Fitting Of Short-term Greenhouse Climate Based On Cluster Analysis

Greenhouse microclimate is a quite special environment, which is isolated from outside synoptic environment by the special structure and covering material (transparent, heat-preservation, etc.) of modern greenhouse. To maintain the good greenhouse microclimate is conducive to achieve the ultimate goal of the improvement of crop quality, high yield and low power consumption. The development and regulation of modern greenhouse are symbols of a modernization of the country, so research of modeling on the greenhouse microclimate has great significance to greenhouse construction, production and development of agriculture. Therefore, it is meaningful to take research in modeling greenhouse microclimate, which is one of the topics in973research project (No.2010CB955905-1).At present, there are two methods, that is, system identification modeling and mechanism modeling, to model the greenhouse microclimate at home and abroad. We find that selection of cluster center is static, thus leading to increase the number of iterations and that cluster analysis can not guarantee that the result is optimal through evaluating cluster analysis. This paper based on the system identification of the input and output data is to improve the performance of the cluster analysis algorithm by reference to the artificial neural network approach. The details are as follows:Firstly, using the fuzzy control theory constructs the function of greenhouse climate. It based on system identification is to fit greenhouse climate with the use of clustering the sample data, which creates the relationship with cluster analysis.Secondly, the paper combines the respective advantages of two commonly used methods of the Partition-based clustering, and it also enhances the performance of the cluster analysis algorithm by changing the selection of the cluster center.Thirdly, artificial neural network has the ability of self-learning to obtain the law of sample data through training. Therefore, it can re-cluster the sample data by using the output of cluster analysis.Finally, the paper confirms the superiority of algorithm by using simulation test, which selects fitness as measurement criteria with the use of Matlab.