改进粒子群优化神经网络在变压器故障诊断中的应用

变压器绕组早期故障的诊断是实现安全生产、避免大事故的技术前提。由于变压器器身振动信号包含有丰富的信息,所以可以通过监测变压器振动信号来预估绕组的状况。笔者首先利用小波包分解原理将变压器振动信号分解到不同的频段中,然后计算各频段的能量熵值,并将其作为BP神经网络的输入向量,同时利用改进粒子群算法(IPSO)对BP神经网络进行优化。最后利用训练好的BP神经网络对变压器进行故障诊断。试验结果表明:与传统BP神经网络法和PSO-BP神经网络方法相比,该方法克服了BP神经网络的一些缺陷,具有较快的收敛速度和较高的诊断精度,对变压器绕组的早期故障具有良好的预测能力。

Application of Neural Network Based on Improved Particle Swarm Optimization in the Condition Monitoring of Transformer

Incipient fault diagnosis in windings is the technical prerequisite for safety operation.The vibration signal of transformer winding is correlative with the conditions of winding compression and loosening,and vibration method is an effective method used for monitoring the conditions of winding.In this paper,wavelet packet is used to decompress the vibration signal into different frequency bands.Wavelet packet-energy entropy(WP-EE)is then extracted to construct characteristic vectors of signals and is used as an input of the back propagation(BP)neural network,which is optimized by improved particle swarm optimization(IPSO).Finally,the fault diagnosis is accomplished by the BP neural network based on IPSO.The experimental results show that the IPSO-BP improves the BP neural network generalization capacity,and the convergence of method is faster and forecast accuracy is higher than that of the traditional BP neural network and PSO-BP.Therefore,the method can be used to forecast the fault diagnosis of transformer’s winding.