This　paper　proposes　a　deep　learning　algorithm　to　diagnose　ship　faults　in　order　to　improve　the　accuracy　and　diagnostic　efficiency　of　ship　fault　diagnosis.　90%　of　the　large　number　of　unlabeled　ship　operational　data　samples　are　selected　for　model　training,　and　the　remaining　10%　is　used　for　model　testing.　We　optimize　model　parameters　and　improve　the　accuracy　of　the　deep　learning　model　for　fault　diagnosis　classification.　Hidden　layer　functions　are　used　to　extract　multilayer　data　features　and　perform　feature　fusion.　Gain　values　are　used　to　define　ship　faults,　primary　faults,　secondary　faults,　and　tertiary　faults.　Finally,　we　use　the　soft-max　classifier　to　fault　output　the　fault　and　get　the　fault　level　output.　The　experimental　results　on　the　ship　simulation　fault　dataset　show　that　compared　with　other　traditional　algorithms,　the　accuracy　of　the　fault　diagnosis　of　this　method　is　greatly　improved,　and　the　simulation　result　is　92.5%.　Experiments　show　that　the　model　based　on　deep　learning　algorithm　for　multi-layer　feature　fusion　training　can　better　meet　the　needs　of　ship　fault　diagnosis　under　complex　systems.