Upcoming trend prediction from historical time series using Bayesian estimation
Yizhou Xu, Jiandong Wang · 2022 41st Chinese Control Conference (CCC) · 2022
Prediction of an abnormal condition is absolutely critical to avoid both human injuries and equipment damage. A new data-driven predictive technique is proposed for predicting the upcoming trend in the current time series. The proposed method has two important features: (i) trends can be predicted at the beginning of trend variation, which can help industrial operators take timely actions; (ii) the prediction of trends is given with quantitative probabilities and credibility intervals. By contrast, existing prediction methods either focus on predicting specific values of time series with short predictive step size, or cannot provide a quantitative measurement of prediction results. The proposed method firstly captures qualitative trends in the current and historical time series, and divides these trends into several clusters. Secondly, the current and historical time series are transformed into trend sequences according to the trend clusters. Finally, the proposed method choose some subsequence of recent trends in the current trend sequence and predicts the probability of the alarm states based on the identical subsequences in the historical trend sequence. A Bayesian estimator is exploited to calculate the posterior probabilities and their credibility intervals for the alarm states. The optimal predicted alarm state is the one with the largest lower bound of posterior probability as a quantitative measurement of prediction reliabilities. The effectiveness of the proposed method is illustrated via numerical examples based on a well-accepted benchmark of the Tennessee Eastman process.