Research And Application Of Multivariate Time Series Forecasting Based On Broad Learning System

Time series are ubiquitous in real life,which is a common form of data expression with a wide range of applications.It utilizes various mathematical models or algorithms to dig out the inherent characteristics of sequences to further guide production and life,and it has gradually become academic and industrial hotspots,especially under the condition of the big data and cloud computing era,time series are increasingly showing characteristics such as high dimensionality,long data,high correlation,and information redundancy.Therefore,it is of great significance to study how to predict multivariate time series by means of new and efficient analysis and calculation methods.The main research work and innovations of this thesis are as follows:Firstly,this thesis proposes an improved general regression neural network algorithm for multivariate time series prediction.The general regression neural network algorithm is used to predict the multivariate time series,with less adjustment parameters and saving computing resources.Then,it combines with the particle swarm optimization algorithm to find a suitable smoothing factor more accurately,the experimental results show that the prediction accuracy of the algorithm is significantly improved compared with the existing mainstream models.Secondly,this thesis designs a multivariate time series prediction algorithm based on Restricted Boltzmann Machine and broad learning system.The broad learning flattened neural network structure and the pseudo-inverse method are adopted to avoid the problems caused by gradient descent in deep deep learning,thus speeding up the calculation speed.At the same time,the Restricted Boltzmann Machine is used in the mapping layer.Through unsupervised learning,the reduced dimension or deepening mapping units are obtained.Compared with typical RBM,GBRBM has more advantages in processing real-valued input data sets,broadening the scope of input data applications,thereby improving the broad learning system method and effectively improving the algorithm prediction efficiency.Finally,in this thesis,a multivariate time series prediction algorithm based on gated recurrent unit and broad system learning is proposed.The algorithm is further improved on the basis of the previous chapter.The Gated Recurrent Unit is added to the enhancement layer to memorize historical data,which strengthens the effect of previous input on the current output and establishes the correlation before and after the time series.Experimental results show that the prediction effect is greatly improved.