Power Transformer Fault Diagnosis Method Based On Intelligent Information Processing Theory

As China's power grid develops to the direction of highly automation, people's livelihood demands on security and reliability of long lasting electricity power supply increase. Condition based maintenance system based on on-line monitoring and fault diagnosis urgently need to be established. A power transformer is one of the key equipments in the power system, therefore its operating condition directly affects the power system running state. Research of power transformer fault diagnosis methods has important theoretical value and broad engineering application prospects, and it is important for the power system to raise operational safety and reliability. Research of power transformer fault diagnosis methods is also meaningful for the power transformers to reduce maintenance costs and economic loss caused by their faults. Power sectors in domestic and abroad attach importance to research of power transformer fault diagnosis methods.Power transformer fault diagnosis methods include power transformer condition assessment, fault diagnosis and fault forecast three objectives. Power transformer condition assessment uses monitor characteristic parameters of power transformer to evaluate its running status, and then formulates the appropriate maintenance strategy. Power transformer fault diagnosis is to determine its fault type and fault location. Power transformer fault prediction forecasts power transformer running trends based on available data. Dissolved gases analysis method analyzes composition, content and gas production rate of hydrogen, methane, ethane, acetylene, ethylene and other gases dissolved in transformer oil to determine power transformer latent fault types and its location. DGA is an international general technical means for power transformer fault diagnosis. Intelligent information processing theory provides new ideas to power transformer fault diagnosis method for its development and progress. Power transformer fault diagnosis data are small samples, and have less information. It is an important problem to be solved in power transformer fault diagnosis technology that how to deal with and utilize such data.This thesis began with present research of the power transformers condition assessment, fault diagnosis and fault prediction, therefore proposed a power transformer fault diagnosis technic system consist of the power transformers condition assessment, fault diagnosis and dissolved gases concentrations prediction. In the thesis, the research of the power transformers condition assessment, fault diagnosis and dissolved gases concentrations prediction algorithms based on intelligent information processing theory was designed for characteristics of power transformer data. The thesis also developed a power transformer fault diagnosis software platform and built a power transformer fault diagnosis database. The main research contents included the following:⑴Research of key technologies in intelligent information processing theoryIn this thesis, cloud model, gray system theory, artificial immune system, genetic algorithm, support vector machine and other key technology features were studied and analyzed to solved problems of power transformer condition assessment, fault diagnosis and dissolved gas content prediction. These theories were experimented on power transformer fault diagnosis to analysis their features. The power transformers fault diagnosis technical program based on intelligent information processing theory was proposed. Research of key technologies in intelligent information processing theory provided theoretical basis for the power transformers fault diagnosis.⑵Research of power transformer condition assessment algorithmPower transformer condition assessment algorithm proposed in the thesis optimized evaluation index system of power transformer condition, which includes dissolved gases analysis, electrical test, insulating oil characteristics analysis and qualitative indicators. Gray cluster analysis based on cloud model white function combined with improved weighted gray target theory evaluated transformer condition was researched under the power transformer condition evaluation index system. Cloud model is good at analyzing fuzzy, random uncertainty problems. Gray cluster analysis has advantages of solving small samples and poor information gray system. Gray cluster analysis based on cloud model white function is suitable for fuzziness, randomness and gray of power transformer condition assessment. Related standards and priori knowledge about power transformer index both optimized cloud model parameters. For power transformer condition assessments only based on dissolved gases analysis data, an improved weighted gray target theory whose center of the gray target is the normal power transformer DGA data was proposed, and its boundaries of condition classification was optimized by power transformers fault severity distribution priori knowledge. Qualitative indicators were determined by expert scoring system. Experiments show that power transformer condition assessment model combine gray cluster analysis based on cloud model white function with improved weighted gray target theory is able to correctly evaluate power transform condition, raises the scientific and objective of power transformer condition assessment, and has practical significance on power transformer condition assessment and other assessment areas.⑶Research of power transformer fault diagnosis algorithmBiological immune system firstly judges whether a cell is autologous cell or not. Then, non-autologous cell is deal with high affinity antibody. The power transformers fault diagnosis algorithm addressed in the thesis was composed of 2 cascade classifiers, which simulated biological immune system process. Firstly, primary classifier determined whether a power transformer is normal. Support vector machine classified power transformer status into normal or fault two categories because power transformer fault diagnosis data was small samples and had poor information. Both support vector machine kernel function and function parameters affect the results of support vector machine classification. Genetic algorithm is global search optimization method. Support vector machine kernel function selected radial basis function. Genetic algorithm optimized parameters of radial basis function. Secondly, best memory antibody set of artificial immune algorithm were trained. Gray relational analysis measured the affinity between antibody and antigen, which was called gray artificial immune algorithm. In gray artificial immune algorithm, priori knowledge of power transformer fault diagnosis constructed vaccine set. High-frequency mutation operator based on the dynamic vaccine was constructed. The best memory antibody set was trained depending on the fault type. Finally, 5 neighbors integrated decision making method diagnosed detail type of power transformer fault according to the best memory antibody set. The 5 neighbors integrated decision making method integrated diagnosed fault type according to both vote number and rank location. Experiments indicate that the power transformer fault diagnosis algorithm combined genetic SVM with gray artificial immune algorithm based on dynamic vaccine can diagnosis power transformer single and multiple faults effectively and improves diagnosis accuracy.⑷Research of dissolved gases concentrations prediction algorithmPower transformer dissolved gases concentrations prediction is the basis for fault prediction because it provides data for power transformer fault prediction. The thesis analyzed the characteristics of power transformer oil dissolved gases concentration data varying with time, and then designed a prediction model composed of two parallel models. For DGA data with single peak, gray Verhulst model forecasted power transformer dissolved gases concentrations. Gray Verhulst model and its improved forms were compared by experiments. For DGA data with monotonically increasing feature, Double-Layer GM(1,1,ρ) Metabolism model was used to predict power transformer dissolved gases concentrations. In Double-Layer GM(1,1,ρ) Metabolism model, GM (1,1,ρ) model forecasted both original data and residuals. Markov chain controlled the polarity between the two predictions. The metabolic parameter was analyzed. Experiments demonstrate that the dissolved gas concentrations prediction algorithm based on gray Verhulst model and DLGMM(1,1,ρ) model can fully reflect the dissolved gases concentration variation, therefore it has higher prediction accuracy.⑸Power transformer fault diagnosis software platform development and power transformer fault diagnosis database constructionMATLAB GUI and M language developed power transformer fault diagnosis and prediction software platform, and achieved main functions such as power transformer condition assessment, improved three-ration method and the Gaussian gray cluster analysis fault diagnosis, dissolved gases concentration and fault prediction, save new data and printout based on former power transformer condition assessment, fault diagnosis and dissolved gases concentrations prediction algorithms. Access database software built power transformer fault diagnosis database according to power transformer condition assessment, fault diagnosis and dissolved gas concentrations prediction data features. The power transformer fault diagnosis database included power transformer condition assessment sub-database, fault diagnosis sub-database and dissolved gases concentrations prediction sub-database. Corresponding attributes, tuples and primary keys of each relation sheet in sub-database were designed. Window, query and report operations were realized.This thesis systematically researched power transformer condition assessment, fault diagnosis and dissolved gases concentrations prediction, and detailed studied and analyzed key technologies of intelligent information processing theory. The thesis developed power transformer fault diagnosis and prediction software platform, and constructed power transformer fault diagnosis database. Research work in the thesis has important theoretical significance and practical value on development of power transformer fault diagnosis method. The intelligent information processing algorithms proposed in the thesis also have some reference value and broad engineering application prospects to solve related problems.