Graph Convolutional Networks for probabilistic power system operational planning

Abstract

Probabilistic operational planning of power systems usually requires computationally intensive and time consuming simulations. The method presented in this paper provides a time efficient alternative to predict the socio-economic cost of system operational strategies using graph convolutional networks. It is intended for fast screening of operational strategies for the purpose of operational planning. It can also be used as a proxy for operational planning that can be used in long term development studies. The performance of the model is demonstrated on a network inspired by the Nordic power system.