Abstract:

This study focuses on the impact of meteorological elements on flue-cured tobacco leaf weight and aims to develop a tobacco leaf weight prediction model based on artificial intelligence algorithms. The research covers data from over 50 tobacco-growing areas in Guizhou Province from 2010-2024, including meteorological and actual tobacco leaf weight data. It analyzes the correlation between meteorological factors and tobacco leaf weight, selects significantly correlated factors, and uses various AI algorithms to build prediction models. NuSVR and SVR algorithms show significant advantages in tobacco leaf weight prediction, with low mean squared error, high stability, and adaptability. Prediction errors vary by leaf position, being lowest in lower leaves, moderate in middle leaves, and highest in upper leaves. From April to August, errors show a downward trend, and during April-September, error fluctuations are small for all three positions, indicating sustained meteorological impacts. During 2020-2024, prediction errors decreased yearly, reflecting model optimization. The study shows that combining meteorological data with AI methods enables reliable June predictions of annual tobacco leaf weight with similar accuracy to September data, offering valuable insights for precision production and smart management in the tobacco industry.

Key words: tobacco leaf weight, meteorological elements, artificial intelligence algorithms, prediction models, NuSVR, SVR