Research on electric vehicle (EV) driving range prediction method based on PSO-LSSVM

Electric vehicle (EV) driving range directly reflects EVs' performance, safety, reliability and economy. EV has gained wide attention in recent years. However, most of researches are carried out under ideal conditions and the existing methods have numerous drawbacks. This paper presents a novel prediction method based on a least squares support vector machine (LSSVM) model with parameters γ and σ² optimized by particle swarm optimization (PSO). The main parameters which cannot be obtained directly by drivers such as days, temperature, depth of discharge (DOD) of battery pack are used for training model to predict EV driving range. Furthermore, the performance of PSO-LSSVM model is illustrated by statistical parameters (RE and AARE). AARE of training data and testing data is 1.99% and 5.99% respectively. The results suggest that the model has a stability, generalization ability and reliable predictive performance to predict EV driving range. Meanwhile, the results can also provide a guidance for drivers to grasp and manage their EVs' health conditions and predict the driving range.

I. Introduction

As the environmental problems and energy crisis are becoming more and more serious, electric vehicles (EVs) are considered to be the most significant green transportation alternative for the foreseeable future [1]. However, because the energy density of commonly used lithium-ion batteries are far lower than the fuel like gasoline or diesel [2], limited driving range is the biggest obstacle of developing EVs. Drivers are seriously concerned about EV driving range because EV driving range can directly reflect EVs' performance, safety, health and reliability. Therefore, it is significant to predict EV driving range and analyze the primary factors determining EV driving range.