Research On RUL Prediction Method Of Lithium-ion Battery Based On Neural Network

Traditional energy depletion and environmental pollution have caused the concern on new energy field.Among all kinds of new energy batteries,lithium-ion batteries are widely used in many industries especially in the field of electric vehicles because of their excellent advantages such as small size,high energy density,long life cycle,zero emissions,and pollution-free.However,in practical applications,due to the influence of its own chemical changes and strict use conditions,lithium-ion battery life often fails to reach the predetermined life value.System failures and safety accidents caused by life failure of battery pose a great threat to people’s life and property.In order to realize real-time monitoring and management of battery operating status and avoid safety accidents,the system equipment is equipped with a battery management system(BMS)to evaluate the state-of-health(SOH)of the battery and predict the remaining life(RUL).An accurate prediction model based on battery degradation characteristics is absolutely essential for a precise estimation of battery performance.Therefore,from the perspective of online application of lithium-ion battery BMS system,this paper is devoted to the research of RUL prediction method of lithium-ion battery with its monitoring data.The specific research contents mainly include:(1)Aiming at the online measurement difficulties of direct health indicators(HIs)such as capacity and impedance,an adaptive indirect HI construction method based on intelligent algorithm is proposed.A number of potential HIs that can reflect the degradation of the battery are selected from battery measurement parameters.Then,the principal component analysis(PCA)algorithm and the sparse auto-encoder(SAE)neural network are designed to fuse potential HIs respectively.The correlation between the two type fusion HI and the capacity are verified by Spearman rank correlation coefficient.The experiment proved the effectiveness of the proposed method.(2)For the difficulties in establishing accurately mechanism model caused by the high nonlinearity and dynamic characteristics of lithium-ion battery system,a RUL prediction method based on the nonlinear auto-regressive network with exogenous Inputs(NARX)model is developed.The model can effectively learn the nonlinear and non-stationary time series by setting the delay unit and external feedback structure.Simultaneously,in order to solve problem that existing data-driven based prediction method cannot directly fit the RUL prediction for other batteries at the same type.A RUL prediction strategy for the same type modeled by using all life cycle data of a battery is proposed,which realizes RUL prediction for the same type battery when the consistency is not strong.(3)Aiming at solving the difficulty that the general neural network model cannot learn the long-term dependence of battery degradation,a long short-term memory(LSTM)neural network is proposed to construct a single lithium-ion battery RUL prediction model.Experiments show that the model can obtain more stable and accurate RUL prediction results,which greatly improves the problem of error accumulation such as NAXR and other methods.(4)Aiming at the problem of local rapid rise and fall of battery degradation data caused by short-term recovery of battery capacity during the cycle of charge and discharge of lithium-ion battery,a LSTM neural network model with adaptive sliding window is proposed for lithium-ion battery RUL prediction.The model adds a window that can adaptively change the number of input according to the battery degradation curves in the LSTM network structure.It can not only learn the long-term time dependence characteristics of battery data,but also better capture the characteristics of short-term local recovery,greatly improving the accuracy and reliability of the prediction results.Experiments show that the model is less affected by the prediction starting point and has strong adaptability and robustness.