基于机器学习的遥测系统故障预测技术研究

发布时间：2018-06-25 12:01

本文选题：遥测系统 + 故障预测　；参考：《中国航天科技集团公司第一研究院》2017年硕士论文

【摘要】：遥测系统的故障预测是一个多个学科的交叉领域。指标预测技术和故障识别技术是实现目前遥测系统故障预测系统的关键。故障预测系统通过分布在遥测系统中的多种传感器以及相应的软硬件接口,能够实现对遥测系统当前状态的检测并获取用于故障预测的原始数据,通过支持向量回归(support vector regression,SVR)、支持向量机(Support Vector Machine,SVM)等数据分析和预测技术,能够实现根据历史指标数据和当前的指标数据判定遥测系统当前的故障状态,并实现对未来故障的预测。本文在经典遥测系统的故障识别和预测技术基础之上,针对故障预测系统架构的设计与实现、指标预测、基于预测指标的故障判别等问题,设计并实现了一个基于历史指标数据的故障预测系统架构;设计并实现了基于SVR的指标预测方法;利用指标预测结果,采用机器学习的方法进行了故障状态的判别:1.针对遥测系统的现状和故障预测的功能需求,设计并部分实现了可以基于历史指标数据的故障预测系统。2.对指标预测过程中的指标数据的采集与清洗、SVR模型的训练、单指标预测和多指标预测等关键技术进行研究和改进。在数据的采集与清洗方面,采用遥测设备传感器获取的指标数据作为数据源,并将数据采用图的方式进行数据的可视化。在单指标预测方面,通过时间窗口分段选取指标数据进行SVR的指标预测,并对精度较高的单步预测和时间跨度较大的多步预测进行了研究和对比。进行了基于时间一致性的多指标预测。3.对目前的遥测系统故障判别方法进行研究,提出了一种基于机器学习的故障判别方法。与现有方法相比,该方法对故障的判别仅依据历史故障数据,能够在不改变方法的情况下对多个型号的遥测设备进行故障判别。4.本文使用Python语言实现了故障预测系统的软件部分。实现的系统能够依据历史的指标数据对未来指标进行预测,预测的结果可以通过可视化的方式进行显示。在指标预测的基础上,使用机器学习的方法对未来的故障状态进行判别。最终通过指标预测和对未来指标数据的故障判别,实现了对遥测系统的故障预测。
[Abstract]:The fault prediction of telemetry system is a cross-disciplinary field. Index prediction technology and fault identification technology are the key to realize the current telemetry system fault prediction system. The fault prediction system can detect the current status of the telemetry system and obtain the original data for the fault prediction through a variety of sensors distributed in the telemetry system and the corresponding hardware and software interfaces. By using support vector regression (support vector), support vector machine (SVM) and other data analysis and prediction techniques, the current fault state of telemetry system can be determined according to historical index data and current index data, and the future fault prediction can be realized. On the basis of the fault identification and prediction technology of the classical telemetry system, this paper aims at the design and implementation of the fault prediction system architecture, the index prediction, the fault discrimination based on the prediction index and so on. A fault prediction system architecture based on historical index data is designed and implemented. An index prediction method based on SVR is designed and implemented. According to the current situation of telemetry system and the functional requirements of fault prediction, a fault prediction system .2. based on historical index data is designed and partially implemented. This paper studies and improves the key techniques of index data acquisition and cleaning SVR model, single index prediction and multiple index prediction in the process of index prediction. In the aspect of data acquisition and cleaning, the index data obtained by telemetry device sensor is used as the data source, and the data is visualized by the way of graph. In the aspect of single index prediction, the index prediction of SVR is carried out by selecting the index data from time window, and the single step prediction with high precision and the multi step prediction with long time span are studied and compared. The time consistency based multiple index prediction. 3. A fault discrimination method based on machine learning is proposed in this paper. Compared with the existing methods, this method can distinguish the faults of multiple types of telemetry equipment only according to the historical fault data, and can be used to judge the faults of multiple types of telemetry equipment without changing the method. In this paper, the software of fault prediction system is realized by Python language. The system can predict the future index according to the historical index data, and the predicted results can be visualized. On the basis of index prediction, machine learning is used to judge the future fault state. Finally, the fault prediction of telemetry system is realized by index prediction and fault discrimination of future index data.
【学位授予单位】：中国航天科技集团公司第一研究院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：V556;TP181

【参考文献】