基于改进BAS-BPNN的Stewart平台位姿正解

Forward position and posture solution of Stewart platform based on improved BAS-BPNN

针对目前Stewart平台位姿正解难以兼顾高精度和实时性的问题,设计出满足平台精度要求且运算时间尽量少的正解方法是研究并联平台运动学的关键。首先建立平台运动学反解模型求得BP神经网络(BPNN)的训练集,然后调整步长策略和引入惯性权重改进天牛须搜索算法(improved BAS,IBAS),提升算法性能,通过改进后的算法优化BP神经网络的初始权阈值,反复学习训练至最小误差,最后选取一段位姿变化进行仿真验证。仿真结果表明,IBAS-BPNN正解法在选取的100组位姿点中的最大位姿误差小于0.6%,运算时间为31.9 s,通过与遗传算法(GA)、粒子群算法(PSO)和天牛须搜索算法(BAS)优化的BP神经网络模型进行对比分析,验证了IBAS-BPNN模型在精度和运算速度方面具有优越性。算法模型有效提升了Stewart平台位姿正解的精度和实时性,为平台的运动学和控制提供了理论和方法参考。

Aiming at the problem that the forward solution of Stewart platform′s position and posture is difficult to have both high accuracy and real-time, it is the key to study the kinematics of parallel platform to design a forward solution method that meets the requirements of platform accuracy and minimum calculation time.Firstly, the platform kinematics inverse solution model was established to obtain the training set of BP neural network.Secondly, the step size strategy was adjusted and the inertia weight was introduced to improve the beetle antennae search algorithm to promote the performance of the algorithm.The initial weight threshold of BP neural network was optimized through the improved algorithm, and the training was repeated to the minimum error.Finally, a section of posture change was selected for simulation verification.The simulation results show that the maximum position and posture error of the IBAS-BPNN forward solution method in the selected 100 groups of points is less than 0.6%,and the calculation time is 31.9 s.Through comparison and analysis with the BP neural network model optimized by genetic algorithm(GA),particle swarm optimization(PSO) and beetle antennae search algorithm(BAS),the superiority of IBAS-BPNN model in accuracy and calculation speed is verified.The algorithm model effectively improves the accuracy and real-time of forward position and posture solution method of Stewart platform, and provides a theoretical and methodological reference for the kinematics and control of the platform.