Research Of Time Series Classification Model Based On BP Neural Network And Semi-supervised Learning

Time series data is commonly found in production and real life.Excavating the information contained in it has important reference value for decision-making in agriculture,industry,and economy.Time series classification is an important branch of time series data mining,and it is also the basis of other mining tasks.Traditional classification methods generally require a large amount of labeled data.The training of a large number of labeled data can make the classifier achieve a higher classification accuracy.However,the acquisition of labeled data often requires a lot of time and manpower.At the same time,in actual applications,there is usually a large amount of unlabeled sequence data,and these unlabeled sequence data also contain data distribution information.The semi-supervised learning method can use the unlabeled sequence data to construct a classifier with higher classification accuracy in the case of less labeled data,which greatly reduces the time and effort consumed by the artificial labeling.In addition,the traditional time series classification methods generally require more complicated feature extraction work and affect the efficiency of the final classification method.BP neural network can realize more complex nonlinear mapping,more convenient to extract feature representation from original data,with self-learning ability and certain generalization ability.In view of this,in order to make full use of unlabeled sequence data,this paper improves a semi-supervised learning algorithm and combines BP neural network to develop a time series classification model based on BP neural network and semi-supervised learning.The work content is as follows:(1)The basic theory and related algorithms in the classification of time series are summarized,and the purpose and related algorithms of semi-supervised learning are introduced and analyzed.(2)By analyzing the time series data distribution and the iterative process of autonomous training algorithm,an improved semi-supervised classification algorithm is proposed for the problem of premature stop of autonomous training algorithm in semi-supervised learning of time series.The algorithm performs different rounds of iterative labeling for different data distributions,eventually marking all unlabeled data,effectively avoiding premature stopping,and enhancing the generalization ability of the model.In the semi-supervised classification of time series,this algorithm not only has higher classification accuracy,but also has higher classification recall and classification F1 metrics.(3)For the traditional time series classification methods,the tedious feature extraction work and the problem of poor classification effect when there are only a few mark data are used.By analyzing the characteristics of the BP neural network and the naive Bayes classifier,the BP neural network is used.The nonlinear mapping ability and classification ability of naive Bayes classifier under a small amount of labeled data,the feature extracted by BP neural network is input into Naive Bayesian classifier,which can effectively solve the problem of traditional time series classification algorithm.At the same time,the classification method is combined with the improved semi-supervised learning method,and a time series classification model based on BP neural network and semi-supervised learning is proposed.With less mark data,this model can fully and correctly use unlabeled data information to improve classification accuracy.