An Enhanced Harris Hawks Optimization Algorithm for GRU Hyperparameter Tuning: Application to Mooring System Tension Prediction

The accuracy of tension prediction for mooring systems affects the safety and efficiency of production planning for offshore platforms as well as ship systems. With the development of deep learning, tension prediction based on neural networks has been widely used and has shown excellent performance. However, the performance of neural networks depends largely on the setting of hyper-parameters, and frequent manual adjustments of hyper-parameters are required in use to obtain the best neural network performance results, which is undoubtedly time-consuming and affects the performance of the task at the same time. To address this problem, this paper proposes a Harris-Hawkes optimization algorithm (ADLS-HHO) based on an adaptive two-layer search strategy and an improved Levy flight mechanism to automatically optimize the GRU hyperparameters. Specifically, ADLS-HHO employs an adaptive two-layer search framework to enhance the dynamic balance between exploitation and exploration, and Levy flights to enhance the global exploration capability while guiding the local search strategy through a dynamically maintained memory pool. Experimental results show that the prediction accuracy of the proposed ADLS-HHO-GRU model is better than that of the conventional model, and in addition, the number of convergence iterations required by ADLS-HHO-GRU is less than that of the existing classical algorithms. These results indicate that ADLS-HHO-GRU provides an effective method while maintaining computational efficiency.