Improving pesticide application using model predictive control with active demand management for precision agriculture

Abstract

The use of robotics in precision agriculture has significantly improved productivity over the years by automating activities such as spraying, harvesting, and sowing. Without robots, the typical way of applying a fixed amount of pesticide over a specific season or time not only results in economic loss via wasted pesticide but also leads to sub-optimal results in terms of pest control leading to low yield and loss of capital in the agricultural industry. Existing techniques for automating pesticide application only focus on the problem of best route selection for the pesticide spraying robots given a predetermined demand for pest without due consideration to the dynamic and complex nature of the interaction between pests and host plants. Moreover, these techniques either consider simplistic scenarios with only one robot or even if they consider a fleet of vehicles, the proposed solution results in poor coordination among the robots resulting in slow field coverage times as well as higher charge or fuel consumption. To solve these issues, this research introduces the concept of active demand management to the agricultural vehicle routing problem (VRP) along with an efficient solution for vehicle routing using model predictive control (MPC). Demand management is introduced and modelled using the mass-spring damper system along with an estimate of the risk of a pest infestation to obtain a state-space model. Analogous to an electromechanical system, the notion of damping is related to pesticide demand, with an objective to reduce demand. The resulting state-space model is then utilized to solve the agricultural VRP for two cases, a single vehicle and a fleet of vehicles. For the single vehicle case, a discrete-time MPC plant model is used to optimize the delivery of pesticide such that the demand is minimized or in other words to reduce the risk of pests while using a greedy algorithm to efficiently route the vehicle to a specific area in an agricultural field. To solve the problem of pesticide application using a fleet of vehicles, an MPC algorithm is converted into a mixed integer linear programming (MILP) optimization problem. The optimization problem in addition considers the charging and capacity constraints for a set of autonomous vehicles (AV) also considers the evolution of charge and pesticide amount carried by the AVs. This results in an overall solution to autonomous mobility on demand in the agricultural industry. Extensive MATLAB/Simulink simulations show that the proposed technique not only results in significant reduction of up to 80% in terms of field coverage time but also results in a reduction of up to 93% in terms of charge consumption as compared to state of the art existing techniques. The percentage improvement achieved demonstrates the advantage of using MPC.