<p>In-cylinder pressure is a key indicator of the combustion process and operating conditions, as it contains rich information about an engine’s state. However, traditional measurement methods are constrained by limited sensor installation space, high maintenance cost, and reduced reliability under high thermal loads and intense vibration environments, making long-term stable operation in complex environments challenging. A cylinder pressure reconstruction method based on multi-source information fusion is proposed to address the aforementioned challenges. The engine’s operating characteristics are comprehensively analyzed by integrating cylinder head vibration, crankshaft torsional vibration, and thermal parameters. Moreover, a neural network model is developed to achieve high-precision reconstruction of cylinder pressure across the full range of operating conditions. Results of the real-machine validation show that the proposed method can accurately reconstruct the cylinder pressure curve under all working conditions, and the average errors of the peak pressure and its corresponding phase are 1.63 % and 0.44 %, respectively. The results validate the effectiveness of the multi-source information fusion strategy under complex working conditions and provide a new technical approach for non-invasive cylinder pressure measurement.</p>