A Hybrid System Energy Management Strategy Based on Inverse Deep Reinforcement Learning

Abstract

The invention discloses an energy management strategy of a hybrid system based on reverse deep reinforcement learning. The strategy includes: using the optimization solution method to calculate the globally optimized SOC result as expert knowledge; creating a reward neural network; using reverse reinforcement learning to learn expert knowledge to obtain parameters of the reward neural network; creating an action neural network and evaluating the neural network; setting the vehicle SOC value before interaction; input the obtained SOC value before interaction into the reward neural network to obtain the reward value; input the obtained SOC value before interaction into the action neural network to obtain the mode allocation ratio; use the mode allocation ratio to interact with the environment, and obtain SOC value after interaction; input the SOC value before interaction, mode allocation ratio, reward value, and SOC value after interaction into the evaluation neural network to obtain the evaluation value; the agent calculates the gradient of each network and backpropagates to update the network parameters until the training is completed. The invention can learn the optimal reward function from expert knowledge, so that the effect of deep reinforcement learning is better.

Description

technical field [0001] The invention relates to the field of hybrid system energy management, in particular to a hybrid system energy management strategy based on reverse deep reinforcement learning. Background technique [0002] The hybrid electromechanical coupling device couples the power of multiple power sources such as the internal combustion engine and the electric motor in a hybrid electric vehicle (HEV) and performs reasonable power distribution, and transmits the power to the drive axle to drive the vehicle. It can be regarded as an automatic transmission system with electric motors by integrating one or more electric motors into the transmission, which is a complex system integrating mechanics, electricity, chemistry and thermodynamics. China's automobile fuel consumption regulations put forward very high requirements for energy conservation and emission reduction of vehicle manufacturers in the next 10-15 years. With the further increase in the pressure of energ...

AI Technical Summary

Problems solved by technology

However, the global optimization energy management strategy is only aimed at solving known working conditions and cannot be applied online. Traditional reinforcement learning methods perform well in solving tasks with limited state and action spaces, but they are difficult to solve in terms of state and action space dimensions. When the problem is high, it seems powerless (Chen Xiliang, Cao Lei, He Ming, Li Chenxi, Xu Zhixiong. A review of deep reverse reinforcement learning [J]. Computer Engineering and Application, 2018, 54(05): 24-35.)

Images