Searching historical data segments for process identification in feedback control loops

Historical Data Segmentation for System Identification: A Comparison of Methods

“…( ) is the disturbance model; ( ) is the process set point; ( ) is the output of the controller (manipulated variable); ( ) is a measured disturbance; ( ) is measurement white noise and ( ) is the process output (controlled variable). Wang et al, 2018).…”

“…Another way to find the potential intervals in Step 1 of the previous methods is proposed in (Wang et al, 2018apud Pettitt, 1979) through a top-down, non-parametric change-point detection algorithm. A top-down algorithm is one that starts with the entire dataset and divides it until no further divisions can be found.…”

“…As explained in (Wang et al, 2018), let us initially consider a data segment of length as ( : + − 1) = ( ( ), ( + 1), … , ( + − 1)). Therefore, in this method, the initial segment begins with the entire data: ( (0), (1), … , ( )), being its length.…”

“…In (Wang et al, 2018), a new method to search data segments suitable for system identification was presented, totally based on statistical methods, in which the authors applied a top-down approach to detect change-points in the data.…”

“…More specifically, the works published in (Peretzki et al, 2011), Huang, 2013a), (Bittencourt et al, 2015), (Ribeiro;Aguirre, 2015) and Kroll, 2017a) are summarized in two methods, both using numerical conditioning as the basis to validate adequate time intervals to obtain a process model operating in closed-loop control. Moreover, the top-down change-point detection presented in (Wang et al, 2018) is proposed as an optional first step of the explained methods. Finally, a generic algorithm structure is presented and its use is exemplified through data from the step response of a pH neutralization process.…”