To address the challenge of quantifying the fire risk associated with 10kV tree faults(TF),the paper analyzes and validates the effectiveness of multidimensional zero-sequence current characteristics in estimating the distribution of flames in TF.Utilizing real-world fault experiment data,a feature vector of zero-sequence current characteristics,including the effective value,mean value,fundamental wave,and the mean values and variances of each harmonic,is constructed to develop a neural network-based model for TF flame distribution estimation.The model is trained on a dataset of 8,170 samples,of which 80%are used for training and 20%for testing,and evaluated through mean squared error and correlation coefficient.The results demonstrate that the neural network model can estimate the flame distribution with high accuracy,with absolute estimation errors within±0.05 and±0.1 reaching 79.99%and 93.27%respectively.Experiments that systematically exclude each feature confirm the positive contribution of the selected features and the robustness of the overall feature set.The findings provide a new research approach for the quantitative estimation of the spatiotemporal distribution of flames in overhead power line TF.
关键词
树木故障/零序电流/炭化路径/神经网络/明火分布估计
Key words
tree fault/zero-sequence current/carbonization path/neural network/flame distribution estimation
维普
万方数据