神经网络在复合绝缘子伞裙优化设计中的应用

Application of Neural Network in Optimization Design of Composite Insulators Shed

针对复合绝缘子污闪问题,提出了神经网络近似模型和遗传算法相结合的优化设计方法。以染污绝缘子沿面最大场强作为优化目标,选取大、小伞直径和伞间距作为设计参数,建立染污绝缘子电场优化模型。利用优化拉丁方采样方法和ANSYS Maxwell获取训练样本,利用BP神经网络对样本集进行非线性拟合,建立神经网络近似模型。遗传算法在求解优化模型时,用BP神经网络对染污绝缘子沿面最大场强和爬电系数进行近似计算,最终得到最优解。结果表明:伞裙优化后,绝缘子沿面最大场强为3.69×104V/m,降低了9.78%。优化结果表明神经网络近似模型具有可行性,为绝缘子伞裙优化问题提供了一个新的思路。

In view of the pollution flashover problem of composite insulator, an optimized design method combining neural network approximate model and genetic algorithm was proposed. Taking the maximum surface electric field strength of polluted insulators as optimization objective, and selecting the diameter of large and small sheds and the spacing between sheds as design parameters, we established an electric field optimization model of polluted insulators. The training samples were obtained using the optimized Latin square sampling method and ANSYS Maxwell, the sample set was nonlinear fitted by BP neural network, and the neural network approximate model was established. When using genetic algorithm to solve the optimization model, BP neural network was used to calculate the maximum surface electric field strength and the creepage coefficient of the polluted insulators approximately, and the optimum solution were obtained. The results show that the maximum surface electric field strength of the insulator is3.69 × 10^4V/m after shed optimization, which reduces by 9.83%. The optimization results show that the neural network approximate model is feasible, and it provides a new method for shed optimization of insulators.