基于模拟退火改进粒子群混合算法的变压器故障诊断

针对传统的变压器故障诊断方法在实际应用中存在的一些不完善性和局限性,笔者将基于模拟退火思想的改进粒子群优化(SAPSO)算法和误差反向传播(BP)算法相结合构成的SAPSO-BP混合算法用于训练神经网络。该混合算法有效克服常规BP和PSO算法独立训练神经网络的缺陷,并应用于变压器溶解气体分析的智能故障诊断。实验诊断结果表明,SAPSO-BP混合算法的收敛速度快于BP及PSO-BP算法,并且具有较高的诊断准确率。

Transformer Fault Diagnosis Based on SAPSO-BP Hybrid Algorithm

Aiming at the imperfection and limitation of the conventional transformer fault diagnosis method in practical applications, the hybrid algorithm which combines the improved particle swarm optimizationalgorithm based on simulation annealing(SAPSO) thought with error back-propagation (BP) algorithm is used to train neural network. The hybrid algorithm can effectively avoid the defects of independently training neural network in conventional BP algorithm and PSO algorithm,, and can be used to analyze dissolved gas in transformer for intelligent fault diagnosis. The experimental results show that SAPSO-BP hybrid algorithm has a faster convergence speed than BP and PSO-BP algorithms, and gains higher diagnosis accuracy.