Construction of digital twin system for in-cylinder combustion and emission of marine engine

Digital twin technology is essential for marine diesel engine optimization, yet purely mechanism-driven methods require excessive computational time and resources, restricting rapid prediction and optimization. This study proposes a hybrid mechanism-data-driven digital twin modeling approach for accurate combustion prediction and simultaneous online optimization of combustion and performance parameters. A complete combustion model from Wiebe parameters to cylinder pressure/heat release rate profiles is first established based on a mechanistic model. For mapping operating parameters to Wiebe parameters, a Snake Optimizer-optimized convolutional bidirectional LSTM (SO-CNN-Bi-LSTM) network is developed using bench test data. This deep learning model is organically integrated with the mechanistic model to construct the digital twin prediction model. For online optimization, the prediction model serves as a virtual engine, employing a closed-loop collaborative combustion strategy and Multi-Objective Snake Optimizer (MOSO) to optimize combustion and emission performance, followed by experimental validation. Results demonstrate that the hybrid-driven model accurately reconstructs and predicts diesel engine combustion processes and output characteristics. The closed-loop collaborative strategy significantly suppresses fluctuations in engine speed and pressure rise rate. Meanwhile, MOSO successfully controls NOx emissions within Tier-III standards. Compared to traditional methods, the proposed framework enables online combustion optimization while maintaining high prediction accuracy.