Model predictive control strategy for precision irrigation towards water saving agriculture

Abstract

The field of precision agriculture has widely utilized drip and capillary irrigations for water saving. However, most of the existing controllers for drip irrigation are not capable of adapting to the changing dynamics of soil, plant and weather effect in real-time. Also, investigations have revealed that capillary irrigation has potential for high water saving capability. Moreover, it suffers under-performance due to inaccurate determination and rapid successes of the internet of things (IoT) and advanced control theory are being integrated for improving monitoring system to achieve precision irrigation. In this thesis, two case studies namely a data-driven model predictive control (MPC) strategy for drip irrigated Cantaloupe plant and a Kalman filter based-PID controller for fibrous capillary irrigated Mustard leaf are proposed. An IoT based cultivation experiment on Cantaloupe plant and Mustard leaf was carried out for data collection on soil, plant, and weather parameters using the drip and capillary irrigations. The data collected are used to develop predictive models using system identification in MATLAB to obtain an accurate model representing the changing dynamics of the systems of both case studies. The predictive model for Cantaloupe cultivation was further used to design MPC and Laguerre network-based MPC for real-time drip irrigation management. To investigate the management of the capillary irrigation system, a Kalman filter based-PID control strategy was developed using the predictive model from Mustard leaf cultivation. The developed model based controllers were then integrated with network of sensor devices using IoT for real-time monitoring of the soil, plant and weather parameters to manipulate the Δh based on plant demand. The effectiveness of the model based controllers was simulated using Simulink and deployed on Raspberry Pi to validate their performance. Three greenhouses (GH1, GH2 and GH3) were used for cultivation experiments to test the performance of the developed control system. The data-driven modelling results for drip irrigation show that the ARX model has better accuracy in terms of MSE of 0.753 with an estimated fit of 91.31%, while a state-space model with an estimated fit of 92.3% was equally identified for the capillary irrigation dynamics. The experimental results show a better performance for the proposed controller deployed in GH2 with 30% savings in water and fertilizer, water productivity index of 23 g/L, and fruit sweetness of 13 Brix greater than existing evapotranspiration-based controller in GH1 for the cultivation of Cantaloupe plant. The Kalman filter based-PID controller for capillary irrigated Mustard leaf in GH3, was able to optimally estimate and control the water supply depth resulting to better water productivity index of 16% and improvement of yield when compared with the fuzzy logic controller deployed in GH3 for benchmarking purpose. Also, the investigation of the computation complexity of the MPC and its suitability for control of irrigation system shows that the Laguerre network-based MPC demonstrated a better computational efficiency when benchmarked with a discrete linear quadratic regulator method for drip irrigation control via simulation. This thesis has contributed to the enhancement of monitoring and advanced control strategies for precision irrigation scheduling towards the realization of improved yield, water saving, and efficient use of fertilizer in agriculture.