Active Ramping Constraint Screening With Bayesian Epistemic Uncertainty‐Aware Modelling for SCUC Acceleration

ABSTRACT

Screening and retaining active ramping constraints can reduce the computational time of security‐constrained unit commitment (SCUC). However, the process of screening ramping constraint activeness from observable operating conditions involves epistemic uncertainty, because ramping constraint activeness is a solution‐induced status in multi‐period SCUC jointly determined by binary commitment transitions, inter‐temporal dispatch trajectories and balancing allocation across generating units. Therefore, explicitly characterising epistemic uncertainty can improve the reliability of active ramping constraint screening. This paper proposes a Bayesian multi‐head attention transformer network (BMHATN) for epistemic uncertainty‐aware active ramping constraint screening to accelerate SCUC. The proposed BMHATN is built on a Bayesian neural network that explicitly characterises the epistemic uncertainty in screening whether a ramping constraint is active or inactive, representing the first attempt to incorporate epistemic uncertainty‐aware modelling into ramping constraint screening. Furthermore, this paper constructs the first publicly available dataset of active ramping constraints in SCUC, with labels fully consistent with the mathematical activeness definition. Numerical results on the IEEE 118‐bus system and a larger realistic 500‐bus system demonstrate that the proposed BMHATN achieves reliable active ramping constraint screening, remains robust under increasing renewable energy penetration and provides SCUC acceleration while preserving the exact optimal solutions of the full SCUC model.