One of the main features of a solar power plant is that its primary energy source, the solar radiation, cannot be manipulated by the control system. Moreover, since the solar radiation changes substantially due to the daily solar cycle and to the atmospheric conditions, significant variations in the dynamics of the plant are observed. Therefore, it is difficult to achieve a satisfactory performance over the whole operating range by employing conventional non-adaptive linear control strategies. The ability of neural networks to learn and generalize based on the input-output behaviour of a given process has had a great impact on the development of nonlinear adaptive models. In particular, due to their inherent ability to incorporate time, recurrent neural networks structures are particularly suited for modelling nonlinear dynamic processes. This work proposed an affine recurrent neural network for capturing the dynamics of a solar power plant. The network is first trained offline, being further improved by means of an online learning strategy using strategy using the Lyapunov and nonlinear observation theories. **Gil** The order of the affine neural network, i.e, the estimation of the number of hidden layer neurons is addressed based on subspace projections and on the formal specificity of the nonlinear model structure. Based on the affine recurrent neural network a nonlinear control strategy is designed using the output regulation theory which provides a framework for deriving stable closed loop systems and asymptotic convergence of the tracking error to zero. Experimental results collected on a distributed collector field of a solar power plant (Plataforma Solar de Almería, Spain) show the effectiveness of the proposed approach.
|Title of host publication||Power Plant Applications of Advanced Control Techniques|
|Editors||Pal Szentannai Eds|
|Place of Publication||Germany|
|Publisher||Verlag ProcessEng Engineering GMbH|
|Publication status||Published - 1 Jan 2010|