Multi-area dispatch driven by data-model fusion dimensionality reduction for AC power flow model

In order to allocate resources over a larger geographical scope, it is essential to cope with the dispatch problems of a multi-area power system. To overcome the drawbacks of conventional decomposition algorithms, i.e., iterative calculation and non-convergence situations, this paper proposes a data-model fusion dimensionality reduction approach for multi-area dispatch. The scale of each area's dispatch model can be reduced by filtering redundant constraints in a decentralized manner, whereas the calculation accuracy of the model will not change. Then, those models can be gathered to a coordinated dispatch center for conducing scheduling plan in the whole system efficiently and accurately. Considering the increasingly serious problem of reactive power allocation in large-scale power grids, a quadratic-convex-relaxed AC optimal power flow model is used for establishing dispatch models. As for dimensionality reduction, compared to other linear constraint screening methods, this paper proposes a model-based redundant second order cone constraint identification method, and apply a data-driven graph neural network model with learnable structural and positional encodings to screen all redundant constraint rapidly to accelerate the coordinated dispatch. To demonstrate the effectiveness of the proposed approach, a multi-area power system composed of China's State Grid (SG) 126-bus systems is adopted for case studies. More system models are also built to further verify the approach's efficiency and correctness.