Research On Construction And Application Of Online Huber-support Vector Regression Algorithm For Big Data

With the advent of the era of big data,the development of big data has brought opportunities to all walks of life,as well as some challenges.In the face of massive and complex big data,how to make good use of big data and analyze it has become a problem for all walks of life.The focus of attention of various industries.Support Vector Machine(SVM)is a new technology in data mining.It has many unique advantages in solving nonlinear and high-dimensional problems.Support Vector Regression(SVR)can handle massive quantities.The data has a high accuracy rate when performing regression prediction.The Huber loss function has a strong tolerance to noise and can better suppress the influence of outliers on the calculation results.Therefore,this article chooses Huber-support vector regression machine as the research object of this article.Sample learning is one of the core problems of support vector regression.The traditional SVR-oriented one-time modeling algorithm is difficult to adapt to the situation where sample increments are provided online.Every time the sample data set changes,it needs to learn from the beginning and re-model.Online learning algorithms can solve this problem well.This article divides online algorithms into incremental algorithms that sequentially increase training samples and online algorithms that add a new sample while deleting an old sample.Therefore,this article focuses on the Huber-SVR-based incremental algorithm and online algorithm model.And apply the model established in this article to the application of short-term(day-based)demand forecasting for shared bicycles.This article has done the following researches:(1)This paper proposes an incremental Huber-SVR algorithm model.Unlike the one-time modeling method,the incremental Huber-SVR algorithm model can continuously integrate the information of new samples into the previously trained In the model,instead of re-modeling all samples,this greatly improves the modeling efficiency.Compared with the incremental RBF model and the incremental ε-SVR algorithm model,the model proposed in this paper has a better prediction effect.(2)Based on the incremental Huber-SVR algorithm model,this paper proposes an online Huber-SVR algorithm model,which can add and delete data in real time.The model combines the incremental algorithm and the decrement algorithm,adding a new sample while deleting an old sample,so that the length of the training window remains the same,and the number of single training samples is fixed,so the online algorithm is more suitable for large data samples.Learning is more suitable for modeling the actual process.(3)There are many redundant data and noisy data in big data,and there are only a few valuable data.How to select truly valuable data from massive big data for data analysis,improve data processing efficiency,and analyze big data Important.This paper proposes a sampling criterion for the maximum prediction variance for incremental Huber-SVR.Compared with the commonly used sequential sampling method,the maximum sampling criterion for prediction variance fits the real data better,and the prediction error is lower.(4)Apply the model established in this paper to the study of short-term(day-based)demand forecasting for shared bicycles.Analyze the problems caused by inaccurate demand forecasts.According to the prediction results of the model,the model proposed in this paper can better predict the short-term demand for shared bicycles,provide a basis for the management of shared bicycle companies and the government to formulate appropriate policies.