Study On Fault Diagnosis Of Oil Immersed Transformer Based On Hybrid Intelligent Algorithm

The oil-immersed power transformer is critical equipment in the power system,and its function is like the heart of the human body.Therefore,the fault diagnosis of the transformer is of great significance.Using dissolved gas analysis(DGA)in oil,using four types of hydrocarbon gases et al as characteristic input quantities,and then using machine learning algorithms for early diagnosis of transformer faults has the advantages of simplicity and economy.At the same time,the(DGA)standard provides various interpretation methods for fault diagnosis and evaluation of insulation service life.However,directly using hydrocarbon gas et al as the input characteristic quantity for fault diagnosis has problems such as incomplete coding and too absolute coding boundary.For such issues,extracted fault features and used as new input features to overcome the shortcomings of incomplete coding and too absolute code boundaries.In addition,the traditional intelligent algorithm also has the disadvantages of slow convergence in the later stage and easy to fall into local optimum.Aiming at the shortcomings of the intelligent algorithm itself,this paper improves the intelligent algorithm.These improvements improve the slow convergence speed in the later stage of the algorithm and easily fall into optimal local performance.Finally,the improved hybrid intelligent algorithm is used to diagnose the fault of the transformer.The main research contents are as follows:Firstly,Aiming at the slow convergence speed in the later period caused by the grey wolf optimizer search mechanism,it is easy to fall into the optimum local problem,modify the control factor and weighted distance of the grey wolf optimizer to improve the convergence accuracy and stability of the algorithm,and further overcome the problem of falling into the local optimum Advantages and disadvantages,The performance of the improved grey wolf optimizer was tested by six commonly used test functions.The improved grey wolf optimizer is compared with other intelligent algorithms.Secondly,the probabilistic neural network(PNN)is suitable for pattern classification,and the smooth factor v between the input and pattern layers greatly influences the PNN model.Used the improved grey wolf optimizer(IGWO)to optimize v in the probabilistic neural network,and the improved hybrid intelligent algorithm(IGWO-PNN)was used for transformer fault diagnosis.Using the three-ratio value as the input feature quantity,the fault diagnosis rate of the IGWO-PNN model reaches 92.5%.Finally,for fault diagnosis directly using hydrocarbon gas or traditional gas ratio as the input characteristic quantity,there are problems such as incomplete coding and too absolute coding boundary.The importance of the candidate gas ratio is scored by random forest(RF)and recombined into five sets of characteristic parameters in order of importance from high to low,which are used as input feature quantities of the model.The feature quantities are input into the improved grey wolf optimizer coupled extreme learning machine(ELM)model and new feature parameters are further optimized.