Brown, William Eric (2020-07). Efficient and Effective Algorithms for Contingency Response in Electric Power Systems with Transmission Switching. Doctoral Dissertation.
Thesis
Optimal operation of the electric power grid is of critical importance to maintain the integrity of the industrial, medical, and defense entities upon which the population relies. In the past, the power grid has primarily been modeled and operated used a fixed configuration. Under such a modeling framework, control on the power grid is exerted only via the use of unit commitment (i.e., which generators are active) and generation dispatch (i.e., how much power is produced at each generator) decisions. However, system operators have another method of control at their disposal, known as transmission switching, where transmission lines are physically switched in and out of the grid. This technique is also referred to as transmission topology control. By leveraging this additional flexibility of the grid given by transmission switching, system operators can obtain significant benefits which are otherwise inaccessible. Unfortunately, transmission switching is currently implemented only in limited capacity. The primary reason for the lack of widespread implementation is because of the computational complexity of the problem. Specifically, the AC optimal transmission switching problem, the optimization model which most accurately models the true behavior of the AC power grid, is a highly nonconvex mixed-integer nonlinear program. Thus, even using state-of-the-art optimization techniques, the ACOTS remains computationally intractable. To address this, the bulk of the transmission switching literature uses either DC-based models or unrealistically small test systems. These facts are problematic because DC-based models have been shown in the literature to be highly inaccurate in short-term operations. In addition, small test cases fail to approximate the complexity of real-world power grids, clouding the possible real-world impact of models and algorithms. Finally, it is clear in the literature that most algorithms simply cannot meet the requirements of real-time operations for large systems, motivating the need for more efficient techniques. This dissertation develops two techniques to address the above concerns, adding to the body of knowledge and pushing transmission switching towards real-world implementation. Specifically, we develop solution approaches for two variants of the ACOTS within the particular application of contingency response. That is, for two different problem contexts, we seek to identify optimal transmission switching actions when a portion of the grid (e.g., a set of generators and/or transmission lines) fails. First, we develop a mixed-integer linear optimization model which accurately reflects the complexities of the ACOTS and is competitive with the state-of-the-art in terms of solution time. The aim of the model developed herein is to identify transmission switching and redispatch actions to prevent load shed caused by contingency events. To do so, we accelerate an existing model by developing three constraint relaxations that dramatically decrease time-to-solution with no decrease in accuracy. Moreover, our model was found to almost always identify AC optimal transmission switching actions while dramatically reducing necessary solution time when compared with the base model. Second, we develop a data mining approach to address a variant of the ACOTS which minimizes post-contingency AC power flow violations. Our methodology, which utilizes a sophisticated guided undersampling procedure consisting of imbalanced-data classification approaches, significantly outperforms the state-of-the-art heuristics. Our data mining approach, which is computationally inexpensive and vetted on real-world AC power system data, addresses three issues from the literature that limit the practical viability of transmission switching: computational complexity, AC system impact, and real-world system impact. These facts, combined with the performance of our approach, demonstrate the strength of data mining techniques for this
