The Optimal Power Flow (OPF) problem is a fundamental building block for the optimization of electrical power systems. It is nonlinear and nonconvex and computes the generator setpoints for power and voltage, given a set of load demands. It is often solved repeatedly under various conditions, either in real-time or in large-scale studies. This need is further exacerbated by the increasing stochasticity of power systems due to renewable energy sources in front and behind the meter. To address these challenges, this paper presents a deep learning approach to the OPF. The learning model exploits the information available in the similar states of the system (which is commonly available in practical applications), as well as a dual Lagrangian method to satisfy the physical and engineering constraints present in the OPF. The proposed model is evaluated on a large collection of realistic medium-sized power systems. The experimental results show that its predictions are highly accurate with average errors as low as 0.2%. Additionally, the proposed approach is shown to improve the accuracy of the widely adopted linear DC approximation by at least two orders of magnitude.

The paper studies how to release data about a critical infrastructure network (e.g., a power network or a transportation network) without disclosing sensitive information that can be exploited by malevolent agents, while preserving the realism of the network. It proposes a novel obfuscation mechanism that combines several privacy-preserving building blocks with a bi-level optimization model to significantly improve accuracy. The obfuscation is evaluated for both realism and privacy properties on real energy and transportation networks. Experimental results show the obfuscation mechanism substantially reduces the potential damage of an attack exploiting the released data to harm the real network.

The growing dependence of electric power systems on gas-fired generators to balance fluctuating and intermittent production by renewable energy sources has increased the variation and volume of flows withdrawn from natural gas transmission pipelines. Adapting pipeline operations to maintain efficiency and security under these dynamic conditions requires optimization methods that account for substantial intraday transients and can rapidly compute solutions in reaction to generator re-dispatch. Here, we present a computationally efficient method for minimizing gas compression costs under dynamic conditions where deliveries to customers are described by time-dependent mass flows. The optimization method uses a simplified representation of gas flow physics, provides a choice of discretization schemes in time and space, and exploits a two-stage approach to minimize energy costs and ensure smooth and physically meaningful solutions. The resulting large-scale NLPs are solved using an interior point method. The optimization scheme is validated by comparing the solutions with an integration of the dynamic equations using an adaptive timestepping differential equation solver, as well as a different, recently proposed optimal control scheme. The comparison shows that solutions to the discretized problem are feasible for the continuous problem and also practical from an operational standpoint. The results also indicate that our scheme produces at least an order of magnitude reduction in computation time relative to the state of the art and scales to large gas transmission networks with more than 6,000 kilometers of total pipeline. The online supplement is available at https://doi.org/10.1287/ijoc.2018.0821 .

We address the problem of power system restoration after a significant blackout. Prior work focus on optimization methods for finding high-quality restoration plans. Optimal solutions consist in a sequence of grid repairs and corresponding steady states. However, such approaches lack formal guarantees on the transient stability of restoration actions, a key property to avoid additional grid damage and cascading failures. In this paper, we show how to integrate transient stability in the optimization procedure by capturing the rotor dynamics of power generators. Our approach reasons about the differential equations describing the dynamics and their underlying transient states. The key contribution lies in modeling and solving optimization problems that return stable generators dispatch minimizing the difference with respect to steady states solutions. Computational efficiency is increased using preprocessing procedures along with traditional reduction techniques. Experimental results on existing benchmarks confirm the feasibility of the new approach.

Distributed constraint solving are useful in tackling constrained problems when agents are not allowed to share his/her private information to others and/or gathering all necessary information to solve the problem in a centralized manner is infeasible. With these two limitations, distributed algorithms solve the problem by coordinating agents to negotiate with each other. However, once information is exchanged during negotiation, the private information may be leaked from one agent to another. We propose and design a framework based on Valuation of Possible States (VPS) to evaluate how well a distributed algorithm preserves the totality of all private information onthe entire system when solving distributed constraint optimization problems, by allowing the uses of different aggregators aggregating agents' individual privacy loss. Two classes of aggregators: idempotent aggregators and risk based aggregators are proposed. We further proposed generalized inference rules to infer privacy loss of individual agents. We implement our work on four distributed constraint solving algorithms: Synchronous Branch and Bound (SynchBB), Asynchronous Distributed Constraint Optimization (ADOPT), Branch and Bound ADOPT (BnB-ADOPT), and Distributed Pseudo-tree Optimization Procedure (DPOP). Preliminary experimental evaluations on two benchmarks, Distributed Multi-Event Scheduling Problem (DiMES) and Random Distributed COP, comparing the four algorithms are performed.

Minimax Weighted Constraint Satisfaction Problems (formerly called Quantified Weighted CSPs) are a framework for modeling soft constrained problems with adversarial conditions. In this paper, we study the effects of a value ordering heuristic in solving ultra-weak solutions on top of the alpha beta tree search with constraint propagation. The value ordering heuristic is based on minimax heuristics from adversarial search, which selects values for variables according to the semantic of quantifiers by considering the problem as a two-player zero sum game. In practice, implementing the heuristic requires costs approximations, and we devise three heuristic variants: HUnary, HBinary, and HFullBinary to approximate costs. In particular, we observe that combining these heuristic variants with consistency notions can achieve a better efficiency and a further reduction of search space. We perform experiments on three benchmarks to compare the effects on applying these heuristic variants, and confirm the feasibility and efficiency of our proposal.

Soft constraints are functions returning costs, and are essential in modeling over-constrained and optimization problems. We are interested in tackling soft constrained problems with adversarial conditions. Aiming at generalizing the weighted and quantified constraint satisfaction frameworks, a Quantified Weighted Constraint Satisfaction Problem (QWCSP) consists of a set of finite domain variables, a set of soft constraints, and a min or max quantifier associated with each of these variables. We formally define QWCSP, and propose a complete solver which is based on alpha-beta pruning. QWCSPs are useful special cases of QCOP/QCOP+, and can be solved as a QCOP/QCOP+. Restricting our attention to only QWCSPs, we show empirically that our proposed solving techniques can better exploit problem characteristics than those developed for QCOP/QCOP+. Experimental results confirm the feasibility and efficiency of our proposals.