A multi-target regression-based method for multiple orders remaining completion time prediction in discrete manufacturing workshops

Abstract

Accurate prediction of multiple orders remaining completion time (MORCT) is crucial in the make-to-order production model. It enables managers to keep track of production status, make timely decisions, and ensure on-time delivery of orders. However, dynamic production environments, characterized by constantly changing order quantities and relationships, as well as the special temporal features of the production process, pose challenges to existing prediction methods. To address these issues, this paper proposes a novel framework based on multi-target regression. First, production data are collected and standardized from various sources using multiple data transfer protocols. The input dataset is then constructed and dynamically adjusted to accommodate changes in order quantities and priorities. Finally, a prediction model named DMTR-LSA is developed to effectively handle the specific temporal relationships in the production data by integrating long short-term memory (LSTM) and self-attention mechanisms. A case study in a real production workshop demonstrates that the proposed method supports simultaneous prediction of multiple orders. It outperforms existing methods on several evaluation metrics, reducing the average prediction error by more than 8.9%. These results highlight the practical value of the proposed method for predicting MORCT in dynamic production environments and its potential impact to enhance the production decision-making process.