Thermal Stability Safety Assessment Framework of Power System Based on Transfer Learning Convolutional Neural Network

In the context of the dual-carbon initiative, the grid-connected capacity of new energy continues to rise. As operation modes and network topologies evolve, they progressively enhance the thermal stability and security of the power system. This study introduces a framework for assessing power system thermal stability security using TL-CNN (transfer learning and convolutional neural network). We construct a nonlinear CNN with multiple hidden layers and utilize node and line information as the input for the assessment. Using the gradient descent method, supervised pretraining is realized, ensuring an efficient initialization for our neural network model. The deep architecture inherent to CNN is then harnessed to delineate a mapping relationship between input samples and the thermal stability security assessment. As we encounter significant alterations in system operation modes and network topologies, it's imperative to maintain consistency. To address this, the architecture of the pre-trained model—including its network structure, convolutional layer, and pooling layer parameters—remains constant. Only the fully connected and classification layers are transitioned to the new model. This structured approach enables adaptive assessment of the power system's thermal stability security state across diverse operational scenarios. The proposed method was evaluated in a simulation using the New England IEEE-39 node systems.