Evolutionary Computation In The Optimization Of Crop Growth Models

Modern Digital Precision Agriculture is a whole modern intelligent management decision-making system which is supported by information technology and able to accurately count the real time requirements of the crop considering the providing capability, place variation and change rule of environment, and with orientation, timing, and ration accurately provides the matter and conditions needed by the crop, in order to get high yield, high quality and good efficiency of agriculture. The nicety crop growth simulation model is the base of the intelligent management and decision-making system and an important part of the development of Modern Digital Precision Agriculture. Usually, it needs to re-estimate and optimize the model parameters when a model is used in different conditions. The parameter optimization of crop growth models is a multi-dimension, nonlinear and complicated optimization problem. Most modern optimization means, for example Gauss-Newton, Simulated Annealing and simple Evolutionary Computation, all exist premature convergence, limit of the number of parameters, precision limit or other disadvantages.Analyzing model structure characters, the particularity of crop growth model parameters optimization, and Evolutionary Computation characters, this thesis aims at the greenhouse cucumber growth model of Shanghai Agriculture Academy of Sciences and brings an improved Evolutionary Computation algorithm: Species Compete-Die out algorithm. Simulating the natural and human rule of the alternant and combinative process between the independent evolution in every species and the evolution among different species, based on Niche thought and floating-point genes, this algorithm model classifies all population into some species through picking-up chromosome characters and Dynamic Clustering, adaptively adjusts fitness and the number of species, and makes a combination of evolution in every species and among all species to reduce the possibility of premature convergence. Besides, Species Compete-Die out algorithm adaptively adjusts mutation step, in order to improves local searching performance and accelerates convergence efficiency .The algorithm result on the platform of PC by simulation data shows a better performance comparedwith simple Evolutionary Computation and a good stability.The research of this thesis not only provides a parameter optimization algorithm model for greenhouse cucumber growth model, but also provides a new method for the parameters re-estimate and optimization of different crops(wheat, cucumber, tomato)growth models applying to different environment(greenhouse, farm).