CAJ | 학술논문

为了提高径流预报的精度，采用一种基于粒子群和遗传的混合方法同时优化人工神经网络结构、连接权和偏置，在进化过程中采用训练样本和验证样本共享适应度技术，并以此建立径流预报模型。通过对柳州径流实例分析，并与离子群优化的人工神经网络模型、遗传进化的人工神经网络模型和时间序列模型方法对比，研究结果表明，该方法学习能力强、泛化性能高和有效提高系统预测的准确率，为获得更高预测精度的径流预报提供了一种有效的建模方法。
위료제고경류예보적정도，채용일충기우입자군화유전적혼합방법동시우화인공신경망락결구、련접권화편치，재진화과정중채용훈련양본화험증양본공향괄응도기술，병이차건립경류예보모형。통과대류주경류실례분석，병여리자군우화적인공신경망락모형、유전진화적인공신경망락모형화시간서렬모형방법대비，연구결과표명，해방법학습능력강、범화성능고화유효제고계통예측적준학솔，위획득경고예측정도적경류예보제공료일충유효적건모방법。
In order to improve the accuracy of runoff forecasting, a hybrid algorithm combining PSO and GA algorithm with optimizing artificial neural network structure, connection weights and bias was proposed and used to establish a runoff forecasting model. This hybrid algorithm adopts training samples and validation samples to share fitness in the evolutionary process. The algorithm was com-pared with two forecasting models including PSO-ANN and GA-ANN through the actual examples of Liuzhou runoff forecasting. The results show that the new approach has strong learning ability and high generalization performance and can improve the accuracy of forecasting system effectively. Thus, it is an effective modeling method to get high precision of runoff forecasting.