Study On Mid-long Term Load Characteristic Analysis And Forecasting Method

Study on mid-long term load characteristics aims to grasp the variationand trend of power load, which has important significance to the planningand operation of power systems. So it is necessary to explore a new loadcharacteristic analysis method and propose improvements to loadforecasting method.In this thesis, load characteristic indexes are defined and the function ofload characteristic indexes is discussed. Based on these indexes, the loadcharacteristics of a real power grid are analyzed as an example. Then, thefunctional data analysis method is introduced into the analysis of loadcharacteristics to give a functional perspective on the dynamic changes ofload over time. By using the functional principal component analysis,differences between load characteristic indexes of various areas and seasonshave been revealed. The peak load is one of the most widely used loadcharacteristic indexes, so the peak load forecasting method based on supportvector machine is studied. On the basis of traditional SVM predictionmethod,improvements are proposed:adopting fuzzy membership coefficientsto indicate the importance of various data samples, optimizing modelparameters by using particle swarm algorithm,and using correlation analysisto select input variables of training samples. Typical daily load curvescontain a wealth of information, from which indexes like load factor can beobtained. Considering that traditional fractal method applies only to shortterm load curve forecasting, improvements on it are discussed: firstly, theadjustment vector is defined to reflect the growing information between long term power load curves; secondly, the "timing shift" technique isadopted to optimize the initial iteration point. With above steps, theapplication field of fractal method can be extended to mid-long termforecasting. Satisfactory forecast result has been got after applying theimproved models to forecasting load characteristics of a real power system.