Agent-based computational economics have become one of the most important electricity market study methods. However， as a method in its infancy， there are many problems need to be solved. These include the experiment design method and simulation result analysis method. The related concepts about agent-based electricity market simulation are firstly presented， and then， three electricity market simulation experiments are designed， including the market model， agents' learning model， experiment purpose， and experiment schemes. The market simulated in the experiments is a day-ahead market with four generators and the agent-learning model is a modified Roth-Erev reinforcement model. According to the simulation results of the first two experiments， the problem of designing iteration and session numbers is discussed. According to the last experiment， the contradiction in the calibration of a parameter in Roth-Erev model is revealed and its reason is analyzed from the mechanism of reinforcement learning.