Intelligent Algorithm Basin Flood Forecasting System Modeling And Software Integration System

Parameter's calibrations of watershed hydrologic forecasting models and flood real-time correction are very important and difficult jobs. Traditional optimal methods to calibrate model parameters need artificial interference and can't entirely consider main characteristic indexes of flood process at the same time. Up to now, mere has not an effective method that correct real-time floods for conceptual hydrologic forecasting model with complex structures and numerous parameters. The advance of Fuzzy Sets Theory (FST), Artificial Neural Networks (ANN) and Genetic Algorithms (GAs) and their application to various fields, provide new approach and probability to resolve parameter's calibrations and flood real-time correction. The successful trials on resolving parameter automatic calibration and real-time correction have been obtained and effective models have been established through combining FST, ANN and GAs with traditional hydrologic forecasting methods. These models are verified using long serial hydrologic data of a large-scale watershed and the results show that these models are very effective. These researches open new ways to real-time forecasting and provide a good application instance. The main research contents are listed in the following:1. GAs is applied to parameter automatic calibration of the Xinanjiang Model, which has been widely applied in humid and semi-humid region in China. The calibration of the model consists of runoff yield parameters optimization and flow concentration parameters optimization. The runoff yield parameters optimization is based on a simple GAs. A new GAs, where the membership degree of alternative is denoted as the evaluation function of the GAs and is obtained by fuzzy optimal model with limited alternatives and multi-objectives, is presented and applied to optimize flow concentration parameters. Also, the rehabilitation strategy is used to deal with nonlinear constrained problem in flow concentration stages. The methods presented herein can automatically and quickly obtain satisfied parameters of hydrologic forecast model. The better simulation results can be obtained when flood peak discharge, peak time and overall water balance are chosen as objectives than single objective. 34 historical floods from 12 yearsin Shuangpai Reservoir are applied to calibrate the model parameters and 11 floods in recent two years are utilized to verify these parameters. Results of study and application show thatGAs not only can improve forecast accuracy but also is an efficient and robust means.2. The flood phenomenon is complex, fuzzy and stochastic, but also there exist some rules among these phenomenons. ISODATA (Iterative Self-Organizing DATA), is used to classify floods. These classified floods are the input data of parameter classification calibrations in the Chapter 5.3. ISODATA cannot be applied to the large sample and real-time online classification. An effective flood classification model, which is based on ANN, is presented. Flood classifier can be established by training samples. In order to find the most effective flood classification model, standard BP algorithms and its improved algorithms such as momentum method, based on strategy of momentum and adaptively adjusted learning rate, Polak-Ribievre method of conjugate gradient, BFGS method of Quasi-Newton and Leveberg-Marquardt have been tested using long serial hydrologic data of a large-scale watershed. The ANN algorithms mentioned above in training and verifying samples can give satisfied results, but there exist great differences in time. Finally. PNN (Probabilistic Neural Networks) is also used to classify floods. Whole results show that both BP and PNN models are feasible and effective in solving the problems of flood classification. The above researches are the basis of establishing real-time correction model.4. Two new methods about watershed real-time flood correction are presented. The first method is flood classification calibration, first classifying floods using ANN and then calibr...