基于改进PSO-BP神经网络的变压器故障诊断

引入动态变异操作来优化粒子群算法,同时将改进的粒子群优化算法和误差反向传播的算法相结合,构成混合算法,用于训练人工神经网络,并将该混合算法应用于变压器的故障诊断.仿真结果表明,该算法具有较快的收敛速度和较高的计算精度;诊断结果表明,该算法有利于提高变压器故障诊断的正确率.

Transformer Fault Diagnosis Based on the Improved PSO-BP Neural Network

Conventional artificial neural networks still have problems of easy plunging into local minimum, slow convergence rate and bad generalization capacity. Conventional particle swarm optimization (PSO) has advantages of fast convergence rate, easy to find the global optimal solution and strong versatility features. But at the same time the PSO also has problems of likely premature convergence, low searching accuracy rate and low efficiency at later iterations. To solve these problems, this paper introduces a dynamic mutation operation to the PSO, and combined the improved PSO and back propagation (BP) algorithm to form hybrid algorithm for training artificial neural networks. Applying the hybrid algorithm to the transformer fault diagnosis, the results show that this hybrid algorithm has a faster convergence speed and higher accuracy. And the diagnostic results show that this hybrid algorithm can improve the accuracy of the transformer fault diagnosis.