Robust Interval Prediction of Intermittent Demand for Spare Parts Based on Tensor Optimization.

Area of Science:

• Operations Research
• Data Science
• Manufacturing Engineering

Background:

• Aftermarket services for large manufacturing enterprises rely on accurate spare parts demand prediction for inventory and quality management.
• Intermittent spare parts demand exhibits random fluctuations and outliers, challenging traditional time series forecasting methods.
• Existing methods struggle to capture evolutionary patterns and provide reliable predictions for noisy, intermittent data.

Purpose of the Study:

• To propose a robust interval prediction method for intermittent time series of aftersales spare parts demand.
• To address the challenges of random fluctuations, outliers, and intermittent characteristics in demand data.
• To enhance the reliability and accuracy of spare parts demand forecasting for aftermarket services.

Main Methods:

• A sequence-smoothing network utilizing tensor decomposition (Tucker decomposition) and a stacked autoencoder to denoise demand data.
• An alternating optimization algorithm to extract evolutionary trends from intermittent series and optimize feature representations.
• An adaptive interval prediction algorithm with dynamic updates for point and interval forecasting.

Main Results:

• The proposed tensor optimization method effectively captures the evolutionary trends of intermittent series, outperforming traditional methods.
• Improved prediction accuracy, especially for small-sample intermittent series, was demonstrated using real-world aftersales data.
• The method provides reliable, elastic prediction intervals, mitigating issues caused by data distortion.

Conclusions:

• The tensor optimization-based robust interval prediction method offers a novel solution for accurate and reliable forecasting of intermittent spare parts demand.
• This approach enhances intelligent planning and decision-making in practical maintenance and aftermarket services.
• The method's ability to handle noisy and intermittent data provides a significant advancement in demand forecasting.