Learning-Enabled Optimization for Security-Constrained Unit Commitment with Energy Storage

Fast and intelligent scheduling is essential for the safe and stable operation of modern renewable energy power systems. To address the complex Security-Constrained Unit Commitment (SCUC) problem, this paper introduces a rapid solution algorithm based on the learning-enabled optimization. Using the SCUC model that incorporates energy storage, we integrate machine learning techniques within a mixed-integer programming framework, proposing a "prediction + decision" approach. Turing statistics are employed to build a confidence-based training set, and a multi-layer neural network classifier is used to predict logical strategies in real time, enabling fast online decisions. Simulation results on the IEEE 9-bus system demonstrate that this method outperforms traditional solvers and other data-driven algorithms in terms of speed, success rate, and optimal value. The learning-enabled decision-making algorithm provides high-quality approximate solutions, improving decision-making speed by more than an order of magnitude.