Probabilistic Renewable Energy Forecasting by Considering Spatial-Temporal Correlation
Renewable energy forecasts can help reduce the amount of operating reserves needed for the system, reducing costs of balancing the system, and improving the reliability of the system. Conventional deterministic forecasts might not be sufficient to characterize the inherent uncertainty of renewable energy. Probabilistic forecasts that provide quantitative uncertainty information associated with renewable energy are therefore expected to better assist power system operations. Meanwhile, studies have shown that the integration of geographically dispersed and correlated wind/solar farms could reduce extreme power output, which is referred to as smoothing effect. In addition, power produced from one wind/solar farm at different times is typically temporally correlated. The impacts of spatial-temporal correlation between wind/solar farms on renewable energy forecasting are not well studied in the literature. This dissertation aims at mitigating power system uncertainty by improving probabilistic renewable energy forecasting accuracy utilizing spatial-temporal correlation modeling. In this dissertation, a variety of methods, such as predictive distribution optimization, ensemble learning, deep learning, and scenario generation-based methods, are developed to assist spatial-temporal correlation modeling, and improve probabilistic forecasting accuracy. Computational experiments indicate that the presented study can provide scholars and engineers with critical insights to the usage of spatial-temporal modeling in probabilistic renewable energy forecasting and anomaly detection, and also serve as a valuable reference for practical industry forecasting systems.