论文部分内容阅读
在分析非线性河道洪水预报方法中常用BP神经网络不足的基础上,采用具有快速收敛和更有效非线性逼近能力特性的小波神经网络.为适应洪水演进的时变特性,将所建立的用于河道洪水预报的小波神经网络与自回归实时校正模型耦合,校正值为小波神经网络预报值与自回归模型预报误差之和.自回归实时校正模型的参数通过自适应衰减因子递推最小二乘动态更新以提高校正效果.将该方法应用于西江高要断面洪水预报,计算结果验证了其有效性.
Based on the analysis of the shortages of BP neural network commonly used in nonlinear river flood forecasting methods, a wavelet neural network with fast convergence and more effective nonlinear approximation ability is adopted. To adapt to the time-varying characteristics of flood evolution, The wavelet neural network of river flood forecast is coupled with the autoregressive real-time correction model, and the correction value is the sum of the prediction error of the wavelet neural network and the autoregressive model. The parameters of the autoregressive real-time correction model recursive least square dynamic Which is used to improve the correction effect.Application of this method to the forecast of flood in the section of Xijiang River in Xijiang River validates its effectiveness.