Integration of data mining techniques and heuristic knowledge in fuzzy logic modelling of eutrophication in Taihu Lake

Abstract

A fuzzy logic (FL) model was developed in this study to predict algal biomass concentration in the eutrophic Taihu Lake, China. Common to any FL model, definition of membership functions and induction of inference rules remain difficult. They conventionally rely on the use of “heuristic knowledge”, which usually seems to be not enough to fulfil practical requirement or even unavailable. In this fuzzy model, a method combining data mining techniques with heuristic knowledge is developed. It used principal component analysis (PCA) to identify the major abiotic driving factors and to reduce dimensionality. Self-organising feature map (SOFM) technique and empirical knowledge were applied jointly to define the membership functions and to induce inference rules. As indicated by the results, the fuzzy model successfully demonstrated the potentials of exploring “embedded information” by combining data mining techniques with heuristic knowledge. The developed method had also been introduced to the European Commission project Harmful Algal Bloom Expert System (HABES) which involves 13 institutes and universities from 9 EU countries. The objectives of this paper are to illustrate how these techniques are integrated and how the developed fuzzy model is applied to predict the algal biomass concentration in the eutrophic lake.