Research on the ART-based Case Matching and the Rail Intelligent Data Diagnosis Technology
|School||Nanjing University of Technology and Engineering|
|Keywords||Similar cases matching case retrieval ART neural network rail transport|
Matching method with similar cases (CBR) is a new mode of reasoning in the last twenty years, the growing difference between the rule-based reasoning. It is an important way of problem solving and learning based on the accumulated knowledge, emphasizing the attention of human knowledge and experience accumulated in the past, and the wisdom of our predecessors. Adaptive resonance neural network (ART) is a non-teacher feedback artificial neural network learning, online learning, it makes many of the advantages of mutual cooperation ties between the CBR and ART, in many ways, so that the two are complementary sex. Therefore, the use of ART techniques and models to achieve CBR can obtain good Reasoning effect. This paper details the ART neural network CBR case retrieval, case updates reuse adjust or amend the case, case complementary assessment of learning, and case database creation and maintenance optimization, optimized ART2model, ART2learning principles and algorithm process.Based on the above theory this paper, the urban rail train major equipment status data the ART2detection model, the use of the laboratory diagnosis of bearing test rig simulation generates data reveal a the ART2model for the identification of major equipment train state four structure by different threshold vigilance parameter classification results and comparative test generated a lot of data on the the ART2model to overcome the feed forward neural networks prone to slow convergence, easy to fall into local minimum point disadvantage, in terms of mass data processing With a fast and accurate clustering effect.Meanwhile, a new demand analysis of ideas, the development of urban rail traffic flow model, as well as a new algorithm based the ART2model to predict urban rail traffic congestion-related events, the algorithm traffic flow model and ART2model observer classifier, traffic flow model observation data and estimates of actual traffic data residuals sequence, open classification the ART2model of the residual sequence, to forecast the occurrence and severity of congestion-related events. By simulation test, indicating that the ART2model not only can identify known types of events, will identify an unknown type, a while to work, the edge learning detection system can provide decision support for urban rail transport safety and energy conservation.Finally, to summarize a the ART2model for application in CBR advantages and disadvantages, the next stage of research work to provide a vision.