Deep Learning Algorithm for Preliminary Siting of Substations Considering Various Features in Distribution Network Planning

2018 2nd International Conference on Power and Energy Engineering

Feng, Liang^1 ; Cui, Can^1 ; Ma, Runze^2,3 ; Wu, Jian^1 ; Yang, Yang^1 ; Zhang, Xiaolei^4 ; Liu, Shengyuan^2

Economic and Technology Research Institute, State Grid Shandong Electric Power Company, Jinan, China^1

College of Electrical Engineering, Zhejiang University, Hangzhou, China^2

Powerchina Huadong Engineering Corporation Limited, Hangzhou, China^3

State Grid Shangdong Electric Power Company, Jinan, China^4

关键词: Convolutional Neural Networks (CNN);    Distribution network planning;    Input features;    Multichannel data;    Planning process;    Power grid operations;    Quality of power supply;    Siting and sizings;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/192/1/012032/pdf
DOI  :  10.1088/1755-1315/192/1/012032

来源: IOP

【 摘 要 】

Substation siting and sizing planning is one of the important contents of distribution network planning, which directly affects the results of subsequent distribution network planning, and affects the quality of power supply and the economy of power grid operation. Given this background, a deep learning algorithm for preliminary siting of substations in distribution network planning is proposed in this work. Features related to the principle of siting of substations are extracted and multichannel data characterization are utilized. Then, the features are integrated into a convolutional neural network (CNN), which is one of deep learning algorithms, based on actual geographical relationships. Next, the preliminary siting of substations for the subsequent planning process is completed. Finally, the validity of the proposed algorithm considering different input features is demonstrated on a distribution network of one certain province in China by case studies and comparisons. The simulation results show that the proposed deep learning algorithm for preliminary siting of substations is more accurate with more input features, and is better than shallow learning algorithms, thus can be employed to preliminary siting of substations in distribution network planning.