Efficient and Scalable Methods for Multi-Stage Transmission Expansion under Uncertainty

点击翻译按钮获取中文摘要

A critical component of the electric power system is the underlying network of high voltage transmission lines that connect geographically dispersed generation and load. The transmission network achieves two important objectives: reducing the cost of energy by providing access to low-cost generation and maintaining reliability by enabling many alternative generation sources and transmission routes to serve load centers. Although utilities, and more recently regional transmission operators (RTOs), have long engaged in transmission planning, the current context requires planning for larger regions and over longer time horizons. The uncertainties to which any new transmission lines should be robust to changes in include the types and locations of generation as well as variations in the spatial and temporal distribution of load, both driven by rapid changes in technologies, market factors, and regulations. The potential sunk cost of transmission lines that do not anticipate future conditions, as well as the expected benefit of properly located transmission additions, can be valued in the millions of dollars. The transmission planning tools commonly used by utilities and RTOs are well suited for near-term tactical planning when uncertainties are manageable, but are not appropriate to long-term planning when the range of possible future conditions becomes large. This project develops new methods for transmission planning over several decades and across a wide range of possible futures.Specifically, the research team will develop two alternative algorithms for solving multi-stage stochastic transmission expansion problems: (i) Multi-stage schemes for smoothed nonconvex problems, and (ii) Monte-Carlo and Importance Sampling-based Q-Learning schemes. In the context of both schemes, convergence properties will be analyzed and error bounds will be developed. Furthermore, both sets of schemes will be implemented within a high performance computing environment (such as a network of computing nodes) with an emphasis on asynchronous implementations. As part of the project, the team will collaborate with the planning group at PJM Interconnection and apply the methods developed to their network, consisting of approximately 16,000 buses. The application of stochastic analysis will help to identify long-term congestion issues that should be anticipated, and provide an initial list of candidate lines that would be robust to the large set of future scenarios. More broadly, the development and adoption of these methods will enable planners for regional power systems across the nation to identify crucial additions to the infrastructure that can reduce energy costs, maintain reliability of energy supply, and enable and enhance the ability of new generation technologies to reduce environmental impacts. The project will also make contributions to education at the graduate and undergraduate levels. Finally, this project will occur within the Penn State Initiative for Sustainable Electric Power Systems, which organizes workshops and collaborations with the power industry and other academic institutions.