Considering the characteristics of the non-intrusive load identification model, this paper studies a high-dimensional feature optimization modeling based on edge computing architecture. This method uses the NILM terminal to collect load data, window to detect events, generate physical feature data of load, and transmit it to the regional edge server. The edge server constructs high-dimensional features and inputs them into the load identification model, and sends the recognition result to the terminal. In this paper, principal component analysis is used to find the relationship between high-dimensional feature space and load identification model, and to generate feature subsets suitable for specific models. The results show that this method greatly improves the training efficiency and recognition accuracy of the model, and provides a new research way for non-intrusive load identification.