Research On Learning Differential Evolution Algorithm And Its Application In Robot

During the development of robot intelligence,evolutionary computation is a very essential technical tool.Differential evolution algorithm(DE),as one of the important representatives of evolutionary computation algorithms,has been broadly used to solve complex optimization problems in the field of robotics.However,these optimization problems are usually characterized by high dimensionality and nonlinearity,etc,and the existing DE algorithms still suffer from the problems of falling into local optimum,weak local search ability,slow convergence speed and low accuracy when solving problems.These limitations restrict the performance of the algorithm and hinder its application and expansion.Hence,designing and developing more efficient DE algorithms to improve the accuracy and efficiency of robot control systems is of great significance for promoting the development of intelligent robots.In this paper,learning strategies are the major focus and modeling mechanisms are supplementary.Firstly,a learning strategy based on the worst individual in the population is investigated to reduce the risk of the algorithm being trapped in a local optimum.Secondly,to improve the convergence accuracy of the algorithm,a biased central spiral learning strategy based on all individuals is designed.Then,a switchable refracted oppositionalmutual learning strategy is embedded into the population initialization and iteration process to achieve a balance between global search and local exploitation capabilities.Finally,those strategies are embedded in the DE algorithm framework,and three learning-based DE algorithms are proposed,which are applied to different robot optimization problems.The main work is listed as follows:(1)Aiming at the problems of falling into local optimum and slow convergence of the DE algorithm,an optimal-worst dynamic opposite learning DE algorithm based on chaotic initialization(CDOWDE)is proposed by integrating the Logistic-Chebyshev chaotic mapping strategy,parameter adaptive mechanism and optimal-worst dynamic opposite learning.Then,the experimental results on the CEC2005 functions show that the CDOWDE algorithm has better convergence and optimization capabilities.Finally,the CDOWDE algorithm is utilized to solve the robot arm inverse kinematics problem,and the comparative results indicate that the CDOWDE algorithm has a stronger solving ability and better stability,which further verifies the optimization effect of the presented CDOWDE algorithm.(2)Aiming at the problems of low convergence accuracy and weak local search ability of the DE algorithm,a biased-center spiral learning DE algorithm based on Bézier-culture collaboration(BCSDE)is developed by incorporating the Cubic-Sine chaotic mapping strategy,normative knowledge strategy,Bézier search mechanism and biased-center spiral learning strategy.Then,the experimental results on the CEC2014 functions demonstrate that the BCSDE algorithm performs better.Finally,the BCSDE algorithm is applied to solve the parameter identification problem of the dynamic model of the flexible robotic arm,and the results show that the BCSDE algorithm has higher identification accuracy and better stability,further verifying the engineering feasibility of the developed BCSDE algorithm.(3)Aiming at the drawbacks that the DE algorithm is difficult to balance global search ability and local exploitation ability,a refracted oppositional-mutual learning DE algorithm based on the Bernstein strategy(BROMLDE)is presented by synthesizing the refracted opposite-learning strategy,mutual learning strategy and Bernstein strategy.Then,the experimental outcomes on the CEC2015 functions show that the BROMLDE algorithm is effective and efficient.Finally,the BROMLDE algorithm is applied to solve the mobile robot path planning problem,and the results display that it has higher planning accuracy and stability,further verifying its engineering applicability.This paper aims to address the shortcomings of the DE algorithm in solving complex optimization problems of robots and gradually conducts research work on the learning DE algorithm to improve its solving capability,the experimental results verify the effectiveness of the proposed algorithm.Firstly,the CDOWDE algorithm is used to solve the robot arm inverse kinematics problem.Then,the BCSDE algorithm is utilized to alleviate the problem of dynamic parameter changes caused by the motion or load of the robot arm,which is helpful to the accurate and stable motion of the robotic arm.Finally,considering that the cooperation between the robotic arm and the mobile robot in the actual environment can efficiently complete monotonous and repetitive tasks,where path planning of mobile robots is particularly important,thus,the BROMLDE algorithm is used to provide better path planning.Overall,the algorithm proposed in this paper can provide some insights for subsequent task collaboration between robotic arms and mobile robots.