Intelligent bidirectional control of combustion phase in dual-fuel engines

The development of intelligent ships imposes increasingly stringent requirements on engine combustion control, a domain in which traditional methods exhibit significant limitations. This study focuses on natural gas engines ignited by a diesel micro-pilot injection strategy and proposes an intelligent, bidirectional combustion phase control framework to achieve end-to-end optimization via a perception-decision-execution pipeline. First, the Gram-Schmidt orthogonalization method is employed to decouple control parameters and quantify their individual contributions to the combustion phase (CA50). This facilitates a three-tiered control strategy: Level 1 establishes baselines for pre-injection and main injection timing; Level 2 compensates for environmental disturbances through intake flow adjustment; and Level 3 optimizes the operating boundaries of fuel parameters. Second, a hybrid GS-DResNet-Boost model, which integrates deep residual networks with gradient boosting, is developed to achieve high-precision CA50 prediction, yielding a mean absolute error (MAE) of 0.099°CA. Furthermore, a reverse parameter recommendation system based on the Optuna framework is designed to further investigate parameter independence and coupling mechanisms. This system inversely derives optimal control parameters from a target CA50 value through a multi-objective search, ensuring stable combustion under dynamic conditions with a recommendation error of ≤0.05°CA. Experimental results demonstrate the model's superior performance across diverse operating conditions and reveal intrinsic relationships between CA50, combustion efficiency, and heat release characteristics. This study provides new methodologies for real-time optimization of intelligent ship power systems and offers theoretical foundations for dual-fuel engine combustion control strategies.