Optimizing Natural Gas Consumption in Fertilizer Manufacturing: A Predictive Model Using Random Forest and Linear Regression

This study focuses on developing a predictive model to forecast natural gas (NG) consumption for a fertilizer manufacturing plant with three distinct production plants: A, B, and C. To improve operating efficiency and save expenses resulting from excessive natural gas procurement and underuse, the goal was to attain a minimum prediction accuracy of 90%. Two machine learning models, Random Forest (RF) and Linear Regression (LR), were developed and evaluated based on regression metrics such as R2 and Mean Absolute Error (MAE). The Random Forest model outperformed Linear Regression, with R2 values of 90.56%, 98.11%, and 94.52% for Plants A, B, and C, respectively. The study demonstrates that Random Forest is an effective tool for predicting NG consumption, though some discrepancies in peak predictions indicate the need for further model refinement. Overall, the implementation of this model has the potential to significantly optimize energy use and reduce costs in the fertilizer manufacturing process.