A risk assessment method of multi-form transient voltage stabilities for high-proportion renewable energy sending-end power grids

• Multi-form transient voltage stability assessment indices are proposed to characterize five voltage instability forms. • A future operating sample set generation method for high-proportion renewable energy sending-end power grid is proposed. • A multi-form transient voltage stability assessment model based on TGNN-CNN is proposed. • Simulation results demonstrate that the proposed method could identify weak operating modes, serious faults and weak buses. For high-proportion renewable energy sending-end power grids, various transient voltage instabilities including transient overvoltage, transient low voltage, delayed voltage recovery, repeated low voltage ride through, and sustained low voltage, are seldom comprehensively considered and assessed. To address this issue, a risk assessment method of multi-form transient voltage stabilities is proposed. Firstly, by analyzing the transient voltage instability characteristics of sending-end power grids, multi-form transient voltage stability assessment indices are introduced. Secondly, an active power optimization dispatch model and a reactive power distribution optimization method incorporating multiple energy sources are established. They are integrated with chronological operation simulation to generate future annual operating samples. Thirdly, multi-form transient voltage stability assessment models are constructed considering of the topological and spatial relationships of nodes in the power grid by the Transformer graph neural network (TGNN) and convolutional neural network (CNN) models. A learnable feature fusion factor is used to adaptively adjust the contribution weights of the TGNN and CNN, thereby improving the model’s assessment performance across multiple scenarios. A grading scheme is established to differentiate the transient voltage stability risks. Finally, the effectiveness of the proposed transient voltage stability risk assessment models and the necessity of constructing future operating datasets are validated through a case study on a real provincial power grid.