Large scale grid expansion planning studies are essential to rapidly and efficiently decarbonizing the electricity sector. These studies help grid participants understand which generation, storage, and transmission assets should be built and where they will have the highest cost and emissions impacts. However, expansion planning studies are often too computationally expensive to run for high-fidelity network models with high uncertainty, leading to suboptimal decision making based on coarse proxy models. To address this issue, we introduce a GPU-accelerated solver for electricity grid operations problems using the alternating direction method of multipliers (ADMM). Our solver is fully differentiable, i.e., the solution of the operations problem can be differentiated with respect to any inputs or parameters. This lets us embed this solver in a gradient-based algorithm for long-term grid expansion planning. We show that this framework scales well to large networks with many uncertain scenarios and can even enable interactive planning studies on realistically sized grid models. We demonstrate applications of our method using a large-scale model of the Western U.S. electricity system.
