山东农业大学学报（自然科学版）2024，Vol.55Issue(3)：444-452,9.DOI:10.3969/j.issn.1000-2324.2024.03.017
多源异构数据和注意力门控机制的小麦产量预测
Wheat Yield Prediction Based on Multi-Source Heterogeneous Data and Attention Gate Mechanism
摘要
Abstract
To solve the problem of low precision of traditional single-mode data for wheat yield prediction,we proposed a new method combining multi-source heterogeneous data and attention gating mechanism.Firstly,a feature-level gating strategy was introduced to capture the information variation within each modality.Then,a neural network is used to evaluate the confidence scores within each modality and construct a module for obtaining effective information between modalities.Finally,a space and channel attention gating mechanism module based on Transformer is designed to fully integrate effective information between different modes,so as to obtain the best prediction feature representation.The comparative experimental results show that the proposed method has higher prediction accuracy compared to traditional methods,with RMSE and MAE only reaching 809 kg/hm2 and 522 kg/hm2,respectively,and R2 reaching 0.806.The three evaluation indicators obtained by predicting the wheat yield in Henan province over the past 10 years are relatively stable and demonstrate strong robustness.The ablation experiment also verified that different components in our method can effectively improve the prediction accuracy of wheat yield,and can provide strong data support for relevant departments to make decisions to ensure food security management.关键词
小麦产量预测/多源异构数据/注意力机制/门控机制/特征融合Key words
Wheat yield prediction/multi-source heterogeneous data/attention mechanism/gating mechanism/feature fusion分类
计算机与自动化引用本文复制引用
陈书理,张书贵,赵展..多源异构数据和注意力门控机制的小麦产量预测[J].山东农业大学学报（自然科学版）,2024,55(3):444-452,9.基金项目
国家自然科学基金项目(61702185) （61702185）
河南省高等学校重点科研项目计划(24B520025) （24B520025）
