Identification of ice loads on offshore wind turbines in cold regions using RBFNN-GA

Accurate identification of ice loads is crucial for ensuring the structural safety and service life of offshore wind turbines (OWTs) operating in cold regions. Although full-scale measurements are considered the most reliable method for obtaining ice loads, their long-term application is limited by high costs, harsh environmental conditions, and the complexity of ice–structure interactions. To address this challenge, this study proposes a hybrid machine learning method that combines a radial basis function neural network (RBFNN) with a genetic algorithm (GA) to identify ice loads based on structural dimensions and environmental parameters. High-fidelity data are first compiled from full-scale measurements, physical model tests, and numerical simulations. Input features are then selected through empirical analysis and correlation evaluation. Based on this, an RBFNN model is developed, with its architecture and parameters globally optimized using a GA. To verify the model’s accuracy and engineering applicability, ice–structure collision experiments are performed in a wave flume, and the predicted results are compared with the measured data. The results indicate that pile diameter, ice thickness, and ice drift velocity are key input features influencing the model’s prediction performance. Optimization using the GA significantly improved both accuracy and generalization, reducing the average error by about 28% compared to the traditional RBFNN model. The method proposed in this paper provides an efficient and scalable solution for the inverse identification of ice loads on OWTs in cold regions, addressing the accuracy and stability limitations of traditional inversion techniques.