Research On Indirect Prediction Method Of Remaining Life Based On GSO-ELM For Lithium-ion Battery

As a unit of power supplying parts,lithium-ion battery has many advantages such as small volume,light weight,wide temperature range and high energy ratio,which is the key part of many complex systems and electronic equipments,and then plays an important role in the whole system.During the system working,if the remaining life of lithium-ion battery cannot be monitored so as to be replaced or repaired in time,once the system fails,it will result in the reduction of system performance directly,even may cause catastrophic failures and lead to heavy losses,especially for aerospace system.Therefore,the remaining life prediction of lithium-ion battery has a significant practical value and research significance.Aiming at the remaining life prediction problem of lithium-ion battery,the remaining life prediction model of lithium-ion battery based on extreme learning machine(ELM)optimized by the glowworm swarm optimization(GSO)algorithm was proposed.In consideration of the defects associated with learning algorithm of traditional single hidden layer feed-forward neural network,ELM algorithm was introduced,while the theory of ELM was expounded in details.In the light of the problem that the capacity of lithium battery is difficult to monitor on-line,an indirect health factor method based on time interval to equal discharging voltage difference was proposed and a framework of RUL indirect prediction was established.Choosing experimental data of NASA lithium batteries(B5,B6,B7 and B18)as research object,the actual capacity was firstly extracted.Secondly,the voltage,current,charge and discharge cycles time were chosen at the discharge model of constant voltage,obtaining time interval of equal discharging voltage difference.Lastly,partial correlation coefficient method was used to prove strong correlation between actual capacity and interval of equal discharging voltage difference,which indicated that the RUL indirect prediction method was feasible by using the time interval of equal discharging voltage difference.In order to reduce or eliminate the shortcoming of ELM method,the glowworm swarm optimization(GSO)was introduced and the glowworm swarm optimization extreme learning machine(GSO-ELM)model was established and then used in the framework of lithium-ion battery RUL indirectly prediction.The influence of various parameters of GSO-ELM model on model performance were analyzed and verified in details.Appropriate parameters was chose and then used to predict the RUL,results showed that the GSO-ELM algorithm inherits the advantages of traditional ELM,which had fast learning speed,while compared with ELM,GSO-ELM owns the better tracking effect.From the experimental data(B5),it could be seen that the RUL error using the GSO-ELM algorithm was 2,and the average relative error was 5.36%,which indicated that this algorithm not only reduced the RUL prediction error,but also improved the result stability of the algorithm.