| Forecasting the hourly electricity prices has been a topic of interest since the introduction of deregulation and competition in electricity markets. This thesis introduces the new idea of "threshold forecasting" or "classification" of market prices as a new approach to electricity price prediction, versus point forecasting of prices, where the exact values of future prices are determined. The motivation for this idea is that the accuracy level presented in the literature for point price forecasting methods is usually low and not all market participants need to know the exact values of prices. Support Vector Machines (SVMs) are used as the core classifier in this work, and two alternative classification models are proposed. The classification results are compared with those of other alternative classifiers as well as those obtained using point forecasting methods in the previous literature. These comparisons verify the effectiveness of the proposed models. |
