Non-stationary Signal Forecasting by Neural Network with Modified Neurons

This paper presents the non-stationary power signal forecasting by using a neural network with modified neurons for PJM data set provided by Independent Electricity System Operator (IESO). In this data set, the load information is the sum of power load consumed by three areas, including Allentown, Baltimore and Philadelphia. The historical load and temperature information from year 2003 to year 2008 were studied and simulated. The forecasts of one-day-ahead daily total load and peak load were implemented. In order to find the accurate forecasting results, different combinations of inputs were carried out. In this study, mean absolute percentage error (MAPE) is used as the measurement of forecasting performances.