automation have been compared (Ödük and Allahverdi 2011). A zigBee wireless
sensor network and a fuzzy controller has been proposed recently in order to control
temperature and humidity of the greenhouse environment (Yang et al. 2013).
14.3.2 Neural Control
Artificial neural networks (ANN) have proved to be a powerful tool to solve
several types of problems in different fields where approximation of nonlinear
functions, classification, identification, and pattern recognition are required.
Neural networks are mathematical representations of biological neurons in the way
they process information as parallel computing units. In general there are two
types of neural network architectures: (1) static (feedforward), where no feedback
or time delays exists and (2) dynamic neural networks, whose outputs depend on
the current, or previous inputs, outputs, or states of the network (Demuth et al.
2007).
One of the most widely used neural network architectures is the multilayer
perceptron (MLP), which has been proved to approximate almost any continuous
function over a compact subset of R
n , if given enough hidden layers and neurons
within them (Master 1993). Its mathematical proof, based on the Stone-Weierstrass theorem, can be found in Gupta et al. (2003), and it has also been studied by
many other authors such as Blum and Li (1991) and Hornik (1991). An MLP is a
structure mapping an input space R
n into an output space R
m , by adjusting the
connection weights that link the inputs to each of the neurons in the hidden layers
and up to the output layer. Finding the appropriate number of hidden layers,
neurons within each layer and weight adjusting methods, constitute the neural
network design procedure. Determining a combination of these that best approximates the problem solution is a heuristic process that depends on the problem and
type of data being processed.
Due to the complexity of the crop-greenhouse biosystem, some emphasis has
been put into implementing artificial intelligence techniques that offer the
advantage of modeling nonlinear relationships with a reduced number of input
parameters, in contrast to analytical models that are too complex to be practical
and are difficult to implement. Although artificial intelligent methods, including
neural networks, fuzzy logic, neuro-fuzzy, evolutionary algorithms, or some
combination among them, do not offer an explanation of the physical principles,
they do allow for a prediction of the system dynamics provided enough data are
presented in the training process.
In this way, hybrid artificial intelligent techniques such as neuro-fuzzy are
finding more applicability on complex systems such as the crop-greenhouse biosystem (Elfelly et al. 2010). Neural networks and fuzzy logic offer the advantage
of modeling nonlinear system relationships and add robustness in situations of
imprecise and incomplete data, where uncertainty exists. Some application
examples are in the areas of forecasting, control, optimization, and pattern
14 Control Strategies of Greenhouse Climate
413
sensor network and a fuzzy controller has been proposed recently in order to control
temperature and humidity of the greenhouse environment (Yang et al. 2013).
14.3.2 Neural Control
Artificial neural networks (ANN) have proved to be a powerful tool to solve
several types of problems in different fields where approximation of nonlinear
functions, classification, identification, and pattern recognition are required.
Neural networks are mathematical representations of biological neurons in the way
they process information as parallel computing units. In general there are two
types of neural network architectures: (1) static (feedforward), where no feedback
or time delays exists and (2) dynamic neural networks, whose outputs depend on
the current, or previous inputs, outputs, or states of the network (Demuth et al.
2007).
One of the most widely used neural network architectures is the multilayer
perceptron (MLP), which has been proved to approximate almost any continuous
function over a compact subset of R
n , if given enough hidden layers and neurons
within them (Master 1993). Its mathematical proof, based on the Stone-Weierstrass theorem, can be found in Gupta et al. (2003), and it has also been studied by
many other authors such as Blum and Li (1991) and Hornik (1991). An MLP is a
structure mapping an input space R
n into an output space R
m , by adjusting the
connection weights that link the inputs to each of the neurons in the hidden layers
and up to the output layer. Finding the appropriate number of hidden layers,
neurons within each layer and weight adjusting methods, constitute the neural
network design procedure. Determining a combination of these that best approximates the problem solution is a heuristic process that depends on the problem and
type of data being processed.
Due to the complexity of the crop-greenhouse biosystem, some emphasis has
been put into implementing artificial intelligence techniques that offer the
advantage of modeling nonlinear relationships with a reduced number of input
parameters, in contrast to analytical models that are too complex to be practical
and are difficult to implement. Although artificial intelligent methods, including
neural networks, fuzzy logic, neuro-fuzzy, evolutionary algorithms, or some
combination among them, do not offer an explanation of the physical principles,
they do allow for a prediction of the system dynamics provided enough data are
presented in the training process.
In this way, hybrid artificial intelligent techniques such as neuro-fuzzy are
finding more applicability on complex systems such as the crop-greenhouse biosystem (Elfelly et al. 2010). Neural networks and fuzzy logic offer the advantage
of modeling nonlinear system relationships and add robustness in situations of
imprecise and incomplete data, where uncertainty exists. Some application
examples are in the areas of forecasting, control, optimization, and pattern
14 Control Strategies of Greenhouse Climate
413