The Study Of Fault Diagnosis In Intelligent Substations Based On Data Fusion

Due to the shortage of energy and the increasing demand for energy and other issues have become increasingly prominent,the Chinese government is vigorously developing smart grid,and the intelligent substation construction which is the core of intelligent power grid is also developing rapidly.Considering that the substation is the hub of the power system,the fault diagnosis function of smart substation has a particularly important role in forming a safe and self-healing power network.Firstly,the theoretical and technical methods of fault diagnosis about smart substation at home and abroad are summarized in this thesis,and their respective scope and limitations are pointed out.Then,the data fusion fault diagnosis method based on BP neural network algorithm and DS evidence theory is put forward according to the architecture of smart grid.Next,considering that the fault diagnosis of the power network is only based on the protection and the circuit breaker action information of the primary system before,the electrical quantity is introduced into the fault diagnosis in this thesis.In the meantime,taking into account the new characteristics of the intelligent substation,the secondary system information are fully utilized to researching the data fusion of the multi-information source,And it overcomes the defects that the traditional substation can only get primary system of information to diagnose.Finally,a conclusion that there is a certain temporal relationship between the protection and the action information of the circuit breaker is drawn by analyzing the timing characteristics of the alarm information,and due to the factors such as refusing,malfunctioning and loss of information,the result of the fault diagnosis may be greate impact.In view of this,a method which uses a cyclic neural network insteading of BP neural network is presented in this thesis.The simulation results show that the BP neural network algorithm is effective when the substation action information does not coincide with the timing,and the conclusion that the cyclic neural network diagnosis considering the timing attribute resulting in a significant improvement in fault tolerance is proved through the comparison of the two kinds of neural networks.