Data-driven and model-driven evaluation and prediction methods for load regulation potential

In the context of high penetration of photovoltaics in distribution networks, researching the load regulation potential is crucial for grid safety and refined dispatching. Thus, a data-driven and model-driven evaluation and prediction method for load regulation potential is proposed. Firstly, a multi-dimensional evaluation index system and assessment model for load regulation potential are constructed. A two-step clustering method, including k-means for load feature extraction and self-organizing map (SOM) for regulation potential feature fusion, is introduced. This enables the assessment of the temporal up-regulation and down-regulation potential of various loads. Secondly, an intelligent prediction method that integrates bidirectional long short-term memory (BiLSTM) and dynamic mode decomposition (DMD) is proposed to predict the 15-minute level load regulation potential for the next day. Finally, simulation verification of the proposed strategy is conducted using local load data. The evaluation results confirm the effectiveness of the proposed assessment and prediction methods, demonstrating that the BiLSTM-DMD model can achieve high prediction accuracy.