SHORT TERM FORECASTING OF URBAN LOAD PROFILES

Abstract

Load forecasting is always defined as basically the science or art of predicting the future load of a given power system, over a specified period of time ahead. These predictions may be just for a fraction of an hour ahead for operation purposes, or as much as twenty years into the future. Load forecasting is a vital component forpower system energy management. Precise load forecasting helps the electric utility to make unit commitment decisions, reduce spinning reserve capacity and schedule device maintenance plan properly which results in reducing generation cost and power system reliability. In this report different approaches for short term load forecasting is used for urban electric load profiles. Support vector machine and relevance vector machine concept is applied for the calculation of accuracy and compared. Enough historical large data is needed for learning of the network such that most of the data is used for training and rest small part is used for testing. Cross-validation is carried before learning the network with given data set which finally leads to the increase in the accuracy of the system. Key factors like weather data, special holidays and weekend data are also included in the analysis to increase the accuracy. In short term load forecasting with SVM and RVM which are supervised learning techniques the accuracy level is quite high enough so the mean absolute percentage error resulted is very small