基于反向传播神经网络的载流导体本体温度仿真计算

Simulation Calculation for Current-carrying Conductor Temperature Based on Back Propagation Neural Network

在常规的载流导体本体温度测试中,由于传感器不能直接贴近导体而存在一定的测量偏差。为解决此问题,建立了基于反向传播(back propagation,BP)神经网络的开关柜载流导体本体温度计算模型,利用己有载流导体表层温度和环境温度数据训练BP神经网络,无需考虑载流导体本身的物性参数。以额定电压12 kV、载流1.25 kA开关柜出线室电缆为试验对象运行该模型,在迭代240次左右完成训练学习,计算出导体本体温度为65℃。通过与有限元模型计算结果进行对比,验证了所建模型能准确计算载流导体本体的温度,且建模简单、操作方便、可重复使用。进一步提高模型的计算精度是下一步研究的方向。

There isacertainmeasuringdeviation inconventional testing forcurrent-carryingfailure of the sensor in adjoining to the conductor directly. In order to solve this carrying conductor temperature of the switch cabinet based on back propagation （BP） neural network was established whichwas to use existing data of the conductor surface temperature and environmental temperature to while neglect physical proper t y parame ters of the conductor. Taking out let cable of the swit ch cabinet with 12 kV rated volt - age and 1 . 25 kA current as the experimental object for running this model , af ter 240 t imes of interat ion , the t raining s tudy was finished and conductor tempera ture was 65 Y by calcula t ion. By comparing this result with that of the finite element model , it is verified this model can accurately work out temperature of the current-carrying condple ,convenient in operation and reusable. It is pointed out the next research direction is improve calcula tion precision of themodel.