Critical infrastructures such as power grids, transportation networks, water systems, and others increasingly interact with each other for functional facilities. Therefore, coordinating these interdependent critical infrastructure networks has become paramount for synergistic operation. Since different infrastructures belong to different entities, this dissertation proposes decentralized optimization algorithms tailored to various coordination challenges. First, this dissertation proposes a decentralized framework for coordinating power and transportation networks with electric vehicles. Since the transportation network model is in a mixed-integer program (MIP) form, an Enhanced SD-GS-AL (stands for Simplicial Decomposition, Gauss-Seidel, and Augmented Lagrangian) algorithm that guarantees convergence in the coordination of MIPs with minimal information exchange is proposed. Second, the Enhanced SD-GS-AL algorithm is further improved to tackle the unique challenges of intertwined power and water networks, where boundary variables are discontinuous. Therefore, the mixed-integer boundary compatible (MIBC) SD-GS-AL algorithm is proposed to coordinate MIPs with discontinuous boundary variables. Third, the decentralized algorithms require iterative solving of the optimization models until convergence. The computational requirement of optimization models used in coordinating infrastructures in normal operating conditions is not high, and models are solved quickly; hence, the algorithms converge relatively quickly. However, the models used for modeling emergency conditions, such as restoration models, are computationally heavy. Therefore, the computationally efficient power distribution system restoration approach is proposed via convex hull relaxation and super-node approximation. Fourth, with increasing penetration of variable renewable energy sources (VRES) like solar energy, the necessity to consider uncertainties in the operation and planning of the power grid is heightened. Therefore, the uncertainty introduced by VRES is modeled as a chance-constrained optimization. Moreover, the efficacy of proposed decentralized optimization algorithms in coordinating uncertainty-aware optimization models is validated by coordinating the chance-constrained power grid and other critical infrastructures. The convergence and optimality of the proposed algorithms are also provided in this dissertation.
Qifeng Li, Committee Chair.
Read More