基于遗忘因子和误差修正的水文实

在进行水文预报时，由于影响河道洪水的因素众多，常用的水文预报模型往往不符合实际水文系统的客观规律。对这类系统的参数辨识要求算法具有较强的实时跟踪能力，以适应模拟或预测洪水运动变化过程的要求。利用洪水预报误差信息，对BP网络洪水实时预报校正模型与方法进行了探讨，提出了两种实时预报方法。第一：运用最小二乘递推算法，引入时变遗忘因子实时跟踪模型中时变参数的变化，建立了神经网络在非线性系统中动态的系统输入、输出数据之间的映射关系。第二：利用BP网络模型对误差的可识别性，将模型对输出变量的预报误差再次作为输出变量，对网络可能预报的误差进行实时修正。计算实例表明：以上两种方法提高了神经网络在水文领域的预报精度，给BP神经网络的实时预报方法提供了新的途径。

Real-Time Hydrological Forecasting Method Based on Forgetting Factor and Error Modification

Conventional hydrological models fail to simulate the actual hydrological system because the flood system is usually very complex with many affecting factors. In order to simulate the flood propagation, it is necessary to modify the system parameter identification to possess the real-time tracing capacity. In this paper, two models of real-time forecasting are presented based on BP network. In the first model, the projection from input to output in the non-linear system of neural network is established through introducing a time-variant forgetting factor into BP model based on recursive least square method. In the second model, the BP model is used to modify the output through identifying the error. The calculation results show that methods presented can improve the forecasting accuracy greatly, which provides a new approach for real-time forecasting.