Application of Fuzzy Cognitive Maps To Water Demand Prediction

This article is focused on the issue of learning of Fuzzy Cognitive Maps designed to model and predict time series. The multi-step supervised-learning based-on-gradient methods as well as population-based learning, with the use of real coded genetic algorithms, are described. In this study, a new structure optimization genetic algorithm for fuzzy cognitive maps learning is proposed for automatic construction of FCM applied to time series prediction. The proposed learning methodologies are based on an FCM reconstruction procedure using historical time series. The main contribution of this study is the analysis of the use of FCMs with their learning algorithms based on the multi-step gradient method (MGM) and other population-based methods to predict water demand. The performance of learning algorithms is presented through the analysis of real data of daily water demand and the corresponding prediction. The multivariate analysis of historical water demand data is held for five variables, mean and high temperature, precipitation, wind speed and touristic activity. Simulation results were obtained with the ISEMK (Intelligent Expert System based on Cognitive Maps) software tool. Through the experimental analysis, we demonstrate the usefulness of the new proposed FCM learning algorithm in water demand prediction, by calculating the known prediction errors. The advantage of the optimization genetic algorithm structure is its ability to select the most significant relations between concepts for prediction.