Research On Optimization Strategy Of Electric Vehicles Energy System With Battery Thermal Management

The environment pollution caused by the fossil fuels in the transportation is becoming more and more serious.Several countries,including China,USA,and Japan,have taken measures to optimize the transformation of energy structure and encourage the development of green cars.Currently,electric vehicles（EVs）are the most feasible alternative to fuel vehicles to realize green and low-carbon transportation and maintain energy security.Therefore,EVs have developed rapidly in recent years.However,due to the slow development of battery energy storage technology,the short driving range has become a bottleneck in the promotion of EVs.Especially,in high-speed and/or aggresive driving scenarios,the cruising range of EVs attenuates severely.It is desirable to improve the energy efficiency of EVs.In this paper,we propose an energy system optimization strategy for EVs with battery thermal management based on several reinforcement learning（RL）and deep reinforcement learning algorithms（DRL）,and innovatively combine the battery thermal effect with the design of the energy system strategy,which effectively relieves the problem of low energy efficiency under high load driving conditions.Specifically,we model the EVs’ energy system with battery thermal management based on a thorough investigation of the related theoretical and technical researches,and analyze the energy flow among the various modules of the vehicles.We fully explain the environmental interaction models,such as the battery capacity temperature change model,the driving power calculation model,and the battery heat generation/dissipation calculation model,as well as the optimization goal and limiting conditions.Also,we demonstrate the feasibility of using RL to solve the optimization of EVs’ energy systems,and explores the above optimization problems from both feature extraction and decision-making.We adopt gate recurrent unit（GRU）to extract the characteristics of the velocity sequence considering the serialized structural characteristics of the vehicle driving state and use Q-learning,deep Q network（DQN）,double deep Q network（double DQN）to make the decision.After the iterations for the energy system optimization strategy,the proposed algorithms have the characteristics of an efficient decision-making and a wide compatibility.Finally,we compare the proposed algorithms with the fuzzy control strategy（as the baseline）under four driving test cycles: NEDC,FTP-75,HWEFT and US06,and obtain the optimal decision sequence length settings based on the simulation comparisons and performance evaluations.We also analyze the diversified performance of each algorithm under different test cycles.In particular,the results show that while achieving an energy consumption like that of other algorithms under urban driving conditions,the double DQN additionally saves at least 6.7% energy during aggressive driving,respectively.