Improvement Of Variable Prediction Model And Its Application In Transformer Diagnosis

The variable prediction model is to build a model based on the relationship between the characteristic variables.There are multi-dimensional relationships with multiple characteristic variables,which leads to complex calculations and nonlinear problems.Moreover,the use of least squares method to solve the model parameters requires a large amount of sample data,which makes the diagnosis effect unsatisfactory in a small sample environment.As the core equipment of the power grid,the transformer’s failure will induce a power outage of the power grid.Therefore,it is very important to detect the fault type of the transformer.At this stage,the use of gas content in the oil during transformer operation for fault diagnosis is a mature and efficient diagnosis technology.However,in practice,it is difficult to accurately detect the content of the gas,and the gas emission is associated with it.All the diagnostic methods have problems such as few sample data,non-linearity,and difficulty in parameter optimization.In view of the above problems,we choose to use the variable prediction model as the fault diagnosis model to construct the mapping relationship between the characteristic gases for fault diagnosis,which can make full use of the accompanying information between the characteristic gases.But it also cannot solve the problems of small samples,nonlinearity,and complex calculations.A method of combining cuckoo algorithm with variable prediction model is proposed.The cuckoo algorithm has the characteristics of simple operation,fast optimization speed,and small demand for sample size.It can effectively solve the inferiority of variable prediction models in small samples and complex calculations.In addition,VPMCD can make full use of the accompanying relationship between gas emissions,which greatly improves the accuracy of fault diagnosis.Aiming at the problems of poor solution diversity,easy to fall into local extremes and poor stability in CS-VPMCD,we improved the cuckoo algorithm in the initial position,position update,step factor and discovery probability,and compared it with the unimproved The algorithm is compared with experiments to verify the superiority of the improved method.In order to verify the advantages of CS-VPMCD in small samples,non-linearity and computational complexity,we designed a transformer fault diagnosis experiment of CS-VPMCD,and compared the experimental results with BP,SVM,and VPMCD.In order to verify the advantages of CS-VPMCD in parameter optimization,we designed GA-VPMCD,PSO-VPMCD and VPMCD transformer fault diagnosis experiments,and compared the diagnosis results to verify the effectiveness of the parameter optimization method selected in this paper.