Optimizing large-scale unit commitment

In emerging power grids, the landscape has evolved to include not only traditional utilities but also millions of electricity generation and storage devices, each owned and controlled by different entities. However, due to privacy concerns and varied objective functions among owners, there is a reluctance to share specific details about the capabilities and limitations of these devices. This situation presents a significant challenge when it comes to optimizing the scheduling of power generation and storage devices due to the exponential increase in variables and constraints, as well as data privacy and communication issues among decision makers.

To address these challenges, we have developed a decentralized optimization framework for tackling prosumer-based large-scale unit commitment (UC) problems. This framework divides the problems into smaller subproblems that can be solved independently and in parallel. Each subproblem is optimized individually, and the solutions are combined by exchanging interfacing variables, such as dual variables, to derive an overall solution for the original problem. Decentralized optimization is particularly suited for scenarios where there are numerous loosely connected blocks of decision variables and constraints. In the context of power systems, this is applicable as operational variables and constraints for different generation, storage, loads, and transmission devices are sparsely interconnected.

The proposed approach has demonstrated exceptional performance through the parallel implementation of the decentralized algorithm for solving large-scale UC instances in power systems encompassing over 3,000 buses. Notably, it has achieved computational gains of up to 100 times when compared to traditional optimization approaches for realistic ISO-size UC problems.