基于人工神经网络的河川径流实时预报研究

Real-time Forecasting of Streamflow Based on Artificial Neural Networks

将人工神经网络技术应用于河川径流实时预报，建立起河川径流实时预报的BP网络模型，并针对经典BP算法所存在的缺陷，采用共轭梯度优化和误差反向传播训练算法，使得所建立的BP网络模型的收敛性大为改善，消除和避免了实际应用中可能出现的局部优化问题．利用西大洋水库1975～1995年的入库径流系列资料，对所建立的BP网络模型进行训练和检验，同时探讨了网络结构对网络模型预报结果的影响．通过大量的实际应用和对比分析，表明BP网络模型比HG分析模型和相关图法更优越、更具有实际推广和应用价值．

In this paper, the artificial neural networks method is applied to develop a BP network model forthe real-time forecasting of streamflow. The convergence of the BP network model is greatly improved by con-jugate gradient optimization and the error-reverse-transmission training method, and the local optimizationproblem is resolved. In the end, by using Xidayang Reservoir's inflow data for the period of 1975～1995, theBP network model is trained and examined, and the affect of the network structure on the prediction precisionis discussed. The application of BP network model and its applications in other fields show that BP networkmodel is better, has greater value for popularization than HG analysis model and Relationship Graph method.