Bayesian Analysis Method In Time Series Data

With the development of computing technology and performance,Bayesian statistical inference has been recived more and more attention.It has been successfully applied in the fields of social economy and public service construction as a research tool for prediction and analysis.This study is dedicated to the study of Bayesian analysis methods in time series data,and relies on practical problems to mine useful information from time series data,establishing Bayesian statistical model and making predictions.The main contributions are as follows:First,in the field of public health,a Bayesian statistical model was established to predict the real number of patients with food-borne diseases in real time.In the Bayesian statistical model,the negative binomial distribution was used to fit the true number of patients per day,the conjugate prior framework of the generalized dirichlet distribution was used to fit the right truncation of the delay distribution under the homogeneity hypothesis,and finally,the model was evaluated by the evaluation criteria such as the probability ranking score.From the results,the established Bayesian hierarchical model based on delay adjustment can provide the latest daily information about the epidemic trend by predicting the real daily incidence of diseases.Compared with the original method,it shortens the prediction cycle and improves prediction accuracy.Secondly,in the field of communication,a Bayesian traffic model is established for the traffic load in the wireless network base station to predict the traffic and the number of users in real time.Among them,we establish a poisson regression model with first-order autoregressive for user behavior characteristics with short-term stability,and establish a non-homogeneous linear model for the interaction between user behavior and traffic load.Experimental results show that the proposed Bayesian traffic model can effectively capture these characteristics,which not only has higher prediction accuracy,but also has higher interpretability and more efficient computational performance.In addition,based on the parameters estimated by the Bayesian traffic model,we further use the unsupervised k-means clustering algorithm to reveal the hidden spatial relationships between base stations,cluster the base stations with high similarity into the same category,and the scene recognition accuracy rate is 75%.