Learning-based spatially-cascaded distributed coordination of shared transmission systems for renewable fuels and refined oil with quasi-optimality preservation under uncertainty

This paper focuses on the distributed optimal coordination framework for energy conservation in the emerging shared transmission systems for renewable fuels and refined oil (STS-RRs) while realizing secure operation with uncertain factors during the energy transition. Specifically, we first propose a practical model for distributed coordination of wide-area pump stations considering sequential transmission features in an STS-RR and variable speed pumps with individual piece-wise linear prejudgment functions (PLPFs) to achieve spatially-cascaded splitting. In the pre-schedule stage, to obtain scenarios-and-spatiality-perceiving slopes of the PLPFs for the stations as well as preserving optimality, a spatial gradient learning method, inspired by the approximate dynamic programming, is designed to acquire prior knowledge from error distribution. In the real-time stage, the models are executed by pump stations based on the real-time measurement information. Both stages are implemented in a spatially-cascaded distributed fashion. The proposed framework was validated using two real-world STS-RRs, demonstrating its feasibility, superior performance, full optimality in ideal conditions, and quasi-optimality under stochastic scenarios, along with good scalability.