| Accurate short term load forecasting can help to develop the electricity sector recentlypower generation plan effectively and make the reasonable arrangements for the start and stopof the power plant generating units. It effectively reduces the cost of power generation, andkeep the real-time supply-demand balance of the power grid energy, to improve power systemstability with good social and economic benefits. Especially after the introduction of marketcompetition mechanism in the power industry, it got more and more the concern of theelectrical workers. The support vector machine is a novel machine learning methods. Thesupport vector machine is a novel machine learning methods, can convert the nonlinearproblem to the linear problem, with good performance in pattern recognition and functionregression estimation. But it will run for a long time when you have massive input data toparticipate in the training model.This article first lists a variety of short-term load forecasting algorithm and explainstheir respective advantages and disadvantages. Then it introduces basic theory of supportvector machines and cluster analysis step by step. After the necessary data preprocessing onthe historical load data, it create a data mapping library, and use the fuzzy K-means clusteringto select the similar days in accordance with the climatic conditions and date types, accordingto the social environment and the load characteristics of the Foshan area. Finally, supportvector machine modeling is trained on the JAVA platform, with clustering similar today asinput data and setting reasonable model parameters. Then we try to predict the load(96timepoints) of the Foshan area in March28,2012with this model, and compare its predictedresults to the predicted results of neural network algorithm. The results show that supportvector machine algorithm based on fuzzy K-means clustering can effectively improve theaccuracy of power load forecasting. |
PDF Full Text Request