Transformer and RF Ensemble Learning to Solve Large-Scale Security Constrained Unit Commitment

This paper focuses on the security constraint unit commitment (SCUC) problem in large-scale power systems and proposes a Transformer and random forest (RF) ensemble learning method to solve it. The Transformer model leverages its self-attention mechanism to capture the long-term dependencies in power data, providing preliminary predictions. Subsequently, RF is used to optimize and stabilize the results through stacking and parallel fusion strategies. In the stacking strategy, the output of the Transformer is concatenated with the original features and used as the input for the RF. In the parallel fusion strategy, the independent prediction results of the two models are directly fused using a weighted average, enhancing the overall prediction accuracy and stability. Testing results on the IEEE 118 bus system demonstrate the advantages of the proposed model. Compared with the traditional methods, this model achieves significant improvements in prediction accuracy and solving efficiency.