Time Series Prediction Based On Hybrid Model

Time series prediction can provide people with good decision support,so time series prediction has been widely used in many fields.In recent decades,a large number of researchers have been studying time series analysis models,and they are constantly injecting new methods and ideas into analysis models.With the rise of machine learning technology,the prediction of time series using machine learning has also developed rapidly.However,the prediction of time series is far from enough to satisfy practical applications,and there are still many problems to be solved.This article examines the characteristics of time series and converts time series data into the data types required for supervised learning.It then uses a variety of machine learning models to train and predict the transformed data.Finally,it combines different prediction models by using appropriate combining strategies.To get the final prediction result.In this paper,two different time series collections for large data sets and small sample sets are used to construct hybrid prediction models,and both are compared with a single model to verify the effectiveness of the hybrid prediction model.The research work of this paper mainly contributes to the following three aspects:(1)Based on the characteristics of the data,this paper converts time series data into supervised learning data,and then uses multiple machine learning models to train and predict the data.Finally,on the large data sample set,a support vector regression machine(SVR)Combining multiple models,compared with the traditional single model and weight fusion method,it has better experimental results.(2)On the small sample time series data,this paper converts the small sample multivariate time series into the data structure required for supervised learning,and according to the specific data characteristics(the time series have a certain regularity,there are certain Relationship),carried out multiple data conversions to maximize the use of the characteristics of the data itself,give full play to the characteristics of the data,and facilitate the improvement of machine learning results later.(3)Use different models to train and predict the transformed small sample data,and finally fuse the prediction structure.In order to further improve the prediction accuracy of small sample time series,this article combines the characteristics of the data to expand the small sample data.Using multiple small samples,by mining the relationship between the small sample time series,a relatively large time series sample is obtained.,And used a variety of models to train and predict the expanded sample,and then fuse the prediction results of multiple models.Finally,the prediction results of the augmented data method and the prediction results without the augmented method are fused,so that the accuracy is obtained.2 improvements.The experiment proves that the method has a good effect,and it has won the first place in the competition for providing this small sample time series data set(2019 ninth "Huawei Cup" Chinese University Student Intelligent Design Competition).