جستجوی مقالات مرتبط با کلیدواژه « Data-driven methods » در نشریات گروه « آب و خاک »

Desalination of saline and brackish water as a relatively permanent resource is a reliable solution to compensate for water shortages. The Electrodialysis process, as one of the desalination methods, separates almost unsalted water from saline water and is becoming a solution for water scarcity all over the worldworldwide. Modeling methods allow provide the study of desalination of saline and brackish water by the Electro Dialysisdialysis, as well as predicting the behaviors and process patterns of these systems.  The purpose of this study is to model the Electrodialysis process by Design of Experiments and to evaluateing the GMDH neural network method in estimating the separation percentage and the output flux of the brackish water Electrodialysis cell.

The volumetric flow values in six levels (1, 2.5, 5, 10, 15, and 20 mL min-1), solute concentration in three levels (200, 500, and 1000 mg L-1), temperature in three levels (50 , 60, and 70°C), voltage at three levels (10, 20, and 30 V) and pressure at three levels (200, 400, and 800 Pa) were extracted. In order to model the effect of each above inputs on the separation percentage and output flux of brackish water Electrodialysis cells, The full factorial design and the GMDH neural network were are used to model the effect of each above inputs on the separation percentage and output flux of brackish water Electrodialysis cells. In modeling the output of the Electrodialysis cell output using the GMDH neural network, after determining the input variables, randomization, normalization and segmentation of input and output variables were performeddone. For modeling, 90% of the data (437 samples) were randomly assigned for training and 10% of the data (49 samples) were are used for validation. On the other hand, due to having specific levels for each of the inputs, it was possible to use a Full Factorial experimental design.

The linear results of statistical regression analysis table results showeshowd that there is a significant difference between all simple and interaction effects of the treatments for the response variable of separation percentage and flux at the one percent level. With decreasing volume flow and solute concentration, and increasing temperature, voltage, and pressure, the separation percentage and output flux showed a significant increase. The results of modeling with the GMDH neural network showed demonstrates that in the training step, the prediction accuracy of separation percentage (R2=0.90, MBE=-0.16 and RMSE=7.48) and flux (R2=0.79, MBE =0.001 and RMSE=0.08).  The These criteria same values in the testing step were are 0.85, -3, and 9.37 for separation percentage and 0.78, -0.008, and 0.07 for flux. Correlation of target and output data, proximity of target and output values to y = x diagram, low transgression of error values from zero value and proximity of error distribution (histogram) to normal distribution were obtained.

After determining the significant differences of all simple and interaction effects of treatments for response variables, the means were compared. It is ideal to Hhaveing more separation percentage and output flux response variables was ideal. Therefore, the results of comparing the means showed illustrate that the best volumetric flow level equal to 1 ml / min, input flow concentration of 200 mg / l, temperature of 70 ° C, input voltage of 30 volts and pressure of 800 Pascals, lead to the highest separation percentage and output flux. In addition, with decreasing volumetric flow,  and decreasing solute concentration, and increasing temperature, increasing voltage and increasing pressure, separation percentage and output flux showed a significant increaserise. Also, according to the statistical indicators for the whole data, appropriate correlation, slightly underestimation, and small error were are obtained. ThereforeThus, using the Design of Experiments and GMDH neural network has suitable accuracy for modeling desalination of brackish water using the Electrodialysis process.

During the past decades, landslides have been a significant subject of research as a consequence of their devastated nature. Landslides are common geomorphic processes in mountain areas and are responsible for mass movements involving rock materials, regolith and/or soil debris. for manufacture roads, railways, water pipe line and electric line, the preparation of landslide distribution map is very much significant. Determine the occurrence of future landslides depend on the geological, geomorphological and hydrological processes that led to instability in the past and also at present. To evaluate terrain susceptibility to landslides, a number of different techniques are used, ranging from qualitative assessments based on expert judgment, which are intrinsically subjective to quantitative assessments based on advanced statistical techniques or mathematical models. the steps of methodologies that were applied in the current study, including six steps. Step1. data sources that are used in the current study including data related to field Surveys, historical reports, topographic maps of 1:50,000-scale, meteorological data, geological map of 1:100,000-scale, A digital elevation model (DEM) with the resolution of 30 m £ 30m was extracted from the ASTER GDEM data, The Landsat 8 OLI images with the resolution of 30 m × 30 m. Step2. Preparing the inventory map. In this study, a landslide inventory map with a total of 80 landslide events was provided by the extensive field survey and interpretation of aerial photos. Step3. Landslide-conditioning factors. Step4. Multicolinearity analysis of landslide conditioning factors. In the current study, 12 factors were used as conditioning factors. These include elevation, slope, plan curvature, stream length, distance from streams, topography wetness index, surface area ratio, distance from roads, lithology, distance from faults, rainfall, and land use. Step5. Combination of EBF data driven and AHP knowledge driven models according to the relation between the landslides location and the different datasets. Step5. Validation of models using AUC and SCAI inficators. Results of the spatial relationship between landslide and conditioning factors using the EBF (belief, disbelief, uncertainty, and plausibility) model are shown in Table 3. Comparison between the belief map and the disbelief map showed that belief values were high for areas where disbelief values were low and vice versa. It revealed that high potential of landslide occurrence was for the areas with high degrees of belief and low degrees of disbelief. The high uncertainty values were in the areas with low belief values. Weighting of conditioning factors by AHP showed that parameters of lithology, elevation, distance to road, slope, and rainfall are the most effective prediction factors in landslide occurrence. The consistency ratio shows 0.034 value, which is reasonably good accuracy value, which reflect the high accuracy of ranking consistency between the factors. Due to some shortening of the AHP knowledge driven and EBF data driven models when applied individually in landslide susceptibility mapping, it can be overcome by using ensemble techniques. The AUC results showed that the success rate and prediction rate for ensemble model are 0.872 (78.3 %), 0.903, respectively. results of SCAI values of the ensemble model is desirable, in the high and very high susceptibility classes. The resultant landslide susceptibility map show that the high susceptibility areas are mainly distributed along the northwest to west direction in the study area. this map can provide very useful information for planners, decision makers, and engineers in slope management and land use planning in landslide areas

The investigation of local scour below hydraulic structures is so complex that makes it difficult to establish a general model to provide an accurate estimation for the local scour dimension. During the last decades, Data Driven Methods (DDM) have been used extensively in the modeling and prediction of unknown or complex behaviors of systems One of these methods is Group Method of Data Handling (GMDH), that is a self-organization approach and increasingly produces a complex model during the performance evaluation of the input and output data sets. So, the objective of this study was to investigate the potential of the GMDH method in the accurate estimation of local scouring geometry (maximum scour depth, the distance of maximum local scour depth till Ski-jump bucket and length of local scour) below the Siphon spillway with Ski-jump bucket energy dissipaters for a set of experimental data. 80% of data set was used for the training period and the remaining data set was used for the test period. The average values of MSRE, MPRE, CE and RB for the nonlinear second order transfer function (FUNC1) were calculated to be 0.92, 0.02, 8.74, -0.01; also, for the nonlinear first order transfer function (FUNC2), they were 0.85, 0.02, 10.43 and -0.02, respectively. The results indicated that the performance of FUNC1 was better than FUNC2. Also, the value of the coefficient of determination (R2) for the estimation of local scour dimension using different methods such as s linear regression, nonlinear regression and ANN indicated the high performance of the developed model of GMDH in the accurate estimation for local scour dimensions.

River flow forecasting in rivers, is one of the most important components of hydraulic and hydrological processes in water resource management. For various hydrological applications such as water and sediment budget analysis, operation and control of water resources projects, the accurate information about flow value in rivers is very important. For this reason hydrologists use historical data to establish a function relationship between water level and discharge which is known stage-discharge relation or rating curve (RC). With the recent advancements in artificial intelligence and soft computing in water resource studies, there is a choice of better techniques to modeling hydraulic and hydrological processes. This paper describes the use of the Group Method of Data Handling (GMDH) that is a self organization modeling approach base on data, for stage-discharge relationship modeling at Philadelphia station in Schuylkill River, USA. 12 Different input combinations including the previous stages and discharges with 2 kind of active function are used. The accuracy of model was evaluated using Root Mean Square Error (RMSE), Mean Percent Relative Error (MPRE) and Nash–Sutcliffe model efficiency coefficient (NASH). The values of statistical parameters RMSE, MPRE and NASH for the best model of stage- discharge relationship in this river in validation period are 15.8, 0.303 and 0.999 respectively.