Research on Deep Reinforcement Learning-based Coordinated Control Strategy for Multi-Intelligent Bodies under Software System Architecture

In today's society, a single intelligent body does not meet the needs of complex tasks, and coordinated control of multiple intelligences becomes an important solution.In this regard, this paper carries out the research on the coordinated control strategy of multiple intelligences supported by deep reinforcement learning technology.Aiming at the problems of uneven task distribution and unsatisfactory decision consistency arising from the collaborative decision making of multiple intelligences under the software system architecture, a hierarchical multi-intelligence collaborative decision-making algorithm based on the AC framework is proposed to realize the information exchange and decision-making collaboration among intelligences, so as to improve the efficiency of coordinated control.However, with the increase of the number of multi-intelligents, the algorithm will have the problem of upper and lower level non-smoothness, in order to solve this problem, a multiintelligents collaborative algorithm based on role parameter sharing is designed.Finally, the research scheme of this paper is evaluated and analyzed from multiple dimensions.When the number of intelligences increases by 5, the reward value of this paper's algorithm does not show a decreasing trend, which indicates that this paper's algorithm is able to handle the control coordination problem in the case of a small number of intelligences.When the number of intelligences increases by 15, the original method shows a decreasing trend, while in the multi-intelligence body collaboration algorithm based on the sharing of role parameters, the performance is very bright, which ensures the coordinated control effect of multi-intelligence bodies under the software system architecture.