Research On Online Identification Algorithm Of Non-Intrusive Load Monitoring

The monitoring of load operation status is an important part of electricity management.Through non-intrusive load monitoring(NILM)users can identify the load operation status inside the building with the electricity consumption information measured at the bus.With the help of NILM,users can achieve load monitoring under the premise of installing a small number of sensors.This paper focuses on the study of online non-invasive load recognition algorithm to satisfy the requirements of identifying loads in real-time and dynamic adaptability during practical application.The main research contents are as follows :(1)Summarize and analyze the basic implementation framework and the key steps of nonintrusive load monitoring technology.On this basis,the paper discusses the basic implementation method of online non-intrusive load monitoring technology.Starting from the demand of online non-intrusive load monitoring,the paper summarizes the basic process of online non-intrusive load monitoring;(2)The paper proposes an online non-intrusive load identification algorithm based on unsupervised learning.The proposed load recognition algorithm is based on self-organizing incremental neural network(SOINN),which takes a single load feature vector as input,constructs the neurons of the network,and dynamically adjusts the similarity threshold to modify the weight and number of neurons.So far there have been few researches on the load naming method suitable for unsupervised algorithms.Therefore,the load naming mechanism is researched and a comprehensive naming mechanism that combines automatic naming mechanism and manual naming mechanism is proposed.The results of calculation examples show that the proposed online non-intrusive load recognition algorithm based on unsupervised learning has better classification performance than existing lazy learning methods in terms of classification accuracy and algorithm time complexity.(3)Based on the research of online non-intrusive load identification algorithm based on unsupervised learning,the paper defines the dynamic supervision scenarios of online nonintrusive load monitoring and proposes an online non-intrusive load identification algorithm based on dynamic learning.The algorithm can improve the neuron distribution of SOINN through the semi-supervised learning method with limited labeling information.Also,Through the active learning method,it can select high-value samples for labeling,and further improve the performance of the semi-supervised learning method.The results of calculation examples show that the proposed online non-intrusive load identification algorithm based on dynamic supervised learning can effectively use the information of labeled samples and improve the classification performance of SOINN.