Non-intrusive Load Identification Method Based on the Online Self-organizing Incremental Neural Network

With the development of intelligent electronic technology, the accurate identification of electrical load usage will have extensive user demands in the field of smart electricity. In order to achieve the online real-time accurate monitoring of the electrical equipment, this paper proposed a non-intrusive load identification method based on the online self-organizing incremental neural network (SOINN). This method included two steps, which are the load feature extraction and the load feature classification with the equipment identification. In the process of load feature extraction, a 12-dimensional feature extraction scheme was proposed, which includes the odd harmonics, the mean value, the variance value, the third-order moment, the fourth-order moment, the root mean square current, the peak value of power spectrum, and the trough value of power spectrum. In the second step, a method combining SVM and SOINN for the load feature classification and the electrical equipment identification was proposed to overcome the limitation of the traditional SOINN algorithm in appliance type recognition. The functional algorithms in the proposed method are implemented as executable functional modules for the microprocessor system using C++ programming language. The functional modules were then ported and deployed on the HPS side of the SoC FPGA, achieving collaborative high-speed data communication between FPGA and HPS. Eight types of conventional household appliances were selected as the load identification objects. A hardware experimental platform based on SoC FPGA was built to select the optimal load characteristics. The proposed method was validated for identifying online loads of both single and multiple appliances. Experimental results showed that the above 12-dimensional features were selected as the optimal feature combination for the method proposed in this paper. The recognition rates of both single and multiple appliances using the proposed method were above 95%. The proposed load identification method can effectively and accurately identify both single and multiple electrical appliances. The system has strong implementability, high flexibility, the advantages of online learning, and practical feasibility for practical applications. © 2024 Sichuan University. All rights reserved.