Date of Award


Degree Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering and Computer Science


Sara Eftekharnejad


cascading failure propagation;dynamic line rating;machine learning;optimal power flow;power system steady state analysis;renewable energy forecasting


Over recent decades, there has been a significant increase in the integration of Variable Renewable Energy (VRE) sources into power grids. Notably, solar energy has emerged as the most rapidly expanding source of electricity in the United States, representing over half of the total new electricity generation capacity added to the U.S. grid for the year 2023. This surge plays a pivotal role in the decarbonization process, yet it simultaneously introduces complexities in maintaining the stability of power systems. A key element in addressing these challenges is the development of precise VRE forecasts. The resulting forecast provides grid operators with a short-term (day-ahead or hour-ahead) estimate of VRE, allowing an optimized power dispatch, balancing the generated electricity and load, and preparing necessary measures to protect the grids. Traditionally, the generation forecast is achieved using historical data, such as meteorological observations obtained from local weather stations or remote sensing devices. Despite their utility, traditional forecasting methods are dependent on large-scale historical data to ensure accuracy, and many utilities face constraints in accessing comprehensive weather data. Moreover, traditional forecasting methods struggled under rapidly changing weather conditions, leading to significant discrepancies between forecasted and actual generation. This inadequacy becomes particularly problematic during unexpected weather events, where the lack of accurate forecasting can lead to grid instabilities or inefficiencies in energy distribution. Addressing these challenges requires elaborative studies, which is the focus of this dissertation. It aims to bridge the gap between the existing challenges and effective solutions by focusing on three objectives: (1) Enhancing forecasting efficiency; (2) Improving forecasting robustness; and (3) Applying forecasts in power grids. Initially, a thorough introduction to the evaluation of VRE forecasts is provided, particularly focusing on solar and wind power forecasts. The first objective is to enhance forecasting efficiency through a data-driven deterministic model that reduces data volume requirements for accurate solar forecasting. By employing a dynamic feature selection method, based on deep reinforcement learning (DRL), the model identifies the minimal yet sufficient features for precise forecasts, adapting to various weather conditions. Compared to conventional methods that use all historical weather data, this model achieves greater forecast accuracy using only 25$\%$ of the data. To further improve forecasting robustness, a probabilistic model integrating copula theory with machine learning techniques is developed. This model captures the spatio-temporal correlations among meteorological data under diverse conditions, thus quantifying the uncertainty due to rapid weather changes. Incorporating such correlations into forecasting, the dynamic model has demonstrated up to 60$\%$ higher accuracy under non-sunny conditions compared to state-of-the-art models. Furthermore, the dissertation examines the impact of probabilistic VRE forecasts on the Day-ahead Optimal Power Flow (OPF) in high VRE penetration scenarios. By integrating forecasts with varying accuracies into a Monte Carlo-based probabilistic OPF framework, the study assesses their effect on grid reliability. The research indicates that while improved forecast accuracy marginally enhances OPF reliability, the combination of forecasts with techniques such as Dynamic Line Rating (DLR) significantly improves it. Additionally, an online model for predicting cascading failure propagation in high VRE penetration systems is presented, utilizing copula theory to estimate concurrent line failure probabilities and updating hourly to better account for the impact of VRE fluctuations. Overall, this dissertation contributes to the development of more efficient and robust VRE forecasting models and offers valuable insights for improving grid management and fostering a sustainable energy future dominated by renewable resources.


Open Access

Available for download on Saturday, June 14, 2025