针对BP神经网络在实际应用中存在预测精度低、易陷入局部极值等不足,引入遗传算法(GA)及Adaboost算法对BP神经网络算法进行改进,提出GA-Adaboost_BP年径流预测模型,并构建基于L-M算法改进的BP、GA-BP及Adaboost_BP模型作为对比模型,以云南省龙潭站年径流预测为例进行实例研究。首先,构建基于L-M算法改进的BP预测模型,确定最佳网络结构及相关参数;其次,将所构建的BP模型作为基本模型,分别构建基于GA算法、Adaboost算法及GA-Adaboost混合算法改进的GA-BP、Adaboost_BP、GAAdaboost_BP预测模型;最后,确定年径流影响因子,利用实例前34a和后20a资料对模型进行训练和预测。结果表明,GA-Adaboost_BP模型对实例后20a年径流预测的平均相对误差绝对值和最大相对误差绝对值分别为2.59%、6.69%,预测精度优于BP、GA-BP及Adaboost_BP模型。GA-Adaboost_BP模型兼顾了GA算法及Adaboost算法二者的优点,有效提高了BP神经网络性能,具有预测精度高、泛化能力强等特点,模型及方法可有效用于年径流预测。 In order to overcome the shortcomings of BP neural network in practice, such as low prediction accuracy and easy to fall into local extreme value, genetic algorithm (GA) and Adaboost algorithm are introduced to improve the BP neural network algorithm. A GA-Adaboost_BP annual runoff prediction model is proposed and a LM algorithm improved BP, GA-BP and Adaboost_BP model as a comparative model, taking Longtan station in Yunnan Province annual runoff forecast as an example for case studies. Firstly, the improved BP prediction model based on the LM algorithm is constructed to determine the optimal network structure and related parameters. Secondly, the BP model is taken as the basic model to build the improved GA based on GA algorithm, Adaboost algorithm and GA-Adaboost hybrid algorithm respectively -BP, Adaboost_BP and GAAdaboost_BP. Finally, the influencing factors of annual runoff were determined, and the model was trained and predicted using the data of the first 34 and the last 20 years. The results show that the average relative error and the maximum relative error of GA-Adaboost_BP model are respectively 2.59% and 6.69%, and the prediction accuracy is better than BP, GA-BP and Adaboost_BP model. GA-Adaboost_BP model not only takes advantage of both GA and Adaboost algorithm, but also improves the performance of BP neural network effectively. It has the characteristics of high prediction accuracy and extensive generalization ability. The model and method can be effectively used for annual runoff prediction.