Prediction of Low-Temperature Energy Consumption and Driving Range of Pure Electric Vehicles Based on the CatBoost Algorithm

This study utilizes real road driving data for a specific model of pure electric vehicles to segment and select kinematic segments. Feature parameters are extracted from these segments and subjected to dimensionality reduction analysis. Based on the reduced principal component feature parameters, a particle swarm optimization-based K-means clustering analysis is conducted to classify driving conditions by energy consumption rate. The results reveal that the average energy consumption rate is lowest during medium-speed driving conditions and highest during high-speed driving conditions, with a 7.85% and 12.62% increase, respectively, compared to low-speed conditions. A CatBoost algorithm prediction model is developed based on these principal component feature parameters to forecast lowtemperature energy consumption and driving range. The results indicate a maximum absolute error of 0.52 kWh and an average relative error of 3.46% for cumulative energy consumption prediction, with a final relative error of 0.59% upon trip completion. The maximum absolute error for predicted remaining range compared to real remaining range is 8.18 km, with an average absolute error of 1.77 km and an average relative error of 4.04%.