论文部分内容阅读
BP算法基于梯度下降原理是一种局部寻优算法,在变压器故障诊断应用中网络学习过程收敛速度慢,且易陷入局部极小值。而遗传算法(GA)具有并行计算的特点,可以有效防止搜索过程收敛于局部最优解。将二者结合起来,由GA寻找最优的BP神经网络权值与相应节点的阈值。仿真结果表明:此方法既能快速收敛,又能大大提高避免陷入局部极小的能力,改善了故障诊断的精度和速度。
BP algorithm based on the gradient descent principle is a local optimization algorithm, network learning process in the application of fault diagnosis of transformer convergence slow, and easy to fall into local minimum. The genetic algorithm (GA) with parallel computing characteristics, can effectively prevent the search process converges to the local optimal solution. Combining the two, GA searches for the optimal BP neural network weights and corresponding node thresholds. The simulation results show that this method not only can converge quickly, but also greatly improves the ability to avoid local minima and improves the accuracy and speed of fault diagnosis.