شبیهسازی با استفاده از ماشین یادگیری و رگرسیون چندگانه خطی در مهندسی هیدرولیک.

Background and Objectives Artificial intelligence models as powerful methods in modeling complex nonlinear problems have a significant ability, and this has been proven in numerous articles. Artificial intelligence has been used in various issues, including engineering, medicine, etc. The success of these methods compared to analytical and numerical methods, their easiness, speed, and accuracy. Today, one of the challenges of human life is the issues related to water resources management. This study has investigated the performance of artificial intelligence and regression models in water resources problems. Various researches have been done in the case of modeling and parametric analysis of water resources. However, this study used artificial intelligence (Learning Machine) models to simulate water's qualitative and quantitative parameters. The models used in this study are SelfAdapting Extreme Learning Machine (SAELM), Least Square Support Vector Machine (LSSVM), Adaptive Neuro-Fuzzy Inference System (ANFIS), and Multiple Linear Regression (MLR). Due to the growing global population, one of the most critical challenges is access to safe drinking water. Due to its location in the semi-arid region and low rainfall in our country, Iran, this danger is felt more than ever. One of the serious issues is the salinity leakage into groundwater resources. In this study, an attempt has been made to simulate the leakage of salinity dynamic flow into the freshwater resources of the coastal aquifer, using artificial intelligence and statistical models. Methodology The study area in this work is Mighan Wetland in Markazi province. Annual rainfall occurs in small amounts in this area. According to the statistical results, the maximum and minimum rainfall values range from 461 mm in the northeast to 208 mm in the center of the Arak plain. In this study, qualitative and quantitative parameters: water salinity, total dissolved solids (TDS), chloride ion (Cl), sampling time (t), electrical conductivity (EC), salinity, and groundwater level (GWL) were simulated. This work used ANFIS, LSSVM, SAELM, and MLR models for simulation. In this study, data from 173 months of sampling were used. 80% of the sample size was used for training and 20% for testing models. After simulation and obtaining the results, the Wilson Score method performed the uncertainty analysis without continuity correction. In this method, the prediction error (ei), the mean prediction error (Mean), and the standard deviation of the error are (Se). If the mean error value in predicting the target variable is positive, the model's performance is Over Estimated (OS). Also, if the average value of the model error is negative, the model's performance is Under Estimated (US). Moreover, the Uncertainty Analysis results with a significance of 5% were obtained. Also, five approaches measured the performance of the models. The proposed approaches were: 1) Evaluation of prediction by accuracy chart, 2) Performance evaluation by mathematical indices, 3) Performance evaluation by Uncertainty Analysis by Wilson Score method without continuity correction, 4) Accuracy evaluation by error distribution charts, and 5) Performance evaluation by discrepancy rate (DR) charts. Finally, all the results are given at the end of each section, respectively. Findings The simulation was performed using artificial intelligence and regression models. The simulation results showed higher accuracy of artificial intelligence models. The results showed that different models were successful in predicting water parameters. Approach 1- According to the prediction accuracy charts, 16 charts were drawn, and the most accurate models are depicted in Figures 4 to 7. After modeling, the results showed that the most accurate models in simulating groundwater parameters were the SAELM model in GWL simulation. According to the results, the SAELM model in GWL and EC simulation, LSSVM in TDS simulation, and MLR in Salinity simulation were superior models.Approach 2- According to the performance measurement indices, the results showed that the SAELM model was the best in simulating parameters (EC) and (GWL). Then the LSSVM model was also the most accurate in modeling the (TDS) parameter. MLR was the best model in (Salinity) parameter simulation. Approach 3- Uncertainty analysis was performed based on the Wilson score method. The performance of the models in the simulation showed that the SAELM model's performance was underestimated, and other superior models in the simulation had overestimated performance. Approach 4- The best accuracy was assigned to SAELM and MLR models based on the error distribution diagrams. Approach 5- Based on the DR diagrams, SAELM and MLR models were the most accurate models in the simulation. Conclusion In all five performance approaches, the top models were introduced as follows:According to the performance measurement indices, each model successfully estimated one of the quantitative and qualitative parameters. Finally, study results showed that the SAELM was the best model for simulating EC and GWL parameters. It had the least computational error. Then LSSVM was also the most accurate model in estimating and modeling the TDS parameter. Also, MLR was the best model in Salinity parameter simulation. The uncertainty analysis results showed that the performance of the SAELM was underestimated, and other superior models were overestimated. The best accuracy was assigned to SAELM and MLR models based on the DR diagrams. SAELM and MLR models were the most accurate models based on the error distribution charts.Finally, it should be noted that based on the simulation results, the following predictions are possible: - The graphs gradually showed more insignificant changes in estimating GWL changes by the SAELM model. However, these changes will continue their downward procedure. - Based on the prediction diagrams, EC changes are almost like a time series and will continue similar to past behavior in the future changes. These changes will also occur in less time. - Based on the estimation of the TDS changes, the LSSVM charts behave like time series and seem the future behavior is similar to past time series. The changes occur in the same range. - Based on the estimation of Salinity changes, the MLR showed the changes of this parameter within a higher range. This tolerance occurred with higher values than before. [ABSTRACT FROM AUTHOR]