Transformer Fault Diagnosis And Forecast Research Based On Particle Swarm Optimization

Transformer’s operation condition directly influences the security of the powersystem.Timely and accurately diagnosis and prediction about the power transformercan effectively reduce security accident caused by the hidden trouble.In this paper,it has established fault diagnosis system in turn with modified IECthree ratio method、BP neural network and adaptive particle swarm optimizationneural network based on the dissolved gas analysis. IEC three ratio method is theoldest transformer fault diagnosis method,it can judge fault type simpley,but theaccuracy rate is low. BP neural network has the ability of self learning and strongnonlinear mapping, it can improve the accuracy rate of the fault diagnosis on a certainextent. But the BP neural network is easy to fall into local minima and difficult toselect the appropriate learning rate.Based on analysising particle swarm algorithm principle and exploring the mainparameters influence on the algorithm’s performence deeply, an adaptive particleswarm optimization algorithm is put forward to optimize the BP neural network.Inthis new algorithm,when the fitness value is better than the optimal particle’s fitnessvalue,it can increase the inertia factor,and encourage these better particle’s positiveeffect in particle updating.Conversely,it can decrease the inertia factor to weaken thepoor particle’s effect in particle updating.And using nonlinear dynamic method toadjust the inertia weight and the acceleration factor in the whole iterative process,itcan reach the equilibrium between the globle search ability and the local searchability effectly.The transformer fault diagnosis system based on adaptive PSO-BP, compared with the two other systems,has improved the accuracy rate and reduced theerror accuracy effectively.In order to forecast the transformer fault accurately, a improved particle swarmalgorithm is put forward to optimize the paramerers of RBF neural network based onanalysising RBF neural network’s structural characteristics deeply.It uses sectionrestriction strategy to limit the particle position,and nonlinear dynamic adjustment toadjust the inertia weight and acceleration factor.Applied the improved PSO-RBF intwo kinds of transformer’s fault gas perdiction, compared with the other methods, ithas obtained better prediction precision and reduce the prediction error greatly.