Study On Power Transformer Fault Diagnosis Technology Based On Dissolved Gases Analysis

Power transformer is the most important and expensive equipment. The running reliability of power transformer can ensure directly the safe running of power system. Power transformer fault diagnosis technology includes state assessment, fault diagnosis, and fault prediction. Among all the fault diagnosis technologies, there are many methods in fault diagnosis, but the study method on state assessment and fault prediction has less method.Oil is usually used to insulate and emit heats in the large power transformer. Under the condition of running voltage, the transformer oil and the solid organic insulated material (paper and cardboard etc.) in the oil will decompose because of the multiplicate factors such as electrics, heats, oxidation and partial arc. It will be decomposed to produce a few low molecular hydrocarbons such as firedamp (CH₄), ethane (C₂H₆), ethene (C₂H₄), ethane (C₂H₂) and carbon monoxide (CO), carbon dioxide (CO₂), hydrogen (H₂). Most of the gases will dissolve in the oil. The potential overheated and discharging electricity malfunction in the transformer will quicken the speed of gas producing. In theory, different types of fault decompose different gases. Different degree of the same fault results the capacity of gases is different too. In a certain extent, the component and content of dissolved gases in transformer indicate the insulation aging and malfunction degree of transformer. It can be used as characteristics which can reflect power equipment abnormity. In order to ensure safety and credibility of the transformer, it is important to detect the component, content and speed of gas producing in the oil of transformer, to find out the potential malfunction in the transformer. The paper will study on power transformer fault diagnosis technology based on dissolved gases analysis.(1) Aiming at power transformer state assessment, grey target theory is introduced in power transformer fault diagnosis. The state assessment method of power transformer based on grey target theory is proposed. It takes the health state specification as bull's-eye of grey target. The bigger approaching degree is, the closer power transformer approaches health state. The smaller approaching degree is, the higher power transformer fault degree. The method overcomes randomicity influence taking the serious fault as bull's-eye of grey target, which makes state assessment having rationality and generalization.The grey target theory can assess the state of power transformer without the standard fault model. If the assessment result approached reality fact wants to be gotten, the state specification model of power transformer should be chosen and optimized by plentiful experiments. Based on plentiful statistic data of dissolved gases, the paper proposes a high veracity method of power transformer state model choosing, constructing standard fault model with mean of different degree fault, which improves the stability and currency of assessment algorithm.Based on a great deal of statistic data of power transformer fault type and fault degree, the paper proposes classification strategy combining rank classification with grading classification. This kind of classification strategy not only has rationality but also accords with ordinary understanding custom.In one hundred groups of natural power transformer data, there are ninety-eight groups approaching degree above 0.6(score above 60). Fault judging veracity is 98 %. In one hundred groups of fault power transformer data, there are ninety-six groups approaching degree under 0.6(score under 60). Fault judging veracity is 96%. Experiment results show the application of grey target theory in power transformer state assessment is effective.(2) Aiming at power transformer fault diagnosis, advanced nonlinear homotopy back propagation (ANHBP) algorithm is proposed.Based on a great deal of statistic data of power transformer fault state, the fault state is divided normal running, low temperature superheat, medium temperature superheat, high temperature superheat, low-energy discharge, and high-energy discharge. The dissolved gases in oil have some mapping relation to power transformer fault state. This relation is difficult to express with appointed expression. Based on the former work foundation, the paper confirms power transformer fault diagnosis method based on artificial neural network (ANN). Combining power transformer fault gas and fault style characteristic, the ANN model of power transformer fault diagnosis is established.Combining the common improved arithmetic of BP network and dissolved gases characteristic of power transformer, optimized LM method is confirmed as learning method of ANN based on emulation experiments.Aiming at local minimization problem of BP network, the homotopy method is introduced in learning process of ANN. Advanced nonlinear homotopy back propagation (ANHBP) algorithm is proposed, which improves greatly the ability to avoid getting in local minimization. Through network model training and power transformer fault diagnosis validating, the fault diagnosis rate in full accord with fact is 79.4%. The fault diagnosis rate in general accord with fact is 11.4%. The fault diagnosis rate in incomplete accord with fact is 4.3%. The wrong fault diagnosis rate is 5.0%. Simulation experiments of standard BP arithmetic and nonlinear homotopy BP arithmetic are made individually by 50 times. The succeeding times to avoid getting in local minimization improved from 31 to 46.(3) Aiming at power transformer fault prediction, the paper researches the common fault prediction method and validates the advantage of grey prediction theory in power transformer fault prediction. Based on GM (1,1) model, anadvanced grey prediction GM (1,1,ρ) model has been constructed.Firstly, model data is pretreated with Lagrange average interpolation method. The non-equidistance data constructs equidistance list. Secondly, equal dimension restriction condition is introduced. The disadvantage of grey model is remedied by equal dimension dynamic prediction model. In the process of background value generation, background value is constructed with weighted coefficient. At last, Markov prediction model is introduced to revise the prediction data.The experimental results demonstrate that mean simulation relative error, mean sqare error ratio and infinitesimal error probability are better than GM (1,1)model. The model does not need abundant Statistic data. It only needs four or more data to construct high precision prediction model. which can forecast effectively the gas concentration in oil at a certain future time. Combining the fault diagnosis method proposed in the paper, we can forecast the fault style, fault part, and fault serious degree.(4) Based on state assessment, fault diagnosis, and fault prediction, the paper constructs expert system of power transformer fault diagnosis and designs software flat of expert system. It researches mainly reasoning mechanism based on information fusion of ANN and best practices of comprehensive correlation analysis. The two reasoning mechanisms are independent correspondingly and syncretic with each other, which gives helpful attempt in some degree to solve the shortage of traditional expert system in knowledge acquirement and expression singleness.The power transformer state assessment, fault diagnosis, and fault prediction are an organic whole. The paper proposed the triune research idea in power transformer fault diagnosis technology. It provides theory foundation for power transformer state maintenance. It has important theory meaning and factual application value to reduce maintenance costs and improve power transformer dependability.