Applied Research Of The Optimized Model GM(1,1) In Prediction Of The Concentration Of Dissolved Gases In Transformer Oil

Power transformer is an very important hub for the equipment in the electricity network,and its normal working is the guarantee of the security of grid and reliable operation. Theevent of a sudden failure will may lead the local power grid and even the entireinterconnected grid to the paralysis of operation. Therefore, it is very important to launch andstrive to do fault prediction task of the power transformer actively. Through establishing ahigh-precision forecast model, effective forecasting and analyzing of the gas concentrations ofoil in the transformer is an effective method of detecting the potential failure of transformerearlier. The paper’s major work is to optimize the gray GM（1，1）prediction model,improving the prediction accuracy of the model and researching its application to predict thevarious gases concentrations of oil in the transformer in death. The main studying works andachievements as follows：（1）Inspecting and researching a large number of domestic and foreign relatedliteratures about the subject, and summarizing and analyzing the optimizing of grey modeland the research of its forecast application status in the gas concentrations of oil in thetransformer.（2）By handling the “abnormal”data with generation processing in the modelingsequence, which can make the regularity of the modeling sequence enhanced and meet themodeling conditions. When the abnormal or unrealistic data appeared in the modeling datasequence, it should be analyzed and examined. If it can’t meet the modeling requirements, itshould be deal with the corresponding value generation processing. And then re-establish thegray GM（1，1）prediction model, which can improve the prediction accuracy effectivelythrough forecast and analysis of examples.（3）Increasing the smoothness of the original modeling sequence can enhance theprediction accuracy of the gray model effectively. Based on the theory of the functiontransform, by establishing a data transformation of constructor and rules, and handling withtransformation processing bycos(x~α)of the modeling sequence, which can enhance thesmoothness of the sequence and re-establishing the gray GM（1，1）prediction model basedoncos(x~α). The prediction and analysis of two examples indicates that the established graymodel can further enhance the precision of both in the multi-data and less data sequenceprediction effectively. （4）The establishment of a residual error correction model to correct the original modelcan improve the prediction accuracy of the model effectively. The paper according to thecharacteristics of the residuals data sequence changes and combining the advantages of thegray Verhulst model and dealing the residual data sequence with corresponding processingaimed at making it be similar to the curve of the "S" type or the approximate index lawchanges and establishing the residual correcting prediction model of gray Verhulst. If it can’tmeet the requirements of accuracy, the prediction model should be corrected by the residualmodification processing which can further improve the prediction accuracy of the model.（5）Based on the basics of the gray theory and the optimization ideas of the gray GM(1,1) model, the comprehensive gray GM（1，1）prediction model was established to furtherimprove the prediction accuracy of the model. Through using several models, predicted andanalyzed the several gases concentrations of oil in the transformer respectively and comparedwith the results of the respective prediction. The results show that the model can improve theprecision of the model effectively, but after based on the residuals correction of the grayVerhulst, the effects of the correction method is better than GM residual correction method.Meanwhile, it also shows that the former method can further improve the prediction accuracyand adaptability.（6）The above researchful achievements are summarized and the further ideas andmethods of improvements are presented in the end of the paper.