基于改进粒子群算法的模糊神经网络变压器油色谱监测故障诊断策略

Fault Diagnosis Strategy for Transformer Oil Chromatography Monitoring of Fuzzy Neural Network Based on Improved PSO

针对标准误差反向传播(back propagation,BP)神经网络算法易陷入局部最优、收敛速度缓慢等问题,提出一种基于改进粒子群算法的模糊神经的变压器油色谱故障诊断方法。该方法首先通过模糊编码边界对网络输入模糊化;再结合非线性策略的惯性权重及学习因子改进的粒子群BP网络算法来诊断变压器故障类型,既能平衡全局搜索和局部搜索能力,还可以避免BP神经网络陷入局部最优;最后,采用MATLAB软件对变压器油色谱数据进行仿真,结果表明该方法具有收敛速度快、诊断准确率高、泛化能力强等优点。

Aiming at problems of being easy to fall into partial optimization and slow convergence speed of standard error back propagation neural network algorithm, this paper proposes one fault diagnosis method for transformer oil chromatography of fuzzy neural based on improved particle swarm optimization. This method firstly uses fuzzy coding boundary to realize fuzzification for network input, then combines improved PSO BP network algorithm based on inertia weight and learning factor of nonlinear strategy to diagnose fault types of transformer which can balance global search and partial search as well as avoid partial optimization. At last, it uses MATLAB software to proceed simulation for transformer oil chromatography data and the result shows that this method is provided with merits of fast convergence speed, high diagnosis accuracy rate and strong generalization ability.