With the increasing number of electrical equipment used in ships, the requirements for ship power system are also getting higher and higher, so the stability of ship power system becomes more and more important. As the core of ship power system, ship microgrid has been specially studied worldwide, and the identification of dc power quality disturbance characteristics is the key to improve the dc power quality in ship microgrid. In this study, a new improved hybrid CLSTM deep learning model based on spatio-temporal fusion is proposed for real-time classification and identification of DC power quality disturbances in ship microgrid. And through the simulation of DC voltage deviation, DC voltage fluctuation, DC voltage ripple, dynamic voltage sag, line fault sag and normal six categories of DC power quality disturbance signals for ablation experiments, this study proposes an improved deep learning CLSTM model. Compared with the single LSTM model and CNN model after ablation, the average recall rate, average accuracy rate and total accuracy rate increased by 3.8%, 9.2%, 8.8% and 1.9%, 3.7%, 4.3%. Compared with other latest classification models proposed in other latest references, CLSTM model has also improved in accuracy. Therefore, the CLSTM model proposed in this paper is the ship micro grid DC power quality disturbance identification system provides a new type of improved deep learning model with ultra-high accuracy, stronger robustness and better stability.