Research On Intelligent Fault Diagnosis And Prediction Methods For Urban Lighting System

With the never-ending developments in economic and the wide spread of urbanization in our country,the size and complexity of urban lighting system is growing year by year.Correspondingly,the service quality provided by the system directly affects the safety of pedestrians and vehicles.On one hand,maintaining well service of the urban lighting system req uires timely response to the ma lfunction diagnosis.On the other hand,it requires predictio ns for possible faults in the future so that proper maintenances and repairs can be set.Traditional fault diagnosis of urban lighting system mainly focuses on the electronical features of devices and requires experts' involvements,while the maintenances and repairs of the system lack of accurate fault prediction,and relies more on the blind random inspections.The development of intelligent urban lighting monitoring system,to some extent,provides means to detect phenomenal fault through monitoring run-time data.However,it does not include sufficient analysis on the fault and run-time data,and has no fault predictive ability.Given the circumstances in related researches,the thesis,which is under support of Sichuan science and technology support pro gram titled “research and application demonstration on key technologies in urban green lighting energy saving system(seven strategy emerging)”(Program No.2016GZ0312),summarized the constitution of urban lighting system and its faults origins.Moreover,it designed fault diagnosis models for single lighting node and fault predictive model for corresponding power system,respectively,in the hope of achieving fast deployment and resource saving.The core task of the thesis is as follows.Firstly,the thesis analyzed the massive deployment problem of the fault detection model for single lighting nodes in urban lighting system,and focused on fast response and limited human involvements to build models using extreme learning machine.By analyzing the approximation ability of extreme learning machine with different structures,the thesis incorporated incremental learning process to design fault a detection model with automatical parameters searching process for single lighting node.Secondly,the thesis analyzed the need for building predictive model using run-time data in a real power supply system and external data,and provided a mathematical model construction for the problem.By combining extreme learning machine,autoregressive model and faded sliding window methods,the thesis proposed a predictive model that could utilize the run-time data to perform on-line learning process.Finally,the thesis used data from real urban lighting system to verify the models of fault detection for single lighting node and fault prediction in a real power supply system,and tested the extendibility of the two models by introducing external data in the input.Moreover,the thesis designed and realized an intelligent urban lighting fault diagnosis and prediction system based on the proposed models.The experimental verification and system tests results showed that the two models both have relatively high accuracy for fault detection in single lighting node and fault prediction in a real power supply system,respectively,which fulfilled the contents of the thesis.