Research On Game Dynamics Of Interspecific Interaction Based On Mixed Population

There are a large number of swarm intelligent behaviors in the real world,such as predators’ cooperative roundup,sardine migration,and drone formation systems.These swarm intelligent behaviors are the basis for the survival and stable operation of nature and human society.How to reveal the emergent mechanism and excitation mechanism of swarm intelligent behavior has become one of the most challenging scientific problems in the field of artificial intelligence in recent years.As the basic theory of behavioral decision-making research,evolutionary game theory provides an effective research tool to quantitatively describe individual behavioral decision-making rules in swarm intelligent behavior,analyze and predict swarm intelligent behavior,and has been widely studied and applied in academia.However,with the advent of the era of artificial intelligence,swarm intelligence behavior is increasingly complex and changeable,and traditional evolutionary game theory has been difficult to accurately describe the learning and evolution laws of swarm strategies in dynamic,time-varying,and incomplete information environments.Therefore,how to reveal the dynamic evolution law of swarm intelligence behavior through the intersection of information science,network science,data science and other disciplines is one of the development trends of swarm intelligence research.Based on evolutionary game theory and complex networks,this thesis proposes two behavioral decision-making models for mixed groups,including the following two aspects:(1)A group game model based on mixed decision-making mechanism is proposed.Although evolutionary game theory based on strategy imitation and game dynamics based on reinforcement learning can both be modelled by replicator dynamics,the internal mechanisms of the two dynamics are totally different.Therefore,in this article,we introduce a new game model that considers a mixture of these two game dynamics.Namely,some agents are evolutionary game players(EGPs),while the others are reinforcement learning players(RLPs).We find that the frequency of cooperation is related to the ratio of RLPs and has a hump shape.This is because the two kinds of players play different roles in the evolution process.RLPs learn reciprocity from the environment and play as the medium encouraging EGPs to cooperate.On the other hand,EGPs become more cooperative since defection is no longer the dominant strategy.We model the mixed dynamics by pair-approximation,whose result is consist with that of Monte Carlo simulations.(2)A group game model based on network coupling mechanisms is proposed.One of the principal directions of the research on cooperation comes from the integration of game theory and network science.Mainly based on the coupling mechanism of the interdependent network,investigating the behavior of a heterogeneous population,composed of conservative-driven players and radical-driven players.Conservative-driven players maintain stable coupling strength in a way that minimizes individual risk and avoid overly risky personal choices,whereas radicaldriven ones break the stereotype and take risks for high rewards.Different from the monotonous growth of cooperation demonstrated in previous studies,the model shows the hump like relationship between cooperation and conservative participant density.In addition,in some cases,these heterogeneous populations do not have appropriate competition,but they form a strategic alliance to obtain better evolutionary outcome.Further analysis shows that polarization of coupling strength enhances cooperation.