Hybrid power system energy management method based on A3C algorithm

Abstract

The invention discloses a hybrid power system energy management method based on an A3C algorithm, belongs to the field of hybrid power automobile energy management, and aims to solve the correlation problem of Actor and Critic and the correlation problem of experience playback pool samples through an asynchronous dominant action evaluation method on the premise of ensuring the dynamic property ofan automobile. Rapid convergence of the neural network can be realized on the basis of ensuring the fuel economy of the automobile. The method mainly comprises the following steps: establishing an A3Cagent model; setting the state, action and return of the A3C agent model to obtain a set A3C agent model; obtaining a related training data set, and training the A3C agent model according to the obtained related training data set to obtain a trained A3C agent model; and using the trained A3C agent model to perform energy management of the parallel hybrid vehicle.

Description

technical field [0001] The invention belongs to the field of energy management of hybrid electric vehicles, in particular to an energy management method of a hybrid electric system based on an A3C algorithm. Background technique [0002] Energy is an important material basis for the survival and development of human society. In recent years, with the rapid development of the automobile industry, the problems of energy shortage and environmental pollution have become more and more serious, and the use of energy in automobiles has also attracted the attention of all walks of life. In order to better solve the problems of energy shortage and environmental pollution, hybrid electric vehicles gradually appear in the modern market. Generally speaking, a hybrid electric vehicle consists of two power sources, an internal combustion engine and an electric motor, so an energy management system is essential for a hybrid electric vehicle. Energy management systems can coordinate the m...

AI Technical Summary

Problems solved by technology

The rule-based energy management strategy has low technical difficulty, small amount of online calculation and high real-time performance, so it is widely used in the field of hybrid electric vehicle energy management industry, but the formulation of relevant rules needs to rely on expert experience. The control strategy is more sensitive to working conditions, lacks certain adaptability, poor robustness, and cannot guarantee optimality; the second is the energy management strategy based on optimization, and the energy management strategy based on global optimization has the ability to obtain the global optimal advantages, but at the same time, it has the disadvantages of requiring known global operating conditions and long calculation time, so it cannot be applied to real-time control of automobiles. These strategies are generally used as test benchmarks for other control strategies

Although the existing energy management strategies can achieve good performance, they still have the disadvantages of large amount of calculation and poor adaptability to working conditions; in order to achieve better performance of automotive energy management systems, in recent years, learning-based algorithms have begun to emerge , especially Actor-Critic

[0004] However, the Actor-Critic method also has many problems. The Actor-Critic needs to calculate the Q value through the Monte Carlo method, which requires a complete state sequence and can only iteratively update the strategy alone. Both the Actor network and the Critic network need to be updated at the same time. There is a strong correlation between the two, which leads to difficult convergence. In order to remove the correlation between Actor and Critic, the DDPG algorithm is proposed. This algorithm adopts a double neural network structure and introduces an experience playback pool so that the neural network can Perform update iterations, but the samples in the experience playback pool in DDPG still have certain relevance

Images