面向城市照明系统的智能故障诊断与预测方法研究

发布时间：2018-10-25 17:32

【摘要】：伴随国家经济的持续增长和城市建设的不断推进,城市照明系统的规模和复杂度逐年上升。与之对应的,系统提供的服务质量直接影响着越来越多的行人和车辆安全。保持城市照明系统良好的服务一方面需要及时对系统故障做出诊断,另一方面还需要预判未来可能发生的故障从而安排合理的维护和检修策略。传统的城市照明系统故障诊断方法主要针对设备电气特征,需要专业人士参与,而在系统维护和检修方面缺乏对故障的准确预测,更多地依赖盲目的随机巡检过程。智能城市照明监控系统的出现虽然能一定程度上通过监控运行数据诊断现象型的故障,但缺乏对故障和运行数据之间关联性的分析,也没有提供故障预测的能力。基于这样的研究现状,论文在四川省科技支撑计划项目“城市绿色照明节能系统关键技术研究和应用示范(七大战略新兴)”(项目编号:2016GZ0312)支持下,归纳总结了城市照明系统的组成和故障成因,分别对系统中的路灯节点的故障诊断问题和区域配电系统中的故障预测问题设计了相应的分析模型,以期达到高效快速部署和节约资源的目的。论文的核心工作如下:首先,论文分析了城市照明系统中路灯节点故障诊断所面临的大范围部署问题,并针对快速响应和少人工参与的需求使用极限学习机对该问题进行了抽象建模。通过分析不同结构的极限学习机的近似逼近能力,结合增量学习过程设计了具备自适应参数搜索过程的路灯节点故障诊断模型。其次,论文分析了城市照明系统中通过区域配电系统运行数据和外部数据构建故障预测模型的需求,并对预测模型进行了数学建模。通过结合极限学习机、自回归模型和带衰减的滑动窗口这三类方法,论文实现了能够利用实时产生的数据进行在线学习过程的故障预测模型。最后,论文使用宜宾市城市照明系统运行数据对路灯故障诊断模型和区域配电系统预测模型进行了验证,并通过引入外部数据说明了模型对输入属性的扩展性。此外,论文还基于所提模型设计并实现了城市照明智能故障诊断与预测系统。实验验证和系统实现的结果表明论文所提的两类模型分别在路灯故障诊断和区域配电系统故障预测方面具有较高的分类和预测准确率,实现了论文的研究内容。
[Abstract]:With the continuous growth of national economy and the development of urban construction, the scale and complexity of urban lighting system are increasing year by year. Correspondingly, the quality of service provided by the system directly affects the safety of more and more pedestrians and vehicles. On the one hand, it is necessary to diagnose the system faults in time, on the other hand, it is necessary to predict the possible future failures and arrange reasonable maintenance and repair strategies. The traditional fault diagnosis method of urban lighting system is mainly aimed at the electrical characteristics of equipment and requires the participation of professionals. However, it lacks accurate prediction of fault in system maintenance and maintenance, and relies more on blind random inspection process. Although the appearance of intelligent city lighting monitoring system can diagnose phenomenal faults through monitoring operation data to some extent, it lacks the ability to analyze the correlation between failure and operation data and to provide fault prediction. Based on this research status, the paper is supported by Sichuan Science and Technology support Project "key Technology Research and Application demonstration of Urban Green Lighting Energy-saving system (Seven Strategies emerging)" (Project number: 2016GZ0312). The composition and causes of failure of urban lighting system are summarized and the corresponding analysis models are designed for the fault diagnosis problem of street lamp node and fault prediction problem in regional distribution system. In order to achieve efficient and rapid deployment and save resources. The core work of the thesis is as follows: firstly, the thesis analyzes the problem of the large-scale deployment of the street lamp node fault diagnosis in the urban lighting system. Aiming at the requirement of quick response and less human participation, the paper uses extreme learning machine to model the problem abstractly. By analyzing the approximate approximation ability of LLMs with different structures and combining the incremental learning process, a fault diagnosis model of street lamp nodes with adaptive parameter search process is designed. Secondly, the paper analyzes the requirement of constructing fault prediction model through the operation data and external data of regional distribution system in the urban lighting system, and carries on the mathematical modeling to the prediction model. By combining the three methods of extreme learning machine, autoregressive model and sliding window with attenuation, a fault prediction model which can make use of the data generated in real time for online learning process is implemented in this paper. Finally, the paper validates the street lamp fault diagnosis model and the regional distribution system prediction model by using the operation data of Yibin city lighting system, and explains the extensibility of the model to the input attributes by introducing external data. In addition, the intelligent fault diagnosis and prediction system of urban lighting is designed and implemented based on the proposed model. The results of experimental verification and system implementation show that the two kinds of models proposed in this paper have high classification and prediction accuracy in street lamp fault diagnosis and regional distribution system fault prediction respectively.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP277;TU113.666

【参考文献】