基于极限学习机与概率神经网络的接地网故障诊断

发布时间：2018-04-25 12:06

本文选题：接地网 + 故障诊断　；参考：《湖南大学》2014年硕士论文

【摘要】：发、变电站接地网是电力系统中不可或缺的重要组成部分，接地系统的工作状态直接影响到工作人员的安全、电力系统的稳定运行和电气设备的正常工作。我国主要采用碳钢作为接地体，由于长年埋于地下，接地网导体的腐蚀造成接地网电气性能参数恶化，严重时直接危及电力系统安全运行，因此，研究接地网故障诊断具有重要意义。鉴于此，本文对接地网模型的构建、基于极限学习机的接地网故障定位方法以及应用概率神经网络的接地网腐蚀程度的识别等进行了较深入的研究。本文采用纯电阻线性网络的接地网稳态模型，提出了基于RBF神经网络的接地网故障诊断方法。以支路断裂时的可及节点电压为训练样本对网络进行训练，只需要把待诊断接地网的可及节点电压送入训练好的网络机进行诊断，根据输出结果即可定位断裂支路。仿真结果表明，此方法在定位接地网断裂支路上是可行的。在此基础上，将极限学习机引入接地网故障诊断。对诊断结果进行分析可知此方法解决了采用RBF神经网络的单故障诊断结果误差较大的问题，且受故障位置与激励位置的影响较小，是一种诊断准确率高且稳定性强的方法。同时，考虑到双故障时训练样本的庞大与输入时的误差，对极限学习机进行了改进，即通过观察识别率，排序最有可能的几条故障支路，并引入白噪声扰动。诊断结果表明，改进后的极限学习机不需要获取庞大双故障训练样本，，仅利用单故障训练样本就能较为准确地定位双断点故障，大大提高了诊断效率，并且对输入误差具有较大的相容度。针对接地网故障模式较多分类困难的问题，提出了结合主元分析(PCA)与概率神经网络（PNN）的接地网故障诊断方法。把不同故障模式下的接地网可及节点电压分别先后送入PCA与PNN进行网络训练，根据PNN的输出结果来识别接地网的故障模式。其结果与利用BP神经网络进行接地网的腐蚀程度识别的结果相比较，表明此方法具有更高的故障识别率，更少的收敛步数与更短的训练时间，是一种快速、准确的接地网故障识别方法。
[Abstract]:Substation grounding network is an indispensable part of power system. The working state of grounding system directly affects the safety of staff, the stable operation of power system and the normal operation of electrical equipment. Carbon steel is mainly used as grounding material in China. The corrosion of grounding grid conductors results in the deterioration of electrical performance parameters of grounding grid, which directly endangers the safe operation of power system. It is of great significance to study the fault diagnosis of grounding grid. In view of this, the construction of grounding grid model, the fault location method of grounding grid based on ultimate learning machine and the identification of corrosion degree of grounding grid based on probabilistic neural network are deeply studied in this paper. In this paper, a fault diagnosis method of grounding grid based on RBF neural network is presented, which is based on the steady-state model of pure resistive linear network. When the reachable node voltage is used as the training sample to train the network, only the reachable node voltage of the grounding network to be diagnosed is sent to the trained network machine for diagnosis, and the broken branch can be located according to the output results. The simulation results show that this method is feasible in locating the fault branch of grounding grid. On this basis, the ultimate learning machine is introduced into fault diagnosis of grounding grid. The analysis of diagnosis results shows that this method solves the problem of large error in single fault diagnosis using RBF neural network, and is less affected by fault location and excitation position. It is a method with high diagnostic accuracy and strong stability. At the same time, considering the large number of training samples and the error of input, the paper improves the extreme learning machine, that is, by observing the recognition rate, ranking the most likely fault branches, and introducing white noise disturbance. The diagnosis results show that the improved extreme learning machine does not need to obtain a large number of double fault training samples, and the single fault training sample can be used to locate the double breakpoint fault accurately, which greatly improves the diagnosis efficiency. And the input error has a large degree of compatibility. In order to solve the problem that the fault pattern of grounding grid is more difficult to classify, a fault diagnosis method of grounding grid based on principal component analysis (PCA) and probabilistic neural network (PNN) is proposed. The voltages of reachable nodes in different fault modes are fed into PCA and PNN respectively for network training, and the fault modes of grounding grids are identified according to the output results of PNN. The results show that the method has higher fault identification rate, less convergence steps and shorter training time, and it is a kind of fast method, compared with the result of using BP neural network to identify the corrosion degree of grounding grid. Accurate fault identification method of grounding grid.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM862

【参考文献】