Design Of Electrical Fire Monitoring System For Arc Fault Detection And Load Identification

With the improvement of power application level in society,the types of household electrical equipment and residential power consumption have sharply increased.The electrical fire accident occurs frequently,so the power safety and energy consumption analysis in lowvoltage lines have attracted much attention.Arc fault is one of the important causes of electrical fires under low-voltage lines.At present,the series fault arcs detection equipment in China is not perfect.The electrical fire monitoring system lacks the analysis function of residential electricity.Therefore,this thesis will combine electrical fire monitoring systems with the nonintrusive load identification technique of smart grids.Based on the characteristics of the series arc fault in the low-voltage line and the difference in the information of different household loads,this thesis designed a set of intelligent electric fire monitoring system,which can detect and alarm the arc fault and load identification in low-voltage lines,and analyzes the residents’ electricity consumption behavior to enhance the residents’ awareness of electricity safety and energy saving.In this thesis,we firstly introduce the development and status of electrical fire monitoring system,and analyze the current research status of arc fault detection technology and load identification technique at home and abroad.Based on the principle and types of arc fault generation and functional requirements of the system,STM32F429IGT6,ESP8266 and computer server are used to complete the multi-level detection and analysis in the local area network.According to the overall design of the system,this thesis selects a suitable transformer,makes an automatic arc generator for simulating arc fault,and designs the hardware and software of the monitoring terminal.The Pyqt5 library is used to accomplish these functions of man-machine interface in the server,complete the real-time monitoring,data processing,and algorithm analysis.Then,through the system experimental platform,the current data of different loads(electric fans,hair dryers and laptops)are collected both on the normal operation and arc fault lines to form an arc database.Since the current signal of the series arc fault is a non-stationary signal,it has a large number of high-frequency pulses,so a one-dimensional wavelet convolution kernel neural network is designed to complete the detection and classification of the arc fault,which uses a continuous wavelet convolution layer to replace the first convolutional layer in traditional convolutional neural networks(CNN).This method overcomes the limitation of traditional arc fault detection algorithms that need to set thresholds based on empirical values,and compared with traditional convolutional neural networks,the convergence speed is faster and the amount of parameters is smaller.The experimental analysis on the dataset in this paper shows that the final detection accuracy rate reach 99.29%,which verifies the effectiveness of the method.Finally,in the load identification function,aiming at the limitations of large network parameters,high computational complexity,and insufficient identification accuracy in traditional artificial intelligence load identification algorithms,a novel method of multidimensional data fusion visualization is proposed to aggregate voltage,current and voltage–non-active current trajectory of per load to generate true-color visualization images with smaller size and higher discrimination,as artificial neural network’s input data.The experimental results show that by using true-color visualization images,the load classification is 100% on the dataset collected in this thesis,96.63% for the PLAID dataset and 99.05% for the WHITED dataset with less than 1% of the artificial neural network and calculation amount of the traditional algorithm.