AVOA-optimized CNN-BILSTM-SENet framework for hydrodynamic performance prediction of bionic pectoral fins

Abstract

The hydrodynamic performance of bionic pectoral fins is crucial for improving Manta Ray robots' propulsion efficiency. However, traditional CFD methods are computationally expensive and inefficient. To address this, we propose AVOA-CNN-BiLSTM-SENet, a novel framework for low-cost, real-time hydrodynamic performance assessment. This framework leverages African vulture optimization algorithm (AVOA) to optimize the hyperparameters of a hybrid model that combines Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory networks (BiLSTM), and Squeeze-and-Excitation Networks (SENet), enhancing multi-scale feature extraction and reducing overfitting. Firstly, an Optimal Latin Hypercube Design (Opt-LHD)-based experimental design is formulated, and a hydrodynamic dataset is generated via CFD simulations. Secondly, the dataset is partitioned for model training and optimization within the proposed AVOA-CNN-BiLSTM-SENet framework. For performance evaluation, the trained model is quantitatively analyzed and benchmarked against four models: support vector machine (SVM), random forest (RF), gradient boosted decision tree (GBDT), and backpropagation neural network (BPNN). Experimental results show that AVOA-CNN-BiLSTM-SENet framework outperforms all benchmarks in predicting hydrodynamic performance indicators of bionic pectoral fins, with a maximum relative error of no more than 12.4 %. Therefore, the model exhibits superior prediction accuracy and generalization capability, demonstrating strong potential for real-world hydrodynamic applications.