Research on time series prediction algorithm based on gaussian process mixture model

Aiming at the time series that change in real time and differ greatly in different time periods, a prediction algorithm based on Gaussian process mixture model is proposed. Firstly, the time series is preprocessed and the singularities are removed by density-space noise-containing clustering (DBSCAN). Then, according to the results of the extended Diki-Fowler test, the time series are divided into three categories: constant term, stationary term and non-stationary term. Finally, the prediction of various time series is made based on the Gaussian process mixture (GPM) model, and the performance of the different autoregressive moving average model (ARIMA), support vector machine (SVM) and Gaussian process (GP) model is compared. Taking the time series of government procurement commodity quotes as an example, the results show that both the GP model and the GPM model can output prediction confidence intervals and give the credibility of the prediction results. The advantage of the GPM model is that it can more accurately characterize the differences of time series in different periods, and the prediction accuracy is higher.