Research On Hybrid Energy Storage Management Strategy Based On Driving Pattern Recognition For Electric Vehicles

Energy management strategy is the key technology of hybrid energy storage system for the electric vehicle,which can take advantage of complementary characteristics of Li-ion battery and supercapacitor.This strategy is of great significance for accelerating the power response,reducing battery loss,and improving vehicle economy.While the driving cycle of electric vehicles has an important influence on the energy management strategy of the hybrid energy storage system,there is still a great challenge to allocate the demand power of electric vehicles effectively and reasonably considering the driving cycles.To this end,this paper studies the driving pattern recognition method based on the extreme gradient boosting algorithm,and constructs an energy management model based on long shortterm memory networks to optimize the management strategy of the energy storage system online.The main research contents are as follows:Firstly,to improve the accuracy of driving pattern recognition for the electric vehicle,a driving pattern recognition method based on the extreme gradient boosting algorithm is proposed.A real-vehicle driving data acquisition scheme is developed,and noise reduction pre-processing is performed for the collected driving data.Based on partitioning driving data by driving blocks,feature extraction and dimensionality reduction are conducted based on principal component analysis,and cluster analysis is performed on the reduced-dimensional feature data.The driving blocks in the clustered sub-clusters are then joined together to construct a typical driving dataset.Using the reduced-dimensional features with driving pattern labels,the driving pattern recognition model is established by the extreme gradient boosting tree algorithm,and the accurate recognition of electric vehicle driving patterns is achieved.Secondly,to realize the energy management strategy optimization online,an energy management strategy based on long short-term memory networks is proposed.The multi-objective optimization problem of energy management is formulated to minimize battery degradation cost and electricity cost,and the global optimal solution of energy management is obtained offline by using dynamic programming algorithm.Then the dataset is constructed with the features of load demand current,vehicle speed,acceleration,supercapacitor charge state,and supercapacitor current in the previous moment,and with the corresponding global optimal solution as the label.This dataset is utilized to train three long short-term memory network energy management models for different driving patterns respectively.When the vehicle is driving,the driving pattern is recognized and the corresponding energy model is selected to map the running state to the optimal reference current,achieving the energy management optimization online.Finally,the evaluation of the real driving data set is conducted to verify the driving pattern recognition method based on the extreme gradient boosting algorithm and the energy management strategy based on the long shortterm memory network.The results presented that the proposed method and strategy can extend the battery life and reduce the operating cost of the whole vehicle.