Loading...

groundwater level fluctuations forecasting using conjunction models of Wavelet - Neural-Fuzzy Network (WNF) and Wavelet-Neural Network (WNN) and linear model of ARIMA (case study: Qom plain

Fathi, Bahman | 2012

697 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 43677 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Shamsaei, Abolfazl
  7. Abstract:
  8. Prediction of groundwater level fluctuations is absolutely necessary for the proper manag ement of these precious resources. There are different methods to predict hydrological time series such as groundwater level. Linear models inefficiency in predicting nonlinear and n on-stationary time series cause researchers widely use artificial intelligence techniques such as Artificial Neural Networks (ANN), Fuzzy Inference System (FIS), Genetic Algorithm (GA) and their hybrid models such as Artificial Neural – Fuzzy Inference System (ANFIS). These smart models are capable to simulate nonlinear and non-stationary time series in suitable accu-racy using by a few time and data. By emersion of Wavelet Transformation (WT) as a power-ful tool in the analysis of the non-linear and non-stationary time series and discovering its ca-pabilities, so far a lot of hydrological phenomena have been modeled by it. However, these models rarely have been used in the field of groundwater resources. In this Thesis, the groundwater lev el fluctuations of Qom plain have been predicted using combinational models of Neural Network - Adaptive Fuzzy - Wavelet (WNF) and Neural Network - Wavelet (WNN) and a linear model of ARIMA. At first, the groundwater level, temperature and precipitation time series were d ecomposed into approximation and detail components using by wavelet tran sformation. Thus, these sub-series were used as inputs of ANN and ANFIS models for models training and groundwater level forecasting. Then, these components were composed again in order to r econstruct the original series using wavelet transform. Afterwards, the ARIMA, ANN, ANFIS, WNN and WNF models were validated using obtained results and were compared to gether. The modeling results indicated that the mean value of RMSE for five mo dels of AR IMA, ANN, A NFIS, WNN and WNF in 8 observed piesometers for 1 month ahead fo recas ting is 0.32, 0.29, 0.30, 0.22 and 0.15 meter, For 3 month ahead forecasting is 0.55, 0.49, 0.50, 0.45 and 0.31 meter and for 6 month ahead forecasting is 0.74, 0.64, 0.64, 0.63 and 0.57 meter respectively. To sum up, the nonlinear models were more accurate than linear models and combined models of wavelet transformation were more accurate than non-combined models. Also the WNF models are more capable to predict groundwater level among these models
  9. Keywords:
  10. Wavelet Transform ; Autoregressive Integrated Moving Average (ARIMA) ; Groundwater Level Prediction ; Wavelet - Neural - Fuzzy (WNF)Network ; Wavelet-Neural Network (WNN)

 Digital Object List

 Bookmark

No TOC