multiobjective evolutionary algorithms. Then, in the so-called compatible control
level the control variables from the Pareto optimal fronts are chosen as the input
vector of the control system (Hu et al. 2011b). In this way an optimal control of the
greenhouse system is implemented using a feedback control architecture. A
multiobjective controller optimization by means of evolutionary algorithms was
performed for the greenhouse environment (Hu et al. 2011b). A physical model
including air temperature and humidity ratio of the greenhouse climate was considered. A Proportional-Integral-Derivative (PID) controller was tuned by stating
and solving a multiobjective optimization problem using a multiobjective evolutionary algorithm. More recently, a multiobjective hierarchical control architecture
for greenhouse crop growth has been proposed (Ramírez-Arias, et al. 2012).These
researchers defined the multiobjective optimization problem as a summation of the
maximization of profits (J 1 ), crop quality (J 2 ), and water-use efficiency (J 3 ) during
the optimization interval. The variables present in the objectives are functions of
the air temperature and/or electric conductivity (EC), the measurable disturbances:
photosynthetically active radiation and CO 2 concentration. The solution provides
both diurnal and nocturnal set point trajectories of EC and indoor air temperature
for the control horizon. In order to solve the problem a goal attainment algorithm,
based on a sequential quadratic programming algorithm, was used. It is possible to
specify priorities for the objectives using weights that can be modified at each
iteration. The constraints for the design variables are defined by maximum and
minimum values of temperature and EC coming from experts’ knowledge. The
results showed acceptable trajectories for short- and long-crop cycles. Regarding
the control architecture it has three layers: a layer with a timescale of days or
months solves the multiobjective optimization problem proving setpoints for the
greenhouse crop. A layer with a timescale of days could modify the set points
coming from the long-term layer according to short-term weather predictions in
order to deal with uncertainties associated to the long-term layer. A third layer
with a timescale of minutes or seconds provides the controllers that minimize the
error between the calculated set points and the actual measured variables in the
greenhouse system.
14.3 Intelligent and Soft Computing Control
Artificial intelligence (AI) techniques, intelligent control methodologies, and soft
computing approaches such as fuzzy systems, neurocomputing, and evolutionary
algorithms have been applied recently to design control algorithms for the
greenhouse environment (Occhipinti and Nunnari 1996; Caponeto et al. 2000). In
this section, the main achievements regarding these AI techniques to control the
greenhouse environment are summarized.
14 Control Strategies of Greenhouse Climate
411
level the control variables from the Pareto optimal fronts are chosen as the input
vector of the control system (Hu et al. 2011b). In this way an optimal control of the
greenhouse system is implemented using a feedback control architecture. A
multiobjective controller optimization by means of evolutionary algorithms was
performed for the greenhouse environment (Hu et al. 2011b). A physical model
including air temperature and humidity ratio of the greenhouse climate was considered. A Proportional-Integral-Derivative (PID) controller was tuned by stating
and solving a multiobjective optimization problem using a multiobjective evolutionary algorithm. More recently, a multiobjective hierarchical control architecture
for greenhouse crop growth has been proposed (Ramírez-Arias, et al. 2012).These
researchers defined the multiobjective optimization problem as a summation of the
maximization of profits (J 1 ), crop quality (J 2 ), and water-use efficiency (J 3 ) during
the optimization interval. The variables present in the objectives are functions of
the air temperature and/or electric conductivity (EC), the measurable disturbances:
photosynthetically active radiation and CO 2 concentration. The solution provides
both diurnal and nocturnal set point trajectories of EC and indoor air temperature
for the control horizon. In order to solve the problem a goal attainment algorithm,
based on a sequential quadratic programming algorithm, was used. It is possible to
specify priorities for the objectives using weights that can be modified at each
iteration. The constraints for the design variables are defined by maximum and
minimum values of temperature and EC coming from experts’ knowledge. The
results showed acceptable trajectories for short- and long-crop cycles. Regarding
the control architecture it has three layers: a layer with a timescale of days or
months solves the multiobjective optimization problem proving setpoints for the
greenhouse crop. A layer with a timescale of days could modify the set points
coming from the long-term layer according to short-term weather predictions in
order to deal with uncertainties associated to the long-term layer. A third layer
with a timescale of minutes or seconds provides the controllers that minimize the
error between the calculated set points and the actual measured variables in the
greenhouse system.
14.3 Intelligent and Soft Computing Control
Artificial intelligence (AI) techniques, intelligent control methodologies, and soft
computing approaches such as fuzzy systems, neurocomputing, and evolutionary
algorithms have been applied recently to design control algorithms for the
greenhouse environment (Occhipinti and Nunnari 1996; Caponeto et al. 2000). In
this section, the main achievements regarding these AI techniques to control the
greenhouse environment are summarized.
14 Control Strategies of Greenhouse Climate
411