Horizontal axis tidal turbines (HATT) are promising candidates for hydroelectric power extraction in coastal urban applications. Currently, concerns around the fluid–structure interaction (FSI) performance of turbines limit their industrial deployment while an efficient FSI prediction model can mitigate these concepts. In this paper, we present an innovative predictive model, TurbineNet, designed to accurately forecast the FSI characteristics of diverse composite tidal turbine blade structures, thereby facilitating its performance optimization. The TurbineNet model utilizes blade mesh information as input to predict HATT performance through a neural network framework. By integrating mesh convolution and fully connected layers, the model delineates the hydrodynamic behavior of deformed blades which then couples with the finite element method (FEM) for a comprehensive dynamic FSI analysis of composite material blades. Through the generation and training of a database of deformed blades, the new TurbineNet/FEM model is capable of precisely predicting the hydrodynamic performance under various structural parameters. This approach significantly streamlines the computational process associated with traditional FSI models, reducing prediction times by nearly 18 times compared to static FSI calculations using established platforms like Ansys Workbench, while maintaining high accuracy. Based on the innovative TurbineNet/FEM model, we analyzed the FSI characteristics of three different web structures. The incorporation of web structures can substantially reduce local stress and enhance the structural stability of the blades, thus preventing vibrations. Our high-efficiency predictive FSI tool and results can aid HATTs in increasing their much-needed contribution to stability optimization design. It has the potential to significantly optimize energy conversion in tidal turbines by refining blade designs to efficiently convert the kinetic energy of tidal currents into electrical energy.