Deep learning-based comprehensive monitor for smart power station

With the wider distribution of power substations, monitoring and control of substations at large scale has become more difficult by solely relying on manpower inspection. Smart monitoring systems are increasingly important to realise fast response, low-cost maintenance and autonomous control. In this paper, we develop a novel inspection system based on deep learning and edge computing techniques. Firstly, the on-site video acquisition is completed by drones only when abnormal situations are detected, realising flexible and low-cost inspection. Using deep Q-learning, we design an efficient and reliable navigation algorithm that guides drones to the target location with minimum human intervention. To reduce the response latency and support large-scale data processing, we take the advantages of edge computing and build a high-performance edge system. Moreover, several strategies from algorithm to hardware are proposed to optimise the processing pipeline of constructed edge computing system. The experiment and simulation results demonstrate the reliability and efficiency of our proposed system in the case of autonomous substation monitoring.