Two-Step Partial Least Squares for Monitoring Dynamic Processes

The dynamic partial least squares (DPLS) method is widely used in dynamic industrial process monitoring. In this method, an autoregressive model is employed to describe the dynamic characteristics of a system, and traditional partial least squares (PLS) is applied to analyze correlations among data synchronously. However, DPLS only expands the dimensions of the original data; it cannot express the characteristics of the dynamic model concretely, and its monitoring performance is often unsatisfactory. The present study uses a moving average autoregressive model to describe dynamic processes and then proposes a two-step PLS (TS-PLS) algorithm to solve this problem. Testing on numerical examples and the continuous stirred-tank reactor, cases reveal that the performance of TS-PLS is much better than that of traditional DPLS.