Modeling Non-Steady State Data for PID Controller Tuning in a Cogeneration Power Plant

摘要

A 25 MW combined-cycle cogeneration plant at the University of Connecticut supplies electricity to the entire UConn campus with three natural gas combustion turbine generators and one high pressure steam turbine generator. Low pressure steam is used to provide building heat in the winter and to drive refrigeration compressors for chilled water cooling in the summer.rnThe UConn Cogen plant is not permitted to charge for power it exports to the grid. All imported power cost the University the same as any large utility customer. The automatic control system thus seeks to operate this power plant while constantly fluctuating demand competes with the desire to maintain zero import and zero export of electric power.rnThe highly integrated natural of the thermal cycles in the Cogen plant makes the concept of steady state operation a fleeting occurrence. Yet modern PID loop tuning tools suggest that a measured process variable (PV) should first be steadied before it is bumped so a dynamic controller output (CO) to PV relationship (i.e. dynamic process model) can be established for reliable PID loop tuning.rnThis paper explores a novel method of obtaining appropriate dynamic models for controller tuning without the requirement that the PV first be steadied prior to performing a bump test. With the ability to fit appropriate process models to data in a constantly dynamic state, modern PID tuning tools can once again be employed in challenging process applications such as the UConn Cogen power plant.rnPresented here is the framework of a method for fitting dynamic models to process data collected from processes that are impractical or impossible to settle to steady operation prior to controller tuning. A pressure header control example from the UConn cogen plant is used to demonstrate the technique.