改进的神经网络在变压器故障诊断中的应用

Application of Improved Neural Network in Transformer Fault Diagnosis

研究变压器故障准确诊断问题。通过对变压器油中溶解气体定性、定量地分析可及时发现变压器内部存在的潜伏性故障。但目前神经网络诊断方法存在收敛速度慢、不稳定问题,导致正确率低。为解决上述问题,提出了小生境遗传算法改进的神经网络模型。充分利用小生境遗传算法的搜索能力和神经网络的非线性映射和学习联想能力,用小生境遗传算法优化神经网络的初始权值和阈值,并对网络进行训练和测试。实验结果表明,与传统方法相比,改进模型有效提高了网络收敛速度、稳定性,提高了故障诊断正确率,具有很强的可行性和有效性。

Research the problem of transformer fault diagnosis. The latent fauhs in transformer can be found timely through the qualitative and quantitative analysis of the gas dissolved in transformer oil. But the problems of slow convergence speed and instability exist in the diagnosis method based on neural network at present, which leads to low accuracy rate. In order to solve the problems, an improved neural network model of niche genetic algorithm was presented. It makes good use of searching ability of niche genetic algorithm and the nonlinear reflection and associa- tion learning ability of the neural network, and optimizes the initial connection weights and thresholds of the neural network through the niche genetic algorithm, then trains and tests the network. The result of the experiment shows that, compared with the traditional method, the improved model is effective to improve convergence rate, stability of network and accuracy of fault diagnosis, and has very strong feasibility and validity.