Remaining Useful Life Prediction Method Of Lithium Ion Batteries Based On Extreme Learning Machine

As a green and recyclable power supply equipment,lithium-ion batteries have been widely used in aerospace,aviation and other advanced fields.In order to ensure that lithium-ion batteries can output electricity stably,it is particularly important to predict their remaining useful life(RUL).Because the electrochemical reactions in lithium-ion batteries are complex,it is difficult to accurately describe the degradation process using physical models.In this paper,the RUL of lithium-ion batteries is predicted by the method based on Extreme Learning Machine(ELM),which is well used in RUL prediction of lithium-ion batteries because of its simplicity and high efficiency.It avoids the trouble of traditional gradient descent based neural network and machine learning algorithm represented by support vector machine(SVM).Several measures have been taken to accurately predict the RUL of lithium-ion batteries.Firstly,the collected lithium-ion batteries data of cyclic charging and discharging are deeply excavated to find a number of parameters that can indirectly characterize the performance degradation of lithium-ion batteries.Then,the training and prediction model of the ELM is constructed and compared with the single-parameter model.The experimental results show that the multi-parameter model has a significant improvement in prediction accuracy.However,the structure of ELM has its own characteristics.The ELM occasionally produces results that exceed the normal norm in many prediction processes,which greatly reduces the credibility of single prediction results.Therefore,using genetic algorithm and particle swarm optimization algorithm for reference,a combinatorial optimization algorithm is constructed to optimize the input of the ELM.Finally,the optimized ELM model iscompared with the traditional data-driven algorithm.Through many experiments,it is verified that the model not only has higher prediction accuracy,but also has excellent stability,and can accurately express the working status of lithium-ion batteries in the future.Several indirect parameters which can characterize the performance degradation of lithium-ion batteries were found by digging the charge and discharge data of lithium-ion batteries in depth.And,the indirect parameters used as input parameters of the model.The input of ELM is optimized by improved particle swarm optimization,and the RUL prediction model of lithium-ion battery is constructed.From the statistical data of simulation results,the prediction error of this method is stable at about 2%,which meets the requirements of most working conditions.