Prediction And Pattern Recognition Of Large-scale Spatio-temporal Sequences

The era of big data has provided tremendous impetus for the reform of many industries.In order to acquire and process big data,more and more technologies have been invented and applied,the "Internet of Things" has become a trend.Most Io T data has both time and location tags.In recent years,large-scale spatio-temporal sequences have emerged endlessly.It is one of the important applications of big data to mine the correlation and causality of spatio-temporal sequences based on big data,and to infer and predict accordingly.The research content of this dissertation is the prediction of large-scale low-frequency spatio-temporal sequences and the pattern recognition of large-scale high-frequency spatio-temporal sequences.It mainly makes some improvements and attempts on the model of large-scale spatio-temporal sequences to facilitate the application of large-scale spatio-temporal sequences in various industries.Firstly,this dissertation elaborated the development of the current spatio-temporal sequence prediction models and extensive application scenarios.The commonly used hierarchical statistics model is elaborated,and the STARMA model and the spatio-temporal Kriging model used in this paper are introduced in detail under the framework of the hierarchical model,and a brief formula derivation of its parameter inference process is given.Secondly,this dissertation focus on the prediction of low-frequency spatio-temporal sequences.For small-scale spatio-temporal datasets,the spatio-temporal Kriging technology and the STARMA model are combined by analyzing the limitations of STARMA on the spatial correlation assumption and the singularity of the spatio-temporal Kriging model equations.Sparsifying the spatial-temporal Kriging and using it as the spatial weight matrix of the STARMA model has two advantages.On the one hand,it solves the problem that the spatial-temporal Kriging is difficult to solve the inverse matrix,on the other hand,it breaks through the limitations of the construction of the spatial weight matrix of the STARMA.After that,through analysis,it is discussed that neither model can be applied in large-scale datasets.Accordingly,for large-scale spatio-temporal datasets,through detailed formula derivation and algorithm description,the newly proposed spatio-temporal sequence prediction technology based on singular value decomposition technology and differential autoregressive moving average model with seasonal effect is elaborated.This large-scale low-frequency spatio-temporal sequence prediction method takes the traffic dataset of communication base stations in a certain area of Shanghai as a real research case,compares the prediction accuracy of the new method and the existing classic spatio-temporal model method,and gives the use conditions and application value of the algorithm.Thirdly,in terms of large-scale high-frequency spatio-temporal sequences,this dissertation builds on a piecewise linear regression model based on Group Lasso constraints based on the vibration signal of the tool machining process and uses the synchronization of the tool degradation process signal to detect the synchronous change points in the high-dimensional signal,and then completes The division of different health life stages of the cutting tools,and the application value is verified by comparing the simulation data and the real tool processing signal data with the currently commonly used methods.Finally,by proposing a prediction model for large-scale low-frequency spatio-temporal sequences and a pattern recognition model for large-scale high-frequency spatio-temporal sequences,this dissertation attempts to develop a spatio-temporal sequence model in the context of big data,and applies it in real applications and has achieved better results.