Day-Ahead Natural Gas Demand Forecasting Using Optimized ABC-Based Neural Network with Sliding Window Technique: The Case Study of Regional Basis in Turkey

Abstract

Abstract:The increase of energy consumption in the world is reflected in the consumption of natural gas. However, this increment requires additional investment. This effect leads imbalances in terms of demand forecasting, such as applying penalties in the case of error rates occurring beyond the acceptable limits. As the forecasting errors increase, penalties increase exponentially. Therefore, the optimal use of natural gas as a scarce resource is important. There are various demand forecast ranges for natural gas and the most difficult range among these demands is the day-ahead forecasting, since it is hard to implement and makes predictions with low error rates. The objective of this study is stabilizing gas tractions on day-ahead demand forecasting using low-consuming subscriber data for minimizing error using univariate artificial bee colony-based artificial neural networks (ANN-ABC). For this purpose, households and low-consuming commercial users' four-year consumption data between the years of 2011-2014 are gathered in daily periods. Previous consumption values are used to forecast day-ahead consumption values with sliding window technique and other independent variables are not taken into account. Dataset is divided into two parts. First, three-year daily consumption values are used with a seven day window for training the networks, while the last year is used for the day-ahead demand forecasting. Results show that ANN-ABC is a strong, stable, and effective method with a low error rate of 14.9 mean absolute percentage error (MAPE) for training utilizing MAPE with a univariate sliding window technique.

Analysis of methods and techniques for prediction of natural gas consumption

“…Authors who used neural networks were [24], [27], [12], [29], [34], [14], [43], [15], [40], [39], and [32]. Viet & Mandziuk [24] presented several neural and fuzzy neural approaches.…”

“…Argentina 1 [25] China 8 [28], [10], [15], [18], [20], [41], [21], [ [33], [12], [34], [14], [29], [36], [37], [17], [30], [31], [32], [ [42].…”

“…Demirel et al [34] predicted daily consumption values in the year 2009 based on historical data, Taspinar et al [29] modeled the daily NG consumption using three different models, and Soldo et al [43] used solar radiation as an input variable in order to predict daily NG consumption. Zhu et al [19] tried to predict natural gas demand in United Kingdom, Bai & Li [41] proposed a structure-calibrated support vector regression approach to forecast the daily NG consumption, Akpinar & Yumusak [31] predicted one-day-ahead NG consumption for the year 2012, as well as Panapakidis & Dagoumas [22], Akpinar et al [32] and Akpinar & Yumusak [38].…”

Developing a Data Mining Based Model to Extract Predictor Factors in Energy Systems: Application of Global Natural Gas Demand

“…Time series [24,25,31,[34][35][36][37][38][39][40][41][42][43] Regression [28,[44][45][46][47] Econometrics [48][49][50][51][52] Expert systems and learning models Artificial neural network (ANN) [21,40,[53][54][55][56][57][58][59][60][61][62][63][64][65][66][67][68][69] Genetic programming (GP) [21,24,40,58,65,67,[69][70][71][72]…”

Artificial bee colony algorithm to forecast natural gas consumption of Turkey

“…They stated that SAGA makes forecasts with less statistical error against linear regression. Akpinar et al [14] proposed an artificial bee colony-based artificial neural networks (ANN-ABC) to forecast day-ahead NGD with a lower statistical error. They stated that ANN-ABC is a strong, stable, and effective method with a low error rate of 14.9 mean absolute percentage error for training utilizing mean absolute percentage error (MAPE) with a univariate sliding window technique.…”