MLD–MPC for Ultra-Supercritical Circulating Fluidized Bed Boiler Unit Using Subspace Identification

Abstract

Before carbon capture and storage technologies can truly be promoted and applied, and nuclear or renewable energy power generation can become predominant, it is important to further develop more efficient and ultra-low emission USC units on the basis of leveraging the strengths of CFB technology. In view of this complex system with strong nonlinearity such as the boiler-turbine unit of a thermal power unit, the establishment of a model that is suitable for control is indispensable for the operation and the economics of the process. In this study the form of the nonlinear model after linearization at the steady-state point has been fully considered and an improved subspace identification method, which is based on the steady-state point deviations data, was proposed in order to identify a piecewise affine model. In addition, the construction of the excitation signal in practical applications has been fully considered. The identification results demonstrate that this method has a better adaptability to strong nonlinear systems. The identification normalized root mean square errors of each working condition were almost all less than 10%. On this basis, a framework that is widely applicable to complex system control has been established by combining with the mixed logic dynamic (MLD) model. The canonical form realization was performed in order to transfer the local models into the same state basis. The predictive control was carried out on the boiler-turbine system of a 660-MW ultra-supercritical circulating fluidized bed unit that was based on the above framework. The results indicate that the predictive control performance is closely related to the setting value of the ramp rate and, therefore, prove the effectiveness of the framework.

Keywords

• 660-MW ultra-supercritical circulating fluidized bed boiler unit
• data-driven model
• MLD model
• model predictive control
• subspace identification