Challenge
Controlling the amount of methanol injected into the TAME process is a critical step in the production of gasoline additives. Traditional methods require engineers to manually collect samples and analyze them offline, a process that can take several hours. This delay makes real-time adjustments difficult and limits the ability to maintain stable product quality. To overcome these challenges, a predictive model is needed to estimate methanol residuals accurately and allow for real-time dosage adjustments.
Approach
Optimal time intervals were selected using statistical tests to reflect time-series characteristics, and datasets with similar distributions were separated to ensure model reliability. Process-specific features such as temperature ratios and differences were engineered to capture equipment operation patterns across production lines. Interpolation, resampling, and timelag processing were applied to analyze the correlation between methanol residuals and facility temperature, improving model stability and accuracy.
Value Delivered
The model enabled real-time prediction of methanol residuals using only plant sensor data, reducing offline sampling and analysis time while improving engineering efficiency. Data-driven analysis clarified the relationship between methanol residuals and facility temperature, complementing engineers’ empirical judgment and providing a reliable basis for decision-making to support process optimization.
