Non-Intrusive Load Monitoring Method Based on Color-Coded V-I Trajectories and Multi-Feature Fusion

Non-intrusive load monitoring technology serves as a pivotal method for demand-side energy management within smart grids, facilitating the optimization of energy consumption. At present, non-intrusive load identification algorithms that utilize voltage-current trajectory (V-I trajectory) image classification demonstrate high accuracy. However, devices with similar V-I trajectories are prone to classification errors. The paper optimized such algorithms, and proposed a device feature combination method based on color coding, which fuses V-I trajectory with timing features and power features to improve the accuracy of device identification. Firstly, a sliding window CUSUM algorithm was employed to extract the independent load curve. Then, timing and power features were incorporated into the V-I trajectory map through RGB color coding. Finally, AlexNet neural network was used for training and identification to obtain the final classification results. In this paper, the public datasets AMpds and WHITED were analyzed. The results indicated that the algorithm significantly enhances load identification capability by leveraging the complementarity among various features.