A neural network based short term load forecasting model

Abstract

A novel feedforward two layer ANN neural network based function approximator model is utilized to forecast electric system hourly load. The forecast model is based on a quantitative weight assignment priority factors for the day type and daytime classes, in addition to the daily average temperature. The forecast vector utilizes scaled historical load data for eight day type classes, four day time subclasses as well as load pattern averaged one-hour, six-hour, 24-hour, and 168-hour filtered historical load. To improve the neural network training load, additional variables such as cross correlation and FFT spectra were utilized. To allow for online implementation, the forecast vector was also augmented with the estimated load first and second differential variations. The new ANN-based short term load forecast (STLF) model was tested using two month data sample of the Singapore Public Utilities Board (PUB) historical data. >