عنوان مقاله [English]
نویسندگان [English]چکیده [English]
One of the strategies for efficient use of water resources in agriculture is development of new irrigation systems including pressurized irrigation methods. Field evaluation and accurate data of pressurized irrigation systems performance are considered as the crucial tools for correct operation of these systems in different terms plan. One of the most important performance evaluation criteria in the design of pressurized irrigation systems, such as solid-set sprinkle, is water distribution uniformity index. On the other hand, field measurements of the water distribution uniformity index in different climatic and hydraulic conditions and projects executive specifications require spending too much time and cost. Therefore, use of indirect methods such as intelligent models can be useful. By checking the studies in simulation of water distribution uniformity coefficient in sprinkle irrigation systems, no research was found using of adaptive neuro- fuzzy inference system (ANFIS) method. Therefore, the present study aimed to check the performance of adaptive neuro- fuzzy inference system and compare with results of gene expression programming method in estimating the water distribution uniformity coefficient. This research was done in solid-set sprinkle irrigation system in different climatic, hydraulic and physical conditions.
The research method consisted of two parts: field measurements and simulations by ANFIS and gene expression programming intelligent models. For this purpose, a solid-set sprinkle irrigation system with considering of different arrangements of pipes and sprinklers were designed and performed. Then, 54 field experiments were done to evaluate the performance of the solid-set sprinkle irrigation system. In each experiment, water cans were used to determine the water distribution uniformity coefficient. Input parameters of adaptive neuro- fuzzy inference system and gene expression programming were the combination of: climatic factors (average of temperature, relative humidity, average of wind speed and direction) and physical factors (arrangement and different distances of sprinklers and sprinkler model (amount of output volumetric flow rate)). Output parameter, in all simulations, was the water distribution uniformity coefficient (cu) in percent. 70 percent of obtained field data were used for learning of the models and 30 percent for testing them. In order to compare and evaluate the performance of the intelligent models in estimating the water distribution uniformity coefficient, Pearson correlation coefficient statistics, root mean square error and mean absolute error were used.
Results showed that generally with increasing laterals distance and wind speed, the amount of water distribution uniformity coefficient decreased. Field observations were indicated in the same terms, Ambo’s model sprinkler had the better results in water distribution uniformity than vyr-155. The simulation results showed that use of all input data, including volumetric flow rate of sprinkler, sprinklers distances, wind speed, wind direction, relative humidity and average of temperature as well as considering the effective radius equal one, lead to gain the best results. So in ANFIS model, the maximum amount of correlation coefficient (R) and root mean square error (RMSE) for the test phase were obtained equal to 0.77 and 7.7 %, respectively. By eliminating the parameters of relative humidity and average of temperature from the model inputs, no significant changes were observed in the results. While by eliminating wind speed parameter, results of model's output (including RMSE index) were changed significantly. It can be concluded that the water distribution uniformity coefficient values in the experiments were strongly influenced by wind speed. Also, the best performance of gene expression programming model was related to the combination of the input data including: volumetric flow rate of sprinkler, sprinklers distance, wind speed and direction. So, the maximum amount of correlation coefficient achieved in the test phase was equal to 0.72 and the lowest amount of RMSE was 7.13%. One of the advantages of gene expression programming model comparing with ANFIS model and other intelligent models is offering the optimal mathematical equation between the dependent variable of uniformity coefficient and the other independent variables (inputs). Generally, the performance of methods had little difference between gene expression programming and adaptive neuro- fuzzy. Inference and sensitivity of the models showed the temperatures and wind speed had the lowest and the most effect on water distribution uniformity coefficient changes, respectively. The estimated amounts of checking for water distribution uniformity coefficient indicated that intelligent models, as well as factors effect such as wind speed and sprinklers distances, have been able to simulate the reducing amount of water distribution uniformity.