Implementing fuzzy cognitive maps with neural networks for natural gas prediction

Abstract

The goal of this research study is to test the hardiness of a novel hybrid computational intelligence model in day-ahead natural gas demand prediction. The proposed model combines an evolutionary learned FCM method with a common ANN to construct a cascaded model that leads to high prediction accuracy in most distribution points. The FCM technique is used to provide a model which concepts are used as input nodes in a second-stage ANN model employed to provide the forecast for each gas time series. Learned by structure optimization genetic algorithm, the FCM outputs are fed into an ANN to refine the initial forecast and upgrade the overall forecasting accuracy. The model is applied to five distribution points that compose the natural gas grid of a Greek region, district of Thessaly. This approach enables the comparison of the hybrid model performance on different FCM and ANN structures and on consumption patterns, providing also insights on the characteristics of large urban centers and small towns. © 2018 IEEE.