支持不停电作业的配网可靠性分析算法设计

Algorithm design for reliability analysis of distribution network supporting uninterrupted operation

为了提升配电网络的负荷预测精度,并为配网不停电技术提供可靠性分析的基础,文中对深度学习的时间序列处理算法进行了研究。针对传统循环神经网络(RNN)中神经元结构简单、无法储存长时间序列的结构信息等问题,引入了一种改进的长短期记忆(ILSTM)单元。通过改进算法的门结构来控制状态单元在隐藏层中的信息结构流向,有效避免了迭代过程中时间信息丢失的问题。对于当前算法中参数初始化方面的缺陷,采用了一种自适应惯性权重粒子群算法(APSO),优化了网络的训练效率与稳定性。结合某低压配电网的自动化改造项目对不同负载下的负荷状态加以采集,并基于实测数据对网络进行了训练和测试。仿真结果表明,与RNN网络相比,所提算法的指标均较为理想,且MAE、MAPE分别提升了42.95%和0.07%。

In order to improve the accuracy of load forecasting in distribution network and provide a basis for reliability analysis of distribution network uninterruptible technology,the time series processing algorithm of deep learning is studied in this paper.Aiming at the shortcomings of traditional Recurrent Neural Network(RNN),such as the simple structure of neurons and the inability to store the structural information of long time series,an Improved Long Short-Term Memory(ILSTM)unit is introduced.The unit controls the flow direction of the information structure of the state unit in the hidden layer through the gate structure,which effectively avoids the problem of time information loss in the iterative process.For the difficulty of parameter initialization in the current algorithm,an Adaptive Particle Swarm Optimization(APSO)algorithm is used to optimize the training efficiency and stability of the network.Combined with the automation transformation project of a low-voltage distribution network,the load states under different loads are collected,and the network is trained and tested based on the measured data.The simulation results show that the performance of the proposed algorithm is improved compared with RNN network,in which MAE and MAPE are improved by 42.95%and 0.07%respectively.