多目标分解随机粒子群优化算法及其在直线电机优化设计中的应用

A multi-objective decomposition-based stochastic particle swarm optimization algorithm and its application to optimal design for linear motor

本文提出了一种多目标分解随机粒子群优化算法(MDSPSO).该算法优化过程中,所有粒子按各自固定的权重向量,采用改进Tchebycheff分解方法,将求解多目标非支配解问题转化为求解多个单目标最优解问题;而后每个粒子在以自身位置、个体历史最优参考位置及群体最优参考位置的几何中心为中心,以中心到自身位置为半径的区域内,随机生成一个新的起始位置,并参考当前的速度更新下一时刻的位置.通过对测试函数多次计算得到的数据进行统计分析,表明MDSPSO的收敛性和多样性均优于另外3种对比算法.最后针对直线电机磁路复杂、有限元计算费时的问题,使用神经网络拟合直线电机结构参数与性能的关系作为优化设计的模型,应用MDSPSO算法,优化结构参数.实际测试结果表明,优化后的直线电机推力大、效率高,同时有效控制了其推力波动和生产成本.

This article proposes a multi-objective decomposition stochastic particle swarm optimization （MDSPSO） algorithm. In MDSPSO, every particle has a weighted vector constantly. Then, an improved Tchebycheff decomposition method is applied to decompose the multi-objective problem into some single-objective problems. The reference position of every particle is uniformly generated in the zone with the center which is the geometrical center of its current position, the best previous reference position as well as the swarm best reference position. The radius of this zone is the distance from the center to its current position. Then the particle is updated to the new position according to the reference position and its current velocity. The comparisons with the decomposition-based multi-objective particle swarm optimizer （dMOPSO）, a multiobjective evolutionary algorithm based on decomposition （MOEA/D）, and nondominated sorting genetic algorithm II （NSGA-II） show that the solutions of MDSPSO can be dominated at least with the best diversity. To reduce the compu, tational time by finite element analysis for optimizing the structure parameters of linear motor, artificial neural network is used as the model to evaluate the performance. Finally, MDSPSO is applied to optimize four objectives simultaneously. The practical result is shown that the optimized linear motor has an increased thrust, improved efficiency, reduced fluctuation and manufacturinz cost.