Spare parts consumption prediction model for improving maintenance and operational reliability

In fields such as industrial production, the reliability of mechanical equipment maintenance is affected by the inventory management of maintenance spare parts. Accurately predicting the consumption of maintenance spare parts is of great significance for optimizing resource allocation and formulating scientific maintenance strategies. This paper proposes a hybrid prediction model integrating STL decomposition, iTransformer and TimesNet. This model combines time series decomposition technology with deep learning frameworks and is capable of efficiently handling long-term series data and their periodic characteristics. Through an empirical analysis of the historical data of a certain maintenance spare part in a warehouse over a period of 10 years, the results show that this hybrid model significantly outperforms multiple benchmark models in multiple key performance indicators. The precise prediction ability of this method helps to enhance the scientific nature of spare parts management and contributes new methods and ideas to predictive maintenance and spare parts support work.