Local model predictive controller in a solar desalination plant collector field Claudio O. Ayala a , Lidia Roca b, 1 , Jose Luis Guzman c, * , Julio E. Normey-Rico d, 2 , Manolo Berenguel c , Luis Yebra b a Departamento de Ingeniería Eléctrica, Universidad de Antofagasta, Avenida Angamos 601, Antofagasta, Chile b Convenio Universidad de Almería-Plataforma Solar de Almería, Ctra. Senés s/n, 04200 Tabernas, Almería, Spain c Dep. Lenguajes y Computación, Universidad de Almería, Ctra. Sacramento s/n, 04120 Almería, Spain d Dep. de Automação e Sistemas, Universidade Federal de Santa Catarina, 88040-900 Florianópolis, SC, Brazil article info Article history: Received 1 December 2010 Accepted 30 March 2011 Available online 22 April 2011 Keywords: Dead-time systems Robustness Generalized predictive Control Local models Solar plants Non-linear systems abstract This paper proposes a new predictive control strategy for a distributed collector field of a solar desalination plant. The main purpose of the controller is to manipulate the water flow rate to maintain constant the outlet-inlet temperature gradient in the collectors in spite of disturbances. The controller is based on a filtered Smith predictor generalized predictive control algorithm and a simple procedure to update the linear model used in the predictor as well as the tuning parameters, in such a way that non-linear optimization is avoided. The controller copes with the process non-linearities, constraints, dead time and plant-model mismatch obtaining the desirable performance both, in the reference tracking and in the rejection of strong irradiance disturbances. Simulations and real experimental tests in AQUASOL desalination plant solar field are presented to show the advantages of the proposed controller. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction Numerous industrial thermal processes make use of the avail- able solar resources by means of solar collectors to achieve a desirable process temperature. In some cases, this temperature must be maintained to assure optimal operating conditions in the system. Thus, from a control point of view, the goal in this kind of plants is to maintain the desired outlet solar field temperature using the water flow as the manipulated variable. Nevertheless, these kinds of systems are very sensitive to the daily solar cycle, the clouds level, and the atmospheric conditions. Therefore, all these variables must be taken into account in the controller design procedure [1]. It is possible to find a huge number of papers in the literature related to advanced control techniques in solar fields using model predictive control, adaptive control, gain scheduling, time-delay compensation, optimal control, robust control, fuzzy logic control, or neural network controllers [2]. Nonlinear model predictive control (NMPC) has been also used to control solar collectors plants [3,4]. NMPC technique becomes an interesting option to account for this problem because of the process characteristics. Nevertheless, the optimization problem is more difficult to solve due to non- linearities and then, advanced optimization tools are needed. Several approaches have been proposed in the literature to over- come these optimization problems [5e8]. One of them is to consider a linear optimization based on local linear models of the process [9], where a linear MPC problem is solved at each sample time using a linear model which is re-computed using the infor- mation of the current operation point. A better solution can be obtained if successive linearizations are performed at each sampling time [10]. In [11], a Local Linearization Trajectory (LLT) approach is used, where multiple pre-computed linear models are used to obtain the predictor model needed in the MPC. A net of local models can be also used to capture the process dynamics in different operation points [12]. In this group of controllers it is possible to find also different approaches [13]: (i) a single linear MPC algorithm (LMPC) is used and the model is obtained combining a set of pre-defined linear models computed at the different operation points; (ii) several LMPC, one tuned for each linear model, are used and a combination of them is used to find the final control action. In these two cases the crucial point is the * Corresponding author. Tel./fax: þ34 950 01 56 77. E-mail addresses: cayala107a@hotmail.com (C.O. Ayala), lidia.roca@psa.es (L. Roca), joguzman@ual.es (J.L. Guzman), julio@das.ufsc.br (J.E. Normey-Rico), beren@ual.es (M. Berenguel), luis.yebra@psa.es (L. Yebra). 1 Tel.: þ34 950 38 79 64. 2 Tel.: þ55 48 37 21 76 70. Contents lists available at ScienceDirect Renewable Energy journal homepage: www.elsevier.com/locate/renene 0960-1481/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.renene.2011.03.037 Renewable Energy 36 (2011) 3001e3012