Medium Term Electric Load Forecasting Using TLFN Neural Networks

Abstract

This paper develops medium term electric load forecasting using neural networks, based on historical series of electric load, economic and demographic variables. The neural network chosen for this work is the Time Lagged Feedforward Network (TLFN), which combines conventional network topology (multilayer perceptron) with good handling of time dependencies by means of gamma memory. This is a versatile mechanism that generalizes the short term structures of memory, based on delays and recurrences. This scheme allows smaller adjustments without requiring changes in the general network structure. The neural model gave satisfactory results exceeding those obtained by classical statistical models like multiple linear regression.