Spare Parts Demand Forecasting Method Based on Intermittent Feature Adaptation.

Area of Science:

• Operations Research
• Supply Chain Management
• Machine Learning

Background:

• Demand for complex equipment aftermarket parts is often sporadic and intermittent.
• This intermittent nature leads to insufficient information in single demand series, hindering traditional prediction methods.
• Existing prediction techniques struggle with the unique characteristics of intermittent demand data.

Purpose of the Study:

• To develop an effective prediction method for intermittent demand in complex equipment aftermarket parts.
• To address the limitations of existing methods in handling sporadic demand patterns.
• To improve the accuracy and stability of demand forecasting in manufacturing after-sales services.

Main Methods:

• A transfer learning approach is proposed to adapt intermittent features.
• An intermittent time series domain partitioning algorithm is developed using demand occurrence and interval information.
• Hierarchical clustering is employed to divide series into sub-source domains.
• A weight vector combines intermittent and temporal characteristics for inter-domain learning.

Main Results:

• The proposed method effectively predicts future demand trends for complex equipment aftermarket parts.
• Experimental results on real-world datasets demonstrate significant improvements in prediction stability.
• Accuracy of demand forecasting is substantially enhanced compared to existing methods.

Conclusions:

• The intermittent feature adaptation method offers a robust solution for sporadic demand forecasting.
• Transfer learning effectively leverages information across different demand series.
• The approach provides a valuable tool for optimizing inventory and maintenance strategies in complex equipment sectors.