基于RNN的低压变压器区域日线损率基准测试

输变电线路损耗是输变电阶段的固有现象，其是评价低压变压器区域日线损率的重要指标和基准。当对大数据样本上进行训练时，区域的数量通常非常大，并且线损率数据集包含大量的异常值。为了准确的计算低压变压器区域日线损率，提出了一种具有去噪自动编码器(DAE)多径网络模型的鲁棒神经网络(RNN)方法，利用丢包层、L2正则论和Huber损失函数的优点获得多种不同的输出，并利用比较结果计算出基准值和合理区间，实现了精确评估采样数据集的质量并消除线损率的异常值，从而提高数据检测的稳定性。通过与传统的机器学习模型相比，所提出的RNN具有较好的鲁棒性和准确性。根据所提出的RNN的最终结果，在整个数据点中约有13%的异常值，一个月内线损率无缺失值和异常值的区域仅占20%左右，说明了计电设备可靠性较低。

Reference Test of Daily Line Loss Rate in Low Voltage Transformer Area based on RNN

Transmission line loss is an inherent phenomenon in the transmission and transformation stage, and it is an important index and benchmark for evaluating the daily line loss rate in the low-voltage transformer area. When training on large data samples, the number of regions is usually very large, and the line loss rate data set contains a large number of outliers. In order to accurately calculate the daily line loss rate of low-voltage transformers, a robust neural network (RNN) method with a denoising autoencoder (DAE) multipath network model is proposed. It uses the packet loss layer, L2 regularity, and Huber loss. The advantage of the function is to obtain a variety of different outputs, and use the comparison results to calculate the reference value and a reasonable interval, to achieve an accurate evaluation of the quality of the sampled data set and eliminate abnormal values of the line loss rate, thereby improving the stability of data detection. Compared with traditional machine learning models, the proposed RNN has better robustness and accuracy. According to the final result of the proposed RNN, there are about 13% outliers in the entire data point, and the area with no missing values and outliers in the line loss rate within one month only accounts for about 20%, indicating that the reliability of the metering equipment is low .