AN ARCHITECTURAL FRAMEWORK FOR THE CONSTRUCTION OF HYBRID INTELLIGENTFORECASTING SYSTEMS - APPLICATION FOR ELECTRICITY DEMAND PREDICTION

This paper presents an implemental architectural framework for the con struction of hybrid intelligent forecasters for utility demand predict ion. The framework has been implemented as the intelligent forecasters construction set (IFCS), which supports the intelligent techniques of fuzzy logic, artificial neural networks, and knowledge-based and case -based reasoning. IFCS is also a hybrid programming tool, which allows the developer to implement forecasters by means of object-oriented vi sual programming, knowledge-based programming and procedural programmi ng. The system was implemented on the real-time expert-system shell G2 , with the G2 Diagnostic Assistant (GDA) and NeurOn-Line (NOL) modules . Rules, procedures and flow diagrams are organized into a hierarchy o f workspaces. The modularity of IFCS allows the subsequent addition of other modules of intelligent techniques. IFCS was applied for daily p ower-load prediction in the city of Regina. The power-load data set wa s separated into subclasses, and a neural-network module consisting of backpropagation networks was applied to each of them. The data set wa s those from the neural-network approach. (C) 1998 Elsevier Science Lt d. All rights reserved.