Data-Driven Traction Substations’ Health Condition Monitoring via Power Quality Analysis

Abstract

Electrified railway traction substations are an important part of the transportation system, the health of its operation condition indirectly affects the national economy. Generally, traction substations’ conditions are studied from their power quality, while the nonlinearity of loads and effects from the outside environment are factors mainly affecting the accuracy of condition monitoring. In order to recognize the status of traction substations intelligently and govern them with fast measurements, this paper proposed a data-driven approach for recognizing types of power quality problems, and developed a system with intelligent governance strategies. The proposed approach contains two parts. Firstly, a double discrete Fourier transform (DDFT) algorithm was developed to extract valid feature vectors from power data. Then, a well-known data-driven method, support vector machine (SVM), was applied to build classifiers. Finally, based on classification results, a strategy library for power quality problems was built. Industrial data of a real traction substation in Wuhan, China, was tested for the experiment. Compared with traditional methods, the proposed approach is validated to be useful in improving the classification performance of power quality problems, and fast and effective for governance in traction substations.