تخمين نوسانات تراز آب زيرزميني توسط مدل بهينه موجك-برنامه‌نويسي بيان ژن

Estimation of Groundwater Variations using Optimized Gene Expression Programming model

به­ دليل كاهش منابع آب زيرزميني و خشك­ سالي­ هاي اخير، شبيه­ سازي نوسانات تراز آب زيرزميني از اهميت بالايي برخوردار است. در اين مطالعه، مدل برنامه ­نويسي بيان ژن (GEP) با تبديل موجك (WT) تركيب شد و مدل موجك-برنامه­ نويسي بيان ژن (WGEP) توليد گرديد و تخمين دراز­مدت نوسانات چاه مشاهداتي سراب قنبر در يك دوره 13 ساله بهينه ­سازي شد. در ابتدا داده ­هاي مشاهداتي به دو دسته آموزش (9 سال) و آزمون (4 سال) تقسيم شدند. سپس با استفاده از تابع خود همبستگي، موثرترين تأخيرهاي داده­ هاي سري زماني معرفي شدند. در ادامه با استفاده از اين تأخيرها براي هر يك از مدل­ هاي GEP و WGEP، چهار مدل مختلف تعريف گرديد. با تجزيه و تحليل كليه مدل­ هاي GEP و WGEP، مدل برتر شناسايي گرديد. مدل برتر يا 4 WGEP مقادير تراز آب زيرزميني را با دقت مناسبي شبيه­ سازي كرد. به­ عنوان مثال، مقادير ضريب همبستگي و ضريب بهره وري نش-ساتكليف براي اين مدل به­ ترتيب مساوي با 0/938 و 0/851 محاسبه شدند. مقايسه مدل­ هاي GEP و WGEP نشان داد كه تبديل موجك دقت مدل­ سازي را به­ شكل قابل­ توجهي افزايش داد. به ­عنوان مثال، ضريب عملكرد مدل برتر WGEP در مقايسه با مدل برتر GEP حدودا 14 برابر افزايش يافت. علاوه بر اين، نتايج نشان داد كه تاخيرهاي شماره (t-1)، (t-2)، (t-3) و (t-4) موثرترين تأخيرهاي ورودي بودند.

Groundwater plays a vital role in supplying water demands for different consumptions in dry and semi-dry regions of earth. Iran is considered as an arid and semi-arid region and its groundwater resources have recently shown some significant changes. Owing to the reduction of groundwater resources and recent droughts, simulation of groundwater level variations has significant importance. In some areas of the country of Iran, groundwater levels have been dropped significantly. Therefore, the prediction and simulation of the groundwater level variation are crucially important. In this study, the Gene Expression Programming (GEP) model was combined with Wavelet Transform (WT) to estimate long-term variations of groundwater level (GWL) in the Sarab-Ghanbar observation well over a 13-years period. Firstly, observation data were divided into two sub-samples, 9 years for training and 4 years for testing. Then, the most effective input lags were identified using the autocorrelation function. Next, four different models for each GEP and WGEP method were developed using the lags. The superior model was identified by analyzing all GEP and WGEP models. The superior GEP model simulated the GWL with acceptable accuracy. For instance, the correlation coefficient and Nash-Sutcliffe efficiency coefficient for the model were calculated at 0.938 and 0.851, respectively. A comparison between the GEP and WGEP models showed that the wavelet transforms enhanced the performance of simulation significantly. For example, Variance Accounted For (VAF) index for the best WGEP model was 14 times more than the best GEP model. In addition, the sensitivity analysis indicated that (t-1), (t-2), (t-3) and (t-4) lags were the most influenced input lags.