Abstract: With the increasing frequency of typhoon disasters, the energy system's supply security faces significant challenges. Rapid and accurate online assessment of the system's resilience based on real-time updated typhoon forecasts is crucial in the short term. This paper proposes a model-data hybrid-driven resilience online assessment method by combining mechanistic models with data-driven approaches, overcoming the limitations of traditional evaluation models that are difficult to apply online. Firstly, addressing the lack of historical data for extreme weather, the non-sequential Monte Carlo method is used offline to simulate massive fault scenarios under different typhoon levels, generating a training dataset for neural networks. Secondly, an online resilience assessment model for the energy internet is constructed based on the Temporal Convolutional Network (TCN). The mapping relationship between typhoon information, system status, and recovery rate is extracted through offline training. Additionally, a deep residual shrinkage network improves the residual module of TCN, reducing the impact of redundant features through self-filtering during training. Subsequently, an online calculation method for resilience indicators based on extended cross-entropy is proposed, improving the convergence speed of the indicator variance through iteratively optimized probability density functions. Finally, the resilience of an energy internet in a certain region is online simulated and assessed, validating the proposed assessment method's speed and effectiveness.