Enhanced Air Conditioner Power Identification under PV Disturbances Using a Staged Model

The rapid adoption of distributed photovoltaic (PV) systems and flexible loads, such as air conditioners, has significantly increased the complexity of household energy consumption patterns. Accurately identifying specific load power under the influence of PV generation remains a challenge due to the dynamic interference caused by PV power fluctuations. To address this issue, this paper proposes a staged modeling approach for non-intrusive load monitoring (NILM) tailored to scenarios involving coexisting PV systems and air conditioners. Firstly, a hybrid model combining convolutional neural networks (CNN) and long short-term memory (LSTM) networks is constructed to identify air conditioner power from total load data without PV. Secondly, an LSTM-based model is developed to estimate the non-PV total load by learning the characteristics of PV power using net load data. Finally, the outputs of the two models are integrated to reconstruct the total load or improve air conditioner power identification accuracy. Experimental results demonstrate that the proposed method effectively mitigates the interference of PV power, significantly enhancing air conditioner power identification accuracy and robustness under complex load scenarios.