| 引用本文: | 武新章,郭苏杭,代伟,王泽宇,赵子巍,石博臣,张冬冬.基于特征降维与分块的输电网概率最优潮流深度学习方法[J].电力自动化设备,2023,43(8):174-180 |
| WU Xinzhang,GUO Suhang,DAI Wei,WANG Zeyu,ZHAO Ziwei,SHI Bochen,ZHANG Dongdong.Feature dimension reduction and partitioning based deep learning method for probabilistic optimal power flow of transmission network[J].Electric Power Automation Equipment,2023,43(8):174-180 |
|
| 摘要: |
| 概率最优潮流需要对非线性最优潮流问题进行重复求解，计算量较大，从而限制了其应用。提出一种基于特征降维、分块和深度神经网络辅助预测的最优潮流两阶段求解方法。在第一阶段，提出基于深度神经网络的最优潮流部分关键决策变量的优先辨识策略，以解决深度学习中因特征维度过高而导致的数值湮没问题，进而以最优潮流的结果特征为导向，基于关联性分析和聚类分析挖掘最优潮流输入与输出特征的关联性匹配度，并构建样本数据的分块特征库，以降低学习难度。在第二阶段，利用深度神经网络完成部分关键决策变量的分块映射，基于潮流模型恢复剩余状态变量，并对计算结果不收敛、不满足约束的情况进行修正，以恢复可行性。根据最优潮流两阶段求解方法构建概率最优潮流求解方法。仿真结果表明所提方法在最优潮流、概率最优潮流的求解速度和求解精度上均有较好的表现。 |
| 关键词: 最优潮流 概率最优潮流 深度神经网络 关联分析 聚类分析 潮流修正 |
| DOI：10.16081/j.epae.202302004 |
| 分类号:TM73 |
| 基金项目:国家自然科学基金资助项目(52107082)；广西科技基地和人才专项(桂科AD22080052) |
|
| Feature dimension reduction and partitioning based deep learning method for probabilistic optimal power flow of transmission network |
|
WU Xinzhang, GUO Suhang, DAI Wei, WANG Zeyu, ZHAO Ziwei, SHI Bochen, ZHANG Dongdong
|
|
College of Electrical Engineering, Guangxi University, Nanning 530004, China
|
| Abstract: |
| Probabilistic optimal power flow(POPF) requires repeated solving of the nonlinear optimal power flow(OPF) problem, thus its application is limited by large calculation amount. A two-stage solution method for OPF is proposed based on feature dimension reduction, partitioning and auxiliary forecasting of deep neural network(DNN). In the first stage, a DNN-based priority identification strategy for partial key decision variables of OPF is proposed, which solves the problem of numerical annihilation caused by too high feature dimension in the deep learning, further, guiding by the result characteristics of OPF, the correlation matching degree between input and output characteristics of OPF is extracted based on the correlation analysis and cluster analysis, and the block feature database of sample data is constructed to reduce the learning difficulty. In the second stage, DNN is used to complete the block mapping of partial key decision variables, the remaining state variables are recovered based on the power flow model, and the conditions that the calculation results do not converge or do not satisfy the constraints are corrected to restore the feasibility. A solution method of POPF is constructed according to the two-stage solution method of OPF. The simulative results show that the proposed method has good performance in the solving speed and accuracy of OPF and POPF. |
| Key words: optimal power flow probabilistic optimal power flow deep neural network association analysis cluster analysis power flow correction |
