Dalgacık dönüşümü kullanılarak veriye dayalı akım tahmin modellerinin iyileştirilmesi

Daily streamflow forecasting is conducted in this study using several data-driven models (DDMs): artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS) and support vector machine (SVM). Seven days ahead and one month ahead streamflow being less predictable wavelet transformation (WT) is used as preprocessing to improve the performance of the models. Two new hybrid models were proposed using multi-gene genetic programming (MGGP) and extreme gradient boosting (XGB) as a selection tool to select only the important scales obtained from the continuous wavelet transformation (CWT) to be them imposed into ANN and extreme learning machine (ELM). Hindcast and real forecast experiment are conducted to investigate the performance of the WT-based models in the real forecasting. The results show that daily forecast can be implemented successfully using DDMs and ANN has the highest performance in comparison to ANFIS and SVM. The proposed two models outperformed the models uses discrete wavelet transformation (DWT) as more information can be included in the model. Finally, the performance of the models using WT-based hybrid models for both CWT and DWT in hindcast experiment was found as increases due to the incorrect application of WT. Mostly, WT is applied onto the time series, divided into calibration and testing subsets to be then imposed into DDMs and that sends some future information into the model. In the real forecast experiment, the WT-based hybrid models have less performance than the stand-alone DDMs in which no preprocessing applied.