Sürdürülebilir Kalkınma Amaçları
Yazarlar
Yayınlar
- MakaleAçık Erişim2025Scopus SJRQ2Forecasting daily natural gas consumption with regression, time series and machine learning based methods
Yucesan, Melih; Pekel, Engin; Celik, Erkan; Gul, Muhammet; Serin, Faruk
An effective short-term natural gas forecasting method contributes to social contributions and allows industrial chain elements to function effectively and minimize economic losses. We dealt with a comparative framework on the applicability of different methods in daily natural gas service (NGS) consumption forecasting. In this context, time series, machine learning, evolutionary and population-based approaches, and their hybrid versions are applied to the NGS data. Hybridized approaches are tested in the scope of NGS consumption forecasting for the first time in the literature in this study. The case of Turkey is handled, and its NGS data is used to demonstrate the comparative framework’s applicability. The comparative study is assessed in the lights of common forecasting accuracy measures of mean absolute percentage error (MAPE), R-squared (R2), and mean squared error (MSE). According to each method’s results, the seasonal autoregressive integrated moving average with exogenous regressors (SARIMAX) and artificial neural network (ANN) hybrid model provides the most dominant performance with respect to MAPE. The lowest error was obtained with a MAPE value of 0.357 in this hybrid model constructed under seven neurons in its ANN structure. This model is followed by another hybrid model, autoregressive integrated moving average (ARIMA)-ANN, with a MAPE value of 0.5 under nine neurons in terms of accuracy performance. The worst performance value belongs to the Genetic algorithm-ANN hybrid model with a MAPE value of approximately 26%.
AtıfScopus16WoS19OpenAlex22
İncele - MakaleYalnız künye2025Scopus SJRQ1Seasonally adjusted periodic time series for Mann-Kendall trend test
Güçlü, Yavuz Selim; Acar, Ramazan; Saplıoğlu, Kemal
In this study, temperature data trend analysis is conducted, which is thought to be influenced by global warming, climate change, and local factor impacts. The main objective of this paper is to obtain an acceptable autoregressive correlation value for the MK test. For this purpose, seasonality (periodicity) especially in monthly time series is adjusted. Autoregressive correlation and homogeneity test values decrease after seasonal adjusting, but reasonable results are not achieved. Then, prewhitening procedure is also applied to the seasonally adjusted data. This process resulted in the data becoming both homogeneous and free autoregressive correlation. The final version of the time series is suitable for the MK test. Furthermore, the time series are divided into different time intervals, and the efficacy of the method is investigated. The results demonstrated that the method is suitable for time series with less than 30 years of data. The study demonstrated that the proposed method enhances the reliability of the data. Also, multiplication by 12 (months) allows the MK test with Z score in place of the Student's t-test in short-term data sets. This suggested methodology can be used to identify MK trend conditions in monthly time series. The application is based on monthly and annual average temperature data between 1957 and 2022 from three stations within the Kızılırmak basin (Çankırı, Kırşehir, Sivas stations) and one station within Seyhan Basin (Adana station) in Türkiye. The test results exhibited a significant increasing trend.
AtıfScopus37WoS34OpenAlex41İncele