生西奎,付强,于洋,吴昊,杨凌.基于深度学习GRU网络的配电网理论线损计算方法[J].电测与仪表,2021,58(3):54-59. Sheng Xikui,Fu Qiang,Yu Yang,Wu Hao,Yang Ling.Distribution network line loss calculation methodbased on deep learning GRU network[J].Electrical Measurement & Instrumentation,2021,58(3):54-59.
基于深度学习GRU网络的配电网理论线损计算方法
Distribution network line loss calculation methodbased on deep learning GRU network
由于实际应用中对理论线损的精度要求不高，且传统计算方法所需电气参数较多，计算过程繁琐。因此，文中提出了一种基于深度学习门控循环单元(Gated Recurrent Unit, GRU)网络的配电网理论线损计算方法。首先，为了综合考虑主客观因素，将互信息理论和层次分析法相结合，进而确定所选电气参数对理论线损影响的权重；然后，按照权重的大小构建不同的输入参数集，通过分析选用不同电气参数集时GRU网络的计算误差，确定最优输入参数，组成样本数据集并训练GRU网络；最后，以某地区10 kV配电网为例验证了文中所提方法，分析结果表明，文中所提方法可以代替等值电阻法，更加快速、便捷地对理论线损进行计算。文中所提方法还与传统BP算法进行了比较，结果表明文中所提方法具有更好的计算性能。
英文摘要:
The accuracy of theoretical line loss calculation is not high in practical application, in addition, calculation method requires many electrical parameters and the process is tedious. Therefore, the paper proposes a theoretical line loss calculation method based on deep learning gated recurrent unit (GRU). Firstly, considering the subjective and objective factors, the correlation between the electrical parameters and the theoretical line loss is calculated by the combination of mutual information and analytic hierarchy process. Then, different input parameter sets were set according to the size of weight. Through the analysis of the calculation error of the GRU network, the optimal input parameters are determined and form the sample data of training the GRU network. Finally, this method is verified by taking a 10 kV distribution network as an example. The results show that this method can replace the equivalent resistance method and calculate the theoretical line loss more quickly and conveniently. Compared with the traditional BP algorithm, this method has better performance.
