A Multi-Model Collaborative Estimation Method For Electric Vehicle Driving Range

Electrification is the future of the automotive industry.Compared with fuel vehicles,the driving range of electric vehicles is shorter and charging is more inconvenient,making "mileage anxiety" one of the main problems for drivers and hindering the further promotion of electric vehicles.Therefore,it is vital to accurately predict the driving range of EVs under actual conditions to guide users to make reasonable travel plans and eliminate drivers’ mileage anxiety.Therefore,aiming at estimating the driving range of EVs accurately,this paper divides the EVs range estimation into three modules: battery state estimation,vehicle future travel conditions prediction and vehicle travel energy consumption calculation.A data-driven modeling approach is proposed to achieve a dynamic estimation of the microscopic range of EVs through multi-model collaboration.The specific research contents are as follows.As the energy supply side,battery is one of the crucial parts of EVs.The accurate estimation of the battery state is the basis of the accuracy of EVs range estimation.Firstly,based on of the real vehicle 32960 remote communication data,the Light GBM algorithm is used as a model framework to establish a high-precision battery SOC and SOH state estimation model based on real vehicle travel battery data in this paper.Secondly,this paper extracts the key features that affect the future driving conditions of EVs with the help of correlation analysis methods based on the time-ordered characteristics of vehicle speed and establishes a Long Short-term Memory Neural Network(LSTM)model for future driving condition prediction of EVs accordingly.By introducing our designed algorithm of autonomous window adjustment,the model can predict the vehicle driving conditions with high accuracy in a longer period in the future.Again,with the help of the fuzzy cluster analysis and Stacking model fusion method,this paper establishes an energy consumption estimation model based on dynamic driving conditions of EVs.Using kernel principal component analysis and fuzzy C-mean clustering,we cluster the vehicle data at the driving working condition level to obtain principal component features that have a good characterization of the original driving data characteristics and use the working condition feature parameters related to the above principal components as input variables for the energy consumption calculation model.Using the electric vehicle power consumption rate per unit mile as a scale,we optimized the model’s parameters.The results show that the estimation accuracy and robustness of the model are significantly improved compared with the traditional energy consumption estimation model.Finally,by combining the available residual energy from the battery SOC and SOH estimation models,the driving information from the vehicle driving condition prediction model,and the energy consumption obtained from the vehicle energy consumption model under this operating condition,we obtain an accurate estimation of the microscopic range of EVs and an EV range estimation framework based on the multi-model collaboration.By inputting the EV data under real travel conditions into the above multi-model collaborative estimation framework,we find that the framework achieves good estimation results for high,medium and low SOC data segments.The accuracy of the range estimation is significantly improved compared with the current estimation level in the industry,thus demonstrating the effectiveness and stability of the multi-model collaborative estimation method for EV range reported in this paper.