Interpretable Hybrid Experimental Learning for Trading Behavior Modeling in Electricity Market

Abstract

A modern electricity market is essentially a complex network, characterized by complicated interactions among cyber communications, physical systems, and social agents. Trading behavior modeling has always been complicated in the physical system-based market. In this paper, trading behavior modeling in the electricity market is solved by a data-driven method combining experimental economics and machine learning, called Hybrid Experimental Learning (HEL). Based on the historical and experiment simulated data, HEL models the trading behavior by a machine learning generative model which will be interpreted by a post hoc interpretation approach. Taking a simulated electricity market based on the trial spot market rule in Guangdong, China as an example, a generative adversarial network (GAN) is employed to generate the offering strategies of a gas generator. Local interpretable model-agnostic explanation (LIME) as a post hoc interpretation approach is applied to explain the relationship between the output of GAN and some of the inputs of HEL, which can be described as offering mechanisms for the gas generator.