| At present,greenhouses have become important agricultural facilities in many areas,especially the areas north of the Yangtze River.Suitable temperature and humidity in the greenhouse will promote the growth and development of crops,improve quality and yield,and how to regulate the microclimate in the greenhouse to create the best for crops.Optimal growth environment has become the main research content at present.The microclimate environment of the greenhouse is coordinated and controlled by the external climate(atmospheric environment)of the greenhouse and various controllers(controllers such as temperature and humidity)inside the greenhouse.The precise prediction of climate change inside the greenhouse is essential for realizing effective climate control.Therefore,the establishment of an accurate temperature,room temperature and humidity prediction model is a prerequisite for achieving temperature,room temperature and humidity control,and greatly improves the intelligent level of greenhouse control.Based on the historical data of the external climate of the greenhouse,the historical data obtained by the internal related controller,and the target temperature and humidity,the greenhouse temperature and humidity prediction model is constructed based on the neural network,and further based on the prediction model,the greenhouse control model is designed and developed.First,21600 sets of data of 150 days in a life cycle of tomato are preprocessed.Aiming at the changing trend of greenhouse temperature and humidity,three kinds of data are constructed using artificial neural network(ANN),autoregressive neural network(NARX)and long-short-term memory network(LSTM).A greenhouse temperature and humidity prediction model,and then the model is trained,and the performance of the model is compared from two aspects of different time steps and different training data sets.At the same time,in order to realize the real-time prediction of the greenhouse,an automatic learning algorithm is given.The comparison experiment between the online model trained by the automatic learning algorithm and the model trained with offline data shows that the prediction performance of the online model has been improved,and the real-time temperature and humidity of the greenhouse are realized.predict.For temperature prediction,ANN uses a 30-day data set to make a 30-minute advance temperature prediction with the highest accuracy,R~2 is 0.99;for humidity prediction,LSTM uses a 20-day data set to make a 30-minute advance humidity prediction with the highest accuracy,R~2 is 0.97.The ANN temperature prediction model is used to construct an optimized ventilation control model based on the output feedback neural network(OFNN).The cost function is calculated according to the difference between the predicted temperature after 30 minutes and the target temperature,and the switches of multiple top and side windows are determined and output Control signal to achieve optimal ventilation control.Experiments show that OFNN is better than P-band control.Using the ANN temperature prediction model and the LSTM humidity prediction model,the temperature and humidity control model is constructed,and the cost required for control is calculated through the cost gate,and the lowest cost control combination is obtained.The greenhouse energy saving control is realized in the three seasons of spring,summer and winter.Simulation and field experiments prove that OFNN has better control performance and lower energy consumption than P-band.A prediction model of temperature,room temperature and humidity was constructed through neural network,which accurately predicted the temperature and humidity in the greenhouse in the next 30 minutes.On this basis,a temperature and humidity control model was developed,and a cost function was added to optimize the control signal and find the optimal combination.The smallest control cost realizes the best regulation of temperature,room temperature and humidity,and improves the intelligence of greenhouse control. |
PDF Full Text Request