| Search analyzes the problem of predicting annual pea load for electrical utilities and provides a measure of risk or uncertainty on the prediction. A review of the literature dealing with this topic revealed that the overwhelming number of works in electrical peak load prediction centered on the short term, i.e., hourly or daily predictions of peak load. Further, it was found that methods for estimating prediction intervals or assigning risk to the point estimate were not discussed. The work to date on this topic can be classified by the method applied to solve the problem. These methods are: (1) regression based approaches; parametric and non-parametric, (2) stochastic or probabilistic approaches, such as time series analysis, (3) neural networks or expert systems, (4) fuzzy logic based approaches, and (5) econometric models. A thorough review along with application of the regression based approaches and the stochastic model is presented herein. The major contribution of this work is to present three new approaches for predicting annual peak load and for assigning a measure of risk or uncertainty to the predicted quantity. These now approaches are (1) The Bootstrap, (2) Extreme Value Theory, and (3) Parametric Survival Models. Each approach is detailed and applied to predict peak load in successive years. Data is made available for this project by the Power Service Company of Oklahoma and includes daily peak load readings and the maximum and minimum temperature recorded on each day for all days between 1982 and 1998. Finally, the results using each model are compared and a summary of conclusions is presented. |
