AN ANALYSIS OF WEATHER FORECASTS IN THE CONTEXT OF ELECTRICITY USE

The development and performance of an electricity load-forecasting model, with meteorological parameters as forcing terms, is investigated. This study aims to answer the question of how to assess what a good forecast is and how it can be achieved. Econometric techniques including ordinary least squares estimation, autocorrelation corrections, and the Autoregressive Conditional Heteoroskedasticity (ARCH) model, are applied to meteorological and electricity load data from Philadelphia and Northern Illinois, including Chicago, to develop a statistical load-forecasting model. Once the model is trained on observed meteorological and load data, Eta and NGM Model Output Statistics are used as input data to assess the skill of the load model. The root mean squared error is used as the primary assessment tool of a good forecast, expressed as a percentage of the average load. Both NGM and Eta statistics exhibit error between 6.5% and 10% of the average load, with Eta MOS performing the best in Chicago and NGM MOS performing the best in Philadelphia.