Structurally constrained hybrid modeling for cross-year prediction and mechanistic interpretation of marine biofouling evolution
Year: 2026
Type: article
Abstract: The stable long-term prediction of marine biofouling is of great significance for the risk assessment of biofouling and the formulation of anti-fouling strategies. However, existing studies still suffer from limitations in prediction targets, cross-year extrapolation capability, and model interpretability. To address these issues, this study constructs a Fouling Severity Index (FSI) based on 2019-2024 shallow-sea immersion experiments conducted in the nearshore waters of Sanya, providing a stable prediction target. On this basis, a Structurally Constrained Evolutionary Regression (SEER) model is proposed, in which an empirical evolutionary trend is introduced as a structural constraint and combined with environment-driven residual modeling to achieve robust prediction of long-term biofouling evolution. Cross-year extrapolation results demonstrate that the SEER model significantly outperforms Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) models in predictive performance. Further ablation experiments and stage-wise analyses confirm the rationality of the model structure and reveal that biofouling responses to environmental factors exhibit pronounced stage dependence. Overall, this study provides a stable and interpretable modeling framework for long-term marine biofouling prediction, offering methodological support for long-term fouling state assessment and risk identification.
Source: Ocean Engineering
Institution: Zhejiang University
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
SDG: Life below water
Open Access status: hybrid
Estimated APC paid: $4,150
OpenAlex ID: W7162096684