Prediction of Tobacco Moisture Content Based on Improved Support Vector Regression Model

Drying plays a pivotal role in the production of tobacco. Unlike many other drying processes, tobacco drying exhibits significant nonlinearity and is influenced by numerous factors, rendering accurate predictions challenging. In light of these characteristics, in order to improve the accuracy of tobacco moisture content prediction, in this study, an improved support vector machine regression model was used to simulate the complex dynamic process in the tobacco drying process. Addressing the inherent limitations of differential particle swarm optimization (DPSO), we introduce the chemotaxis operation of bacterial compartmentalization algorithm to guide particles to move to optimal positions during searching. Additionally, we propose particle rotation to improve diversity of search space by changing particle position to facilitate their escape from local optimization and avoid premature convergence of particles. Ultimately, the SVR-QBDPSO model is established by integrating support vector regression (SVR) and Quaternion bacteria-specific chemotaxis differential particle swarm optimization (QBDPSO). The experimental results demonstrate that this model outperforms other models when applied to the production data of tobacco.