基于改进粒子群BP网络的变压器故障诊断方法

A Study of the Transformer Fault Diagnosis Method Based on Improved PSO -BP Neural Networks

粒子群优化（PSO）算法是基于群智能的全局优化技术，它通过粒子间的相互作用，对解空间进行智能搜索，从而发现最优解。本文对基本粒子群算法进行改进，并将改进粒子群优化算法与误差反向传播（BP）算法结合起来构成的改进PSO—BP混合算法用于训练人工神经网络，对变压器故障进行诊断。实践结果表明：改进PSO—BP算法有效地解决常规BP算法学习网络权值和阈值收敛速度慢、易陷入局部极小等问题，具有较快的诊断速度和较高的诊断精度。

The particle swarm optimization （PSO） algorithm is a swarm - intelligence based global optimization technique. It searches for solution space through the interaction of particles to find out optimum solution. This paper improves the PSO algorithm, and applies the improved PSO - BP algorithm （constituted by combining improved PSO algorithm with back propagation （BP） algorithm） to the training of artificial neural network tice shows that the improved PSO - BP algorithm to diagnose transformer fault. The result of praceffectively solves the problems of the conventional BP algorithm that it has a slow convergence rate and is susceptible to local minimum in leaming the weights and thresholds of the networks, and has a quicker diagnosis speed and higher diagnosis precision.