| The world is facing some serious problems,such as environmental pollution,global warming and depletion of oil resources,new energy vehicles have become the key research object of major automobile manufacturers.With the development of fuel cell technology,fuel cell vehicles give full play to the advantages of zero emission and energy saving for new energy vehicles,which has attracted researchers’ extensive attention at home and abroad.Energy management strategy uses optimization algorithms to realize the optimal power distribution of energy sources,and effectively improve economy,power and drivability of vehicle.Deep reinforcement learning algorithm can effectively deal with multi-objective energy management optimization problems,and has significant advantages in optimization effect,real-time and selflearning;Therefore,this paper will carry out the research on multi-objective energy management of fuel cell vehicle based on deep reinforcement learning.The specific research contents are as follows:Firstly,the vehicle powertrain modeling is studied.The longitudinal dynamics model and the transmission components’ control oriented mathematical model are built;The factors affecting the durability of power source are analyzed,and the fuel cell degradation model and battery aging model are established;The basic principle of global optimization algorithm-dynamic programming(DP)is described,realizes the DP algorithm to solve the optimization problem of energy management,and analyzes the working characteristics of each power source under the global optimization strategy.Secondly,research on the deep reinforcement learning-based energy management strategy.Based on the concept and principle of reinforcement learning,the basic theories of deep Q-network(DQN)algorithm and deep deterministic policy gradient(DDPG)algorithm are described respectively,and constructs the energy management framework of fuel cell vehicle;The off-line training results of DQN and DDPG under UDDS are compared and analyzed,and the trained agent is applied to the untrained combined driving cycle to verify the adaptability of the proposed strategy.Thirdly,research on soft actor critic(SAC)-based life-conscious energy management strategy.The basic concept of SAC algorithm is described,and the advantages and disadvantages of SAC algorithm and traditional deep reinforcement learning under the energy management framework is analyzed;Based on the fuel cell degradation model and battery aging model,the multi-objective energy management optimization problem of hydrogen consumption,SOC maintenance,fuel cell degradation and battery aging is constructed,the compromise scheme between power source aging and hydrogen consumption is determined,the results of SAC-based life-conscious energy management are analyzed,and the effectiveness of the proposed strategy under combined driving cycle is verified.Finally,the research on the collaborative energy management considering air conditioning system.Based on the thermodynamic equilibrium theory,the airconditioning system model and cabin heat load model are established,the collaborative energy management framework considering the air-conditioning system is established,and the problem of collaborative energy management is solved by SAC algorithm;The results of collaborative energy management under high-temperature refrigeration and low-temperature heating are compared and the effectiveness of the proposed strategy is further verified under the combined driving cycle. |
