Non-intrusive Load Identification Based on the Improved Voltage-Current Trajectory with Discrete Color Encoding Background and Deep-Forest Classifier

Abstract With the development of non-invasive load monitoring, we can monitor household appliances' category, operation status, and electricity consumption. Voltage-Current (VI) trajectory feature significantly improves load identification accuracy by representing the voltage and current waveform of appliances in images. However, it cannot reflect power information and has low pixel utilization. To solve this problem, we proposed an improved VI trajectory feature with discrete color encoding background. First, we added motion and momentum information to original VI trajectory images through color encoding. Then, the active and reactive power information was discretized using the Chi2 method, and the result was added to the background's invalid pixels. Further, we proposed a deep-forest-based VI trajectory classification method to solve the problem of model redundancy of existing image recognition methods. We also discussed the data imbalance in the VI trajectory recognition problem and proposed a balancing algorithm based on the PixelCNN++ model. The result of case studies shows that the proposed improved feature can effectively improve the classification accuracy. Compared with the advanced image recognition classifiers based on CNNs, the proposed deep forest classifier has higher accuracy, faster speed, and stronger robustness. Moreover, the proposed PixelCNN++ data balancing method is more robust and can generate realistic VI trajectory samples.