Construction and Implementation of Electric Power Deep Learning Computing Service Platform
Jipu Geng, Xiaoyan Zheng, Xiaohan Lai, Lei Zhao, Yibai Xu · 2023
In order to accurately and effectively identify the types and locations of abnormal information in the power industry, an electric power deep learning computing service platform is constructed to provide intelligent inspection services. The platform adopts hierarchical technology architecture, based on Docker technology, realizes the abstract pooling of many kinds of computing resources, and provides the management interface of heterogeneous resources such as GPU, NPU and so on according to the specific requirements of deep learning field, so as to realize the unified management and scheduling of heterogeneous resources, and provide an efficient, flexible and scalable distributed computing service supporting environment for deep learning computing tasks. The platform layer provides data center, development center, training center, model center, service center and other subsystems, and built-in a variety of learning frameworks and model algorithms to support external algorithm model integration. It really realizes the one-stop development ability of deep learning tasks. In the process of intelligent inspection of electric power, the online recognition and detection of a variety of abnormal information is realized, which provides a technical basis for on-line identification, supervision, prevention and control, comprehensive management of abnormal behavior.