Advancements in the application of neural networks and fuzzy logic for short term load forecasting

Short Term Load Forecasting (STLF) is an essential function in electric power system operations and planning. Forecasts are needed for a variety of utility activities such as generation scheduling, scheduling of fuel purchases, maintenance scheduling and security analysis. Depending on power system characteristics, significant forecasting errors may lead to either excessively conservative scheduling or excessively risky scheduling. These can induce heavy economic penalties. Numerous models, including time series, regression, and expert systems, have been proposed and, in a limited sense, used to generate forecasts.; Recently, artificial neural networks (ANN) have been applied for the task of STLF, and show some promising results. Most references dealing with ANN applications to load forecasting use a neural network model called "Back-propagation" (BP). The BP model may not be the best suitable neural network model for load forecasting. This dissertation investigates the feasibility of several promising ANN models for STLF. Results show that the ANN model called, "Radial Basis Function" is a better alternative for STLF.; One of the major short comings of ANN based load forecasting methods is that they require precise information about the input weather variables. Usually, power utilities do not receive exact weather inputs, instead they receive forecasted weather readings. This dissertation describes a novel probabilistic approach to include the uncertainties of these weather inputs. This dissertation also desribes a feasibility study of the use of fuzzy logic for STLF. It has been observed that the accuracy of load forecasts from a fuzzy logic based method is very similar to that of the ANN based methods.