Abstract:

The yield of ramie has a high correlation with the climate factors during the growth period， and the final yield can be effectively and accurately predicted by constructing a ramie yield prediction model based on climate variables. The BP neural network has strong data analysis capabilities and is widely used in crop yield prediction modeling. However， traditional BP neural networks have problems such as low accuracy and poor robustness. The sparrow search algorithm （SSA） can be used to optimize the BP neural network model. Based on the fiber yield， fresh skin yield and climate data collected in ramie long-term positioning experiment from 2010 to 2019， this study analyzed the changing trend of climate factors in 10 years and their impacts on the perennial ramie yield by comparing the performance of BP neural network model and the optimized SSA-BP neural network model in predicting ramie yield to determine the best prediction model. It showed that there were extremely significant correlations between the yield of ramie and 4 meteorological factors including the seasonal average temperature， the average seasonal extreme maximum temperature， the seasonal extreme minimum temperature， and the seasonal average sunshine hours， among which the ramie yield had the highest correlation with the seasonal average temperature. SSA algorithm could effectively optimize the BP neural network. R² of the ramie fiber yield prediction model and fresh skin yield prediction model based on SSA-BP were 0.591 3 and 0.679 1， respectively， which were higher than that of the ramie fiber yield prediction model （R²=0.405 7） and fresh skin yield prediction model （R²=0.551 8）. Therefore， the SSA-BP model could predict ramie yield more scientifically and reasonably， which was of great guiding significance for field management and overall plan of ramie production.

Key words: yield forecast, climatic factor, sparrow search algorithm, BP neural network

摘要：

苎麻产量与生长期间的气候因子具有极高相关性，基于气候变量构建的苎麻产量预测模型能够有效精准预测最终产量。BP（back propagation）神经网络具有强大的数据分析能力，在作物产量预测建模中得到广泛应用，然而传统BP神经网络存在精度低、鲁棒性差等问题，可采用麻雀搜索算法（sparrow search algorithm，SSA）对BP神经网络模型进行优化。基于2010—2019年苎麻长期定位试验采集的纤维产量、鲜皮产量和气候数据，分析气候因子在10年内的变化趋势及其对多年生苎麻产量的影响，对比构建的BP神经网络模型及优化后的SSA-BP神经网络模型预测苎麻产量的性能，确定最佳的苎麻产量预测模型。结果表明，苎麻产量与季平均气温、季极端最高气温均值、季极端最低气温均值、季日照时数均值4项气候因子具有极显著相关关系。SSA算法能有效优化BP神经网络，基于SSA-BP的苎麻纤维产量预测模型和鲜皮产量预测模型的R²分别为0.591 3和0.679 1，高于BP神经网络的苎麻纤维产量预测模型（R²=0.405 7）和鲜皮产量预测模型（R²=0.551 8）。因此，SSA-BP模型能够更加科学、合理地预测苎麻产量，对于苎麻生产的田间管理及统筹规划具有重要指导意义。

关键词: 产量预测, 气候因子, 麻雀搜索算法, BP神经网络