基于HGANN的全方位推进器运动学正解研究

Forward Kinematics of the Variable Vector Propeller Based on HGANN

针对潜器全方位推进器的调距机构运动学位置正解求解高度非线性、计算速度慢、准确率低的特点,提出了一种改进的混合编码遗传神经网络算法(HGANN)。算法兼具了遗传算法的全局寻优能力和神经网络对于非线性映射的强大逼近能力,同时由于采用了二进制和浮点数混合编码方案及3层的染色体结构对遗传神经网络算法进行了改进,优化了网络结构和权值矢量,解决了遗传神经网络算法计算过程中短基因组实际交叉、变异机会过小的问题,使后代种群具有更好的多样性,结合Solis & Wets算子生成后代的方法丰富了遗传搜索空间,加快了收敛速度。仿真结果表明,HGANN算法有效地加快了遗传算法的收敛速度,提高了调距机构的位姿精度。

The Variable Vector Propeller （VVP） Controllable Pitch Machine （CPM）＇s unique structure ex- ists several problems in its Forward Kinematics （FK） solution, including non-linear equation, low speed of computation and low accuracy. This article proposed modified Hybrid Encoding Genetic Algorithm Neural Network （HGANN） for solving the FK problem of the CPM. The algorithm possesses overall optimization seeking ability as well as formidable approaching ability of the neural network to the non-linear mapping. Meanwhile, this paper applies the hybrid encoding scheme of binary system and real number as well as 3 chromosomic structures so as to modify the Genetic Algorithm Neural Network （GANN）algorithm, optimize the network architecture and the weight vector and diversify the descendant population by solving the problem of the lack of opportunity in short genome team actual overlapping and variation in computation process. In addition, the combination of genetic algorithm with progeny generated by Soils ＆ Wets operation enriches the heredity search space and speeds up the convergence rate. Simulation and experimental results show that the HGANN algorithm proposed in this article effectively speeds up the convergence rate of genetic algorithm and enhances the posture precision of VVPCMP＇ position.