Research On Time Series Data Mining And Application To Short-Term Load Forecasting

Sort-term load forecasting (STLF) is an important task in electric dispatcher automation system. It is the important foundation of the study on electric system planning problem, economical running and dispatcher automation. The effective algorithm of STLF and its application are studied in this paper.According to the characteristic that electric load is a kind of time series, this paper puts forward a new short-term forecasting algorithm based on data mining technology of time series. Aiming to solve the problem of low precision caused by building forecasting models at a series of time points independently, a new idea which is named load-series-forecasting (LSF) is proposed. LSF builds forecasting models in a series of time slices so as to catch the shape and change pattern of the load between corresponding interval, It includes two key parts, the one is load series analysis based on clustering, the other one is load forecasting based on local association classification.In the study of load series analysis, a new dimension reduction method for load series based on feature points is proposed, then a hierarchical clustering based on density (HCD) is applied to find similar load pattern and special load pattern. Further more, HCD is also regarded as the pretreatment for farther data mining.In the study of load series forecasting, a more accurate method named classification based on local association rules (CLAR) is proposed to mine association rules between influence factors and load series. Rules are stored in rules prefix general list (RPG-List), which reduces the storage and facilitates rule matching. Then a classifier based on rules is constructed. By considering factors such as intending weather, the load is forecasted effectively.The implementation of the system proves that short-term forecasting model based on data mining technology of time series is feasible and effective.