A data-driven approach for high-speed train interior sound quality prediction using WOA-EEMD and NOA-XGBoost

High-speed trains are one of the important public transportation methods in China. As the number of passengers increases, the importance of interior noise comfort has gained attention. Traditional methods for subjective evaluation of interior noise often suffer from low efficiency and high time consumption. To overcome these limitations in assessing the interior sound quality of high-speed trains, this study introduces a comparative grade scoring method for subjective evaluation. Furthermore, it proposes a feature extraction approach based on ensemble empirical mode decomposition (EEMD) optimized by the Whale Optimization Algorithm (WOA), a bio-inspired metaheuristic that simulates the bubble-net hunting strategy of humpback whales, to enhance feature extraction. The extracted features are then integrated into an Extreme Gradient Boosting (XGBoost) model, whose hyperparameters are fine-tuned using the Nutcracker Optimizer Algorithm (NOA), a recently developed nature-inspired algorithm that mimics the seed storage and retrieval behaviors of nutcrackers to achieve global optimization. The results indicate that the proposed comparative grade scoring method enables accurate subjective evaluation of interior sound quality in high-speed trains. The intrinsic mode functions (IMFs) obtained through the WOA-EEMD approach preserve the frequency features of the original signal, confirming that interior noise in high-speed trains is primarily composed of mid- and low-frequency components, which exhibit strong correlations with subjective perception. The XGBoost model optimized using the NOA algorithm, when combined with loudness, sharpness, and other sound quality features, demonstrates high accuracy in predicting interior sound quality.