Research On Protection Of Traction Transformer And Controllable High Voltage Reactor Based On Random Forest Algorithm

The development of large modern power grid puts forward higher requirements for fault diagnosis and relay protection. Traditional relay protection mainly uses some feature quantity to judge and has certain limitation. At the same time, it’s possible to use some advanced algorithms for fault diagnosis and relay protection with the continuous upgrading of hardware and software.A data-mining model named random forest algorithm used for fault diagnosis and relay protection is studied in this dissertation. Firstly, the random forest algorithm is introduced. Then, types of failure in oil-immersed equipment were diagnosed by random forest based on the analysis of dissolved gas in oil. Seven combinations of original gas concentration and six characteristic gas ratio methods proposed in some literatures are used as the input of random forest. The result of random forest diagnosis is good and some conclusions are obtained, the best combination of characteristic parameters is found.This dissertation studies the differential protection of traction transformer and establishes the V/X wiring traction transformer model in Simulink. Various simulation cases consisting of inrush current, internal faults and other situation have been simulated. This dissertation proposes the differential protection scheme of traction transformer based on random forest. The original waveform data of differential current is used as the inputs of random forest. At the same time, this dissertation studies the backup protection named overcurrent protection started by low voltage that commonly used in traction transformer, uses the original waveform data of characteristics used in overcurrent protection started by low voltage as the inputs of random forest, and proposes backup protection scheme of traction transformer based on random forest. Herein, random forest is trained and tested by using data acquired from simulation, and the accuracy of random forest classification is high.The random forest algorithm is also applied to the protection of magnetically saturation controllable reactor (MSCR). The basic structure of MSCR is briefly introduced and the MSCR model that combining working winding and control winding is established in Simulink. In the MSCR protection scheme based on random forest proposed, the combination of the original voltage and current waveform data is used as the inputs of random forest. Simulink model is used to obtain data to train and test the random forest. The accuracy of random forest classification shows good results.The protection scheme based on random forest identifies the fault status by mining waveform feature. It can simplify the protection configuration. The accuracy of random forest classification is high. Some minor faults can be recognized accurately.