Abstract:
Electricity is one of the most essential component of human life. Maintain ing and operating electric power systems is a complex task. Short term electricity load forecasting plays an important role in operation of electric systems. Plenty of methodologies have been applied to perform short term load forecast. In this study, performances of several machine learning methodologies such as random forest, sup port vector machines and gradient boosting method are analyzed in short term load forecasting. Performances of these methods are compared with performances of three naive methods, several time series methods and linear regression. All methods are tested on a dataset which includes hourly electricity load information of a region of a country and hourly temperature values of cities in that region. For machine learning methods and linear regression, a feature set is constructed based on calendar, temper ature and past load information. Minimum absolute percentage error(MAPE) is used as performance metric to compare methodologies. According to results of experiments, machine learning methods showed better performance than conventional methods in short term load forecasting.
