Research on Prediction Model and Optimization of Enterprise Material Procurement Management Based on Global Linkage

Abstract Against the backdrop of increasing uncertainty in the global supply chain and simultaneous rise in procurement costs and delivery risks for enterprises, the traditional material procurement model is facing severe challenges due to information silos between departments, deviation in demand forecasting, and dynamic changes in supplier relationships; especially in the semiconductor manufacturing industry, high capital investment, long lead times, and high demand volatility can lead to significant capacity losses due to procurement decision-making errors. Therefore, this article constructs an enterprise material procurement management optimization framework based on a global linkage mechanism, aiming to use intelligent means to connect procurement, production, and warehousing data, and achieve multiple goals of cost reduction, quality assurance, and stable supply. The framework includes three technological innovations: ① integrating a Bayesian convolutional neural network and multi-head attention mechanism, using device image features and sensor timing data to predict material remaining life, and reducing the root mean square error of semiconductor key consumables demand prediction to 2.12 × 10⁶ (23.7% lower than LSTM), p < 0.01); ② designing a multi-objective optimization engine that combines the NSGA-II algorithm with a supplier maturity evaluation system that includes six indicators such as timely delivery rate and defect feedback rate to achieve Pareto equilibrium of procurement cost, quality defect rate, and delivery delay rate; the experiment shows that the total procurement cost is reduced by 17.4%, and the supplier complaint rate is reduced to 3.2%; ③ proposing a Global Linkage Rule Set (GLR), which coordinates the data flow of production, warehousing, and procurement through dynamic weights. In the empirical study of semiconductor manufacturing enterprises, it reduces emergency procurement frequency by 42% and increases inventory turnover by 29%. This study provides a reusable methodological framework and validated technical cases for the intelligent transformation of supply chains in capital-intensive industries such as semiconductor manufacturing.