A lightweight model based on improved PSO-SVR for predicting tobacco paper pot seedling growth

This study aimed to develop an improved PSO-SVR model for the growth of tobacco paper pot seedlings under small-sample conditions. The model integrated environmental factors and substrate formulations to explore the nonlinear relationship between the environment and seedling growth. Two independent seedling cultivation experiments were conducted, yielding 480 observational records nested within 24 independent experimental plots and four growth stages. These data included key indicators such as root number, plant height, shoot dry weight, and chlorophyll a content. The experiment used three different plant ash ratios (20%, 40%, and 60%) and a control group. One-way analysis of variance (ANOVA) was used to identify sensitivity differences among the indicators. An improved PSO-SVR model was established. This model used adaptive inertia weight, dynamic learning factors, and a hybrid kernel function to improve parameter optimization and the model’s ability to fit nonlinear data. The results showed that the PSO-SVR model performed best on the independent validation set. The R 2 values for root number, plant height, shoot dry weight, and chlorophyll a content were 0.923, 0.976, 0.958, and 0.919, respectively. Its prediction accuracy was much higher than that of MLR, Lasso, kNN, RF, GBM, SVR, MLP, and GPR models. The experiments also showed that a proper amount of plant ash (20%) could improve root activity and shoot growth, while a high amount (60%) inhibited early growth. This study provides a high-accuracy prediction framework for small-sample seedling production. It offers a reliable theoretical basis and technical guidance for substrate optimization, seedling quality assessment, and managed nursery cultivation.