Research on the Liquor Yield Prediction Model Integrating Hybrid Kernel Support Vector Regression and Dung Beetle Optimizer

To enhance the efficiency of liquor production and reduce subjective errors caused by manual operations, this study explores the relationship between production process parameters and yield to achieve accurate yield prediction. Based on actual liquor production data, a Multi Kernel Support Vector Regression (MKSVR) prediction model is established using the Stacking model fusion method to improve the model’s generalization ability. Additionally, the Dung Beetle Optimizer (DBO), which has stronger global optimization capabilities, is used for parameter optimization to further enhance model performance. The results show that the root mean square error (RMSE) and coefficient of determination for the multiple linear regression model were 0.1295 and 0.6987, respectively; for the DBO-optimized radial basis function, the RMSE and coefficient of determination were 0.0824 and 0.7268. The swarm intelligence optimization model exhibits smaller prediction errors than traditional machine learning methods. The RMSE and coefficient of determination for the DBO-MKSVR model were 0.0738 and 0.7680, respectively. Compared to single-kernel function prediction models, the fusion model demonstrates higher data adaptability and optimal prediction performance, enabling accurate forecasting of liquor yield. This approach is significant for the research and development of automated liquor production technologies.