Comprehensive quality grading and dynamic prediction of physicochemical indicators of maize during storage based on clustering and time-series prediction models

Abstract

Maize is a vital cereal crop, and its quality is influenced by various environmental factors during storage. To achieve scientific management of maize quality and maintain optimal conditions, this study examined 26 sets of maize samples stored under different environmental conditions (temperature, humidity, and initial moisture). Six key physicochemical indicators—bulk density, fat acidity, germination rate, malondialdehyde (MDA), protein, and starch—were selected to represent maize quality. Based on these indicators, a clustering algorithm was employed to grade maize quality and develop a comprehensive classification standard tailored for high-quality storage outcomes. On this basis, time-series prediction models for quality indicators were developed, utilizing 360 days of time-series data to dynamically predict indicator values. Comprehensive quality levels were calculated to evaluate storage performance. The results demonstrated that the K-Means++ algorithm achieved the best clustering performance with a Silhouette Coefficient of 0.7446 and a Davies-Bouldin index of 0.4336, effectively distinguishing maize quality variation across different storage environments. Moreover, the RNN-based time-series prediction model yielded a Mean Absolute Percentage Error (MAPE) as low as 1.28 % under a 21-day time window, confirming its high accuracy and suitability for practical storage management. This research provides scientific guidance for maize storage management, offering technical support for reducing quality loss during storage and extending the storage period of high-quality maize, with significant practical implications.