Neural Network Modeling and Nonlinear Model Predictive Control for a Gas Fired Boiler Simulator

摘要

ALSTOM Power Plant Laboratories (PPL) has been working on power plant controls with optimization using dynamic process simulators. In the 2007 ISA POWID conference, a paper was presented on the application of linear model identification and linear model predictive control (LMPC) to steam/water system in ALSTOM's dynamic simulator for a gas-fired boiler (GFB). As continued research and development work into nonlinear model predictive control (NMPC) for power plants, a nonlinear auto-regressive with external variable (NARX) model was constructed. The model was based on an artificial neural network (ANN) and trained using data that was generated from the GFB simulator by using designed multi-level excitation signals. The NARX model was validated using additional test data and then used as the basis for the design of an NMPC controller. The controller was then connected to the GFB simulator via OPC connection to carry out control simulations. The controls performance of the NMPC is stable and comparable to the studies with LMPC. The simulation results show that the NMPC has potential to improve plant control particularly in rapid load changes for wide load range of operations.