Plug-load Classification Based on CNN from V-I Trajectory Image using STM32

Non-intrusive load monitoring (NILM) is a solution for manage energy consumption because cheaper and easier to implement than other methods. NILM can be applied to classify the type of electric loads which enables more economical energy consumption and a more efficient home grid. Classification using deep learning algorithms has proved the most outstanding performance but with disadvantages, it has high complexity. As a result, there is a growing interest in deploying optimal deep learning networks on low-power processor microcontrollers. In this paper, we proposed deep learning architecture Convolutional neural networks (CNN) implementation on microcontrollers. The benefit of using implementation CNN load classification into a small microcontroller is lower power, lower cost, and more privacy compare to CPU/GPU with consideration of latency. Optimal CNN architecture can be obtained by using VI trajectory as feature extraction. The experimental result Plug-load inference can be run in the microcontroller STM32F4 with latency is 85.92 ms and 99% accuracy and will sufficient to run in real-world applications.