A stacking denoising auto-encoder with sample weight approach for order remaining completion time prediction in complex discrete manufacturing workshop

Accurate order remaining completion time (ORCT) prediction provides an essential criterion for dynamically triggering the adjustment of production plans and establishment of dispatching strategies, which helps to improve the plan rationality and production efficiency, thus guaranteeing delivery orders on time. With the extensive deployment of Industrial Internet of Things in the workshop, the data for ORCT prediction is perceived in real-time. However, data quality, knowledge inconsistency, and data distribution variation make ORCT prediction more difficult. Hence, a stacking denoising auto-encoder with sample weight (SW-SDAE) method is proposed to improve the robustness and applicability of ORCT prediction. Firstly, a four-layer SDAEs is constructed to extract high-level and robust features. Secondly, a dynamic updating method of sample weight for regression prediction is designed to guide the training of prediction model parameters and improve the prediction accuracy. Thirdly, model-based transfer learning is employed to adapt to the data distribution change over time and ensure the prediction applicability. Finally, different prediction models are applied to an actual case for comparison. The experimental results show that the proposed prediction method is effective for ORCT prediction and superior to other methods.