SHORT TERM LOAD FORECASTING WITH MULTILAYER PERCEPTRON AND RECURRENT NEURAL NETWORKS

Muhammad Riaz Khan - Čestmír Ondrůšek

   The ability of the Elman recurrent neural network (RNN) to model the short-term load forecasting (STLF) problem is investigated in this paper. Its performance in a competition is then contrasted with that of a multilayer perceptron (MLP) network. It is postulated that the load can be modeled as the output of some dynamic system, influenced by a number of weather, time and other environmental variables. RNN exhibiting inherent dynamic behavior can thus be used to construct a forecasting model for this dynamic system. Due to a nonlinear dynamic nature of this model, the behavior of the load prediction system can be captured in a compact and robust representation. This is illustrated by the performance of the Elman RNN model for the short-term forecasting of the nation-wide load for the Czech Electric Power Utility (ČEZ). Both techniques have been trained and tested on the data provided by ČEZ and promising results have been obtained.

Keywords: short-term load forecasting, multilayer perceptron, Elman recurrent neural network, backpropagation learning algorithm

[full-paper]