Adaptive online improvement method for deep reinforcement learning type control strategy

• Main Authors: CHEN JIAXIN, LIN LIHONG, YANG WEI, LI JIACHENG, TANG XIAOLIN
• Format: Patent
• Language: Chinese; English
• Published: 02.04.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to an adaptive online improvement method for a deep reinforcement learning type control strategy, and belongs to the crossing field of new energy vehicles and artificial intelligence algorithms. The method comprises an initialization training stage of an Actor strategy network, a Critic network pre-training stage of a multi-agent environment, an adaptability improvement stage of the multi-agent environment, an online stage of a mature Actor strategy network, an adaptability improvement stage of the Actor strategy network and an adaptability preparation stage of the Actor strategy network. Aiming at a hybrid electric vehicle and a deep reinforcement learning type energy management strategy, the self-adaption of the deep reinforcement learning type energy management strategy is improved, and the problem that due to a special mechanism of a machine learning type strategy, an environment model in a training stage always has a certain difference with a real environment, and the training effic