پيش‌بيني آورد آبي رودخانه‌ ي گرگانرود با استفاده از روش‌هاي تركيبي آنتروپي تقريبي و شبكه هاي عصبي مصنوعي

Combined Approximate Entropy Model and ANNs to Predict Inflow at Gorganrood River

يك از مباحث كليدي در مدل هاي پيش بيني كننده ي سري زماني، تعيين تعداد تاخير در مدل پيشنهادي است. در نوشتار حاضر، تعداد تاخير مدل سري زماني پيش بيني كننده ي آورد رودخانه با استفاده از معيار آنتروپي تقريبي و روش شبكه هاي عصبي به دست آمده است. روش كار به اين صورت بود كه ابتدا رابطه بين بلوك هاي تاخير و آنتروپي تقريبي در ايستگاه هاي آب سنجي آق قلا و قزاقلي در رودخانه ي گرگانرود بررسي و مشخص شد كه بهترين تعداد تاخير در ايستگاه هاي آب سنجي آق قلا و قزاقلي به ترتيب برابر 12 و 10 عدد جهت رسيدن به يك مدل جامع پيش بيني كننده ي آورد رودخانه است. در ادامه، نيز دو مدل ARIMA و شبكه هاي عصبي به همراه ميزان بلوك تاخيري به دست آمده براي پيش بيني دبي رودخانه ي گرگانرود استفاده شد. نتايج نشان از كارايي مدل هاي توسعه داده شده در منطقه ي مورد نظر است، كه مسئله ي مذكور با توجه به نتايج به دست آمده در ايستگاه هاي مورد نظر براي برنامه ريزي آينده مي تواند كارساز باشد.

Prediction of river inflow along with other parameters such as sediment load, flood magnitude, and so on plays an important role in Water Resources Management planning and reservoir operation program. In this regard, many attempts have been devoted so far by the researchers to predict inflow or flow discharge in the rivers. One of the most important rivers in the Golestan province that highly influences climate changing is Gorganrood river. Assessment of chaotic behavior of time series like inflows of rivers by the entropy measure is an important facility in Water Resources Management projects. In addition, this tool can be employed to extract the number of embedding dimensions to predict time series by models like the ARIMA and ANNs. To this end, this paper employed the capability of Approximate Entropy (ApEn) measure as one of the famous models to capture irregular behavior of a time series. Then, ARIMA and ANNs models are implemented to predict monthly inflows at Gorganrood River as the biggest river in Golestan province. The models are developed by using AQqala and Ghazaghli hydrometric stations gathered data. Final results show that at AQqala and Ghazaghli stations, to have an informative predictor model, the number of embedding dimensions must be set to 12 and 10, respectively. In addition, it is concluded that the developed models are accurate enough to be applied in another period of the times in the studied case. Sensitivity analysis of the ARIMA and ANNs models versus various embedding dimensions proves extracted values of embedding dimensions obtained by the ApEn. In addition, the evaluation of the ApEn curves proves the effect of the constructed dams like Voshmgir, Golestan, and Boustan dams on the environmental process and river behavior. The sensitivity analysis versus embedding dimensions clarifies the effects of these parameters over ARIMA and ANNs models.