Research On Temperature And Humidity Control Mechanism Of Northern Greenhouse Based On RBFNN

Greenhouse environment models are the important basis of greenhouse structural design and environmental control. Greenhouse is a complex system with characteristics of strong coupling, long delay and nonlinear. Therefore, mechanism analysis methods difficultly established its models. This caused high-energy consumption for greenhouse production and unsatisfactory control effect. Neural network (NN) modeling technology can flexibilly acquire parameters and nonlinear characteristics of greenhouse. Now research objects on greenhouse environmental model and control mostly lay in southern region of china. Similar study on northern greenhouse was deficient. High-energy consumption for northern greenhouse production restricts its development, temperature difference between day and night in spring and autumn is great, variables strongly couple each other, modeling and control for it is very difficult. Environmental control mechanism of northern greenhouse was furtherlly studied in this dissertation. Main topics are as follows:According to the problem that mechanism analysis methods difficultilly establish accurate model of greenhouse. Because radial basis function NN (RBFNN) has such characteristics that simpler structure and approximating any nonlinear procedure. RBFNN structure and different optimal algorithms were combined to learn and optimize NN structure and parameters, northern greenhouse models were established in this dissertation. OLS algorithm has such characteristics that smaller coumputing amount, quicker operation speed and designing simpler NN structure. When OLS algorithm designs structure of NN, hidden centers selected are related to orthogonalized order. Therefore, it can not establish the simplest structure of NN.In fact, RBFNN performances are only related to hidden centers selected, not related to its'order. Hence, selection of hidden centers is thought as combination optimization problem. ES mainly use selection and variation operation during evolutionary, this is fit for optimizing hidden unit centers. ES algorithm was applied to search the optimal hidden centers and its number in sample set. RBFNN model of northern greenhouse was established.If stronger noise data were included in sample set, NN may overfit these data to decrease its generalization. Regularizing method is an effective measure to improve NN generalization. But regularizing factor must be optimized. NN models based on OLS and ES algorithm were designed after spreads of RBF were assumed. Accuracy of spreads influences NN performance. Therefore, PSO algorithm was used to optimize regularizing factor and spreads, ROLS algorithm used optimal results to design model structure in this dissertation.The output layer of RBFNN is linear layer, its output weights are usually acquired by linear least squares algorithm. However, greenhouse is a large-scale system with multiple input-output. If strong correlation among input variables exists, or data quantity is deficient for calculating cost, this will increase error of model regression coefficients, and decrease model accuracy. PLS algorithm can select and recombine the input information, effectively resolve the problems mentioned above. Hence, PLS algorithm was applied to extract principal components from the input variables. These components are orthogonal and complementary. The model was regressed by these components. These principal components and hidden centers were hierarchically determined.The development of northern greenhouse was restricted by energy consumption for production. Now control accuracy of temperature in greenhouse was usually emphasized, but energy consumption was ignored. Large mount of energy is saved by means of the characteristics of integrating temperature. Humidity control greatly restricts application of integrating temperature control. According to growing habit of phalaenopsis aphrodite and control scheme of integrating temperature, integrating temperature control schemes of northern greenhouse environment during spring were studied in this dissertation. These lay theoretical basis for controlling greenhouse environment and energy-saving production.In accordance with the problems of traditional and modern control theory applied on greenhouse environment control, according to the model and control schemes acquired in this dissertation, NN controller was designed to regulate the temperature and humidity in northern greenhouse. NN controller regulates all actuators according to the bias of temperature and humidity. OLS algorithm was applied to accelerate NN learing. The effect of humidity on integrating temperature control was perfectly resolved.Environmental model and controller of northern greenhouse are studied in this dissertation. These research results provide theoretical and technical basis for healthy development and sustainable production of greenhouse agriculture in northern region.