Data-driven approach to calculating nodal net load forecasts in power systems analysis, A

Abstract

The electric power grid is one of the largest and most complex systems ever built for the benefits of humanity. Most of the modern world relies heavily upon this system, and its unavailability may result in severe economic, safety, and security issues. Ensuring the electric power grid maintains an adequate level of reliability is a challenging task that usually begins well in advance of real-time system operations. This task of ensuring grid reliability typically begins in system operations with the Operational Planning Analysis (OPA). TheOPA is a study required by the North American Electric Corporation (NERC) Reliability Standards. The goal of that study is to assess whether planned operations for the next day will exceed any operating limits or present any potential reliability threats to the system. The next day is also the same time frame in which the electricity markets begin to procure resources needed to meet the following day's demand or forecasted load. The foundation for these studies is the load forecast. The accuracy of it can determine the next day's system requirements, e.g., generation and system outages. The problem is that the load forecast is typically done at the system level, and for power system engineers, the accuracy of the data is needed at the most granular level, the load bus. A bus or node level forecast does not exist in the industry today, and this research aims to develop a method based on the historical data accumulated by the utility companies. This research also describes the concept of net load, which is the resultant value of load and resources real-time system operators monitor at the transmission system level. This dissertation discusses the current practice of the utilization of load forecast data in the electric utility industry. It proposes a new data-driven framework based on machine learning to predict net loads at the node or bus level.