Research On Battery Temperature Prediction Method Based On Data-driven Electric Bus

Under the state’s key support for electric vehicles,the development of electric vehicles has advanced by leaps and bounds.The bottleneck of battery technology has always been the key to hindering the development of the entire electric vehicle industry,and the state prediction of batteries is one of the major difficulties.In the actual use of pure electric vehicles,the temperature of the power battery is a very important factor,which will affect the charging and discharging efficiency,life and safety of the battery,so it is important to accurately predict the temperature of the power battery.In this paper,the temperature prediction method of electric vehicle power battery is carried out by taking the pure electric bus battery data as the research object.Based on data-driven method,a pure electric bus power battery temperature prediction model is proposed to effectively predict the temperature of the battery to improve the performance and reliability of the battery.Here are the main works of this article:(1)Firstly,this paper summarizes the methods for estimating battery temperature at home and abroad,compares and analyzes the advantages and disadvantages of various types of methods and the difficulties existing in current research,and determines the research direction of this paper on this basis.This paper introduces the battery data and temperature data used in the experiment,matches the weather data with the battery data,and then preprocesses the experimental data and extracts the charging fragment data of the electric bus during the charging process.Finally,the MIC-based segmentation strategy was adopted,and it was determined that the average MIC of the fragments was 0.812 when the segmentation length was 50,and the overall correlation between the fragments was the best.(2)Secondly,after comparing the characteristics of different types of lithium-ion batteries and the technical routes of different manufacturers,this paper takes lithium iron phosphate batteries as an example to conduct an in-depth analysis of their charging principles,charging characteristics,temperature characteristics and factors affecting temperature.The analysis results show that the main factors affecting the temperature during battery charging include charging time,state of charge and charging voltage.Then,by fusing CNN and RNN neural networks,a pure electric bus battery temperature prediction model based on CNN-LSTM is established,and the factor of ambient temperature is comprehensively considered,and the input feature vector of the model is composed by combining the five characteristics of charging time,voltage,maximum temperature,minimum temperature and SOC,and the model is trained and hyperparameter adjusted.(3)Finally,the robustness of the temperature prediction model is verified,and the ablation experiment proves that the CNN-LSTM model proposed in this paper has better performance than the other three prediction models,and the short-term temperature prediction model has good predictive ability.The average value of RMSE for the battery temperature prediction results of eight electric buses was 0.774,the average value of MAE was 0.6348,and the average value of R2＿Score was 0.9798.By adopting the improved loss function,the prediction accuracy of the maximum and minimum temperature change trend of the LSTM-based maximum temperature prediction model is improved,and the prediction results using the improved loss function are0.181 R2＿Score 4 and 0.0232 lower than the original MAE on average.