In order to address the issues of network bandwidth limitation and data transmission delay caused by cloud-based cen-tralized inference mode in image intelligent recognition for the power industry,this paper explores and integrates research in two direc-tions:cloud-edge collaborative application and lightweight compression of artificial intelligence deep learning network models.By comparing and analyzing the cloud-edge collaborative architecture system and mainstream artificial intelligence deep learning network model compression methods,a cloud-edge collaborative application mode based on parameter quantization model compression is pro-posed.The feasibility of the application mode is verified by business application alerts and data conclusions through on-site testing in certain intelligent distribution station.The cloud-side model reuse reduces the human and material resources required for developing additional edge-side models,and the on-site application effect demonstrates that this application mode has high demonstration and promotion value.
关键词
云边协同/模型压缩/边缘计算/参数量化/人工智能/物联网
Key words
cloud-edge collaboration/model compression/edge computing/parameter quantization/artificial intelligence(AI)/internet of things(IoT)
维普
万方数据