A Multi-Protocol Smart Sensing Method Based on Lightweight Deep Neural Network for Edge Computing

Correctly recognizing protocol types of packets from different communication channels and implementing protocol analysis is fundamental to data acquisition for edge computing in power Internet of Things. Such studies have so far been largely restricted to specified type of protocol, because only when the type is known can the protocol be parsed. Targeting more protocol types requires methods that can automatically recognize various types of protocols with both high recognition rate and high efficiency. To be suitable for edge computing in power Internet of Things, this paper proposes a light-weight deep neural network model named PPRNet trained with data frames from protocols of different types. The experimental results show that the technique is very effective on sensing 6 protocol types applied in power IoT containing DL/T645-2007, Q/GDW1376.1, and IEC62056. This model allows us to sense different types of protocol when previously unknown data frames entering in a system, and overcome the dilemma of economically deploying an artificial intelligence system to an edge device. This method has the ability of learning and expanding, can be extended to identify and apply other application protocols. It may ultimately help speed up the recognition of protocol types, and promotes the interconnection and interoperation of access devices and the plug-and-play.