Research on Rice Safety Risk Prediction Model Under High Temperature and Humidity Environments

ABSTRACT Under high temperature and humidity environments, the pollution characteristics of rice exhibit instability and high complexity, which makes traditional prediction models face challenges of insufficient stability and robustness in feature screening and outlier processing. Therefore, this paper proposes a rice safety risk prediction model of OREDRVFL (Outlier‐robust Ensemble Deep Random Vector Functional Link Network) improved by the FAOA (Fitness‐Distance‐Balance‐Arithmetic Optimization Algorithm). First, the MI (Mutual Information) method is employed to screen a series of key risk factors. Second, the AOA (Arithmetic Optimization Algorithm) is improved by the FDB (Fitness‐Distance‐Balance) strategy. Then, regularization constraints and sparsity modeling are used to construct the OREDRVFL network, and FAOA is employed to optimize its number of hidden layers and regularization parameters. Finally, experimental verification is carried out using the detection data of a major rice‐producing province from 2022 to 2023. The results show that the improved MI‐FAOA‐OREDRVFL model significantly outperforms traditional models in terms of indicators such as root mean square error (RMSE = 0.4300), mean absolute error (MAE = 0.2900), and correlation coefficient ( R 2 = 0.8800). Under noise interference, its RMSE remains stable at 0.1700, verifying the high precision and strong robustness of the model, and providing technical support for the prediction of rice safety risks under high temperature and humidity environments.