Scholarlink

<p>Supply chain demand forecasting plays an important role in maintaining the normal operation of enterprises. To address the limitations of traditional supply chain demand forecasting algorithms in accuracy and flexibility, this manuscript puts forward a supply chain demand forecasting model based on the Sequence-to-Sequence model and the Convolutional Neural Network. The manuscript combines the Convolutional Neural Network, the Long Short-Term Memory neural network, and the sliding window method for feature extraction and feature construction. The Random Forest algorithm is then employed for feature selection and preliminary forecasting, followed by a Sequence-to-Sequence model—built on Long Short-Term Memory neural networks—for demand forecasting. An attention mechanism is further incorporated to enhance the Sequence-to-Sequence model and improve forecasting accuracy. To comprehensively evaluate the performance of the model, two complementary experimental scenarios were designed: one used historical data from a complete supply chain covering three types of mask products in a certain region for performance verification, and the other selected operational data from agricultural product enterprises with different demand characteristics, such as fruits, vegetables, and eggs, for application effect analysis. The two types of datasets represent typical demand patterns for industrial products and agricultural products, respectively, forming effective complementarity in terms of demand fluctuation characteristics, seasonal sensitivity, and promotion response. In the performance analysis experiment, the median Nash efficiency coefficients of the proposed model reach 0.846 for the training set and 0.852 for the test set. Both values are much higher than those of the comparison models and are closer to 1. These results show that the model delivers better forecasting performance. In the practical application experiment, the proposed model achieves forecasting accuracies of 93.30% (maximum, normal periods), 88.25% (minimum, promotion periods), and 89.05% (minimum, off-season periods). All values are much higher than those of the comparison models and show smaller declines, which indicates that the model produces higher forecasting accuracy and better stability. The average demand variation index for fruits reaches 0.02, and the demand variation index for vegetables and eggs reaches 0.01. All values are much lower than those of the comparison models and show higher forecasting accuracy. The experimental results show that the proposed supply chain demand forecasting model achieves stronger accuracy, steadier behavior, and overall superior forecasting performance compared with the benchmark models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Supply chain demand forecasting method based on Seq2Seq with attention

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月28日

查看全文
  • Xiaoyan Ma,
  • Jian Xue

摘要

Supply chain demand forecasting plays an important role in maintaining the normal operation of enterprises. To address the limitations of traditional supply chain demand forecasting algorithms in accuracy and flexibility, this manuscript puts forward a supply chain demand forecasting model based on the Sequence-to-Sequence model and the Convolutional Neural Network. The manuscript combines the Convolutional Neural Network, the Long Short-Term Memory neural network, and the sliding window method for feature extraction and feature construction. The Random Forest algorithm is then employed for feature selection and preliminary forecasting, followed by a Sequence-to-Sequence model—built on Long Short-Term Memory neural networks—for demand forecasting. An attention mechanism is further incorporated to enhance the Sequence-to-Sequence model and improve forecasting accuracy. To comprehensively evaluate the performance of the model, two complementary experimental scenarios were designed: one used historical data from a complete supply chain covering three types of mask products in a certain region for performance verification, and the other selected operational data from agricultural product enterprises with different demand characteristics, such as fruits, vegetables, and eggs, for application effect analysis. The two types of datasets represent typical demand patterns for industrial products and agricultural products, respectively, forming effective complementarity in terms of demand fluctuation characteristics, seasonal sensitivity, and promotion response. In the performance analysis experiment, the median Nash efficiency coefficients of the proposed model reach 0.846 for the training set and 0.852 for the test set. Both values are much higher than those of the comparison models and are closer to 1. These results show that the model delivers better forecasting performance. In the practical application experiment, the proposed model achieves forecasting accuracies of 93.30% (maximum, normal periods), 88.25% (minimum, promotion periods), and 89.05% (minimum, off-season periods). All values are much higher than those of the comparison models and show smaller declines, which indicates that the model produces higher forecasting accuracy and better stability. The average demand variation index for fruits reaches 0.02, and the demand variation index for vegetables and eggs reaches 0.01. All values are much lower than those of the comparison models and show higher forecasting accuracy. The experimental results show that the proposed supply chain demand forecasting model achieves stronger accuracy, steadier behavior, and overall superior forecasting performance compared with the benchmark models.