基于非线性自回归的盾构机时间序列数据分割

Time series segmentation of tunnel boring machine based on nonlinear autoregression

盾构机是现代地铁隧道挖掘的重要设备,系统构成复杂,很难通过较为简单的物理-物理模型进行整机的描述随着检测技术完善,5G技术发展,使盾构机能够输出大量数据。对于盾构机来说,输出的大量数据部分是与时间有关的时间序列,而时间序列的实时分割是分析盾构机所处状态的重要手段。传统分割方式多基于线性自回归分割,且只针对单一数据进行分割,这往往不能满足工程运用。为此,文中提出一种将多元非线性自回归与动态规划相结合的方法,实现基于多元非线性自回归的动态规划分割。在非线性自回归算法中,使用最小二乘法进行拟合得到自回归模型,通过模型得到拟合数据,与样本数据求方差得到每个时间段的误差后进行动态规划,找出最小误差值的分段位置,达到最优分割。

Tunnel Boring Machine(TMB)is the most important equipment in modern subway tunneling.The system structure is complicated,so it is difficult to describe the whole machine by a simple physical-physical model.With the improvement of detection technology and the development of 5G technology,the shield machine can obtain a large amount of data.For TMB,a large amount of output data is part of the time series related to time,and the real-time segmentation of time series is an important means to analyze the state of TMB.The traditional segmentation methods are mostly based on linear autoregressive segmentation,and only for a single data segmentation,which is often unable to meet the engineering application.Therefore,a method combining multivariate nonlinear autoregressions with dynamic programming is proposed to realize the segmentation of dynamic programming based on multivariate nonlinear autoregressions.In the nonlinear autoregressive algorithm,the least square method is used for fitting to get the autoregressive model.After the fitting data obtained by the model and the variance of the sample data are obtained to get the error of each time period,the dynamic programming is carried out to find the segmenting position of the minimum error value,and the optimal segmentation can be achieved.