Learning to Estimate Package Delivery Time in Mixed Imbalanced Delivery and Pickup Logistics Services

Abstract: Accurately estimating package delivery time is essential to the logistics industry, which enables reasonable work allocation and on-time service guarantee. This becomes even more necessary in mixed logistics scenarios where couriers handle a high volume of delivery and a smaller number of pickup simultaneously. However, most of the related works treat the pickup and delivery patterns on couriers' decision behavior equally, neglecting that the pickup has a greater impact on couriers' decision-making compared to the delivery due to its tighter time constraints. In such context, we have three main challenges: 1) multiple spatiotemporal factors are intricately interconnected, significantly affecting couriers' delivery behavior; 2) pickups have stricter time requirements but are limited in number, making it challenging to model their effects on couriers' delivery process; 3) couriers' spatial mobility patterns are critical determinants of their delivery behavior, but have been insufficiently explored. To deal with these, we propose TransPDT, a Transformer-based multi-task package delivery time prediction model. We first employ the Transformer encoder architecture to capture the spatio-temporal dependencies of couriers' historical travel routes and pending package sets. Then we design the pattern memory to learn the patterns of pickup in the imbalanced dataset via attention mechanism. We also set the route prediction as an auxiliary task of delivery time prediction, and incorporate the prior courier spatial movement regularities in prediction. Extensive experiments on real industry-scale datasets demonstrate the superiority of our method. A system based on TransPDT is deployed internally in JD Logistics to track more than 2000 couriers handling hundreds of thousands of packages per day in Beijing.

Analysis of TransPDT: Transformer-Based Prediction Framework for Mixed Logistics Services

The paper introduces TransPDT, a transformative model tailored for estimating package delivery times in logistics scenarios characterized by a substantial imbalance between delivery and pickup tasks. In this context, predicting delivery times accurately is crucial due to the demanding constraints and logistical complexities couriers face, balancing a high volume of deliveries with a limited number of pickups. This model leverages several advanced techniques, including Transformer architectures and multi-task learning frameworks, to address these specific challenges.

Core Challenges and Model Architecture

Key challenges addressed by TransPDT include:

1. Complex Spatio-Temporal Interdependencies: The interwoven nature of spatio-temporal factors—such as location proximity and remaining time constraints—affect courier routing decisions significantly.
2. Limited Pickup Impact Modeling: With fewer but more time-sensitive pickup tasks, capturing their intermittent impact on delivery routes is vital but challenging due to limited data frequency.
3. Courier Mobility Patterns: Existing approaches rarely explore couriers' spatial movement regularities, even though these are crucial to understanding delivery behaviors.

TransPDT introduces a Transformer-based approach to capture the intricate spatio-temporal dependencies within couriers' routes. Specifically, the model comprises three main components:

• Spatio-Temporal Correlation Extraction: Utilizes dual Transformer branches to separately model historical package routes and future pending package sets. This separation allows the model to focus on distinct patterns inherent in both completed and awaited tasks.
• Pickup Influencing Pattern Learning: Implements a novel pattern memory via attention mechanisms to address imbalance and learn minor pickup influence patterns with limited samples. This module excels in emphasizing and storing pickup impact profiles which are sporadic yet significant.
• Multi-task Prediction Leveraging Spatial Mobility: Incorporates LSTM layers for dynamic route prediction and integrates pre-existing courier movement regularities into task assessments. This hybrid layer synergizes database-recorded mobility knowledge with machine-generated predictions for enhanced time estimation precision.

Performance and Evaluation

The framework's effectiveness is demonstrated against a suite of baseline models, showing substantial improvement in key predictive metrics:

• RMSE and MAPE indicate decreased prediction errors, confirming TransPDT's ability to more accurately gauge delivery time fluctuations compared to traditional statistical and machine learning models.
• Enhanced Hit-Rate@3 and Location Mean Deviation metrics illustrate that route predictions align closely with actual courier decisions, a testament to the model's robust handling of complex routing environments.

Practical Implications and Deployment

The deployment of TransPDT into JD Logistics' operational framework highlights its operational value. Through real-world application, the system delivers actionable insights for logistics station managers, crucial for task allocation and ensuring timely service. It improves delivery punctuality by approximately 0.68%, which translates into substantial financial savings and customer satisfaction levels across an expansive courier network.

Prospects for Future Developments

Going forward, TransPDT paves the way for innovations in logistics AI, where enriched environmental data sampling and broader user-feedback loops might further advance predictive accuracy. Potential improvements include fine-tuning attention mechanisms within the transformer to better account for predominant influencing factors, such as highly dynamic urban traffic conditions.

In conclusion, the TransPDT model not only provides theoretical advancements but also exemplifies practical enhancements in logistics through the nuanced application of AI methodologies in multifaceted delivery contexts. This paper substantiates a scalable model adaptable across varied logistical paradigms, setting a precedence for future research exploring AI's transformative role in hybrid operational scenarios.