Low-dimensional Modeling And Fast Optimization For Greenhouse Environment Based On Proper Orthogonal Decomposition

Greenhouse is a kind of semi closed and multi-phase coupled complex distributed parameter system.The spatial and temporal distribution and change of its internal environmental parameters are important factors affecting crop growth and energy consumption.In this paper,aiming at the contradiction between the accuracy and complexity of greenhouse microclimate model and the difficulty of controller design,the key problems such as high-resolution modeling and parameter optimization of greenhouse environment are studied by introducing model reduction method.The main contents are as follows:(1)The spatial information of the key parameters(velocity field,temperature field and carbon dioxide distribution)of greenhouse environmental system is obtained by CFD method,and the original parameter space is projected onto the optimal subspace by the eigenorthogonal decomposition(POD)tool,According to the field greenhouse physical model,the greenhouse CFD simulation model based on airpak3.0 platform is established,and the temperature and wind speed sensor array is built to verify the accuracy of CFD and dimensionality reduction model.(2)In order to further reduce the approximation error of the reduced order model,BPSO algorithm is used to optimize the grid points generated by CFD.The results show that the distribution of mesh nodes screened by BPSO is still relatively uniform,and the fitness function converges during the screening process.Reconstruction of the temperature field according to the selected grid node set can improve the modeling accuracy and prove its effectiveness.(3)A fast optimization strategy based on low dimensional model is proposed considering the spatial effects of the changes of greenhouse environmental system parameters.This method combines the pod parameter subspace with the multi-objective optimization algorithm,and constructs the pod subspace of temperature field,air flow field and carbon dioxide distribution by extracting the CFD results.On this basis,the second generation of non-dominated sorting evolutionary algorithm(NSGA-II)is used to optimize the above parameters.In the process of optimization iteration,multidimensional interpolation is used to solve the environment response of pod subspace.Compared with the traditional method,the optimization strategy has the advantages of high spatial resolution,small computation and good real-time performance.