SOE And SOH Estimation Algorithms For Ternary Lithium-ion Batteries Based On Particle Filter

As the environmental pollution caused by the use of fossil energy is becoming more and more serious and it is decreasing day by day,the electrochemical energy storage system has become one of the important alternative energy sources.In the electrochemical energy storage system,Ternary lithium ion battery has attracted wide attention and been applied to the automotive power supply due to its advantages such as durability,stability,low cost and small environmental impact.Therefore,it is necessary to study the related state indexes of ternary lithium ion batteries.In the battery State,SOE(State of Energy)represents the current available remaining Energy of the battery,and its research is of great significance to the estimation of the remaining mileage of the vehicle and the optimization management of the battery.SOH(State of Health)represents the current Health State of the battery.With the gradual reduction of the Health State of the battery,the capacity is getting smaller and smaller.Therefore,the estimation of battery SOH is directly related to the life of the battery.The research of this paper is based on lithium battery SOE and SOH as the core,and mainly carries out the following work:(1)The establishment of ternary lithium ion batteries model and its parameter identification.In this paper,the battery experiment platform was briefly introduced at first,and then the battery SOE-OCV experimental characteristics were studied at 0℃,25℃ and 40℃ respectively.The results show that at different temperatures,the curve error of SOE-OCV becomes smaller and smaller as the value of SOE becomes larger and larger.Then,in order to simulate the actual operation of the vehicle,three working conditions of UDDS,NEDC and HWFET were introduced to study the dynamic characteristics of the battery.After completing the above experimental process,this paper establishes the corresponding equivalent circuit model according to the charging and discharging characteristics of the battery.By comparing the advantages and disadvantages of different battery models,it is found that the second-order RC model is convenient to operate and can better reflect the characteristics of the battery.Therefore,it is chosen as the equivalent circuit model of the lithium battery in this paper.After the cell model is established,the battery parameters are identified offline by Simulated Annealing(Simulated Annealing)under the condition of pulse experiments.The experimental results show that the maximum error of terminal voltage is 95 m V,and the average error is less than 5m V.However,basic SA process identification has too many parameters and consumes too much time.Therefore,an improved SA is proposed in this paper.By optimizing the identification equation and controlling the annealing parameters,a more simple identification process can be realized.Although the maximum and average error of terminal voltage identified are higher than that of basic SA,the time consumption is significantly reduced.(2)Theoretical derivation of particle filter and estimation of battery SOE.In the previous section,we first introduce the Bayesian Filter from the point of view of probability and statistics.Aiming at the complex integral problems existing in the process of Bayesian Filter in the process of nonlinear system,we use the Monte Carlo method to give the derivation process of Particle Filter in the continuous stochastic system.Based on the nonlinear characteristics of lithium batteries,PF was used in this paper to conduct SOE online estimation of batteries under UDDS dynamic conditions.Experimental results show that when the initial SOE error is 5%,the average SOE error of the basic PF estimate is 5.7%.Then,the number of particles is added for further verification,and the average error is reduced by 1%,without significant improvement.Considering the reason,it may be that the quality of prior particles is poor,so the method of assisted particle filtering(AVPF)is adopted,which combines the terminal voltage observation value to improve the quality of prior particles.The experimental results show that the average error is reduced to 3.65%,the error eventually converges to close to 1.5% over time.By analyzing the particle filtering process,it is found that increasing the number of particles and improving the prior particles cannot solve the particle degradation problem in the later stage of filtering,so an intelligent optimization algorithm-simulated annealing particle swarm particle filtering(SA-PSOPF)is used to update the particles in the later stage in real time and change the characteristic of singularity,so as to improve the accuracy of SOE estimation.The method is verified by experiments under the same dynamic conditions as the standard PF.The average error is 2.5%,and the error converges to close to 0.5% with the passage of time.The results prove that it has achieved a good estimation accuracy.(3)Adaptive joint estimation of battery SOE and SOH.The previous SOE estimation in this paper was carried out under the research of the standard PF and its improved PF,but there are also obvious problems,such as poor error convergence effect and long convergence time.Considering these problems,this paper introduces the adaptive particle filter(APF)from the point of view of adaptive particle number and adaptive noise variance.Based on this algorithm,experimental verification was carried out under three dynamic conditions.Compared with the standard PF and its improved algorithm,the adaptive particle filter made the SOE estimation error stable at about 1%in about 3000 seconds.In the process of SOE estimation,the health status of the battery will also change over time,which will also affect the accuracy of SOE estimation.Therefore,based on APF,a joint estimation model of SOE and SOH is proposed.Experimental verification was carried out under the same three dynamic conditions.The experimental results show that compared with the single particle PF,the joint estimation reduces the time of error stable convergence,which is less than 200 seconds on average.At the same time,in the estimation algorithm of SOH,the average error is about 5%.