Advanced Optimization and Cost Estimation for Utilities and Interruptible Customers

点击翻译按钮获取中文摘要

ECS-9726577 Luh Scheduling power systems and performing inter-utility power transactions are important activities to supply reliable, economic, and clean power. The economics of scheduling and transaction decisions, however, are significantly affected by uncertainties, including inaccurate forecasted demand, uncertain unit availability, and the random occurrences of bilateral power transaction opportunities. The impact of uncertainties propagates to utilities' industrial customers with interruptible load, as their prices or consumption limits may be changed forcing them to alter their resource allocation and personnel commitment. To reduce the cost of electricity through effective managing uncertainties and to provide reliable price estimates for industrial customers with interruptible load, a coalition of University of Connecticut (UConn), Northeast Utilities (NU), Taylor & Fenn, and Praxair is formed. The specific objectives of the proposed research are: 1 . To develop an effective stochastic-fuzzy optimization method for the short-term scheduling and transaction problem to minimize cost, reduce solution variance, and manage risks. 2. To further improve the above results by developing a new generation of optimization approaches based on a synergistic combination of "Lagrangian relaxation" and neural networks for better and faster resolution of larger and more complicated problems - problems covering a larger number of units, with three levels of reserve, or with a longer planning horizon. 3. To develop methods to accurately and robustly estimate the costs for providing bundled or unbundled electricity products and services, and to explore other means for a utility to better serve its industrial customers with interruptible load. The optimization of large-scale, mixed-integer uncertain systems such as the scheduling and transaction problem considered here for near-optimal cost with low variance is theoretically important and practically challenging. The foundations needed t o be laid for and the potential benefits of neural networks for such applications are intriguing and fascinating. Based on our excellent results obtained thus far, we believe that the proposed research shall not only help utilities cut costs, reduce variance, and provide better products and services for interruptible customers, but also open up a new generation of neural optimization approaches for other applications in power systems and beyond.