Severe weather is the leading cause of power outages in the United States, leading to tremendous economic and social costs. Given days-ahead weather forecasts, hardening power lines can significantly mitigate cascading outrage risks, shorten the time to restore electricity, and therefore improve power grids weather resilience. Due to resource constraints, such as time and/or budget, it is important to identify the optimal hardening plan with the constraints. Given the large number of power lines on the grids, searching for the optimal subset of power lines of hardening is challenging. In this research, a novel searching approach for optimal hardening plan with practical constraints is studied. The method takes the advantages of high-dimensional data analysis from statistics and discrete optimization via stochastic simulation from operations research and makes fundamental theoretical and algorithmic advances to optimize power grids hardening plans and reduce cascading power outage risks in severe weather. Because of the broad economic and societal importance of power grids, the research has broader impact on the welfare and security of the country.
This project develops a model-free dimension reduction method to improve the computational efficiency of discrete optimization via simulation for improving power grids weather resilience through optimal power line hardening. The dimension reduction method ranks the subsets of the transmission lines according to power loss caused by their disconnection from power grids. The new method does not need to assume a specific statistical joint model between power loss and all considered line combinations and only uses the marginal information of the line combinations, and thus are generally applicable for hardening planning. A new stochastic search algorithm that exploits the dimension reduction capability is then proposed to reduce the size of the effective search space given resource constraints when preparing for severe weather conditions. To improve computational efficiency, the stochastic search algorithm uses an informative Gaussian mixture prior to incorporate dimension reduction results while achieving asymptotical convergence and constructs a hierarchical sampling distribution using dimension reduction results. The model-free dimension reduction and stochastic search algorithm are generally applicable to a variety of other disciplines, e.g., the protection of other critical civil infrastructures such as road networks.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.