Abstrakt: |
Dez dam has an essential role in controlling the floods and water supply in the plains of Khouzestan. Therefore, accurate and reliable prediction of the flow input to the reservoir is crucial to make decisions about the exploitation and management of this reservoir. In this research, a novel WAVELET-ARIMA-NARX model was developed to predict the monthly inflow to Dez dam reservoir. The WAVELET method was used to break down the time series into sub-series and better analyzing, the ARIMA model was used to model the linear component of the analyzed series, and the NARX model was used to model the error resulting from the WAVELET-ARIMA model. The NDVI index was also used to measure the accuracy change of the WAVELET-ARIMA-NARX hybrid model. The results showed that the prediction performance of the WAVELET-ARIMA-NARX hybrid model has improved significantly compared to the ARIMA model. In such a way according to the RMSE criterion, the prediction accuracy in the verification and training stages has decreased by 74% and 82%, compared to the ARIMA and WAVELET-ARIMA model, respectively. Also, the NDVI parameter along with the average temperature and rainfall of the basin as an input of the NARX model has increased the accuracy of the model. In the model with 10 neurons in the hidden layer, compared to the two-parameter model of rainfall and NDVI with 15 neurons, in evaluation section, the MAE values have decreased from 27.2 to 18.5 and the RMSE from 0.45 to 0.26. These values indicate the importance and impact of simultaneous consideration of three parameters on forecasting accuracy [ABSTRACT FROM AUTHOR] |