Inter-day energy storage expansion framework against extreme wind droughts based on extreme value theory and deep generation models

The worldwide occurrence of wind droughts challenges the balance of power systems between energy production and consumption. Expanding inter-day energy storage serves as a strategic solution, yet optimizing its capacity depends on accurately modeling future renewable energy uncertainties to avoid over- or under-investment. Existing approaches that use the historical extreme scenario set (HESS) to represent future conditions are contentious due to potential inadequacies in forecasting future extreme scenarios (ESs), including those on a decadal or centennial scale. This study addresses the issue by proposing an advanced energy storage expansion framework that leverages Extreme Value Theory (EVT) and a novel Deep Generative Model, namely the Diffusion Model. To model the extremes in a principled way, this work leverages EVT to establish a severity-probability mapping for wind droughts, guiding the training process of the Diffusion Model. This model excels in generating ESs that accurately reflect the distribution of real-world extremes, thereby significantly enhancing the predictive capacity of HESS. Case studies on a real-world power system confirm the method's capacity to generate high-quality ESs, encompassing the most severe historical wind droughts not included in the training dataset, thereby facilitating resilient energy storage expansion against unforeseen extremes.