Short-term load forecasting using a Gaussian process model: The influence of a derivative term in the input regressor

Abstract

This paper establishes a regressive methodology, applied to the electrical distribution sector, to predict the next hour load. The approach followed is based on a Gaussian process model. To create a simple structure for the input regressor only a few instances of the endogenous variables, based on the homologous and contiguous values, were included. Information that allows to evaluate the influence of a derivative term in the quality of the forecasts is also included in the input vector. In order to assess the performance of different load situations the model was tested on a real-load case-study. The case study includes three different electrical distribution substations, representative of typical load consumer's patterns, namely the residential, non-residential and the service sector. The results obtained are in accordance with the values normally accepted in the sector.