基于深度学习方法集成的配-微电网分布鲁棒优化调度

汪奕宏; 刘继春; 邱高; 周昊; 何佩芯

汪奕宏,刘继春,邱高,周昊,何佩芯.基于深度学习方法集成的配-微电网分布鲁棒优化调度[J].电测与仪表,2025,62(6):35-44.
WANG Yihong,LIU Jichun,QIU Gao,ZHOU Hao,HE Peixin.Distributed robust scheduling of distribution-microgrid based on deep learning method integration[J].Electrical Measurement & Instrumentation,2025,62(6):35-44.

基于深度学习方法集成的配-微电网分布鲁棒优化调度

Distributed robust scheduling of distribution-microgrid based on deep learning method integration

DOI：10.19753/j.issn1001-1390.2025.06.004

中文关键词: 深度学习集成配电网多微电网分布式鲁棒机会约束调度优化

英文关键词:deep learning integration, distribution network, multi-microgrid, distributed robustness, opportunity constraint, scheduling optimization

基金项目:国家自然科学基金资助项目(U2066209)

作者中文名	作者英文名	单位
汪奕宏	WANG Yihong	四川大学电气工程学院
刘继春	LIU Jichun	四川大学电气工程学院
邱高	QIU Gao	四川大学电气工程学院
周昊	ZHOU Hao	四川大学电气工程学院
何佩芯	HE Peixin	四川大学电气工程学院

摘要点击次数: 26

中文摘要:

针对配电网与多微电网耦合系统调度中分布式电源出力不确定性以及运行效率低下等问题，提出基于深度学习方法集成的配电网与微电网互联系统分布鲁棒机会约束运行优化调度模型。构建了基于概率输出的支持向量机、贝叶斯神经网络、深度信念网络的微电网可再生能源和负荷不确定性概率集合；建立了D-S(dempster-shafer)证据理论信息集成框架，提出了基于Kappa系数与准确率权重的证据修正方法，进而得到更高精度不确定性概率集合，得出源荷功率的概率分布模糊集。接着，建立配电网与多微电网两阶段滚动调度优化模型，即第一阶段预调度模型和第二阶段实时调控模型，第一阶段以实现配电网与多微电网区域全局运行经济最优为优化目标进行能量预分配；阶段二为实时运行调控阶段，考虑微电网内新能源实时出力的不确定性对分布式电源进行实时调控；两阶段鲁棒经济调度模型采用改进列约束生成方法(column-and-constraint generation, C&CG)和交叉方向乘子(alternating direction multiplier method, ADMM)和结合的内外双环算法分布式求解。仿真结果表明，有效提高了源荷预测不确定性下配-微电网市场安全可靠运行，提高了互联系统的新能源消纳率及经济收益。

英文摘要:

Aiming at the problems such as the uncertainty of distributed power output and the low efficiency of operation in the coupled system scheduling of distribution network and microgrid, an optimized scheduling model of Branch-bar operation with chance constraint based on the integration of deep learning method for distribution network and microgrid interconnection system is proposed. The uncertainty probability set of renewable energy and load of microgrid is constructed based on probabilistic output support vector machine, Bayesian neural network and deep belief network. The D-S evidence theory information integration framework is established, the evidence correction method based on Kappa coefficient and accuracy weight is proposed, the evidence is revised from the output and load power of renewable energy, and then, the uncertainty probability set with higher precision is obtained, and the probability distribution fuzzy set of source load power is obtained. The two-stage rolling scheduling optimization model of multi-microgrid is established, namely, the first-stage pre-scheduling model and the second-stage real-time regulation model. In the first stage, the energy pre-allocation is carried out to achieve the optimal global operation economy of multi-microgrid region. The second stage is the real-time operation control stage. Considering the uncertainty of the real-time output of new energy in the microgrid, the proposed two-stage robust economic dispatching model adopts column-and-constraint generation(C&CG) and alternating direction multiplier method(ADMM) combines the column and constraint generation algorithm and joint target cascade analysis algorithm for distributed solution. The simulation results show that the safe and reliable operation of the distribution-microgrid market can be effectively improved under the uncertainty of the source load prediction, and the new energy consumption rate and economic benefits of the interconnected system can be improved.

查看全文查看/发表评论下载PDF阅读器

关闭