Instant delivery time prediction model based on deep learning

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Sept.2024, Vol.47 No.9: 1248-1254, 1274

DOI: 10.3969/j.issn.1003-5060.2024.09.014

Authors

DING Xiang $ ^{1,2} $, NI Liping $ ^{1,2} $, HAN Lu $ ^{1,2} $

(1. School of Management, Hefei University of Technology, Hefei 230009, China; 2. Key Laboratory of Process Optimization and Intelligent Decision Making of Ministry of Education, Hefei University of Technology, Hefei 230009, China)

Abstract and Keywords

Abstract: Based on the goal of achieving higher quality of instant delivery services, a model for instant delivery time prediction is constructed with multi-feature learning. Firstly, the impact of different factors on the instant delivery process is fully considered, and multiple features are represented using geographic hashing and graph embedding to facilitate the input of different models subsequently. Then, the association relationships between multiple features are learned by a combination of multi-head self-attention (MHSA) and residual connections, and the spatial relationships between delivery nodes are extracted using convolutional neural network (CNN), thus achieving sufficient learning and extraction of different features. Finally, the features extracted by different modules are fused and input to the multilayer perceptron module to realize the prediction of instant delivery time. The comparative experiments on real instant delivery dataset show that the proposed prediction model can effectively learn various features and association relationships, and the prediction effect is better.

Keywords: instant delivery; time prediction; multi-head self-attention(MHSA); convolutional neural network(CNN); deep learning

作者信息

丁翔 $ ^{1,2} $，倪丽萍 $ ^{1,2} $，韩露 $ ^{1,2} $

(1. 合肥工业大学管理学院, 安徽合肥 230009; 2. 合肥工业大学过程优化与智能决策教育部重点实验室, 安徽合肥 230009)

摘要和关键词

摘要: 为了提高即时配送服务水平,文章针对即时配送时间预测问题,提出一种分特征学习的预测模型。首先考虑不同因素对配送过程的影响,使用地理哈希、图嵌入等方法对多种特征进行表示;然后使用多头自注意力模型(multi-head self-attention, MHSA)和残差连接组合的方法学习多种特征间存在的关联关系,同时利用卷积神经网络(convolutional neural network, CNN)对配送节点间存在的空间关系进行提取,实现对不同特征的学习;最后将提取的特征进行融合,输入多层感知机实现对配送时间的预测。在真实即时配送数据集上的对比实验表明,该文提出的预测模型能够有效学习各类特征及关联关系,预测效果更优。

关键词: 即时配送；时间预测；多头自注意力模型(MHSA)；卷积神经网络(CNN)；深度学习

Funding

国家自然科学基金资助项目（62276146）

合肥工业大学学报(自然科学版)

Navigation