Optimising Fungicide Applications on Winter Wheat using Genetic Algorithms

Abstract A genetic algorithm is used in a decision support system to select the combinations of chemicals and the timing of successive treatments for the optimal control of fungal diseases in winter wheat crops, using a simulation model to predict the performance of different treatments. The search space is large and discrete, making the use of conventional optimisation methods impractical. Furthermore, the user requirements specify that the method must supply lists of near-optimal solutions, which fits with the use of populations of solutions in the genetic algorithm. Substantial improvements in the performance of the algorithm were obtained by tuning the fitness, selection, reproduction and replacement methods for the optimisation of short-term and long-term decisions. These also ensured rapid convergence in the former and prevented premature convergence in the latter. The algorithm has proved to be effective at finding optimal and near optimal solutions within an acceptable time. When compared with exhaustive searches for cases where this is possible (short-term planning with restricted choices), it typically finds 5–8 of the top 10 plans and a similar number of the next 10. The results of the system in field and user trials have been good.