A hybrid neuro-fuzzy approach for greenhouse climate modeling

Greenhouse climate is a nonlinear time variant multi-input multi-output system with delay time and non-minimum phase. Because of the variety of parameters and strong coupling, developing a physical model based on thermodynamic principles is rather difficult. Having the ability of universal approximations, Artificial Neural Networks (ANN) can be well adapted to model the nonlinear behavior of greenhouse climate. However, a random selection of the initial parameters makes their convergence slow and suboptimal. Fuzzy logic makes it possible to solve this problem due to its capability to handle both numerical data and linguistic information. In this paper, a hybrid neuro-fuzzy approach based on fuzzy clustering is proposed in modeling a greenhouse climate built upon the experimental data. In the first stage, the nearest neighborhood method generates the necessary fuzzy rules automatically. Then, the cluster centers were used as the initial condition for the applied neural network trained and optimized using the Self-Organized Feature Mapping (SOFM) algorithm. The simulation results have shown the efficiency of the proposed model.