Research On Power Transformer Fault Prediction Based On Genetic Algorithm And Grey Theory

1. IntroductionElectric power system is developing for EHV, large power network, high capacity, automation with the rapid development of power industry and national economy in China. Power transformer is one of the key equipments of power system. The quality of its running performance will directly impact on the safe operation of power system. Once the power transformer has accident, its direct and indirect economic losses are enormous. Therefore, it is important for raising reliability and safety of power supply and reducing maintenance costs and economic losses that research power transformer fault forecast technology.Current high voltage and capacity power transformer at home and abroad usually uses oil-filled transformers. Oil-filled power transformer internal insulation material is mainly the transformer oil, insulating paper and paperboard etc. A-class insulation materials. Insulation materials in heat, electricity and so on will gradually age and deteriorate in the course of normal operation, and decompose into hydrogen and low molecule gases. Most of these gases are dissolved in transformer oil. When there are potential overheating faults or discharge faults, gases production and gases production rate will gradually increase. Gases accumulate in transformer oil, and precipitate bubbles after convection, diffusion and dissolving until saturation. So, creation mechanism of gases dissolved in power transformer oil has close relationship with power transformer oil and insulation materials. Gases dissolved in power transformer oil can be characteristic features of transformer faults. Periodic analyzing gases composition, gases content and rate of gases dissolved in power transformer oil can find latent faults inner power transformer and judge weather it endanger the safe operation of power transformer. According to GB/T7252-2001《Guide to the analysis and the diagnosis of gases dissolved in transformer oil》, Dissolved Gases Analysis (DGA for short ) method analyze the capacity of H2,C2H2,C2H4,C2H6,CH4,CO and CO2 and determine types of faults. GB/T7252-2001《Guide to the analysis and the diagnosis of gases dissolved in transformer oil》recommend characteristic gases method and three-ratio method. Some experts and scholars propose ANN, fuzzy mathematics, SVM etc. based on DAG. There are some shortcomings in these methods.2. Research ContentGray system theory is mainly used in systems analysis, evaluation, modeling, forecasting, decision-making, control and optimize. Because it is uncertainty that relationship between gases dissolved in power transformer oil. It has no clear qualitative and quantitative descriptions which fault produces certain gases. So, power transformer fault system is a gray system that part information is known and part information is unknown. The essential of transformer fault diagnosis is a whitening process of a grey system.Genetic algorithm is highly parallel, randomized, adaptive search algorithm inspired by biological mechanisms of natural selection and evolutionary development. GA can realization global search in complex, multi-peak, non-linear and non-minimal space. Genetic Algorithm has advantages of simple, common and robust, and searches not rely on high-order gradient information. Because of its robustness, Genetic Algorithm particularly suited to dealing with complex and nonlinear problems, and it widely applies in production and daily life.In this paper, we use Gray system theory combined with genetic algorithm to realize power transformer dissolved gases content prediction and power transformer fault diagnosis.(1) power transformer dissolved gases content prediction based on variable weight grey Verhulst(ρ) modelPower transformer dissolved gases content prediction is the prerequisite of power transformer fault forecast based on DGA. Grey GM (1,1) model, grey GM (1,1,ρ) model and grey Verhulst model were mastered after in-depth study of gray prediction models. Power transformer dissolved gases content data were statistical analyzed. These data don't monotonically increase with time change, but reach a peak at a certain time and then followed by possible fluctuations. Power transformer dissolved gases content data are single-peak feature. Grey GM (1,1) model are for sequence with exponential law, only describes monotonic change process. So, Grey GM (1,1) model and grey GM (1,1,ρ) model are not the most suitable model for power transformer dissolved gases content prediction. Grey Verhulst model is suitable for non-monotonic swing sequence or S shape sequence. So, this paper chooses grey Verhulst model to predict power transformer dissolved gases content according to its single peak feature, and compares the forecast results with grey GM (1, 1) model and grey GM (1,1,ρ) model forecast results. Experiments demonstrate that grey Verhulst model is more suitable for power transformer dissolved gases content prediction. This paper improved grey Verhulst model the same way as grey GM(1,1,ρ) model. There are 2 optimization rules of choosing background function parameterρof grey GM(1,1,ρ) model. They are prediction error rule and posteriori error rule. These 2 rules are introduced into grey Verhulst model to choose background function parameterρ. Genetic algorithm has features of simple, commonly used, robust, highly parallel, randomized and self-adaptive. So, genetic algorithm optimizes grey Verhulst model background function parameterρ. Genetic algorithm parameters are chosen by experiments.Comparison between forecast of grey Verhulst model optimized by 2 different rules shows that these 2 rules are not the most suitable rule for optimizing background function parameterρ. Variable weights proposed by this paper are determined based on 2 rules results x?e ( k + 1)and x?c ( k + 1). The new model is called variable weights grey Verhulst(ρ) model. The variable weights related with grey Verhulst(ρe) model and grey Verhulst(ρc) model prediction error ratio. Experiments on power transformer dissolved gases content prediction indicate that variable weights grey Verhulst(ρ) model proposed by this paper has higher accuracy , is suitable for power transformer dissolved gases content prediction and show a new way to power transformer dissolved gases content prediction. Meanwhile, this model can be transferred into other fields.(2) power transformer fault diagnosis based on grey relation gradePower transformer fault diagnosis is the key technology of power transformer fault forecast. Power transformer fault forecast technology based on DGA analyzes power transformer dissolved gases content prediction data to diagnoses fault type, so the future statue and future fault type of power transformer will be gained. Grey relation grade is the basis of gray system theory. Grey relation grade can analyze small sample, lack information grey data, and make up the shortcomings of the number and strict law of samples of mathematical statistics method. Grey relation grade and its improved forms are applied in power transformer fault diagnosis and have high precision of diagnosis and easy to calculation etc. advantages. Compared with other grey relation grade gain method, improved grey relation grade and grey area relation grade have better performance on power transformer fault diagnosis. So, this paper chooses improved grey relation grade and grey area relation grade to diagnose power transformer fault based on power transformer fault standard spectrum. Experiments shows that power transformer fault diagnosis based on improved grey relation grade and grey area relation grade are validity and practicality. This paper also proposed a preliminary conception of synching improved grey relation grade and grey area relation grade. The new results are gained from improved grey relation grade and grey area relation grade results. The 2 results of different grey relation grades are weighted and balanced in the new synthetic grey relation grade. The weights of improved grey relation grade and grey area relation grade results should be determined through a lot of experiments.(3) power transformer fault forecast based on grey system theoryFirst, variable weights grey Verhulst (ρ) model predict contents of gases dissolved in power transformer oil. Second, the prediction results of gases contents are sent to fault diagnosis mode based on improved grey relation grade and grey area relation grade to diagnose the type of fault. Experiments analysis shows that prediction results of real data are consistency with the prediction results of fault forecast algorithm in this paper and demonstrates that power transformer fault forecast algorithm proposed in this paper is effective and practical.3. ConclusionFor power transformer fault prediction contained in the power transformer oil dissolved gas concentration prediction and fault diagnosis of power transformer , grey Verhulst model and grey relation grade analysis methods on grey system theory are used, and combined with genetic algorithm to optimize grey Verhulst model parameter, finally variable weights grey Verhulst(ρ) model is put forward in this paper. Fault forecast experiments demonstrates that variable weights grey Verhulst(ρ) model combine with improved grey relation grade and grey area relation grade can forecast future fault type based on existing power transformer oil chromatographic data, and it has great theoretical significance and practical value for power transformer safe and reliable operation.