Predicción del Precio de la Electricidad en la Bolsa mediante un Modelo Neuronal No-Lineal Autorregresivo con Entradas Exógenas

Abstract

The forecasting of the monthly average stock price of Colombian electricity was carried out by means of a nonlinear autoregressive exogenous model. Such a model considers as exogenous inputs the demand, the ratio between hydraulic and thermal generation, the probability of El Niño phenomenon and the daily available energy volume; and as an autoregressive variable the monthly average stock price. For training, validation and testing, data corresponding to the period January 2003 to March 2014 were considered. Furthermore, the effectiveness of the model was tested with data for the period April 2014 to February 2015.This shows that the incorporation of historic data of prices and variables that depend on hydrologic conditions, such as El Niño phenomenon, allows reproducing the dynamics of Colombia's energy prices, even for forthcoming periods.