基于CNN-GRU的配电网中压馈线合环电流预测研究

Research on Closed-Loop Current Prediction for Medium Voltage Feeder in Distribution Network Based on CNN-GRU

摘要: 为提高传统配电网中压馈线合环电流估算或预测准确性和适应性，本文提出了一种基于卷积神经网络(Convolutional Neural Network, CNN)和门控循环单元(gated recurrent unit, GRU)的配电网中压馈线合环电流预测方法。首先，利用数据采集与监视控制系统获取历史负荷数据、电网结构参数以及运行方式等数据并进行预处理。其次，将预处理后的海量数据按时间滑动窗口构造为连续的特征矩阵作为输入，最后，利用CNN-GRU混合模型建立输入特征与合环电流的映射关系，生成基于CNN-GRU的中压馈线合环电流预测模型，进而实现其回归预测。借助DIgSILENT/PowerFactory和MATLAB 2020a软件，案例分析在贵州某城市配电网中展开，预想场景仿真和3组6条馈线合环试验结果初步表明所提的基于CNN-GRU的数据驱动方法能提升合环电流估算或预测的准确性和适用性，与其他模型CNN、GRU相比R2值分别由79.91%，87.7%提高到99.68%，验证了所提的模型具有较高的准确性和可行性，相关结论与讨论对配电网智能化技术研发有一定参考价值。

Abstract: In order to improve the accuracy and adaptability of estimating or predicting the closed-loop cur-rent of MV feeders in traditional distribution networks, this paper proposes a method of predicting the closed-loop current of MV feeders in distribution networks based on Convolutional Neural Net-work (CNN) and Gated Recurrent Unit (GRU). Firstly, the data acquisition and monitoring control system is used to obtain and pre-process the historical load data, grid structure parameters and operation modes. Secondly, the pre-processed massive data are constructed as continuous feature matrices according to the time-sliding window as input. Finally, the CNN-GRU hybrid model is used to establish the mapping relationship between the input features and the loop current to generate the CNN-GRU-based medium voltage feeder loop current prediction model, and then realize its re-gression prediction. With the help of DIgSILENT/PowerFactory and MATLAB 2020a software, the case study is carried out in a city distribution network in Guizhou Province. The simulation of the envisioned scenario and the results of three sets of six feeder loop closing tests initially show the ef-fectiveness and adaptability of the proposed method, and the related conclusions and discussions are of reference value for the development of intelligent technologies for distribution networks.