Abstract
Wheat is a vital staple food crop, and its post-harvest storage is paramount to maintaining its quality. However, conventional grain storage methods frequently impede the ability to promptly and accurately predict and assess quality changes. Moreover, most storage systems are ineffective in dealing with the impact of temperature and humidity fluctuations on wheat quality, which can potentially lead to quality degradation during storage. To address these challenges, this paper proposes a dual model system of "prediction-evaluation", which integrates a dynamic quality prediction model based on SGCNiFormer with an evaluation framework based on K-Smeans clustering to establish a closed-loop mechanism from quality prediction to storage effect evaluation. The system incorporates a graph convolutional network (GCN) and a dynamic gating module, enabling precise simulation of the multidimensional evolution of wheat quality under the interaction of moisture and temperature. The experimental results demonstrate the superiority of SGCNiFormer in time-series prediction tasks, while the K-Smeans method establishes a wheat quality grading standard with physical interpretability. This integrated method provides a systematic theoretical framework for optimizing storage parameters and offers substantial support for intelligent grain storage management.