Reliable Sensing and Smart Control Techniques in Agricultural Irrigation

Abstract

The freshwater resource is quite limited and is being stressed due to growing human population and climate fluctuations. Agricultural irrigation is a major consumer of fresh water and therefore plays a critical role in potential water savings. The limited water resource requires the irrigation water use efficiency to be much higher than ever before. Moreover, irrigation plays a pivotal role in producing good quality and yield of crops of which importance is extremely vital to the public’s subsistence. However, conventional irrigation systems and scheduling methods are oversimplified often resulting in not only a huge amount of water loss but also a reduction in crop productivity. By deploying soil moisture sensors in the field, the irrigation water application amount can be based on the real-time soil water content and therefore unnecessary irrigation can be avoided. By this way, the water use efficiency (WUE) of irrigation can be significantly improved. However, the success of such a mechanism heavily relies on the assumption that the sensors can reliably convey representative soil moisture information with acceptable accuracy. A reliability-driven soil moisture sensing methodology is developed and discussed in this dissertation. It includes a genetic algorithm based optimization technique and a fault detection technique. The results indicate that the proposed methodology considerably improves system reliability in terms of mean time to failure (MTTF). To further improve the WUE and automate the irrigation management, a deep reinforcement learning based irrigation optimization approach and an automated scheduling method for fixed-zone irrigation systems were developed to automate the irrigation process and achieve precise water application. The deep reinforcement learning irrigation optimization approach automatically determines the optimal or near optimal water application for an individual irrigation zone, while the automated scheduling method arranges irrigation tasks of a large number of irrigation zones in a multi-zonal system. The scheduling method prevents water hammer and assures that the system operates under a set of pre-defined hydraulic constraints. It can save time and reduce errors as compared to solutions based on manual computation and control. Simulation results show that the proposed irrigation optimization approach and the scheduling method can optimize irrigation control for each zone, increase water use efficiency and achieve fully automated scheduling for multi-zonal management scenarios. A real-time data acquisition system and a web-based micro-irrigation controller applying these ideas are built. The data acquisition system can collect real-time environmental data. Based on these data, one can do manual control through our controller to turn on and turn off the specific irrigation management zone. The controller can also work in a hybrid mode doing automated scheduling given the water application amounts of zones and in a fully automated mode using the deep reinforcement learning to get the optimized water application amounts.