Skip to Main Content
This study discusses non-linear optimal feedforward control of solar collector plants. The control problem is challenging, with non-linear disturbance effects and long input-dependent transport delays. The controller operates in a model-based predictive control framework, without feedback and constraints, to highlight feedforward controller performance gains. To reduce the computational complexity state extension is avoided when modelling the time delays. The motivation for this is provided by a proof of the equivalence between state extension and direct prediction in optimal control, for non-linear multiple-input-multiple-output (MIMO) systems with time delays in disturbances and controls. The solar collector controller also uses flow-dependent sampling to reduce the time variation of the delays. To obtain an accurate model for feedforward controller design, a black-box recursive prediction error identification algorithm was used for modelling of the non-linear plant using measured data. Experimentally, accurate feedforward control was obtained when the controller design was tested with simulation, using measured disturbances from the plant. About two-thirds of the control effort needed appears to be available by feedforward control only. In order to evaluate the algorithms in difficult conditions, the evaluation was performed on data obtained during a day with partly cloudy weather. This caused large flow variations, and consequently large time delay variations.