Using an Adaptive Neuro-fuzzy Inference System in the Development of a Real-Time Expert System for Flood Forecasting

Abstract

AbstractThis chapter describes the development of a prototype flood forecasting system provided in a real-time expert system shell called COGSYS KBS. Current efforts on the development of flood forecasting approaches have highlighted the need for fuzzy-based learning strategies to be used in extracting rules that are then encapsulated in an expert system. These strategies aim to identify heuristic relationships that exist between forecast points along the river. Each upstream forecast point automatically produces extra knowledge for target downstream forecast points. Importantly, these strategies are based on the adaptive network-based fuzzy inference system (ANFIS) technique, which is used to extract and incorporate the knowledge of each forecast point and generate a set of fuzzy “if–then” rules to be exploited in building a knowledge base. In this study, different strategies based on ANFIS were utilised. The ANFIS structure was used to analyse relationships between past and present knowledge of the upstream forecast points and the downstream forecast points, which were the target forecast points at which to forecast 6-hour-ahead water levels. During the latter stages of development of the prototype expert system, the extracted rules were encapsulated in COGSYS KBS. COGSYS KBS is a real-time expert system with facilities designed for real-time reasoning in an industrial context and also deals with uncertainty. The expert system development process showed promising results even though updating the knowledge base with reliable new knowledge is required to improve the expert system performance in real time.KeywordsExpert systemadaptive neuro-fuzzy inference system (ANFIS)COGSYS KBSflood forecasting