粮食安全是“国之大者”，是国家安全的基础，是经济安全的底线。小麦是我国主要的粮食品种，面粉作为小麦的主要加工产品，其价格波动与小麦市场密切相关，是反映粮食市场供需变化的重要指标。准确预测面粉价格对稳定消费市场、保障国家粮食安全具有重要意义。本文基于农业农村部重点农产品市场信息平台中2023年11月至2025年11月的日度面粉价格数据，系统构建并比较了ARIMA、GM(1,1)、LSTM及Transformer四种时间序列预测模型。通过序列分析揭示价格波动的平稳性及非线性等复杂特征，为后续差异化模型的选择与构建提供依据。再基于误差倒数平方的加权融合策略，构建Ensemble融合模型。实证结果表明，单一模型在预测性能上各具优势：ARIMA与GM(1,1)在刻画整体趋势上表现稳健，而LSTM与Transformer在捕捉非线性波动方面作用显著。而集成各模型优点、弥补单一方法局限构建的Ensemble融合模型，综合评估平均绝对误差（MAE）、均方根误差（RMSE）和平均绝对百分比误差（MAPE）等关键指标均表现优异，其表现显著优于任一单一模型，展现出更高的预测精度与稳定性。多模型融合策略在面粉价格预测方面具有显著有效性与实用价值，可用于粮食及其加工品市场的价格预测研究。

Food security is a fundamental priority for the nation, the cornerstone of national security, and the baseline for economic security. Wheat is a primary grain crop in China, and flour, as its main processed product, has price fluctuations closely linked to the wheat market, serving as a crucial indicator reflecting supply and demand changes in the grain market. Accurately predicting flour prices is of great significance for stabilizing the consumer market and ensuring national food security. This paper is based on the daily flour price data from the Key Agricultural Product Market Information Platform of the Ministry of Agriculture and Rural Affairs from November 2023 to November 2025, systematically constructing and comparing four time series forecasting models: ARIMA, GM(1,1), LSTM, and Transformer. Sequence analysis reveals complex characteristics of price fluctuations such as stationarity and non-linearity, providing a basis for the subsequent selection and construction of differentiated models. Then, based on a weighted fusion strategy using the reciprocal of squared errors, an Ensemble Integration Model is constructed. Empirical results indicate that individual models each have their own advantages in predictive performance: ARIMA and GM(1,1) perform robustly in depicting overall trends, while LSTM and Transformer play a significant role in capturing non-linear fluctuations. The Ensemble Integration Model, which integrates the advantages of each model and compensates for the limitations of single methods, performs excellently in comprehensive evaluations of key metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Its performance is significantly superior to any single model, demonstrating higher prediction accuracy and stability. The multi-model fusion strategy has significant effectiveness and practical value in flour price forecasting and can be applied to price prediction research for grain and its processed product markets.