【摘要】 为了提高故障电缆测距模型的预测精度,对已提出的基于零序直流原理的模型进行理论分析,得出消除电网电压波动和过渡电阻对检测电流的影响的测距方法。利用神经网络的联想记忆功能,对模型中不易测得且会发生变化的量进行动态网络辨识;同时根据电缆分支的多样性,引入了并行神经网络和基于减聚类的模糊C均值聚类算法,避免聚类中心陷入局部最优。建立多分支故障电缆距离预测动态模型,通过仿真表明该模型具有良好的预测效果。更多还原

【Abstract】 To improve the prediction accuracy of fault cable location model based on zero sequence DC, a way which can eliminate the effect of the power grid voltage fluctuation and transition resistance on the detective currents is proposed. Then, the neural network associative memory function is used to dynamically identify parameters that are difficult to measure and prone to changes. Moreover, due to the diversity of power grid, parallel neural network and fuzzy c-means algorithm based on subtractive clustering are introduced to avoid clustering centers to be trapped into local optima. Finally, the location prediction dynamic model is established for multi-branch fault cables. Simulation shows the above-mentioned model has good predictive performance.更多还原