PATTERN RECOGNITION APPROACH AND ARRAY PROCESSING FOR DISTRIBUTED SOURCE IDENTIFICATION IN WATER POLLUTION SYSTEMS (ENGINEERING, ENVIRONMENTAL SCIENCE)

The research described in this dissertation is directed to the development of a methodology for the identification of input functions in distributed parameter systems and more specifically of pollution sources in water pollution systems. Two major challenging problem areas, i.e., river pollution systems and aquifer pollution systems, are discussed as examples. Conventional identification methods such as the regularization method among others have several crucial drawbacks. In particular they need restricted assumptions on pollution sources and involve a large mount of computation time. In order to overcome these difficulties, a pattern recognition approach including feature extraction and signal processing is introduced. Coherence functions and the normalized correlation function are employed as feature vectors to extract the original pollution pattern from the measurement data with the presence of high-level noise. Conventional identification methods are then employed to specify the extracted pollution sources more precisely. The entire identification procedure is executed by a host/peripheral array processor to improve computational speed. In particular, performance evaluation of the partial differential equations, the calculation of the feature vectors, the calculation of the conventional identification method of the identification process, are performed using DEC VAX-11/750 and CSPI Mini-Map array processor. The Monte Carlo method is introduced and executed to solve the partial differential equations. In order to demonstrate the verification of the entire identification procedure, several numerical examples are analyzed with the aid of simulations. Evaluations of the identification procedure are made by varying the noise level of the measurement data.