改进BP神经网络算法在中小流域洪水预报中的应用研究

摘要

Dingan river watershed,one of the primary tributary of Wanquan River and located at the cen-tral part of Hainan Province,is a typical middle and small watershed.Aiming at solving the problems of prediction results unsmooth and in risk presenting outliers when using BP neural network flood forecas-ting,the multi-time synthesis algorithm and smoothing algorithm are proposed considering the character-istics of hydrological processes.Selecting the Dingan River watershed of Hainan province as the study ar-ea and adopting multiplayer feed-forward BP neural network are to build several different plans for com-parison and analysis.The results show that the proposed method can be used to compensate for the defi-ciency of the original algorithm in improving the accuracy of flood forecasting ,with a good practical val-ue as a useful reference for the traditional forecasting methods.%定安河流域位于海南省的中部，是万泉河的一级支流，属于典型的中小流域。针对利用BP神经网络进行洪水预报时预报结果不平滑、冒异常值等问题，在考虑水文过程性质的基础上，提出了多时段综合算法和修匀算法。选取海南省定安河流域作为研究区域，采用深层前向BP神经网络，构建多组预报方案进行对比分析。结果表明，本文所提方法可以弥补原有算法的不足，提高洪水预报精度，作为传统预报方式的有益参照，具有较好的实用价值。