榆林学院学报

针对跨境农产品物流行业样本数据少，影响因素多等情况，提出一种基于GRA-PSO-LSTM的跨境物流需求预测模型，以解决传统预测方法在非线性、非平稳性处理中的局限性。基于农产品跨境物流需求特征，选取农产品供给、社会经济发展水平、物流基础设施与服务水平、市场规模与消费能力四个维度共17个典型影响因素指标，据各指标与物流需求的灰色关联分析构建影响因素指标体系；针对2004-2023年我国农产品跨境物流数据，利用PSO算法优化LSTM网络进行预测。实证结果表明：相较于传统模型，PSO-LSTM模型的MSE、MAE、RMSE分别降低至0.0025、0.0369和0.0501，验证了模型在跨境农产品物流需求预测中的有效性，利用已训练模型对未来五年物流需求进行预测，对优化跨境农产品供应链资源配置，提升国际贸易竞争力具有重要意义。

To address the challenges of limited sample data and multiple influencing factors in cross-border agricultural product logistics, this study proposes a GRA-PSO-LSTM-based prediction model to overcome the limitations of traditional methods in handling nonlinear and non-stationary data. Based on the characteristics of cross-border agricultural logistics demand, the research selects 17 typical influencing factors across four dimensions: agricultural supply, socio-economic development level, logistics infrastructure and service quality, market scale, and consumption capacity. A grey relational analysis is conducted to establish an influencing factor index system. Using 2004-2023 cross-border agricultural logistics data from China, the PSO algorithm optimizes the LSTM network for prediction. Empirical results demonstrate that compared to traditional models, the PSOLSTM model reduces MSE, MAE, and RMSE to 0. 002 5, 0. 036 9, and 0. 050 1 respectively, validating its effectiveness in predicting cross-border agricultural logistics demand. The trained model's application to forecast logistics demand for the next five years holds significant implications for optimizing cross-border agricultural supply chain resource allocation and enhancing international trade competitiveness.