多负载回路串联型故障电弧智能诊断及选线方法研究

Research on Intelligent Diagnosis and Route Selection Method of Multi-load Circuit Series Fault Arc

低压用电系统线路结构复杂,负载形式多样,因线路老化、接触不良等原因发生串联型故障电弧时,现有的故障电弧断路器通常只能对单相单负载回路进行保护,难以检测到下级线路发生的故障。因此,在AC 220 V下搭建了家用多负载回路串联型故障电弧试验平台,开展各支路及干路故障下的故障电弧试验,获得不同工况下发生故障电弧时的干路电流数据,并将其作为样本建立模型样本库来训练建立的一维卷积神经网络。经过大量训练,建立的家用多负载回路串联型故障电弧诊断及选线模型在验证集上的准确率达到了98.6%。随后使用测试集中的连续电流数据对模型的鲁棒性、抗扰性及泛化性进行测试,最后将训练好的模型导入树莓派中进行运行时间测试,结果表明其检测时间满足UL标准的要求,但在投入实际应用前,仍需要对模型做进一步优化使得其更适合在嵌入式系统上运行。

The circuit structure of the low-voltage power system is complex,and the load forms are various.When a series arc fault occurs due to line aging,poor contact,etc.,the existing arc fault circuit breaker can usually only protect the single-phase single-load circuit,and it is difficult to detect the lower level line failure.Therefore,a household multi-load circuit series arc fault experiment platform is set up under AC 220 V to carry out arc fault experiments under various branch and trunk faults,and obtain the trunk current data when fault arcs occurred under different working conditions,which are used as samples to establish a model sample database to train the established one-dimensional convolutional neural network.After a lot of training,the established household multi-load circuit series arc fault diagnosis and line selection model has an accuracy rate of 98.6%on the validation set.Then,the continuous current data in the test set is used to verify the robustness,immunity and generalization of the model,and finally the trained model is imported into the Raspberry Pi for running time test.The results show that the detection time meets the requirements of the UL standard.However,before it is put into practice,the model still needs to be further optimized to make it more suitable to run on embedded system.