融合注意力机制与卡尔曼滤波的LSTM模型在农村网络零售额预测中的应用  

Application of LSTM Model Integrated with Attention Mechanism and Kalman Filter in Rural Online Retail Sales Prediction

作　　者：丁汉韬 宋瑾钰[1] 

机构地区：[1]浙江理工大学计算机科学与技术学院(人工智能学院),浙江 杭州

出　　处：《电子商务评论》2025年第3期1140-1150,共11页E-Commerce Letters

摘　　要：在信息技术不断进步和农村基础设施日益完善的背景下,电子商务已逐步渗透到农村市场,并成为推动农业现代化的重要动力。然而,农村电商的发展仍面临诸多挑战,如商品质量参差不齐、专业人才缺乏、物流效率低下等问题。为了更准确地预测农村电商的市场趋势,为电商企业和政策制定者提供科学的数据支持,本研究提出了一种基于长短时记忆网络(Long-Short Term Memory, LSTM)的时间序列预测模型,并结合注意力机制(Attention Mechanism)和卡尔曼滤波(Kalman filter)技术,对2024~2026年农村网络零售额进行预测。实验结果表明,引入注意力机制和卡尔曼滤波后,模型的预测精度显著提升,均方误差(mean-square error, MSE)、均方根误差(root-mean-square error, RMSE)和平均绝对误差(Mean absolute error, MAE)均有所降低。研究结果为农村电商的市场趋势分析和政策制定提供了科学依据,具有重要的现实意义。With continuous advancements in information technology and the gradual improvement of rural infrastructure, e-commerce has increasingly penetrated rural markets, becoming a key driver of agricultural modernization. However, its development still faces numerous challenges, such as inconsistent product quality, a shortage of skilled professionals, and low logistics efficiency. To more accurately predict market trends in rural e-commerce and provide scientific data support for e-commerce enterprises and policymakers, this study proposes a time series prediction model based on Long Short-Term Memory (LSTM) networks, incorporating an attention mechanism and Kalman filter techniques to forecast rural online retail sales from 2024 to 2026. The experimental results indicate that the introduction of the attention mechanism and Kalman filter significantly improves the model’s prediction accuracy, reducing the mean square error (MSE), root mean square error (RMSE), and mean absolute error (MAE). The findings provide a scientific

关 键 词：农村电商 时间序列预测 LSTM 注意力机制 卡尔曼滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]