• A risk-aware scenario screening method is proposed for facilitating the holistic energy management. • It integrates data-driven clustering with its downstream use into a unified modeling framework. • It reformulates the operational decision-making problem as a neural network by surrogate modeling. • It improves the risk sensitivity and computational efficiency of scenario screening under unexpected renewable fluctuations. High renewable energy penetration introduces complex uncertainties and elevates the risk of inadequate power supply. Effective scenario screening is therefore essential to ensure that representative scenarios adequately capture operational risks. However, conventional scenario generation methods often overlook how scenario variations impact system operation, resulting in suboptimal support for decision-making. To this end, we propose a risk-aware operational scenario screening method, assisted by a surrogate model for system operation, to effectively consider the final use in the generation of representative scenarios. Specifically, we formulate the risk-aware scenario screening problem as a two-stage mixed integer optimization problem and convert it into a neural network composed of an operation module and a clustering module. For the operation module, the explicit function between scenarios and the risk metrics is deduced by multiple parametric linear programming and constructed in the form of a neural network. For the clustering module, the discrete scenario assignment is addressed using a straight-through Gumbel-Softmax estimator in a differentiable manner. The clustering process directly assigns scenarios without being compared with the entire dataset, thereby enabling efficient computation in an online decision-making manner. Case studies on the modified IEEE 6-bus and 118-bus systems demonstrate the effectiveness of the proposed method. Scenario screening and online decision-making framework.
Zhou et al. (Sun,) studied this question.