Open AccessA deep learning approach for power system knowledge discovery based on multitask learning
Author(s) -
Huang Tianen,
Guo Qinglai,
Sun Hongbin,
Tan ChinWoo,
Hu Tianyu
Publication year - 2019
Publication title -
iet generation, transmission and distribution
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.92
H-Index - 110
eISSN - 1751-8695
pISSN - 1751-8687
DOI - 10.1049/iet-gtd.2018.5078
Subject(s) - deep belief network , computer science , deep learning , artificial intelligence , task (project management) , machine learning , electric power system , artificial neural network , multi task learning , process (computing) , stability (learning theory) , transfer of learning , power (physics) , engineering , physics , quantum mechanics , operating system , systems engineering
Power system security assessment is an important and challenging problem. Large variations in loads and power generation present increased risks to the secure operation of power systems. This study proposes a distributed deep network structure for power system security knowledge discovery based on multitask learning to monitor and control power grids more properly and effectively. First, a deep neural network structure based on the deep belief network (DBN) is designed to non‐linearly extract deep and abstract features layer‐by‐layer for total transfer capability (TTC) regression tasks. Then, a distributed training algorithm for the deep structure is developed to accelerate the training process. Furthermore, multitask learning is adopted by grouping and training‐related tasks together to improve the task performance. Finally, the accuracy and efficiency of the deep structure are evaluated using the Guangdong Power Grid in China. The simulation results demonstrate that the proposed model can outperform the existing shallow models in terms of accuracy and stability and can meet the requirements of online computing efficiency.