Fault diagnosis of track circuit based on improved sparrow search algorithm and Q-Learning optimization for ensemble learning

Due to the complexity and randomness of jointless track circuit faults, it is difficult to improve the performance evaluation index of fault diagnosis with a single model. There are several problems for the use of ensemble learning, such as the structure and parameters of each base learner are blind design. The weights of each base learner combination are difficult to be assigned in the ensemble model. In order to address these issues, a new fault diagnosis approach of track circuit based on the improved sparrow search algorithm and Q-Learning optimization for ensemble learning was proposed. The proposed method can organically combine ensemble learning with computational intelligence and reinforcement learning to fully exploit the fault characteristics of track circuit and improve the performance evaluation index. Firstly, the use of convolutional neural networks, long and short-term memory networks and multilayer perceptron deep learning models, as well as support vector machines and random forests traditional machine learning models, together constitute an ensemble learning base learner, which addresses the shortcomings of a single learning model and ensures the diversity of ensemble learning. From the perspective of automated machine learning, the improved sparrow algorithm is used to optimize the structure and parameters of ensemble learning model to overcome the problem that its structure and parameters are difficult to determine. On this basis, reinforcement learning Q-learning was introduced to optimize the combined weights of each base learner in the ensemble model. The combined weights of each base learner of ensemble learning were determined intelligently. Finally, the error was obtained by comparing the prediction results of the ensemble learning model with the real results. The BP neural network was used to compensate and correct the prediction results, which further improves the fault diagnosis performance evaluation index of track circuits. The simulation experiments show that the performance evaluation indexes such as accuracy, precision, recall and F1 value of track circuits fault diagnosis are further improved by using our proposed method. © 2023, Central South University Press. All rights reserved.