Intelligent control of agriculture production in greenhouses

Guesbaya, Mounir (2022) Intelligent control of agriculture production in greenhouses. Doctoral thesis, Université Mohamed Khider Biskra.

Text PhD Thesis_merged - Final version for the University of Biskra.pdf Download (16MB)

Abstract

The agricultural greenhouse system has undergone significant developments in recent years. Greenhouse microclimate is the phenomenon under study in this work. Its modelling and control processes are complex tasks to be performed mainly due to the strong nonlinearity of the phenomenon. In this thesis, a set of contributions in greenhouse microclimate modelling and control, including implementing computational intelligence algorithms, have been accomplished. The second chapter briefly describes the experimental greenhouses used in this thesis. Initially, due to the lack of an experimental greenhouse, a wooden-structured polyethene-covered greenhouse prototype was constructed and used as a small-scale nursery under arid climate conditions (moderate desert climate) in Meziraa, Biskra, Algeria. A low-cost microcontroller-based data acquisition system with a wireless connection was designed (hardware and software) and installed in the greenhouse with several low-cost sensors. It was used to gather instant information on the essential inside and outside climate variables. A dataset of five days was successfully acquired for modelling, estimation and experimental validation purposes. Secondly, a metal-structured polyethene-covered commercial-sized experimental greenhouse under Mediterranean climate conditions was exploited. It is located at “Las Palmerillas” Experimental Station, a property of the Cajamar Foundation in Almería, Spain. It is equipped with all the necessary professional sensors, actuators and data acquisition systems. A set of sufficient reliable datasets of fifteen days were obtained in different agri-seasons and used for different purposes such as microclimate modelling and control, online parameter estimation and real-time experimental validation. In the third chapter, two contributions were achieved. Firstly, a grey-box model for greenhouse temperature prediction under moderate desert climate conditions has been proposed. This contribution stands on reformulating a white-box model to make it independent of the availability of accurate values of the static parameters of its elements. The model has become less complicated by alleviating the coupling between its parameters, which makes it easier for the identification algorithm to find the optimal parameter values. A variant of the Particle swarm optimisation algorithm (PSO) called Random Inertia Weight PSO (RIWPSO) was used to identify the parameters of the proposed model by calibrating it against the experimental data. The constructed greenhouse prototype has been used to validate the proposed temperature model. The simulation results show that particle swarm optimisation has successfully achieved the desired optimality. The experimental validation process has confirmed the suitability of this model to be implemented to study and predict the greenhouse temperature, and it has emphasised the successful prediction with satisfactory accuracy. Secondly, an enhanced variant of the bio-inspired metaheuristic Bat Algorithm (BA) has been proposed and called the Random Scaling-based Bat Algorithm (RSBA). The proposition includes modifying the exploitation of the standard BA by randomly making the scaling parameter changes over the iterations. It has been dedicated to the same task of calibrating the proposed thermal grey-box model. It has been assessed as the same as PSO, primarily on the same simulated greenhouse temperature model with the assumed parameters. The simulation results have shown the superiority of the proposed RSBA compared to the standard BA in terms iv of convergence and performance accuracy. To experimentally investigate the proposed RSBA algorithm, the same experimental dataset from the greenhouse prototype has been used. The obtained prediction results are found to be in good agreement with the measured ones, which show the effectiveness of the proposed RSBA in identifying the real greenhouse thermal model. Finally, a comparative study was conducted between the RSBA and the RIWPSO. The BA has shown a faster convergence than PSO at the start of optimisation, but its convergence speed was reduced at the end. BA and PSO have shown superb performance in accurately finding the optimal solutions. However, PSO has shown a superior performance than BA in terms of time consumption regarding the problem of interest. Greenhouse microclimate modelling is a difficult task mainly due to the strong nonlinearity of the phenomenon and the uncertainty of the involved physical and non-physical parameters. The uncertainty stems from the fact that most of these parameters are unmeasurable or difficult to measure, and some are time-varying, signifying the necessity to estimate them. As the first contribution in the fourth chapter of the thesis, a methodology for online parameter estimation is proposed to estimate the time-varying parameters of a simplified greenhouse temperature model for real-time model adaptation purposes. An online estimator is developed based on an enhanced variant of the Bat Algorithm called the Random Scaling-based Bat Algorithm. It allows the continuous adaptation of the internal air temperature model and the internal solar radiation sub-model by estimating their parameters simultaneously by minimising a cost function, intending to achieve global optimality. Constraints on the search ranges are imposed to respect the physical sense. The adaptation of the models was tested with recorded datasets of different agri-seasons and on a real greenhouse in real time. The evolutions of the time-varying parameters were graphically presented and thoroughly discussed. The experimental results illustrate the successful model adaptation, presenting an average error of less than 0.28 °C for air temperature prediction and 20 W m−2 for solar radiation simulation. It proves the usefulness of the proposed methodology under changing environmental conditions. Natural ventilation flux is an important variable to measure or estimate for its significant effect on greenhouse microclimate modelling and control. It is commonly known that it can be mathematically estimated depending on the type and dimension of the greenhouse and its vents and, most importantly, on the vents opening percentage. However, most commercial greenhouses are not equipped with an automatic vent opening system which obligates the grower to perform manual control, in addition to the lack of vent position sensors, due to economic and management reasons. It leads to the absence of the control signal variable representing the vents opening percentage necessary for ventilation flux estimation. This issue has been encountered in this work after attempting to implement the developed adaptive microclimate model based on the online parameter estimator through an IoF2020 platform (internet of food and farm) in a set of commercial greenhouses with manually controlled vents located in Almeria province, Spain. To cope with this issue, the estimation of ventilation flux without using the vent opening percentage was investigated. As a second contribution in the fourth chapter, a virtual sensor for greenhouse ventilation flux estimation is proposed. It has been developed using a nonlinear autoregressive v neural network with exogenous inputs based on principal component analysis using the available measured data and the evolutions of the heat fluxes representing the greenhouse energy balance. Preliminary results show an encouraging performance of the virtual sensor in estimating the ventilation flux with a mean absolute error of 0.41 m3 s-1.

Actions (login required)

View Item