Malignant Load Detection Method And System Design

Non-intrusive Load Monitoring(NILM)is a technique that monitors and analyzes the total electricity consumption data in a certain area and disaggregates it into individual load information,it plays a key role in smart grid construction and electrical fire monitoring field.In recent years,the number of electrical fires caused by the use of malignant loads with fire risks has increased year by year,posing a great threat to fire safety.Therefore,a set of malignant load detection system is designed by applying the NILM technique to the electrical fire monitoring system,corresponding research on the load identification algorithm and the malignant load detection algorithm is conducted under the actual application of the system,which is used to avoid electrical fires caused by illegal use of malignant loads.Firstly,the significance of the electrical fire monitoring system and malignant load detection research are introduced,and the research status of the electrical fire monitoring system and NILM technique is shown.The load data acquisition module is built,the load identification algorithm and the malignant load detection algorithm are implemented in the edge device,and the design and implementation of the system software and hardware are completed.Then,according to the requirements of the malignant load detection system,a single load identification algorithm with multi-dimensional data visualization is implemented.Aiming at the limitations of large image redundancy,low recognition rate and large required network scale for the existing load data visualization methods to realize load identification,this algorithm combines three different load signatures of the load data into a small true-color image,so that the image contains load features of multiple dimensions,which improves the information density of the image.At the same time,the network model required by this algorithm is also small,and the recognition accuracy on the PLAID dataset and the WHITED dataset has reached 98.78% and 99.50% respectively.The experiment verifies the effectiveness of this algorithm through the combined analysis of different load signatures,feature visualization and analysis of the impact of network scale on performance.Finally,in view of the changing application scenarios of the system and the multi-load detection situation in real applications,a malignant load detection algorithm based on continuous learning is designed to further improve the system function.The algorithm uses Orthogonal Weights Modification(OWM)combined with the Context-dependent Processing module(CDP)to realize the load detection function,and sequentially learns the load characteristics of each type in the multi-load data for a single network,using Stochastic Gradient Descent(SGD)and OWM optimize network respectively.The results show that the former presents the problem of catastrophic forgetting,and the detection accuracy is only51.38%,while the detection accuracy of this algorithm is 83.09%,which is close to the ideal83.63% accuracy of the fixed application scenario.The results verify the effectiveness and universality of this algorithm.