Research On Damage Prediction For Insulation Joints Based On Rough Set Theory And Multi-class Support Vector Machines

Insulation joints are an important part of the track circuit. When insulation joints are damaged, traffic accident is possible to be happened, causing economic loss, even endangering the safety of the lives and property. Therefore, it is very necessary to regularly test and judge degrees of insulation joints to make the field technicians understand different damaged degrees of insulation joints and take reasonable maintenance and replacement, both to ensure traffic safety and reduce maintenance costs. Based on track circuit test data, rough set theory and support vector machine(SVM) algorithm are applied to damage prediction for insulation joints to classify the current insulation degrees. Moreover, according to the various states, some maintenance advices are given. The main contents are as follows.Firstly, the basic theory of insulation joints is deeply studied. The electrical characteristics and influence factors of the ends of insulation joints are analyzed based on track circuit test data. Rack circuit voltage, limited resistor voltage, sending voltage, rail joints, broken trough, rail voltage, insulation, inside voltage and outside voltage are selected as condition attributes, which are extracted to complete feature extraction. And according to the insulation state decision rules, insulation damage information system is constructed.Secondly, in allusion to continuous and non numeric attributes, an information entropy discretization algorithm is applied to get discrete data. Meanwhile, an improved principal component heuristic reduction algorithm is proposed to attribute reduction of discretization information system to get the optimal reduction set, achieving data dimensionality reduction and the characteristics of the secondary extraction. The simulation results show that the prediction indexes of insulation joints are advanced by reduction.Thirdly, a hybrid sampling balanced algorithm is proposed to solve the unbalanced samples issue. The simulation results show that balanced data sets can improve the prediction precision of the classifier.Finally, a damage prediction model for insulation joints is proposed, which is based on rough set theory and multi-class support vector machines(MSVM). According to track circuit test data of a signalling depot to instantiated prediction, and compared with the damage prediction model for insulation joints based on MSVM, the simulation results show that rough set theory and SVM algorithm are applied to damage to insulation joints classification prediction has a higher prediction accuracy.