基于深度强化学习的混合能源系统多目标优化研究

Research on Multi-objective Optimization of Hybrid Energy System Based on Deep Reinforcement Learning

在新能源技术高速发展的背景下,本文提出了一种基于深度强化学习的多目标优化方法,对混合能源系统进行配置和设计。首先,设计了一种基于循环神经网络的深度强化学习方法,对系统配置方案优化过程中计算耗时的目标值进行近似估计,以在线优化的方式实现能源系统运行控制策略的快速求解;其次,以经济性、可靠性、环境效益作为优化指标,采用多目标进化算法对混合能源系统进行优化配置,计算得到满足用户不同偏好的帕累托解集;最后,以离网型混合能源系统作为研究对象验证了该方法的有效性,所提方法明显优于多种传统优化设计方法。

With the rapid development of new energy technologies,a multi-objective optimization method based on deep reinforcement learning was proposed for the configuration and design of hybrid energy systems.Firstly,a deep reinforcement learning method based on recurrent neural networks was designed to approximate the time-consuming process of objective calculation in the optimization of system configuration schemes.The energy system operation control strategy could be optimized in a fast manner by means of online optimization.Secondly,taking economy,reliability and environmental benefits as optimization objectives,a multi-objective evolutionary algorithm was used to optimize the configuration of hybrid energy system,and Pareto solution set satisfying different user preferences was calculated.Finally,an off grid hybrid energy system was taken as an example to verify the effectiveness of this method.The proposed method was obviously superior to various traditional optimization methods.