An Attention-Enhanced Deep Learning Method Tailored for Non-IntrusiveLoad Monitoring in Air Conditioning Systems

The increasing energy demand for air conditioning systems presents substantial economic and environmental challenges, making effective air conditioning load monitoring crucial.Traditional intrusive load monitoring techniques were costly and challenging to implement, whereas non-intrusive load monitoring (NILM) technology offered a cost-effective alternative.This study constructed the CN-AC-NILM dataset, customized for the characteristics of Chinese electrical appliances and cultural usage patterns, and subsequently developed an RNN-attention model for air conditioning systems, with a focus on evaluating the impact of the attention mechanism on the model's performance.Initially, the optimal input feature set for the air conditioning load disaggregation prediction model was determined using the forward feature selection method based on the dataset, followed by an evaluation of performance differences among various RNN models.Subsequently, it was observed that in the comparative analysis of baseline models applied to air conditioning system for NILM, LSTM exhibited higher accuracy than GRU, and bidirectional models outperformed unidirectional models.Furthermore, after introducing the attention mechanism in the baseline models, the average accuracy of RNN models increased by 21.09%, accompanied by a 5.33% increase in average training time.Finally, the effects of attention mechanisms and bidirectional structure on the models were compared, revealing that the introduction of the attention mechanism outperformed the bidirectional structure, and the LSTM-Attention exhibited the best overall performance for NILM of the air conditioning systems.Notably, Attention mechanisms enhanced model interpretability through the visualization of attention weights.By improving load disaggregation accuracy and interpretability, this attention-enhanced model enables more precise energy management for air conditioning systems.It reduces electricity costs, optimizes energy use, and supports demand-side management, offering substantial economic savings and lowering environmental impact, especially in high energy-demand settings.