Research On FPGA Adaptive Evolution Technology Based On Reinforcement Learning

In recent years,with the rise of artificial intelligence research,evolvable hardware targeting hardware intelligence has achieved preliminary results in many fields such as aerospace,adaptive systems,hardware acceleration,etc.and has great research potential.However,neither evolutionary algorithms nor hardware implementation platforms have unified standards for evolutionary hardware,so it is necessary to conduct various research attempts.The genetic algorithm commonly used in evolvable hardware has the problem of premature convergence,which leads to the low success rate of hardware evolution.Reinforcement learning has dynamic adaptability and can be used to improve genetic algorithm.In addition,hard-core microprocessor plus FPGA architecture,and reconfigurable evolution circuit architecture based on dynamic reconfiguration technology,is the mainstream way to implement evolvable hardware.Based on the research of reinforcement learning to improve genetic algorithm,this dissertation designs and implements FPGA-based evolutionary system based on dynamic reconfiguration with ZYNQ SoC as the platform.The main research contents are as follows:Firstly,on the basis of studying the evolving hardware mechanism of FPGA and the theory of genetic algorithm,the premature convergence problem of genetic evolutionary algorithm is analyzed.Aiming at solving the premature convergence problem,the genetic algorithm is improved by using reinforcement learning.An improved RGA algorithm suitable for evolving hardware is proposed,which lays a solid theoretical foundation for the design and implementation of adaptive evolutionary system of FPGA.Then,based on the analysis of the existing research on the implementation scheme of the evolution technology of FPGA,using the ZYNQ SoC which is integrated with ARM processor and FPGA as the development and verification platform,the adaptive evolution system of FPGA based on dynamic local reconfiguration is designed and implemented,and the specific hardware and software design flow of the evolution system is given.The reconfigurable repetition based on dynamic local reconfiguration is elaborated in detail.The design and implementation process of the chemical circuit,as well as the software program design and implementation process of RLGA algorithm.Finally,experimental verification and result analysis are performed.The excellent performance of RLGA algorithm is verified by simulation experiments.The FPGA adaptive evolution experiment is carried out with the evolutionary polynomial calculation circuit.Through a large number of experimental statistics,the results show that the FPGA adaptive evolution system based on RLGA algorithm has excellent evolution performance.