Forecasting Sugarcane Yield Using (μ+λ) Adaptive Evolution Strategies

Abstract

Sugarcane is a very important crop in the sugar industry. However, the annual amount of harvested sugarcane is oftentimes uncertain, posing risks to sugarcane mills in terms of raw material supply. The forecast for the sugarcane yield would allow the mills to plan sugar production accordingly. This paper proposes (μ+λ) adaptive evolution strategies, which generate equations for accurately forecasting the sugarcane yield. Our proposed scheme combines the advantages of two algorithms: genetic algorithm and evolution strategies. Specifically, the genetic algorithm is good for determining patterns of forecasting equations, while the evolution strategies are used to tune the equations’ coefficients. The test data is collected from sugarcane farmers in 24 provinces of Thailand during 2010–2014. The equations obtained by the proposed method are 80 % accurate on average, outperforming the previous method (back propagation neural network) in all data set.