عنوان مقاله [English]
The development of intelligent irrigation systems with the aim of providing the plant with water in a timely and adequate manner is an important strategy to increase the quantity and quality of agricultural products with minimal water consumption. On the other hand, determining the amount of water required by the plant largely depends on an accurate estimate of evapotranspiration in vegetation. In this study, neural network model was used to estimate the evapotranspiration in a lettuce hydroponic rotary culture system. The actual evapotranspiration of lettuce in the hydroponic cultivation system was estimated in order to design the irrigation system with the help of fuzzy logic model. Cultivation time was 30 days and data sampling time from temperature and humidity sensor was 10 minutes. According to the desired results obtained from the evaluation of the planted crop and water consumption, the efficiency of fuzzy model was proved. Therefore, it was used as a criterion for validating the neural network model of this study. The data volume for the neural network model was about 4500, which was randomly divided into three parts, 70% (training), 15% (evaluation) and 15% (test). Different ANN structures were evaluated to find the most suitable neural network architecture. The best result was achieved with BR algorithm with three hidden layers in an 8-10-10 topology and tansig transfer function in three hidden layers and output layer. For this architecture, the absolute error and coefficient of determination were 0.43 and 99.98%, respectively. Furthermore, considering a simple single layer network (one hidden layer), the BR algorithm with 8 neurons and logsig and tansig transfer functions for the hidden and output layers, were selected as the best model. The error and coefficient of determination of this structure were 0.79 and 98.84%, respectively. According to the sensitivity analysis, humidity and temperature were the most important parameters in predicting evapotranspiration, respectively.