Fateme Bayat; Hesam Ghodousi; Kazem Shahverdi
Abstract
Introduction: The rapid growth of population, agriculture, urban and industries has led to increasing water demand and competition for its consumptions. The promotion of agricultural water productivity has the main effect on improving water consumption. Water delivery and scheduling methods are important ...
Read More
Introduction: The rapid growth of population, agriculture, urban and industries has led to increasing water demand and competition for its consumptions. The promotion of agricultural water productivity has the main effect on improving water consumption. Water delivery and scheduling methods are important to increase the flexibility of irrigation systems. Among different available methods, the on-request water delivery has higher flexibility than the rotational one and doesn’t need the high cost of automatic systems. The appropriate adjustment of the structures and their operational instructions between successive requests is a function of discharge variation, time interval between operations, coincidence of different request, physical condition of canal and structures and hydrodynamic behavior of the flow, which is a complex task. To obtain the performance of the recently utilized method, i.e., FSL (Fuzzy SARSA Learning), it is necessary to compare it to a traditional method like Artificial Neural Network (ANN). In this research, data from the east Aghili canal was trained for programming water delivery and distribution using MLP (Multi-Layer Perceptron) and RBF (Radial Basis Function) networks of ANN with the on-request method. Finally, the results of the FSL and ANN models were compared.Material and methods: In this research, the MLP (Multi-Layer Perceptron) and RBF (Radial Basis Function) networks of ANN were used to determine the procedure for exploiting the operational instructions of the on-request method in the east Aghili canal, in Khuzestan Province, using its flow and gate opening data. In this research, 70%, 15%, and 15% of data were used to train, test, and validate the model, respectively. The correlation coefficient and root mean square error were used for determining the better method. Modeling of the canal was done using the Irrigation Canal Conveyance System (ICSS) hydrodynamic model. To evaluate the MLP, RBF, and FSL outputs, maximum and average errors of water depth, adequacy, efficiency, equity, and dependability were used.Results and discussion: The operational instructions were determined using the MLP in March 2017 in the east Aghili canal, and were compared to the corresponding determined operational instructions using FSL. According to the obtained results, it was observed that the MPA index in the ANN method in the first and second block of this channel, respectively were 0.952 and 0.919 and in the case of using the FSL method, these values were equal to 0.996 and 1. Also, the MPF index in the simulation using the ANN in both blocks were equal to 1 and in the case of FSL, these values were equal to 0.999 and 0.971. The maximum error of MAE of water level in the first and second block of the study, respectively were equal to 9.2 and 3.8 % and in the case of using the FSL method, these were equal to 5.5 and 7.4 %. The results showed that the MLP was better than the RBF to determine the operational instructions. The MAE and IAE indicators were minimum, and the water delivery indicators were close to their desired values according to the Molden and Gates (1990) criteria. Aldo, it was revealed that the FSL was better than the MLP, however, the MLP results were valid and can be used in practice.Conclusion: In this research, the ANN model was used for determining operational instructions using MATLAB. The training was done using the MLP and RBF using the east Aghili canal data. The ICSS was used for simulating the canal. The results showed that the MLP was better than RBF, and the FSL model was better than the MLP as well. However, both of them can be used in practice.
