本文着重探讨了针对数据的非线性拟合的两种方法:时间序列和人工神经网络在对黄河三门峡段1470~1989年的年径流量数据的分析和预测上的效果，同时通过遗传算法优化BP网络的初始权重，建立了GA-BP模型改进算法，并使用均方根误差RMSE、Nash-Suttc1iffe Coefficence(NSC)和平均相对误差em作为模型效果评定指标，对两个模型的可行性、可靠性和准确性进行验证和评价。比较得出:ARMA模型能够充分运用数据，计算速度快，对模型参数有动态确定的能力，精度较好;ANN模型具有很强的对复杂非线性问题进行识别与处理的能力;而GA-BP模型加快了网络收敛速度，提高预报精度，并避免了局部极小问题，实现了遗传算法和BP网络的优势互补。两种方法有各自的特点，如果合理的使用将有利于更好地进行水文预报。

Two popular methods, Time Series Analysis and Artificial Neural Network (ANN) model, on nonlinear fit are mainly discussed on the analysis of 1470~1989 Sanmenxia yearly runoff data for the runoff forecasting in this paper. A Genetic Algorithm-Back Propagation (GABP) model is proposed by optimizing the initial weights of BP model in order to overcome the disadvantages of BP model. The root-mean-squared error (RMSE), Nash-Sutcliffe coefficient (NSC) and average relative error are selected to evaluate model performances, including their feasibility, reliability and accuracy. It shows that ARMA model can apply data fully with high computation speed. It can identify model parameters dynamically then have high precision. AN has sufficient ability to discern and handle complicate nonlinear problems. GABP training speeds up the convergence process, reduces the time of iterative and get more accurate results while avoiding local minimum. The GABP model has the advantages of both Genetic Algorithm (GA) and BP. It is more convenient for using their features equitably in hydrologic forecasting.