Font Size: a A A

Research Of Appling Data Mining To Intelligent Modeling Of Short-time Load Forecasting Of Power Systems

Posted on:2005-07-26Degree:MasterType:Thesis
Country:ChinaCandidate:X H LiuFull Text:PDF
GTID:2132360152967385Subject:Probability theory and mathematical statistics
Abstract/Summary:PDF Full Text Request
Load forecasting plays an essential role in power industry. Economic and reliableoperation of a power plant depends significantly on the accuracy of load prediction withdevelopment of power markets. Many researchers have proposed several dozens oftheories and method about load forecasting since 1980s. Among these theories, thepopular thought is to seek for math model and algorithm, which can simulate thenonlinear mapping function of load change and the important factors pertaining to it. Ifthe model exists, only the factors should be demanded. Artificial Neural Network modelis prevalence for that approaching ability. According to the publication, more and moreprecise but intricate models and algorithms were founded, while data analyses werereceived relative less concerned. In fact, the prediction precision hardly satisfies withoutaccurate factor analysis. A point of view supported by this dissertation is to give much more concern ondata analysis than before. Owing to the ability of distilling connotative knowledge andinformation from abundant data, Data mining can be used to set up the intelligent loadprediction model. Much knowledge, including the influencing factors to load change,the law of load change with factors change, the training samples of networks and inputsamples for prediction corresponding to the predicting day, will be distilled byinstruction of the many different meta-templates. Besides, the database of models andalgorithms has been suggested here and been shared by data mining engine and functionsimulating. The idea of disjunction between models and algorithms is proposed in acreative way. With it, much efficiency, flexility and resource saving will be made.Trough the analysis to the tested and forecasted records made by various combinationsof patterns, models and algorithm, the load forecasting model will be founded. A novelclustering method ---- Ranking Means Cluster is proposed in this dissertation. Thismethod is able to avoid local optimal solution with traditional K-means clusteralgorithm and it can also decide the number of clusters to be classified into. With ouralgorithm, the central vectors of hidden layers in RBF models can be computed and thenodes number and RBF infrastructure can also be decided. Moreover, a new interactivelearning scheme is proposed here to choose network parameters. The learning samplesare categorized into training samples and testing samples, which lead to stable networkstructure by adjusting the power values and radius.
Keywords/Search Tags:Power systems, Load forecasting, Mathematics Model, Data Mining, Artificial Neural Network, Clustering
PDF Full Text Request
Related items