Prediction Model of Transpiration Rate of Strawberry in Closed Cultivation Based on DBN-LSSVM Algorithm

A theoretical basis for irrigation of greenhouse crops will be provided by the establishment of a prediction model for transpiration rate of strawberry leaves in solar greenhouse of closed cultivation. This paper selects strawberry in solar greenhouse of closed cultivation as the research object. With sufficient water supply conditions, the deep belief network and least squares support vector regression (DBN-LSSVM) have been used to establish a prediction model for transpiration rate of strawberry leaves in solar greenhouse of closed cultivation, thus predicting the transpiration rate of strawberry through greenhouse environmental parameters. First, the multi-scale feature vectors of meteorological parameters in greenhouse have been extracted by using the deep belief network (DBN) method to eliminate the correlation of variables, thus improving the predictability and generalization ability of the model. The extracted feature vectors have been used to train and optimize the LSSVM model, finally obtaining the prediction model of transpiration rate of strawberry leaves in solar greenhouse of closed cultivation, which have been compared in experiments with the traditional BP neural network and LSSVM model.The results indicate that when training samples become a certain amount, the prediction accuracy and regression fitting degree of DBN-LSSVM can be higher than that of the two traditional models. It performs best with the largest coefficient of determination R-c(2) of 0.972, smallest root mean square error RMSEC of 0.623.In the case of several training samples involved in modeling, the prediction of the model performs better than that of BP neural networks, but slightly lower than that of the LSSVM model. With the training sample size increasing, the prediction accuracy and regression fitting degree of the model have been also steadily improved and significantly superior to the traditional model. The transpiration rate model of strawberry leaves have been established to realize the prediction of leaf transpiration rate through the basic meteorological parameters in greenhouse with high simulation accuracy and obtainable parameters. As a preferable exploration of the research on transpiration rate simulation in short time scale, it is of certain theoretical significance and excellent application prospect. (C) 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.