Bahman Kermanshahi - Environmental Energy Engineering Lab. Department of Electrical and Electronics Engineering
Ken Nagasaka - Tokyo University of Agriculture and Technology, JAPAN

This paper presents the application of an Artificial neural Network (ANN) for short-term load forecasting (STLF) of an urban area in Japan (Tokyo Metropolitan Area). The basic reason for STLF is to predict the loads of next 24 hours ahead. This will allow a utility to match its generation capabilities to the expected demand. Although, there are several methods to determine such forecasting, however a neural network is selected in this paper. The reason is because the ANN can correspond to the problems, which have highly nonlinearity among their inputs and outputs. The ANN mode can forecast next 24 hours loads (one day) in advance. Presentl y, one of the most popular, effective, and easy to learn three-layer back- propagation (BP) models is introduced. 15 to 25 input neurons are chosen based on their impacts on the output loads. They are. actual load, temperature, wind speed and direction There is one neuron in output layer which is corresponded to an hour load of next day. Training and testing of STLF are carried out based on historical data collected for two years (calendar year 1996 and 1997). On the basis of each day's load characteristics, three ANNs are designed for STLF. Also, three types of simulations for STLF have been discussed.

Artificial neural network, Short-term load forecasting, Urban area