基于改进的BP神经网络径流量还原计算研究

Study on Natural Runoff Reduction Calculation Based on Improved BP Network

针对水文系统复杂性与非线性的特点,加入动量项因子对BP神经网络进行改进,加快收敛速度,将自然因素、人为因素、自然与人为混合因素分别作为输入因子,构建了大沽夹河天然径流量还原计算方案,用逐项还原法的结果验证对比选出最佳方案。结果表明:①经过改进的BP神经网络收敛速度明显加快,由平均的6 028次迭代优化到1 782次迭代;②以降雨量、蒸发量和实测径流量为输入因子的第三种方案模拟误差最小,适用于大沽夹河流域天然径流量还原计算。

Aiming at the characteristics of the complexity and nonlinear of hydrological system,the BP neural network model was improved by the introduction of momentum factor to accelerate the convergence rate,established three kinds of schemes based on different input factors which were natural factors,human factors and synthetic nature and human nature to restore natural runoff calculation in Dagujia River and selected the best scheme by comparison and analysis of successive reduction method.The results show that the improved BP neural network can accelerate the convergence rate and the iteration time is optimized from the average 6 028 to 1 782.Getting the third scheme which used the rainfall,evaporation and measured runoff as the input factors is the best scheme in the simulation error and it is suitable for the natural runoff reduction calculation in the Dagujia River basin.