Fault Identification Method Of Coal Mine Power Supply System Based On Improved SDSAE

The safe and stable operation of underground coal mine power supply system is of great significance to maintain the safe production of coal mines and guarantee the life safety of underground workers.A fault identification method based on improved stacked noise-reducing sparse self-encoder is proposed for the problems of indistinct characteristics of various types of faults and electrical quantities between fault lines in underground coal mine power supply systems,large noise content and small data sample volume.Through theoretical analysis,the time-domain steady-state characteristics of neutral-unearthed systems with different types of faults occurring at the same location and different faults occurring on different lines are investigated,and a multi-dimensional data sample construction method containing the time-domain three-phase voltage,current and zero-sequence current after the fault enters the steady state is proposed.An end-to-end fault identification model based on SDSAE network is proposed to establish a mapping relationship between fault data and fault types and fault lines,which avoids the fault identification performance of the model being affected by human factors.In view of the small size of the fault data sample set of the actual power system,the SDSAE model is improved by introducing the method of constraining the loss function by the Marxian distance similarity metric,which constrains the metric distance between the original input data and the reconstructed data through the Marxian distance construction penalty term,so that its fault discrimination capability on the small sample data set can be improved and the intra-class diversity and inter-class similarity between samples can be alleviated to a certain extent.The model is also able to mitigate the impact of intra-class diversity and inter-class similarity on the accuracy of the model.At the same time,the activation function is optimised to avoid gradient disappearance during the model training process.The base data set is obtained through the simulation model of ungrounded neutral coal mine power supply system established based on the electromagnetic transient simulation software PSCAD/EMTDC,in which the influence of different fault conditions on the performance of the fault discrimination model is fully considered,and the data pre-processing operation is used to avoid the sample data with too large a difference affecting the discrimination accuracy of the model.A greedy layer-by-layer pre-training and fine-tuning two-stage training method is used to complete the training of the improved SDSAE model,so as to realize the automatic extraction of fault features and fault recognition function.By means of experiments,the correctness of the improved SDSAE fault discrimination method using the Marxian distance similarity measure constraint is demonstrated,and the fault discrimination accuracy is higher than that of deep neural network algorithms such as traditional SDSAE and SSAE.The paper has 37 figures,6 tables and 76 references.