I. INTRODUCTION
Load Forecasting is very important to the power system operation for economic and security reasons 131. If an accurate estimate of the power system load is available with leading times of at least 24 hours, then it is possible to coordinate the energy generation economically, by selecting the more adequate generating unites to supply the load. Security analysis of the power system uses prdcted loads to assess the state of the system in advance, so that future contingencies may be prevented. Moreover, a load forecasting system is also important as an analytical tool used in electrical power systems recovery. Many authors have already addressed the use of Artificial Neural Networks (ANNs) to the load forecasting problem [1,2,4]. However, most of the works is this area consider 0-7803-2972-4/96$5.00@1996 IEEE short term load forecasting (STLJ), whlch is the load prediction with a lending time of up to 24 hours. Besides, weather variables (m;unly temperature) have been commonly used in most of these applications. At ICA (Research Centre for Applied C0mputatic)nal Intelligence of PUC-Rio, Brazil), we have used the Backpropagation Neural Network [5] in a multi-step procedure to predict the load with leading times of up to 744 hours. Forecasting is made by feeding the Neural Network with the past loads and the with hour of the day. We have adopted four different Neural Nets to predict the load of dfferent groups of week days. Temperature data is not used due to the characteristic of the consumers supplied by the electric company in study. Our objective is to show that accurate prebctions can be obtained using tlvs approach and to investigate the performance limit as we increase the predctions' leading time. The following section presents the load series previ.ous analysis that was made in order to identify the most suitable Neural Network topology. In section 3 we present the cho'sen Neural Network topology and section 4 presents the case studes and results. The conclusions of the work are presented in section 5.