Various stakeholders (grid operators, regulators) are placing increasing emphasis on evaluating and monitoring the system’s response to even multiple ‘N-k’ contingencies. In addition, the penetration of renewable sources makes the system’s operating point increasingly uncertain in the context of operation planning. Therefore, control methodologies are important for the best system response to N-k contingencies, in the presence of uncertainties on both the occurrence of the contingencies themselves and load and renewable generation forecasts.

The report proposes the formulation of a corrective-predictive risk-based control with probabilistic constraints for handling N-k contingencies. The control—formulated as an efficient mixed integer linear programming problem—suggests preventive and/or corrective actions (implemented before and after the occurrence of a contingency, respectively) ensuring the optimal combination of the costs of preventive measures and the expected costs of corrective measures and ‘failure to secure,’ with probabilistic security constraints. The tests—carried out by first implementing preventive measures only—show that control manages multiple contingencies that can also cause the network to be separated into islands. In the next year of research, the costs associated with these tests will be compared with those obtained after the implementation of the corrective measures.

The forecasting accuracy of renewables and loads influences the operating costs to limit the risk; for this, the production forecasts of some wind farms—made by a producer—are compared with the forecasts made using advanced methods developed in RSE: the forecasting errors on wind production depend on the accuracy of weather forecasts, the spatial resolution of the models and the orographic conformation of the location of the farms.

Assessing the benefits of control on resilience over the long term requires the analysis of multiple operating points representative of actual system operation: it is therefore essential to define a set, albeit limited, of representative points. Thus, k-means clustering and fast forward method (FFM) techniques are analyzed, as well as two metrics designed to define the distance of scenarios in terms of ‘propensity to cascade.’ The metrics are based on the load level of the branches and the electrical robustness of the nodes, respectively.

Tests on the 24-node IEEE network show that the mean square errors between the cumulative probability distribution of the network load, obtained from the original set of scenarios, and the cumulative distribution, obtained from the reduced set of operating points resulting from k-means clustering, are lower than those obtained using the FFM.