arXiv subjects

Yeesian Ng

Publications and source records attributed to Yeesian Ng.

3 recordsLinked to original sources

arXiv2019

Learning for Constrained Optimization: Identifying Optimal Active Constraint Sets

In many engineered systems, optimization is used for decision making at time-scales ranging from real-time operation to long-term planning. This process often involves solving similar optimization problems over and over again with slightly modified input parameters, often under tight latency requirements. We consider the problem of using the information available through this repeated solution process to directly learn a model of the optimal solution as a function of the input parameters, thus reducing the need to solve computationally expensive large-scale parametric programs in real time. Our proposed method is based on learning relevant sets of active constraints, from which the optimal solution can be obtained efficiently. Using active sets as features preserves information about the physics of the system, enables interpretable models, accounts for relevant safety constraints, and is easy to represent and encode. However, the total number of active sets is also very large, as it grows exponentially with system size. The key contribution of this paper is a streaming algorithm that learns the relevant active sets from training samples consisting of the input parameters and the corresponding optimal solution, without any assumptions on the problem structure. The algorithm comes with theoretical performance guarantees, and is known to converge fast for problem instances with a small number of relevant active sets. It can thus be used to establish the practicability of the learning method. Through extensive experiments on the Optimal Power Flow problem, we observe that often only a few active sets are relevant in practice, suggesting that the active sets is the appropriate level of abstraction for a learning algorithm to target.

math.OC↗

arXiv2018

PowerModels.jl: An Open-Source Framework for Exploring Power Flow Formulations

In recent years, the power system research community has seen an explosion of novel methods for formulating and solving power network optimization problems. These emerging methods range from new power flow approximations, which go beyond the traditional DC power flow by capturing reactive power, to convex relaxations, which provide solution quality and runtime performance guarantees. Unfortunately, the sophistication of these emerging methods often presents a significant barrier to evaluating them on a wide variety of power system optimization applications. To address this issue, this work proposes PowerModels, an open-source platform for comparing power flow formulations. From its inception, PowerModels was designed to streamline the process of evaluating different power flow formulations on shared optimization problem specifications. This work provides a brief introduction to the design of PowerModels, validates its implementation, and demonstrates its effectiveness with a proof-of-concept study analyzing five different formulations of the Optimal Power Flow problem.

math.OC↗

arXiv2018

Statistical Learning For DC Optimal Power Flow

The optimal power flow problem plays an important role in the market clearing and operation of electric power systems. However, with increasing uncertainty from renewable energy operation, the optimal operating point of the system changes more significantly in real-time. In this paper, we aim at developing control policies that are able to track the optimal set-point with high probability. The approach is based on the observation that the OPF solution corresponding to a certain uncertainty realization is a basic feasible solution, which provides an affine control policy. The optimality of this basis policy is restricted to uncertainty realizations that share the same set of active constraints. We propose an ensemble control policy that combines several basis policies to improve performance. Although the number of possible bases is exponential in the size of the system, we show that only a few of them are relevant to system operation. We adopt a statistical learning approach to learn these important bases, and provide theoretical results that validate our observations. For most systems, we observe that efficient ensemble policies constructed using as few as ten bases, are able to obtain optimal solutions with high probability.

math.OC↗