Multi-agent Affective Decision Learning Method With Its Application In Flow Intelligent Transportation

Traditional computer technology can no longer meet the development needs of artificial intelligence.Particularly,there need more intelligent decision learning methods to solve various complex problems.Distributed technology enables agents to collaborate and learn from each other,and their ability to deal with complex decision-making problems is greatly improved,which has become a research hotspot in the field of intelligent control.However,when Multi-Agents complete decision tasks that require frequent interaction,there are generally problems such as poor learning ability,preference evaluation,and low group consistency.In recent years,affective computing has provided a new method for solving such complex decision-making problems with its advantages that can quantify the emergence process of group wisdom and the convergence process of decision consistency.At present,the research on affective computing mainly focuses on affective recognition and expression,and there is little discussion on how to use affective mechanisms to make better decisions.In addition,the development of agents is limited to the improvement of logical reasoning ability,rarely involving individual emotional changes and group emotional interactions.It is worth mentioning that when facing large-scale multi-agent system modeling problems,the feedback of the affective mechanism will help to assess the current environment better and help agents make beneficial decisions,along with making the decision-making system more intelligent and efficientTo this end,this paper proposes a multi-agent decision learning method under the emotional interaction mechanism,develops a multi-layer affective computing model based on decision preference,and gives the mapping relationship between the agent's emotional change and behavior preference;then,a new kind of decision consistency and scheme entropy indices are defined to reflect the convergence process and group consistency of the decision,and prove the feasibility and advantages of solving traditional group decision problems through numerical examples;finally,Under the framework of reinforcement learning,the paper gives the definition of the affective reward function inside the agent and establishes a kind of emotion-driven reinforcement learning model.Finally,the proposed decision learning method is applied in the Flow intelligent traffic control platform.Compared with the traditional traffic control and reinforcement learning methods,verifying that the affective interactive decision method proposed in this paper can improve the learning speed and traffic fluency of the agents in the Flow platform.Reinforcement learning has also noticeably improved the sparse rewards that are common in complex scenarios.This work provides a group affective interactive decision method for traditional group decision-making problems for the multi-agent system,and at the same time establishes an affect-driven multi-agent reinforcement learning method based on Flow intelligent transportation system,which gives a new approach to construction of future urban intelligent transportation system new ways.