Adaptive Distributionally Robust Service Composition and Optimal Selection Problem in Cloud Manufacturing

Service composition and optimal selection problem (SCOSP) in cloud manufacturing are crucial tasks. However, due to insufficient historical data or accurate forecasting methods, making unbiased decisions for this problem often faces challenges in addressing uncertainties. In this paper, we address the problem of service composition and optimal selection within the framework of adaptive distributionally robust optimization. In particular, we design an event-dependent ambiguity set associated with manufacturing capability in different events, which combines the 1-Wasserstein metric with the box support set to effectively capture the distributional ambiguous information for each event. To solve SCOSP exactly, we reformulate adaptive distributionally robust SCOSP into the mixed integer programming model. In the end, we conduct a series of numerical experiments to assess the value of incorporating event-dependent distributional information and to evaluate the robustness of the model.