Comparative Analysis of ANN and SVM Models Combined with Wavelet Preprocess for Groundwater Depth Prediction

Ting Zhou; Faxin Wang; Zhi Yang

Comparative Analysis of ANN and SVM Models Combined with Wavelet Preprocess for Groundwater Depth Prediction

2017

Ting Zhou | Faxin Wang | Zhi Yang

Reliable prediction of groundwater depth fluctuations has been an important component in sustainable water resources management. In this study, a data-driven prediction model combining discrete wavelet transform (DWT) preprocess and support vector machine (SVM) was proposed for groundwater depth forecasting. Regular artificial neural networks (ANN), regular SVM, and wavelet preprocessed artificial neural networks (WANN) models were also developed for comparison. These methods were applied to the monthly groundwater depth records over a period of 37 years from ten wells in the Mengcheng County, China. Relative absolute error (RAE), Pearson correlation coefficient (r), root mean square error (RMSE), and Nash-Sutcliffe efficiency (NSE) were adopted for model evaluation. The results indicate that wavelet preprocess extremely improved the training and test performance of ANN and SVM models. The WSVM model provided the most precise and reliable groundwater depth prediction compared with ANN, SVM, and WSVM models. The criterion of RAE, r, RMSE, and NSE values for proposed WSVM model are 0.20, 0.97, 0.18 and 0.94, respectively. Comprehensive comparisons and discussion revealed that wavelet preprocess extremely improves the prediction precision and reliability for both SVM and ANN models. The prediction result of SVM model is superior to ANN model in generalization ability and precision. Nevertheless, the performance of WANN is superior to SVM model, which further validates the power of data preprocess in data-driven prediction models. Finally, the optimal model, WSVM, is discussed by comparing its subseries performances as well as model performance stability, revealing the efficiency and universality of WSVM model in data driven prediction field.

Mostrar más [+]