A Non-Intrusive Load Identification Method Based on Novel Data Acquisition Terminals and Model Fusion

Due to the randomness and uncertainty of household electricity use, efficient grid management faces challenges. Non-intrusive load monitoring (NILM) technology has become a pivotal solution to understanding the behavior of electricity consumers. However, traditional data acquisition terminals often struggle to balance cost and performance. To address this barrier, this study proposes a novel, low-cost, high-performance data acquisition terminal, which abandons the traditional dedicated chip solution and instead uses a microcontroller to complete all control and data processing tasks. At the same time, by using the Fast Fourier Transform (FFT), the current signal is converted into a frequency domain signal containing rich information such as amplitude and harmonics, providing great convenience for subsequent intelligent algorithm analysis and processing. This study transforms the non-intrusive load identification problem at the algorithm level into a change point detection problem. A proposed fusion algorithm comprises two layers: the first is based on decision tree algorithms XGBoost and LightGBM, used for feature extraction and preliminary classification; the second uses logistic regression algorithms for decoding and outputting results, achieving high-precision load identification. Experimental results show that the method proposed in this study can achieve more than 95% accuracy when dealing with complex scenarios of mixed use of high-power and low-power appliances. Compared with other algorithms, this method shows significant advantages in load identification accuracy.