Resilience-Constrained Economic Dispatch With Graph Convolutional Network

Power system resilience has gained additional concerns in recent years due to increasing levels of extreme events. Various economic dispatch and operation strategies are established to improve the power system resilience, such as fine selection of N-k contingencies and improving the power flow entropy. However, since the resilience indices, basically, are the value of expectation and calculated statistically with numerous failure samplings, the mathematical mapping relationship between system operating points and resilience indices cannot be expressed analytically. As a result, current methods cannot incorporate resilience indices directly into power system economic dispatch constraints and objectives. In other words, current methods can only enhance system resilience in an indirect fashion. In this paper, we propose a novel resilience-constrained economic dispatch (RCED) framework with graph convolutional network (GCN) which can formulate the resilience objective function and constraints for the optimization of specific resilience indices. The proposed RCED framework contains offline and online phases. In the offline phase, cascading failure simulations would evaluate system resilience indices and construct the training set. The mapping relationship between operating points and resilience indices is learned with GCN. Moreover, the proposed GCN is equivalently transformed into a set of analytical equations in mixed-integer linear form. In the online phase, considering system resilience requirements, the proposed RCED model with the corresponding resilience constraints and objectives is solved dynamically. Case studies show the effectiveness and advantages of the proposed RCED method.