Abstract: The promotion of the source-load-storage interaction and participation in the power market facilitates the coordinated development and allocation of various resources. However, the randomness of elastic load is not considered in the quantitative analysis of demand-side adjustable resources. Considering the abundant quantity and strong controllability of air conditioning loads, we investigate a non-intrusive air conditioning load accurate perception system for source-load-storage coordination optimization in this paper. Firstly, to address the issue of wasted computing power caused by full perception of demand-side loads, the event detection of air conditioning load is realized by wavelet transform through analyzing the transient characteristics of air conditioning load. Secondly, the consistency of steady-state characteristics of air conditioning loads and the difference of other loads are structurally expressed through feature engineering, enabling the construction of an exclusive identification model for air conditioning loads through a combination with random forests. Finally, the adjustable capacity of air conditioning in the area is analyzed in real-time through the potential assessment model for air conditioning load regulation. The system’s accuracy and effectiveness are verified through simulation using publicly datasets and actual collected data.