Dynamic Time Series Modeling Method based on Bayesian Learning and Gray Forecasting

As real-life time series data are usually complex, the traditional dynamic time series model cannot achieve satisfactory prediction results. In this paper, we combine gray prediction with dynamic time series to construct a dynamic time series model based on gray prediction to improve the prediction accuracy of dynamic time series. Aiming at the missing data problem often faced in the process of dynamic time series modeling, we design the missing data recovery algorithm for dynamic time series based on Bayesian learning, and model the missing data prediction problem in dynamic time series as a multi-sparse vector recovery problem based on the theory of compressed sensing. Carrying out dynamic time series prediction simulation experiments, the residual sum of squares, residual median error, and average relative error of the dynamic time series model based on gray prediction in this paper are 0.616, 0.307, and 0.297, respectively, which are lower than those of the comparative traditional GM(1,1) model, and the time series analysis model. And in the missing data recovery simulation experiments, this paper's algorithm has a smaller RMSE at any data missing rate, and the RSME value of this paper's algorithm remains lower than 0.2 when the data missing rate reaches the highest 95%.