Stochastic configuration network based cascade generalized predictive control of main steam temperature in power plants

Abstract

The main steam temperature (MST) in power plants suffers from nonlinearity and large time delay, which cause large overshoot and long settling time under widely used cascade proportion integration differentiation (PID) controller. In order to cope with the negative effects, we propose a stochastic configuration network (SCN) based cascade generalized predictive control (GPC) scheme to improve the performance of the MST. A three-layer SCN is employed to model the MST process. The SCN is constructed by two phases, i.e., initial phase and real-time phase. The initial phase determines the structure and primary parameters of the learner model using the stochastic configuration algorithm. The real-time phase employs weighted recursive least squares (WRLS) for building the real-time MST process model for GPC design. Taking into account some constraints of the MST process, Karush–Kuhn–Tucker (KKT) conditions are applied for solving the constrained receding-horizon optimization problem. The derived explicit solutions of GPC avoid the implicit form which usually has to be solved iteratively. Comparative simulations demonstrate the superiority of the proposed SCN based cascade GPC (SCN-CGPC). Finally, the proposed SCN-CGPC is implemented via a standalone external MST control system in a power plant. The effectiveness and practicability are validated with the real-world application.