| Solar greenhouses,mostly without heating equipments,receive solar energy and provide plants with effective radiation for photosynthesis by opening quilts in the daytime in winter,while cut off the cold air and ensure the indoor temperature meeting the needs of the crop growth by covering quilts at nights,provide appropriate environment temperature for anti-season fruits and vegetables.If the quilt is opened too early or covered too late,it would lead to low temperature inside the greenhouse even to cause damage or harvest reduction.In contrast,solar energy would not be utilized efficiently if the quilt is opened too late or covered too early.Aiming at the problem that it is hard to maximize the utilization of solar energy without low temperature damage,this paper developed a temperature prediction model for a greenhouse after opening the quilt in the morning and a lowest temperature prediction model after covering the quilt at night,presented a scientifical time controlling method for opening/covering the quilt,while taking the minimize temperature threshold of the crop growth into consideration.Then provided a scientific method for the control of opening/covering the quilt of solar greenhouses,and further compared the application effect of the method with traditional method by effect verification tests.The results showed that this method expanded the light duration of solar energy without low temperature damage,and was great helpful to improve the crop yields.The main work and conclusions of this paper are shown as follows:(1)Experiment platform establishing and data preprocessing.Based on previous research and collected data,it can be known that if the temperature is below the crop critical growth temperature low-temperature damage will occur.If the light radiation is lower than the light compensation point,photosynthesis could not be done efficiently.An overall test plan was designed according to the environment needs information using an existing greenhouse.Test datas was preprocessed by Kalman filter algorithm,maximum and minimum normalization,correlation analysis method.First,based on the ZigBee Wireless sensor technology and GPRS technology,an experimental data acquisition platform was built by using an environmental monitoring and control system previously developed by the laboratory.And then pre-experiments were carried out to determine the location of the temperature monitoring node:6.5 meters away from the back wall,20 meters away from the west wall,and 1.5 meters away from the ground.Based on the results of the pre-experiment,the main influencing factors of the indoor temperature were determined to be:indoor air temperature,indoor carbon dioxide concentration,indoor air humidity,indoor light radiation,outdoor light radiation,outdoor air temperature.The experiment of covering and uncovering the quilt was designed,and a total of 18144 sets of environmental sample sets were collected.pretreatment the data from experiment,determine that the modelling environmental factors for the indoor real-time temperature prediction model of a solar greenhouse are indoor temperature,indoor light radiation,indoor air humidity,indoor carbon dioxide concentration,outdoor light radiation,outdoor air temperature,and curtain quilt state factors.The indoor minimum night temperature prediction model,and the modeling environment factors for indoor nighttime coldest temperature prediction model are the temperature of the indoor air at the time of the cover,the temperature of the outdoor air,the outdoor minimum temperature at night,and the indoor minimum temperature at night.(2)Study on an indoor real-time temperature prediction model of a solar greenhouse.Taking 20 min as the forecast duration,4536 groups of new model sample sets were obtained.Through comparison experiments,determined the network structure of NARX neural network.Node number is 6-10-1,the hidden layer activation function is tagsig function,the output layer activation function is purelin linear transfer function.The training function is trainlm function,and empirically determine error performance function is mse function.Then the real-time temperature prediction model based on NARX Neural network was established and the model determining coefficient R~2 is 0.99.Finally,the input?parameters of the model quilt condition factors and indoor light radiation were optimized to build the decision model for uncovering the quilt.The results show that the maximum absolute error of the modified NARX neural network temperature prediction model is 0.80℃and the maximum relative error is 5.84%,which can provide a reliable basis for the decision of opening the quilt.(3)Study on an indoor lowest temperature prediction model after covering quilt at night for a solar greenhouse.Only one covering-quilt operation was performed a day,and 62groups of model sample sets were obtained to build the model.,There are only 59 groups of useful sample sets remained.There was no quilt operation in 3 days because of the impact of rain and snow.The penalty parameter C and kernel function parameter g were determained to be 2.83 and 0.06 based on K-CV method.The network kernel function was designed and tested as RBF kernel function.The prediction model of indoor lowest temperature at night was established based on SVR regression,while the model determining coefficient R~2 is 0.96.The covering quilt decision model was built after modifying the outdoor lowest temperature at night.The results show that the maximum error is 0.70℃by using the modified SVR temperature prediction model,and the maximum relative error is 6.10%,which can provide a reliable basis for the decision.(4)Field tests were put based on the control methods of opening and covering the quilt.The total illumination length,radiant heat accumulation,accumulative temperature and yield of the test greenhouse and the comparative greenhouse are compared in the field experiment.The results show that the illumination time increased 15.15%,the radiant heat accumulation increased 48.74%,the effective accumulative temperature increased 19.15%,and the output was increased by 41.04%by using the designed intelligent control system in the 21-day field test. |
PDF Full Text Request