| Summary: | 博士 === 國立臺灣大學 === 土木工程學研究所 === 92 === The objective of the theme is to apply artificial neural networks in hydrosystem, and modify their learning algorithms. Five major parts are included in the study, which are described as follows.
In the part 1, a back-propagation neural network (BPN) with two hidden layers is developed to forecast the typhoon rainfall. First, the model configuration is evaluated using eight typhoon characteristics. The forecasts for two typhoons based on only the typhoon characteristics are capable of showing the trend of rainfall when a typhoon is nearby. Furthermore, the influence of spatial rainfall information on the rainfall forecasting is considered for improving the model design. The semivariogram is applied to determine the required number of nearby rain gauges whose rainfall information will be used as input to the model too. With the typhoon characteristics and the spatial rainfall information as input to the model, the forecasting model can produce reasonable forecasts. It is also found that too much spatial rainfall information cannot improve the generalization ability of the model, because the inclusion of irrelevant information adds noises to the network and undermines the performance of the network.
In the part 2, the radial basis function network (RBFN) is used to construct a rainfall-runoff model, and the fully supervised learning algorithm is presented for the parametric estimation of the network. The fully supervised learning algorithm has advantages over the hybrid-learning algorithm that is less convenient for setting up the number of hidden layer neurons. The number of hidden layer neurons can be automatically constructed and the training error then decreases with increasing number of neurons. The early stopping technique that can avoid over-fitting is adopted to cease the training during the process of network construction. The proposed methodology is finally applied to an actual reservoir watershed to find the one- to three-hour ahead forecasts of inflow. The result shows that the RBFN can be successfully applied to build the relation of rainfall and runoff.
In the part 3, based on the combination of the RBFN and the semivariogram, a spatial interpolation method, named improved RBFN, is proposed. To evaluate the interpolation accuracy of the proposed method, reference surfaces with prescribed semivariograms of different sills and scale parameters are generated. The proposed method as well as two existing methods (ordinary kriging and standard RBFN) is then used in the restoration of these reference surfaces. Among three interpolation methods, the proposed method has the highest interpolation accuracy regardless of the arrangement of sample points. The proposed method is performing well especially when the variance of the reference surface is large. An application of the proposed method to the estimation of the spatial distribution of rainfall also shows that the proposed method can estimate more precisely as compare to the other two methods. The proposed method is recommended as an alternative to the existing methods, because it has a clear principle and a simple structure. Moreover, it provides more flexibility adjusted with stochastic property.
In the part 4, the Self-Organizing Map (SOM) is applied to identify the homogeneous regions for regional frequency analysis. First, the experimental design is applied to test the cluster accuracy of the SOM, K-means method and Ward’s method. These three clustering methods are tested on experimental data sets where one can control the amount of cluster dispersion and know the cluster membership. Among the three clustering methods, the results show that the SOM determines the cluster membership more accurately than other two methods. Finally, the SOM is applied to actual rainfall data in Taiwan to identify homogeneous regions. According to the two-dimensional map, one can find that the rain gauges can be grouped into eight clusters. The heterogeneity test indicates that the eight regions are sufficiently homogeneous. Moreover, the results show that the SOM can identify the homogeneous regions more precisely as compared to the other two methods. Because of unsupervised learning, the SOM does not require the knowledge of corresponding output for comparison purposes. In addition, the accuracy of the SOM is more robust than the traditional clustering methods. Hence, the SOM is recommended as an alternative to the identification of homogeneous regions for regional frequency analysis.
In the part 5, based on the combination of the RBFN and the SOM, a time-series forecasting model is proposed. In the proposed model, the SOM is used to construct the two-dimensional feature map from which the number of clusters (i.e. the number of hidden units in the RBFN) can be figured out directly by eyes, and then the radial basis centers can be determined easily. The proposed model is examined using the simulated time series data. The results demonstrate that the proposed RBFN is more competent in modeling and forecasting time series as compared to the ARIMA model. Finally, the proposed model is applied to actual groundwater head data. It is found that the proposed model can forecast more precisely than the ARIMA model. For time series forecasting, the proposed model is recommended as an alternative to the existing method, because it has a simple structure and can produce reasonable forecasts.
|
|---|
