Explicit representation of knowledge acquired from plant historical data using neural network

Abstract

A causal index, which translates implicit knowledge contained in a neural network into an explicit representation, is proposed. A backpropagation-based learning algorithm which suppresses the nondominant causal relationships to improve association accuracy is developed using the index. The validity of the proposed algorithm is investigated using historical data from a coagulant injection operation in a water-purification plant. Learning of relationships between the operational factor (coagulant injection rate) and influence factors (water qualities and floc image properties) in plant operations is carried out by the proposed algorithm. Improvements in the association ability for unknown conditions and in the reliability of acquired knowledge are realized