论文部分内容阅读
基于代理的计算经济学已成为电力市场研究的一种重要方法,构建智能代理的学习模型是其中的重要研究内容之一。常用的强化学习和信念学习算法各有弊端,为此引进了一种综合了强化学习和信念学习的经验权重魅力值(EWA)算法,将其应用于电力市场仿真研究中,模拟发电商决策行为。基于混合代理和单一代理系统的仿真结果表明,EWA学习算法对市场参与者的行为有更好的描述,在参与者众多的大系统中较Roth-Erev算法更为先进、智能,具备更好的学习性能;EWA算法具备更高的捕捉博弈均衡的能力。
Agent-based computational economics has become an important method in power market research. Building a learning model of intelligent agent is one of the important research contents. Commonly used reinforcement learning and belief learning algorithms have their own drawbacks. To this end, an empirical weighting value (EWA) algorithm based on reinforcement learning and belief learning is introduced, which is applied to power market simulation to simulate the decision-making behavior of power suppliers . The simulation results based on hybrid agent and single agent system show that the EWA learning algorithm has a better description of the market participants’ behavior and is more advanced, intelligent and better than the Roth-Erev algorithm in many large-scale systems Learning performance; EWA algorithm has a higher ability to capture game equilibrium.