触觉传感器非线性补偿仿生算法

摘要

为解决触觉传感器非线性误差大的问题,本文提出了一种基于动态密度聚类改进的自适应多种群遗传算法(IMPGA).IMPGA算法通过对个体相似度的动态聚类分析生成多个子种群,各子种群采用自适应交叉、变异概率并行进化,提高了搜索全局最优解的效率.通过动态邻域搜索策略提高算法局部搜索的能力,通过移民算子保持每个种群的多样性和进化动力.实验表明通过IMPGA算法优化的BP神经网络能够有效减小触觉传感器非线性拟合误差,鲁棒性能好.%To solve the large nonlinear error problem of a tactile sensor,we present an adaptive multi-population genetic algorithm based on dynamic density clustering (IMPGA) in this paper.To improve efficiency when searching for a global optimal solution,the IMPGA generates multiple sub-populations based on a dynamic clustering analysis of individual similarities,adaptive crossovers,and variation probabilities.We improved the local search ability of the algorithm by employing the dynamic neighborhood search strategy and we maintained the diversity and evolutionary dynamics of each population through a migration operator.Our experimental results show that back propagation neural networks optimized by the IMPGA can effectively reduce the nonlinear fitting error of the tactile sensor and are also highly reliable.