The Total Transfer Capability Assessment of Transmission Interfaces Combining Causal Inference and Multi-Task Learning

Power system security assessment has many tasks for different evaluation topics or objects, such as the Total Transfer Capability (TTC) assessment of different transmission interfaces. Some different tasks are related with each other, which is an important characteristic to improve the assessment performance if used properly but easily to be neglected. In this study, a novel architecture of the Causality-based Multi-Task Neural Network (CMTNN) is proposed, which can take the causal relevance between tasks into account while avoiding the introduction of too much extraneous information. First, the mask layer is applied after the input layer to filter irrelevant information task-specifically. Then, the causal analysis based on the Peter-Clark algorithm is performed to find those causally-related tasks. Finally, the causality-based information-sharing mechanism between related tasks is designed in the neural network to share deep information, so as to improve the model assessment performance on those tasks. In this article, the proposed method is applied on the total transfer capability assessment problem of multiple transmission interfaces, and the case study on a real-world regional system with more than 8000 nodes in China validates its effectiveness.