Under the influence of alternating shock loads, the oil of the cutting reducer of cantilever road-header is easy to fail prematurely, which seriously affects the operation of the equipment. Therefore, a fuzzy deep learning model (FDD) based on Dropout layer is proposed to monitor the oil state in real time based on three indicators: viscosity, moisture, and particle number. Firstly, according to each index, the oil state is evaluated fuzzily and divided into four levels. Secondly, the index data is normalized as the input of the deep neural network, and then the ReLU activation function is used to activate the network, and an overfitting neural network is obtained. Then, by using the Dropout layer characteristics, the degree of network fitting is reduced, and genetic algorithms are used to optimize the hyperparameters in the model, thereby building a deep learning network model with high accuracy. Finally, the simulation data is used to train the model, and the actual data is used to verify the model. The results show that the FDD model can quickly monitor the oil state, the av-erage prediction accuracy reaches 97%, and the data loss is only 0.0018. At the same time, it solves the problem that the oil state characterization is difficult due to the inconsistency of multi-index information and the neural network training is difficult with less data in the oil monitoring process.