An Enhanced Distributed Optimization Subject to Multiple Constraints: A Case Study on Entire Ethylene Separation Process

Abstract

This article studies an enhanced distributed optimization algorithm in an undirected topology based on local communication and computation to optimize the sum of local objective functions under multiple constraints. In particular, the adapt-then-combine distributed inexact gradient tracking algorithm (ATC-DIGing-MC) is developed for smooth and strongly convex local functions with multiple constraints. By implementing the adapt-then-combine scheme and using the gradient tracking technique, rapidity and flexibility can be obtained by the ATC-DIGing-MC with uncoordinated step-sizes. Subsequently, the parameter projection strategy is implemented to deal with realistic production limitations in the presence of multiple constraints. Meanwhile, rigorous theoretical proofs and convergence theorem are provided to verify the geometrical convergence rate of the ATC-DIGing-MC. Furthermore, the numerical experiments based on the entire ethylene separation process validate the superior performance of the ATC-DIGing-MC algorithm in terms of time and energy savings compared to established algorithms.