水利水电技术（中英文）

为提高年径流预测的精度,以呼兰河下游兰西水文站1959—2014年的年径流数据作为输出,相应的流域气象数据作为输入,将BP神经网络和支持向量机(SVM)相结合,构建基于马尔科夫链修正的BP-SVM预测模型,运用该模型对呼兰河流域的年径流进行预测。采用哈里斯鹰群算法(HHO)优化支持向量机参数,构建HHO-SVM模型,并进行年径流预测,利用训练好的BP神经网络对年径流进行预测,分别运用马尔科夫链对两种模型的预测结果进行修正,通过最小二乘法确定模型组合权重,将两模型的预测结果进行组合,得到最终的年径流预测值。研究结果表明:HHO-SVM模型预测结果优于BP神经网络预测值;经马尔科夫链修正后,BP神经网络预测值精度提高较大,经最小二乘法组合后的预测结果平均相对误差为11.36%,确定性系数为0.95,合格率达90.91%。哈里斯鹰群算法(HHO)能较好的解决支持向量机参数优化问题,马尔科夫链的修正在一定程度能提高了各个模型的预测精度,提出的混合模型为年径流预测提供了一种新的方法。

In order to improve the accuracy of annual runoff prediction, taking the annual runoff data from 1959 to 2014 of Lanxi Hydrological Station in the lower reaches of Hulan River as the output data and the corresponding watershed meteorological factors as the input data, BP neural network and support vector machine(SVM) are combined to established BP-SVM prediction model based on Markov chain modification, and the presented model is used to predict the annual runoff of Hulan River Basin. First, the HHO-SVM model is constructed by Harris Hawks optimization(HHO) algorithm to optimize parameters of SVM and forecast annual runoff. Second, the annual runoff is predicted by the trained BP neural network. Third, Markov chain is used to modify the prediction results of the two models, and the combination weight of the two models is determined by the least square method. Finally, the prediction results of the two models are combined to obtain the final annual runoff prediction value. The results show that the prediction result of HHO-SVM model is better than that of BP neural network. The accuracy of BP neural network prediction value can be improved greatly which is modified of Markov Chain. The average relative error of the prediction result after the least square method combination is 11.36%, the certainty coefficient is 0.95, and the qualified rate is 90.91%. The HHO can better resolve the problem of SVM parameter optimization, and the modification of Markov Chain can improve the prediction accuracy of each model to a certain extent. In this paper, the proposed hybrid model provides a new method for annual runoff prediction.