Uncertainty Aware Mechanism and Data Fusion Modeling for Ship Air Cooler Prediction

In equipment prediction and health management, mechanistic models face problems with accuracy and stability under complex and changing working conditions. In contrast, data-driven models suffer from the drawback of over-reliance on data. For this reason, mechanism and data fusion-driven methods have emerged, which fully utilize the advantages of mechanism knowledge and data-driven models. Existing fusion-driven methods face the problems of limited expressiveness of data-driven networks and insufficient resistance to data uncertainty. To address these challenges, this paper proposes a fusion-driven modeling approach that takes uncertainty into account, and experimentally validates it in a ship main engine air cooler. First, this paper constructs a high-fidelity mechanism model of the ship air cooler; second, ConvDL is proposed as the data-driven model to enhance the sequence modeling capability; in terms of uncertainty modeling, the cloud model theory is adopted to quantify the uncertainty of the input data; and finally, based on the uncertainty weighting method in multi-task learning, the adaptive weighting mechanism model and data-driven model are contributions. The experimental results show that the proposed method effectively improves the accuracy and robustness of the cooling system prediction model.