The BP Network-forecasting Model For Flood Based On Adaptive Genetic Algorithm

The forecasting of hydrology intelligence in watercourse is very complex. It is a non-linear dynamic course for influenced by all kinds of factors. And it is difficult to confirm the variety and relations of those factors. Artificial neural networks have the quite ability of treat with great complex non-linear dynamic system. Genetic algorithm have the real advantage of looking for the best in the whole room. The thesis surround the two kinds of knowledge of mathematical domain closely, raise the adaptive genetic algorithm, improve the artificial neural network-forecasting model for flood, and unit the two arithmetic, draw a good conclusion by applying it to the forecast of hydrology intelligence exert myself. The real tasks are as the follow.According to the disadvantage of genetic algorithm in searching the best in the whole space all along, slow in convergence velocity, low in efficiency, this thesis proposes animproved genetic algorithm-adaptive genetic algorithm, in which individuals breed onlyin powerful solution space after evolving for several generations and then deleting weak solution space by inducting in adaptive idea. The new method can make efficiency and precision raised and time shortened. And give out the theoretical analyse.According to the complexion of exceeding historical flood peak in the period of forecast of flood. In the artificial neural network-forecasting model for flood water level, the thesis standardize the datum of input and output layers to the interior between 0.2 and 0.8, and leave the space of (0,0.2) and (0.8,1), which make the model can forecast the flood water level easily. In addition, the thesis took the certain coefficient in "Forecasting norm for hydrology intelligence" as the objective function in order to make the forecasting result more clear.According to the static network of artificial neural network, the thesis introduces the method of putting-off cells into the forecasting model of complex watercourse. The model has the ability of identifying time-sequence by distilling the character of time from those swatch patterns.The thesis use the artificial neural network-forecasting model for flood water level and the artificial neural network-forecasting model for flood water level based on adaptive genetic algorithm to forecasting of flood on Beijiang River in Zhujiang Delta. Those models have been tested, the test results show the forecasting model with the reasonable original datum of the input layer element and the reasonable forecasting period can obtain the precision forecasting datum.