic variable sequence , transform it to the original solution space according to the following equation, and transform the ant After random movement, you will get Chaos Ant ; ; Step S8: Update the position of the ant lion: compare the fitness values ​​of the chaotic ants and the normal ants, retain N groups of solutions with smaller fitness values, and use them to update the elite ants. By introducing the logical chaos operator, the algorithm has both global search capability and local optimization capability, which improves the convergence and diversity of the algorithm. Step S9: Introducing Gaussian perturbation operator: During the n consecutive iterations of the algorithm (n is set to 5 in the present invention), if the minimum fitness value remains unchanged, the algorithm falls into local convergence, and the Gaussian perturbation operator is used as external intervention. The condition for falling into the local optimum is written as: ; The Gaussian perturbation operator is as follows: ; It is an updated Antlion. is the new ant colony generated by the Gaussian perturbation operator, Follows a normal distribution, where x is The random number, is the position function, set to: ; m is a random value between 0 and 1; Step S10: Update the position of the ant lion: by comparing and The fitness value of the ant is adjusted according to the local optimal value; Step S11: Optimizing particles: When Optimization completed, the position of the antlion is used to optimize the particles Otherwise, when Go to step S2.

[0009] As a further technical solution of the present invention, in step 4, the average value of the particles is calculated to obtain the formula for state estimation as follows:

[0010] As a further technical solution of the present invention, in step 5, the step of predicting the remaining service life includes: After determining the model parameters through battery data, the IALO-modified PF-based method is used to estimate the starting prediction time. ; By passing the parameter Introducing the formula To obtain the predicted capacity curve and record the RUL prediction.

[0011] The terms involved in the present invention are explained as follows: PF: PF (Particle Filter) is a state estimation technology based on the Monte Carlo method, which is used for state estimation in nonlinear and non-Gaussian systems. Particle filtering represents the posterior probability distribution of the state by randomly sampling a set of particles in the state space, and then updates and resamples these particles according to the dynamic model and measurement data of the system to approximate the true posterior distribution.

[0012] ALO: ALO (Ant Lion Optimizer) is a new heuristic algorithm based on the optimization behavior pattern of ant lions for solving optimization problems. It solves practical optimization problems by imitating the process of ant lions capturing ants. ALO contains two populations, ants and ant lions, and the positions represent multiple optimization solutions. The process of finding the optimal value is similar to that of ant lions capturing ants, and each individual has its fitness. When the fitness of the ant is better than that of the ant lion, the ant lion goes to the position of the ant and captures the ant. In each iteration, all ants move randomly to generate new populations to increase diversity.

[0013] PDF:Probability density function is a method used to describe the probability distribution of a random variable. For continuous random variables, the probability density function represents the probability that the random variable takes a specific value in a given interval. For discrete random variables, the probability density function represents the probability that the random variable takes a specific value near a given point.

[0014] Logistic Chaos Operator: The Logistic Chaos Operator is a method for generating random numbers that combines chaos theory and random walk behavior.

[0015] Gaussian perturbation: Gaussian perturbation is a method used in optimization algorithms to search for better solutions by introducing random perturbations in parameter space. In this method, the perturbation follows a normal distribution (Gaussian distribution) to explore different parameter combinations. This method can help the algorithm jump out of the local optimal solution to find a better global optimal solution.

[0016] Compared with the prior art, the present invention has the following beneficial effects: First, in view of the particle depletion problem existing in the traditional particle filter algorithm, the present invention optimizes the particle distribution by introducing the IALO algorithm, reduces the resampling rate, and alleviates the particle depletion. In addition, the ALO algorithm is improved in view of the convergence speed and local optimality problems existing in the standard ALO algorithm. By introducing the judgment function in the roulette strategy, the premature convergence problem caused by the poor fitness of the ant lion individuals randomly selected in the early stage of the algorithm is avoided, which helps to alleviate the problem that the algorithm is easy to fall into the local optimal state, without the need to set parameters for the proportion of random walks of ants around elite ant lions and ant lions selected by roulette at different periods. By introducing the logical chaos operator, the algorithm has both global search capability and local optimization capability, which improves the convergence and diversity of the algorithm. By introducing the Gaussian perturbation operator to externally interfere with the local optimal solution, the local convergence problem is solved. The proposed IALO method solves the particle depletion problem, so that the prediction accuracy of RUL is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A diagram of a RUL prediction framework provided in an embodiment of the present invention.

[0018] Figure 2 A calculation process diagram of IALO provided in an embodiment of the present invention.

[0019] Figure 3 A complete remaining useful life prediction block diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0021] See also Figures 1 to 3 , a method for predicting remaining battery life, the method comprising the following steps: Step 1: Build a battery capacity aging model: The RUL prediction accuracy based on the PF method depends on an accurate aging model. The battery capacity decay to 80% of the rated capacity is set as reaching the end of life. In the empirical model, the double exponential model has better global regression ability than other models and is widely used in data regression analysis. Therefore, the battery capacity aging model uses a double exponential model, and the double exponential model of battery capacity aging is expressed as:

[0022] Where k is the number of discharge-charge cycles, represents the battery capacity at the kth cycle, are the model parameters to be identified in the aging experiment, where It is related to the internal resistance of the battery. Related to battery aging rate; Considering the complexity and real-time nature of the internal reactions of the aging process, the battery aging process is assumed to be a first-order Markov process and expressed as:

[0023] where g(·) and h(·) represent the state transition equation and measurement equation of the aging model, respectively; represents the state vector at period k, which is composed of model parameters to composition; is the battery capacity data at cycle k; and represents the process and measurement noise of the system.

[0024] Step 2: Set the current number of discharge-charge cycles to k, based on the initial prior probability density , extract the initial state particles , the number of particles N is set to 50, and i=1, 2, …, N; Step 3: Importance sampling based on IALO-PF: Choose the prior PDF as the state PDF, i.e. ; IALO algorithm for iterative optimization , the steps are as follows: Step S1: IALO algorithm parameter initialization: set the maximum number of iterations , the number of ants and antlions n, the dimension dim, the bounds of each dimension and , t is the current iteration; Step S2: Initialize ant lions and ants: Randomly initialize the positions of all ants and all antlion locations ; Step S3: Set the fitness function and record the elite ant lion: Set the fitness function f, calculate the fitness value of the ant lion and sort them from small to large, and then select the ant lion with the smallest fitness value as the elite ant lion individual, recorded as ; Step S4: Select antlions through roulette strategy: introduce judgment function Use the roulette strategy to select antlions , if the position A of the selected ant lion in the population is less than , this choice is considered appropriate; otherwise, the ant lion is reselected by roulette until the condition is met; by adding a judgment function to the roulette strategy, the fitness value of the selected sample is judged, thereby improving the global search ability and avoiding the algorithm from falling into the local optimum; Step S5: Random movement: The ants move randomly around the ant lion, and the minimum and maximum values ​​of the ant walking boundary are expressed as: ; The random walk process is expressed as: ; Step S6: Update the position of the ant: Normalize the position of the ant: , the ant's position will be updated to: ; Step S7: Introduce the Logistic chaos operator: The Logistic chaos mapping expression is: ;in are all in the range of [0, 1]. represents the initial solution, Indicates that by A set of new solutions mapped to the solution space; Set the position of ant i to , normalized Each dimension of can be normalized to Location : ; in Represents the minimum and maximum values ​​of the r-th dimension variable of all ants; Mapping the normalized variables to obtain a chaotic variable sequence , transform it to the original solution space according to the following equation, and transform the ant After random movement, you will get Chaos Ant ; ; Step S8: Update the position of the ant lion: compare the fitness values ​​of the chaotic ants and the normal ants, retain N groups of solutions with smaller fitness values, and use them to update the elite ants. By introducing the logical chaos operator, the algorithm has both global search capability and local optimization capability, which improves the convergence and diversity of the algorithm. Step S9: Introducing Gaussian perturbation operator: During the n consecutive iterations of the algorithm (n is set to 5 in the present invention), if the minimum fitness value remains unchanged, the algorithm falls into local convergence, and the Gaussian perturbation operator is used as external intervention. The condition for falling into the local optimum is written as: ; The Gaussian perturbation operator is as follows: ; It is an updated Antlion. is the new ant colony generated by the Gaussian perturbation operator, Follows a normal distribution, where x is The random number, is the position function, set to: ; m is a random value between 0 and 1; Step S10: Update the position of the ant lion: by comparing and The fitness value of the ant is adjusted according to the local optimal value; Step S11: Optimizing particles: When Optimization completed, the position of the antlion is used to optimize the particles Otherwise, when Go to step S2.

[0025] Step 4: Calculate the average value of the particles to obtain the state estimate;

[0026] Step 5: Predict the remaining service life: After determining the model parameters through battery data, the IALO-modified PF-based method is used to estimate the starting prediction time. ; By passing the parameter Introducing the formula To obtain the predicted capacity curve and record the RUL prediction.

[0027] In the standard particle filtering process, the weights of some particles gradually become very small, resulting in these particles contributing little to the overall prediction results. This phenomenon will increase the deviation of the battery life prediction results and reduce the prediction accuracy. However, the existing particle distribution optimization intelligent optimization algorithm has the problems of slow convergence speed and insufficient global search capability. The traditional ant lion algorithm (ALO) has limitations such as easy to fall into local optimality, slow convergence speed in the later stage, and easy to have premature convergence. In response to this phenomenon, the present invention proposes an improved ant colony optimization algorithm (IALO) to optimize particle distribution. IALO uses the latest measurement values ​​to generate the importance density function, moves particles to the maximum likelihood region, ensures the integrity of particles, and reduces particle degradation.

[0028] The main design ideas of the present invention are: introducing a judgment function to optimize the roulette strategy to avoid the premature convergence problem caused by the poor fitness of ant lion individuals randomly selected in the early stage of the algorithm; introducing the Logistic chaos operator into the standard ALO algorithm, combining the random movement behavior of ants, generating chaotic ants with the randomness and regularity of the chaos operator, combining the global search ability of the chaotic ants with the local optimization ability of the ant lion, improving the convergence speed in the early stage of the algorithm, and increasing the population diversity in the later stage of the algorithm; aiming at the phenomenon of local convergence, a Gaussian perturbation operator is proposed to perform external interference on the local optimal solution.

[0029] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0030] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for predicting remaining battery life, characterized in that: The method comprises the following steps: Step 1: Build a battery capacity aging model; Step 2: Set the current number of discharge-charge cycles to k, based on the initial prior probability density , extract the initial state particles , the number of particles N is set to 50, and i=1, 2, …, N; Step 3: Importance sampling based on IALO-PF; Step 4: Calculate the average value of the particles to obtain the state estimate; Step 5: Predict the remaining useful life.

2. A method for predicting remaining battery life according to claim 1, characterized in that: In step 1, the steps of building a battery capacity aging model include: The battery capacity aging model uses a double exponential model, and the double exponential model of battery capacity aging is expressed as: ； Where k is the number of discharge-charge cycles, represents the battery capacity at the kth cycle, are the model parameters to be identified in the aging experiment, where It is related to the internal resistance of the battery. Related to battery aging rate; The battery aging process is assumed to be a first-order Markov process and expressed as: ； where g(·) and h(·) represent the state transition equation and measurement equation of the aging model, respectively; represents the state vector at period k, which is composed of model parameters to composition; is the battery capacity data at cycle k; and represents the process and measurement noise of the system.

3. A method for predicting remaining battery life according to claim 2, characterized in that: In step 3, the steps of importance sampling based on IALO-PF include: Choose the prior PDF as the state PDF, i.e. ; IALO algorithm for iterative optimization , the steps are as follows: Step S1: IALO algorithm parameter initialization: set the maximum number of iterations , the number of ants and ant lions n, the dimension dim, the bounds of each dimension and , t is the current iteration; Step S2: Initialize ant lions and ants: Randomly initialize the positions of all ants and all antlion locations ; Step S3: Set the fitness function and record the elite ant lion: Set the fitness function f, calculate the fitness value of the ant lion and sort them from small to large, and then select the ant lion with the smallest fitness value as the elite ant lion individual, recorded as ; Step S4: Select antlions through roulette strategy: introduce judgment function Use the roulette strategy to select antlions , if the position A of the selected ant lion in the population is less than , this selection is considered appropriate; otherwise, the ant lion is reselected using roulette until the condition is met; the fitness value of the selected sample is judged by adding a judgment function to the roulette strategy; Step S5: Random movement: The ants move randomly around the ant lion, and the minimum and maximum values ​​of the ant walking boundary are expressed as: ； The random walk process is expressed as: ; Step S6: Update the position of the ant: Normalize the position of the ant: , the ant's position will be updated to: ; Step S7: Introduce the Logistic chaos operator: The Logistic chaos mapping expression is: ;in are all in the range of [0, 1]. represents the initial solution, Indicates that by A set of new solutions mapped to the solution space; Set the position of ant i to , normalized Each dimension of can be normalized to Location : ； in Represents the minimum and maximum values ​​of the r-th dimension variable of all ants; Mapping the normalized variables to obtain a chaotic variable sequence , transform it to the original solution space according to the following equation, and transform the ant After random movement, you get chaos ants ; ； Step S8: Update the position of the ant lion: compare the fitness values ​​of the chaotic ants and the normal ants, retain N groups of solutions with smaller fitness values, and use them to update the elite ants. By introducing the logical chaos operator, the algorithm has both global search capability and local optimization capability. Step S9: Introduce Gaussian perturbation operator: During the n consecutive iterations of the algorithm, if the minimum fitness value remains unchanged, the algorithm falls into local convergence. The Gaussian perturbation operator is used as external intervention, and the condition for falling into the local optimum is written as: ； The Gaussian perturbation operator is as follows: ； It is an updated Antlion. is the new ant colony generated by the Gaussian perturbation operator, Follows a normal distribution, where x is The random number, is the position function, set to: ； m is a random value between 0 and 1; Step S10: Update the position of the ant lion: by comparing and The fitness value of the ant is adjusted according to the local optimal value; Step S11: Optimizing particles: When Optimization completed, the position of the antlion is used to optimize the particles Otherwise, when Go to step S2.

4. A method for predicting remaining battery life according to claim 3, characterized in that: In step 4, the average value of the particles is calculated, and the formula for state estimation is as follows: 。 5. A method for predicting remaining battery life according to claim 4, characterized in that: In step 5, the steps to predict the remaining useful life include: After determining the model parameters through battery data, the IALO-modified PF-based method is used to estimate the starting prediction time. ; By passing the parameter Introducing the formula To obtain the predicted capacity curve and record the RUL prediction.

Images