深度学习与幅值特征相融合的低频采样非侵入式负荷辨识算法

非侵入式负荷监测中虽然高频采样能提高负荷辨识准确率,但对数据采集设备要求高,难以推广,因此,低频采样下负荷辨识方法成为研究热点。以低频采样下负荷投入时的暂态电流波形为特征,采用卷积神经网络算法实现负荷辨识,辨识结果发现CNN对暂态电流波形差异大的负荷辨识准确度高,但是对暂态电流波形相似的负荷识别准确率低,为解决这一问题,在卷积神经网络辨识的基础上,对暂态电流波形相似的负荷,以暂态电流幅值为特征作进一步辨识,以提高辨识准确率。通过使用实测数据进行验证,结果表明所提算法可以很好地克服低频采样下波形特征相似负荷识别准确率低的问题。

Non-intrusive Load Identification Algorithm Based on Deep Learning and Amplitude Feature Fusion at Low Sampling Frequency

Although high-frequency sampling can improve the accuracy of load identification in non-intrusive load monitoring, it requires high data acquisition equipment and is difficult to popularize. Therefore, the load identification method of low sampling frequency has become a research hotspot. With the characteristics of transient current waveform at the time of load start of low frequency sampling, the convolutional neural network algorithm is used to identify the load. The identification results indicate that CNN has high accuracy for load identification with large differences in transient current waveforms, but it is not effective to identify loads with similar transient current waveforms. In order to solve this problem, based on the convolutional neural network identification, the load with similar transient current waveform is further characterized by the amplitude of the transient current to improve the identification accuracy.The verification using actual measured data shows that the proposed algorithm can well overcome the problem of low recognition accuracy of load with similar waveform feature under low frequency sampling.