中文摘要：

径流水位预测是进行洪水监测的重要手段,对于包含详尽信息的广西柳江日径流水位时间序列,采用基于BP神经网络模型进行预报可取得较好效果.如LMBPDH模型采用双隐含层BP网络能加强预测模型输入输出的非线性映射能力,采用Levenberg Marquardt （LM）算法对网络进行训练则能缩短BP网络的收敛时间,改善网络的收敛性能,同时采用实验法确定模型的其他参数使模型获取最佳预报性能.在对柳江近10年日平均水位的预测中,将LMBPDH模型与单隐含层BP神经网络、LM算法以及带适应学习率和动量因子的梯度递减法算法等组合构成的BP神经网络模型,以及遗传算法进化的神经网络模型比较,LMBPDH模型预报稳定性、预报准确率最佳.

英文摘要：

Runoff water level prediction is the important means to flood monitoring. When daily runoff water level time series of Guangxi Liujiang River, which contains detailed information, is forecasted by BP neural network model, good effect can be achieved. For example, LMBPDH model uses BP neural network with double hidden layers to strengthen the nonlinear mapping ability between the input and the output, uses Levenberg-Marquardt algorithm to reduce convergence time and improve the convergence performance of the BP neural network. At the same time, the experiment method is adopted to determine other model parameters to obtain the best forecasting performance. Compared with those models which are composed of single hidden layer, LM algorithm, gradient descending method algorithm with adaptation learning evolved neural network, LMBPDH model is superior to terms of the same evaluation measurements. rate and momentum factor, and the genetic algorithm the other models on forecasting stability and accuracy in Liujiang river runoff water level forecasting; time series ; BP neural network