The health state of the fuel system of an internal combustion engine has an important impact on the performance of the whole engine, but the current internal combustion engine dispenser monitoring parameters have not yet been fully utilized. For this reason, this paper proposes an intelligent diagnostic method based on the fusion of deep learning and anomalous information, which realizes the utilization of the existing dispenser monitoring parameters and the anomalous phenomenon information, leading to improved reliability of the internal combustion engine operation. Mutual information theory is first introduced to realize the automatic grouping of the dispenser monitoring parameters, and a deep learning diagnostic model is constructed for the dispenser monitoring parameters of the internal combustion engine by using the denoising autoencoder and an attention mechanism in combination with a bidirectional gate recurrent unit. This affords a preliminary intelligent diagnosis of the typical faults of the fuel system. Subseguently, considering the auxiliary value of anomalies acquired during the actual operation of the internal combustion engine for fault diagnosis, a Bayesian network is constructed and a Leaky-Noisy-Or model is used to quantify the correlation between anomalies and specific faults, thus optimizing the results of the intelligent diagnosis of faults. Finally, the fuel system fault sample dataset obtained from GT-Power simulation is substituted into the model, and the diagnostic results verify the effectiveness of the proposed method in improving the accuracy of fuel system fault diagnosis in internal combustion engines. This provides a deep learning intelligent diagnosis model based on the monitoring parameters of the dispenser, and also provides a new information fusion pathway for fuel system fault diagnosis, and has important practical application value for the intelligent diagnosis of internal combustion engines.