Non-intrusive load monitoring algorithm for coarse-grained electricity consumption data based on YOLO deep neural networks

Non-intrusive load monitoring can obtain classified load data in a user acceptable way, which is of great significance for the user’s energy use behavior and demand response behavior guidance. Taking into account the increase in the type and number of loads, the cost control of embedded device calculation and storage, as well as the cost control of data acquisition equipment. It is becoming more and more important to construct representative data organization patterns and design accurate and efficient load decomposition algorithms for NILM technology. Since traditional non-intrusive load identification generally relies on fine-grained electricity consumption data (second-level resolution, millisecond level resolution), hardware investment and communication overhead are increased. This paper presents a coarse-grained electricity consumption data decomposition algorithm based on the YOLO object detection model. The algorithm adaptively extracts and reassembles minute level power sequences and labels them automatically. This form of data organization will greatly reduce the footprint of storage space, while reducing hardware investment and even relying on the minute level coarse-grained electricity consumption data of existing energy meters. Through the joint design and optimization of the target detection model, the problem of load identification and classification is transformed into a regression problem, which makes the algorithm have high scalability and portability. Through the validation of NILM public data set, it is proved that this method has good non-intrusive load monitoring performance, improves the processing speed and precision of balance, and has high robustness.