Dual Residual for Centralized Augmented Lagrangian Coordination Based on Optimality Conditions

Centralized augmented Lagrangian coordination (ALC) has drawn much attention due to its parallel computation capability, efficiency, and flexibility. The initial setting and update strategy of the penalty weights in this method are critical to its performance. The traditional weight update strategy always increases the weights and research shows that inappropriate initial weights may cause optimization failure. Making use of the Karush–Kuhn–Tucker (KKT) optimality conditions for the all-in-one (AIO) and decomposed problems, the terms “primal residual” and “dual residual” are introduced into the centralized ALC, and a new update strategy considering both residuals and thus guaranteeing the unmet optimality condition in the traditional update is introduced. Numerical tests show a decrease in the iteration number and significant improvements in solution accuracy with both calculated and fine-tuned initial weights using the new update. Additionally, the proposed approach is capable to start from a wide range of possible weights and achieve optimality, and therefore brings robustness to the centralized ALC.