Research Of Simulation System For Evolutionary Game Theory Based On Multi-agent

Nowadays, Evolutionary Game Theory (EGT) in the field of economics is becoming more and more focused both at home and abroad. The same as EGT, quickly development of computer technology makes economic simulation becomes true and to be another method in research of economics.this thesis introduces EGT, research of EGT both at home and abroad, complex adaptive system, and economic simulation. It mainly does some summary and detail of learning algorithms including reinforcement learning, imitation learning, belief learning and other integrate learning algorithms.First, this thesis mainly constructs a simulation system for EGT based on multi-agent. This system realized has three parties which are game of monomorphic population with two persons, game of two-morphic populations with two persons and game of polymorphic populations with n persons(n is more than two). In every party, you could observe the choice of one agent and whole population evolutionary curves during the simulation underway by adjusting the parameters and the learning algorithms.Second, an example about games between banks and enterprise in credit market is modeled and simulated under imitation learning algorithm. From simulation results, we get ESS of the game model in different parameters. At the same time, the same model under certain parameters set with Q-learning algorithm and experience-weighted attraction (EWA) algorithm is simulated. Form the results, we make a comparison among the three algorithms.The main conclusions and some questions for further research are proposed at last.