关键词: 稻瘟病, 发生面积, 全子集回归, BP神经网络算法, 预测模型, 构建, 信阳市

Abstract: Using meteorological data from the Xinyang City between 2004 and 2021 (excluding 2020), including air temperature, relative humidity, precipitation, and sunshine duration, five key factors influencing rice blast epidemics were identified through correlation analysis. These factors were:The minimum relative humidity in late June, the minimum relative humidity in early May, the minimum temperature in early May, the sunshine duration in mid-June, and the cumulative rainfall in early August. Both all-subset regression and BP neural network algorithms were employed to predict the incidence area of rice blast in the Xinyang City. The results showed that all-subset regression model 1 and model 2 achieved back-testing accuracies of 92.49% and 94.43%, respectively, for the 2004—; 2021 rice blast incidence area, and both yielded a prediction accuracy of 79.68% for the years 2022 and 2023. In comparison, BP neural network models 1 and model 2 achieved back-testing accuracies of 82.72% and 83.55%, respectively, for the 2004—; 2021 period, and prediction accuracies of 98.06% and 95.49% for 2022 and 2023. Based on these results, BP neural network model 1 was identified as the optimal prediction model. Using this model, the predicted incidence area of rice blast in Xinyang City for 2024 was 26 500 hectares(hm²).

Key words: rice blast, incidence area, all-subset regression, BP neural network algorithm, prediction model, construction, Xinyang City