Prediction of Acid and Caustic Consumption in the CIP System of the Beverage Industry Based on Ensemble Learning

This paper focuses on predicting the acid and caustic consumption in the clear-in-place system during beverage production. By tracking and collecting daily data from various units, a foundational dataset was built for modeling and evaluation. First, factory data was preprocessed to generate a dataset for training and prediction. Through in-depth analysis of the daily acid and caustic consumption characteristics, multiple regression models, including multiple linear regression, Bayesian ridge regression, Gaussian process regression, and support vector machine regression, were comprehensively evaluated. Based on these models, an ensemble learning model was developed, and genetic algorithms were used to select the optimal training dataset for the base models. Experimental results show that the proposed method significantly reduces the mean square error and improves prediction accuracy in both static and multi-day rolling predictions compared to conventional regression models.