基于神经网络的HVDC系统故障诊断

Neural Network Based Fault Diagnosis in HVDC System

HVDC(高压直流输电)系统对运行时的稳定性要求比较严格,出现故障时要求能及时分辨故障类型并快速恢复。鉴于神经网络具有很强的非线性分类能力,该文研究了多种不同结构的神经网络在HVDC系统故障诊断中的应用,并对其诊断性能进行了分析。仿真结果表明,神经网络能够很好的完成HVDC系统的故障诊断。

HVDC （High Voltage Direct Current） system has to run in a high stability, when a fault comes up, the system should distinguish in time which fault occurs and then recovers quickly. Neural network can map complex input/output by associations learned from pervious experience, so it is suitable to be used in fault diagnosis. Firstly, an HVDC system model is established according to the first benchmark model under PSCAD （Power System CAD）. Its constant dc voltage is 500kV and constant dc current is 2kA. Then six most common faults are selected： Single Line to Ground, Double Line to Ground, Line to Line Fault, 3 Phase, DC Fault and no Fault. Based on this model, we a lot of experiments are carried out Through the analysis, it is found that some variables have great relations with the six faults. The 3 phase ac voltages （Va, Vb and Vc）, dc line current （Idr） and ground current （IG） are used as inputs obtained from fault simulations in the HVDC system. The dc line current measurement is used for detecting dc line faults. The simulation results are well organized and used under MATLAB to train NN. Four different network configurations （BP, RBF, SOM and LVQ） are tried, and they all give a correct classification. From the result, it is seen that that NN is a good method to realize HVDC fault diagnosis. Further research is going on. More faults will be included in our research and it is tried to complete the whole work in real time.