Comparative study of machine learning models and time series models for predicting natural gas consumption in industries

Accurate prediction of natural gas consumption plays an important role in energy planning, resource management, and formulating industrial development policies. In this study, monthly data on natural gas consumption in the province's industry during the period 1394 to 1403 were examined. In order to evaluate and compare the performance of different prediction methods, three random forest models, autoregressive integrated moving average (ARIMA) model, and nearest neighbors were implemented in the MATLAB software. Also, in order to examine the effect of data clustering on prediction accuracy, the models were developed and evaluated in two independent scenarios, including the use of fuzzy C-Means clustering and the lack of use of FCM. In the data preparation stage, lagged variables, temporal features, and seasonal components were extracted and used in the modeling process.The results showed that the use of fuzzy clustering improved the performance of forecasting models and increased their ability to identify consumption patterns. Comparison of evaluation indicators indicated that the random forest model had the lowest error rate and the highest accuracy in both scenarios and was more capable of modeling complex and nonlinear relationships in natural gas consumption. The ARIMA model ranked second and showed good performance in identifying seasonal trends and patterns, while the KNN model provided less accuracy in forecasting than the other two models. The forecast results up to the horizon of 1410 also indicate a continuation of the increasing trend in natural gas consumption and increased demand in the cold months of the year.The findings of this study show that combining fuzzy clustering with machine learning methods can improve the accuracy of energy consumption forecasting and provide an effective tool for long-term planning, demand management, and decision-making in the energy sector.