The Role of Machine Learning-based Surrogate Models for Wind Power Curtailment Prediction from Electricity Market Data

The proliferation of distributed low-inertia generators combined with the large number of power transactions ruled by complex electricity market dynamics and the increased penetration of non-linear loads make modern electrical grids more vulnerable to dynamic perturbations. Hence, to enhance the power system resilience to these perturbations, it is incumbent to develop enhanced tools for Dynamic Security Assessment (DSA), which try inferring from large aggregated datasets actionable information helpful in defining corrective actions to mitigate the dynamic impacts of severe perturbation phenomena. In this context, the curtailment of renewable power generators is one of the most critical actions that DSA should identify for enhancing the power system stability after severe grid contingencies. To face this challenging issue, this paper explores the potential role of machine learning-based surrogate modeling in discovering the hidden relationships between aggregated grid data describing the electricity market clearing and the corresponding wind power curtailment identified by a real DSA tool. Experimental results obtained by processing the Italian electricity market data are presented and discussed to assess the performance of the analyzed techniques, outlining the most revealing future research paths of this research.