基于时间序列的噪声监测点异常发现

发布时间：2018-05-27 03:23

本文选题：符号化聚集近似 + 相似性度量　；参考：《南京航空航天大学》2015年硕士论文

【摘要】：随着机场噪声污染程度日益严重,以及噪声对环境影响的投诉逐年增多,机场噪声问题成为影响民航业可持续发展的一个重要因素。其中,机场噪声时间序列中的异常情况虽然相对正常情况来说极少出现,但却拥有很大价值,利用它能够及时检修飞机和机场设备来防止问题变得更加严重。通过判断机场内分布在各个区域的监测点上收集的数据是否出现异常,来确定是否存在异常情况。又由于导致监测点失效或数据采集错误的原因有很多,因此还需要对异常情况的原因进行分析。为了能够发现监测点异常的原因,首先需要研究如何判断监测点异常。将机场噪声时间序列降维并符号化,然后运用提出的新度量方法度量处理后的时间序列,最后根据度量结果利用k近邻孤立因子进行异常检测,由此得出机场噪声单监测点时间序列异常检测方法。该方法在大大减少了数据量并提高了计算速度的同时,使用改进的度量方法减小了形态信息被弱化所带来的影响,完成了对异常监测点的初步筛选。根据发现的异常监测点,研究这些监测点出现异常的原因。为了实现这一目标,对于监测到异常数据的监测点,需要先对它们应有的数据进行预测。使用神经网络集成方法,融入组合模型思想提高预测精度,探寻其中的权重计算方法,提出关联监测点神经网络集成组合预测模型,使用这些监测点的极强关联监测点数据训练相应的预测模型,然后预测出异常点数据,作为后续异常原因判断的标准。最后找出异常监测点数据出现异常的原因。提出关联噪声监测点异常发现算法,首先使用上述预测方法,将所得预测值与实测值进行相似度计算,最终判断该点是否异常。若候选异常点此时被判断为异常点,则说明该监测点本身老化损坏;若被判断为正常点,则说明是机场设备或飞机问题。
[Abstract]:With the increasingly serious noise pollution in airports and the increasing number of complaints about the impact of noise on the environment, the problem of airport noise has become an important factor affecting the sustainable development of the civil aviation industry. The abnormal situation in the airport noise time series is relatively rare, but it has great value. It can be used to inspect and repair the aircraft and airport equipment in time to prevent the problem from becoming more serious. By judging whether there are anomalies in the data collected from the monitoring points distributed in each region in the airport, we can determine whether there are abnormal conditions. There are many reasons that lead to the failure of monitoring points or data acquisition errors, so it is necessary to analyze the causes of abnormal situations. In order to find out the reason of abnormal monitoring point, it is necessary to study how to judge the anomaly. The time series of airport noise are reduced and symbolized, then the time series after processing are measured by the proposed new metric method, and then anomaly detection is performed by using k-nearest neighbor solitary factor according to the results of measurement. The method of time series anomaly detection for single monitoring point of airport noise is obtained. This method can greatly reduce the amount of data and increase the speed of calculation, at the same time, the improved measurement method is used to reduce the influence of the weakening of morphological information, and the preliminary screening of anomaly monitoring points is completed. According to the abnormal monitoring points found, the causes of the anomalies in these monitoring points are studied. In order to achieve this goal, it is necessary to predict the data that should be used to monitor the abnormal data. Neural network ensemble method is used to improve the prediction accuracy by incorporating the idea of combinatorial model, and the weight calculation method is explored, and the integrated combined forecasting model of neural network for the associated monitoring points is put forward. The data of these monitoring points are used to train the corresponding prediction model, and then the outliers data are predicted as the criteria for judging the causes of the anomalies. Finally, we find out the reason of abnormal data at the abnormal monitoring point. In this paper, an algorithm is proposed to detect anomalies at the monitoring points of associated noise. Firstly, the similarity between the predicted values and the measured values is calculated by using the above prediction methods, and finally the anomaly of the points is judged. If candidate outliers are judged as outliers at this time, the monitoring points themselves are aged and damaged; if they are judged as normal points, it is an airport equipment or aircraft problem.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X839.1

【相似文献】