人工神经网络方法在径流预报中的应用

采用ＢＰ神经网络模型，以西北内陆河黑河流莺峡年平均出山地表流量为研究对象，对人工神经网络研究方法在干旱区环流径流预报中的应用进行了初步尝试，结果表明该预报成功率较高，证实了人工神经网络方法应用于流量预报领域的可行性，并分析了该方法在预报过程中的优缺性。

A Study on Runoff Forecast by Aritifical Neural Network Model

In spite of that nonlinear problem of hydrological runoff process have been noticed early, hydrologists elaborate various runoff forecast models by means of available experimental methods and dynamic model methods. Traditional models and methods depend on the parameters that have been initially well estimated. Traditional methods with well test interpretation are usually based on a combination of manual and automated techniques. There is obvious limitation in these methods, because the calibration of parameters involves artificial factors. In this paper a new method, named artificial neural network model(ANN), is presented and the Yinluo Gorge is taken as a case study. The runoff in the Yinluo Gorge is selected as an object to test the ANN method, and the back-propagation model, one of the typical artificial neural network models, is applied. How to forecast runoff by using the artificial neural network model and how to reach a fast convergence speed and high accuracy are described by choosing dynamic learning factor and inertia factor. It is revealed that the ANN method might be referred as an effective technique for runoff forecast. It is well known that the neural network model is tolerant to noise in the data. This property provides both advantage and disadvantage. The neural network is effective in recognizing noisy data, and then data smoothing is not necessary. However, the method sometimes recognizes similar something else that not actually belonging to the same pattern. Introducing the knowledge of well test interpretation into the forecast is useful in order to enhance the ability to express the relationship of cause and outcome.