Article Preview
Top1. Introduction
A greenhouse system is a closed environment where some climate variables can be manipulated in order to obtain adequate climatic conditions, for the development and growth of the cultures, using automatic control strategies (Shamshiri & Ismail 2013). The greenhouse environmental control involves the field of control technology, as the way to optimize inside greenhouse climate based on measured variables and acting on greenhouse equipment (Lu et al., 2015). The dynamics of the climatic variables in a greenhouse are very complex. That is due to the presence of nonlinearities, subjected to strong disturbances (measurable and non-measurable ones) and a high degree of correlation among variables (Frausto & Pieters, 2004; Bennis et al., 2008).
Due to the complexity of the real engineering systems, like the greenhouse system, some importance has been put into implementing Artificial Intelligence (AI) techniques including neural networks, fuzzy logic, neuro-fuzzy, evolutionary algorithms, or some combination among them. Although artificial intelligent methods offer the advantage of the capability of capturing essential functional relationships among the data when such relationships are not a priori known or are very difficult to describe mathematically in situations of the collected data are corrupted by noise. Therefore, they had gained importance and successfully applied in large areas, such as modelling, prediction, control, optimization, business, and financial engineering (He & Ma, 2010).
For plants of high complexity, like greenhouse process, it is of main importance to develop accurate models of the plant which will be used to describe the system behaviour. Furthermore, a perfect model is significant for the parmaters tuning of the controller based on the system’s dynamic model, to design a performance control law (Kiranyaz et al., 2009). In this way, neural networks algorithms are a very sophisticated nonlinear modelling techniques used to perform an accurate modelling of the greenhouse system dynamics for temperature or both air temperature and relative humidity, due to its capability of learning and generalization from examples using the data-driven self-adaptive approach, as long as enough data are presented in the training process (Lai & Zhang, 2009).
Normally, when designing and training a neural network, different architectures must be tried before the one that seems effective is found. Obviously, there is no guarantee that the final selected architecture is the best possible one and for large problems this method becomes impractical. In addition, change in other network parameters such as the learning algorithm or the number of epochs affect the best choice of architecture. This interdependence makes it difficult to find optimal architectures for a given problem (Zhang et al., 2007).
Hence, the heuristic optimization methods which evolve network architectures can solve these problems (Chen et al., 2016; Meng et al., 2016). One of the main advantages of heuristic methods is that they convergence to the optimum solution in more short time than others and convergence fewer to local minimum (Shan Ngan &Wei Tan, 2016).
Particle Swarm Optimization (PSO) algorithm is a new population based heuristic optimization method first proposed by Kennedy and Eberhart (Bas, 2016). It’s an effective swarm intelligence optimization algorithm featuring high search speed and high efficiency. Recently, successful applications of PSO algorithm to the optimization problems attract much attention in various problems intelligence optimization (Aladag et al., 2012). Although, PSO algorithm can be used in terms of the learning algorithm to assist in network training phase in order to adapt connection weight and biases adaptation. This benefits to the ANN since its generalization capability can be improved (Mohammadi & Mirabedini, 2014).