A Novel Sliding Window PCA-IPF Based Steady-State Detection Framework and Its Industrial Application

In industrial processes, it is of great significance to carry out steady-state detection (SSD) for effective system modeling, operation optimization, performance evaluation, and process monitoring. Traditional SSD approaches often need to identify process state for each variable and obtain a composite index with sliding window technique, which ignores the variable correlations and is time consuming. Moreover, they can only provide the state of each whole window that slides along data series. To deal with these problems, a novel sliding window principal component analysis-improved polynomial fitting based method is proposed for steady-state detection. In the proposed framework, principal component analysis is first used to eliminate the data correlations and variable noises. Then, the size of sliding window is automatically determined by the data series of the first principal component. After that, SSD is carried out for each selected principal component by 2
^nd
-order improved polynomial fitting. At last, the overall process state is determined by the weighted combination of the SSD results of selected principal components, in which the weight of each principal component is determined by its corresponding contribution of variance. The effectiveness and flexibility of the proposed SSD framework is validated on an industrial hydrocracking process.