A neuro-fuzzy-multivariate algorithm for accurate gas consumption estimation in South America with noisy inputs

This paper presents an adaptive-network-based fuzzy inference system (ANFIS)-fuzzy data envelopment analysis (FDEA) algorithm for improvement of long-term natural gas (NG) consumption forecasting and analysis. Two types of ANFIS (Types 1 and 2) have been proposed to forecast annual NG demand. For each type, several ANFIS models have been constructed and tested in order to find the best ANFIS for NG consumption. Two parameters have been considered in construction and examination of plausible ANFIS models (Type 1). Six different membership functions and several linguistic variables are considered in building ANFIS. Also different value of cluster radius has been used to construct ANFIS (Type 2) models. The proposed models consist of two input variables, namely, Gross Domestic Product (GDP) and Population. All trained ANFIS are then compared with respect to mean absolute percentage error (MAPE), Root mean square normalized error (RMSE) and correlation coefficient (R) using data envelopment analysis (DEA). To meet the best performance of the intelligent based approaches, data are pre-processed (scaled) and finally our outputs are post-processed (returned to its original scale). FDEA is used to examine the behavior of gas consumption. To show the applicability and superiority of the ANFIS–FDEA algorithm, actual NG consumption in six Southern America countries from 1980 to 2007 is considered. NG consumption is then forecasted up to 2015. The ANFIS–FDEA algorithm is capable of dealing both complexity and uncertainty as well several other unique features discussed in this paper.

Highlight
► We use pre-processing and post-processing approaches to eliminate possible noise.
► We use ANFIS model based on minimum MAPE.
► We consider standard input variables of long term NG demand estimation.
► Our algorithm provides more accurate solution than previous approach.
► Our approach handles uncertainty, noise, and non-linearity in the given data set.