Three layer wavelet based modeling for river flow
All existing methods regarding time series forecasting have always been challenged by the continuous climatic change taking place in the world. These climatic changes influence many unpredictable indefinite factors. This alarming situation requires a robust forecasting method that could efficiently...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2019
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/81666/1/SirajMuhammedPandhianiPFS2016.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | All existing methods regarding time series forecasting have always been challenged by the continuous climatic change taking place in the world. These climatic changes influence many unpredictable indefinite factors. This alarming situation requires a robust forecasting method that could efficiently work with incomplete and multivariate data. Most of the existing methods tend to trap into local minimum or encounter over fitting problems that mostly lead to an inappropriate outcome. The complexity of data regarding time series forecasting does not allow any one single method to yield results suitable in all situations as claimed by most researchers. To deal with the problem, a technique that uses hybrid models has also been devised and tested. The applied hybrid methods did bring some improvement compared to the individual model performance. However, most of these available hybrid models exploit univariate data that requires huge historical data to achieve precise forecasting results. Therefore, this study introduces a new hybrid model based on three layered architecture: Least Square Support Vector Machine (LSSVM), Discrete Wavelet Transform (DWT), correlation (R) and Kernel Principle Components Analyses (KPCA). The three-staged architecture of the proposed hybrid model includes Wavelet-LSSVM and Wavelet-KPCA-LSSVM enabling the model to present itself as a well-established alternative application to predict the future of river flow. The proposed model has been applied to four different data sets of time series, taking into account different time series behavior and data scale. The performance of the proposed model is compared against the existing individual models and then a comparison is also drawn with the existing hybrid models. The results of WKPLSSVM obtained from Coefficient of Efficiency (CE) performance measuring methods confirmed that proposed model has encouraging data of 0.98%, 0.99%, 0.94% and 0.99% for Jhelum River, Chenab River, Bernam River and Tualang River, respectively. It is more robust for all datasets regardless of the sample sizes and data behavior. These results are further verified using diverse data sets in order to check the stability and adaptability. The results have demonstrated that the proposed hybrid model is a better alternative tool for time series forecasting. The proposed hybrid model proves to be one of the best available solutions considering the time series forecasting issues. |
|---|