Aiming at the randomness and complexity of jointless track circuit faults, the single diagnosis model has the problems of one-sided extraction features and unreasonable empirical design of model structure. A fault diagnosis method based on intelligent optimization deep network is proposed. Firstly, the fault feature set is established by six voltage detection quantities of the track circuit signal centralized monitoring system. The convolutional neural network (CNN) is used to extract the feature space information, and the long short-term memory network (LSTM) is used to extract the time feature information, so that the features extracted by track circuit fault diagnosis have both spatial and temporal information. At the same time, the genetic algorithm (GA) is introduced to optimize the structure and parameters of the aforementioned deep neural network, and the output weight of the feature level of the two combined networks is further optimized by combining the Q-learning method in reinforcement learning. Finally, the multi-layer perceptron (MLP) is used to fit and correct the classification error of the deep network to improve the fault diagnosis accuracy of the model for the track circuit. The simulation results show that the recognition rate of the fault diagnosis of the track circuit using the intelligent optimized deep network model can reach 99.28% compared with the single model and the refined design combination model. The evaluation index is improved, and the fault diagnosis accuracy is higher. It is proved that the intelligent optimized deep network can further improve the fault diagnosis performance of the track circuit.