Lightweight non-intrusive air-conditioning loads monitoring approach

GAO Yong; YANG Di; ZHANG Ying; ZHANG Jianfeng; REN Yanru; LIU Bo

doi:10.3969/j.issn.1002-9605.2025.11.004

Abstract：Air conditioning load takes the largest proportion of flexible loads in residential housing,and accurate monitoring of its energy consumption is of great significance in realizing the goals of load control,energy saving and emission reduction.Non-Intrusive Load Monitoring(NILM)is a key technology in smart grids and intelligent energy systems.It achieves cost-effective,scalable,and privacy-friendly air conditioner load monitoring by collecting aggregate load signals at the main circuit entry and decoupling the operating states of individual devices.However,existing methods typically suffer from high computational complexity and large model parameters,making realtime processing on edge devices challenging.In this regard,this paper proposes lightweight non-invasive air conditioning load state sensing methods,specifically involving knowledge distillation-based and depth-separable convolution-based load monitoring methods,both of which are applicable to engineering scenarios with different conditions,respectively.The former transfers knowledge from a teacher model to a student model,enabling the student model to reproduce the teacher model's performance while maintaining a lightweight structure.The latter significantly reduces model parameters and computational load by decomposing standard convolution into depth-wise and pointwise convolutions.Experimental results show that the proposed methods,while ensuring load decomposition accuracy,significantly reduce computational complexity and storage requirements,enhance real-time performance and deployability,and provide new ideas and approaches for the practical application of NILM.