A multiobjective-optimal neuro-fuzzy extension to power plant co-ordinated control

Abstract

This paper presents an overall control scheme for wide-range multiobjective-optimal operation of a fossil-fuel power unit. The scheme provides the means to accommodate operating scenarios characterized by multiple operating requirements, and to attain optimal operation along any arbitrary unit load-demand pattern. The proposed strategy builds upon current co-ordinated control schemes, which typically account for simultaneous control of power and steam pressure. First, the scope of co-ordination over the internal processes of the unit is extended by including the control of the drum water level to achieve balanced overall plant operation at all loads. Then, a supervisory reference governor and a neuro-fuzzy feedforward control path are added to complement the already existing multiloop feedback control configuration. The reference governor implements a set-point scheduler based on unit load demand to set-point mappings, which are designed by solving a multiple objective optimization problem. The feedforward control path approximates the nonlinear multivariable inverse static behaviour of the power unit at the optimal operating conditions specified by the set-point mappings through a set of multi-input-single-output fuzzy inference systems, which are designed using a neuro-fuzzy learning paradigm. With this approach, the plant is driven by an arbitrary unit load-demand pattern from which the reference governor specifies the multiobjective-optimal operating conditions for the plant through set-point trajectories; the feedforward control path provides control signals to achieve wide-range manoeuvrability and the feedback control path compensates for uncertainties and disturbances, along and around the commanded set-point trajectories, respectively. Simulation results demonstrate the feasibility of the proposed control scheme.