Structural Prediction Of Multiwariate Time Series Based On Outlier Limination And Application

Multivariate time series data is widely present in various real-world projects such as air quality prediction,stock trend analysis,and mechanical structure reliability assessment.By constructing a predictive model for multiple influencing factors at the same time,multivariate time series forecasting can effectively improve the prediction effect of a single influencing factor time series,and therefore it has significant theoretical and practical significance.However,traditional multi-variable time series prediction methods often ignore the structural information among variables,thereby limiting their prediction accuracy and numerical stability.In view of that,this paper starts from the characteristics of data distribution,introduces time series clustering algorithm,fully exploits the similarity information among variable sequences,and constructs two kinds of multivariable time series structured prediction methods based on support vector machines,which are applied to air quality prediction and actual trends in the prediction of bearing recession trends.The main work and contributions are as follows:(1)Aiming at the problem that existing multi-variable time series prediction methods fail to effectively measure the similarity among sequences and thus affect the forecasting effect,a new multivariate time series structural prediction method based on outlier eliminating is proposedin this paper.This algorithm predicts on the selected multivariate by using structural output characteristic.First,torecognize the relatedness among the sequences,the variable sequences are divided initially by conducting hierarchical clustering based on fuzzy entropy.Second,to further evaluate the similarity of the sequences in the obtained sequence cluster,the principal curve is introduced to calculate the abnormality degree of each sequence,and then the outlier sequence can be eliminated in terms of the value of abnormality degree.As a result,the similar sequences can be distinguished.Finally,for the similar series,multi-dimensional Support Vector Regression is introduced to construct the prediction model,and then the structural prediction for multivariate time series prediction is conducted.Moreover,a theoretical proof is also provided to prove the proposed method has upper bound of the loss of information and lower bound of the reliability,which demonstrate the proposed method is reasonable and feasible from the perspective of information entropy.The computer experiments are conducted on three chaotic time series datasets and four real-life datasets.The results show that the proposed method can e_ectively recognize the inner group structure among multivariable sequences,and then obtain better forecasting accuracy and numerical stability than widely used methods in terms of two different error measurements.(2)For the problem of recursive time series prediction in the prediction of bearing declining trend,which lacks prior knowledge,affects the prediction accuracy and numerical stability,this paper proposes an improved time series clustering algorithm,and builds on this basis the trend towards rolling bearing recession structured recursive prediction method.The algorithm extracts the bearing fault sequence data with similarity offline and recursively predicts the trend of the decline of the target bearing.Firstly,phase space warping(PSW)is introduced to transform fault data into a series of tracking matrices.Dynamic time wraping distance(DTW)distance measures are calculated and hierarchical clustering of time series is performed.The similarity division of bearing declining trend is obtained;secondly,the main curve of each type of bearing fault sequence is constructed;finally,each type of main curve is introduced as a priori knowledge to participate in the test of bearing failure trend prediction.The PHM 2012 data challenge bearing data was used to verify the simulation results.The results show that the algorithm can use the internal structure information of the bearing data to realize the prediction of the unknown bearing recession trend.The research content of this paper provides a new solution for multi-variable time series forecasting.It has a wide range of applications and has good reference for solving multivariable time series forecasting problems in practical engineering problems.