三峡生态环境监测

年径流预测对于减灾和水资源的高效利用具有重要意义。为有效提高年径流预测的准确性，本研究提出一种基于马尔可夫链(MC)优化的BP神经网络模型与支持向量机模型(SVM)相结合的混合数据驱动模型，称为BPSVM-MC。利用屏山站和寸滩站的年径流数据评估了该方法的有效性。分别使用BP神经网络模型和支持向量机模型进行年径流预测，并使用马尔可夫链方法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，最终结合两种模型的预测结果得到年径流预测值。结果表明，所提出的BP-SVM-MC模型优于其他两种方法，耦合模型在年径流预测中表现出优越的预测性能。在屏山站，该模型的合格率达到90.00%，平均相对误差为7.62%，均方根误差为137.52×10~9m³。在寸滩站，模型的合格率达到83.33%，平均相对误差为9.24%，均方根误差为374.58×10~9m³。

Annual runoff prediction is of great significance for disaster reduction and the efficient utilization of water resources. In order to effectively improve the prediction accuracy of annual runoff, this study proposes a hybrid data-driven model based on Markov Chain(MC) optimized Back Propagation(BP) neural network model and Support Vector Machine model(SVM), termed as BPSVM-MC. The effectiveness of the proposed approach is assessed using the annual runoff data from Pingshan and Cuntan stations.Annual runoff predictions are conducted using BP neural network model and SVM model. The predictions from both models are corrected using the MC method separately. The model combination weights are determined through least squares method, and the predictions from both models are then combined to obtain the final annual runoff forecast. The results show that the proposed BPSVM-MC model outperforms the other two methods. The coupled model demonstrates superior predictive performance in annual runoff forecasting. At the Pingshan station, the qualified rate of the proposed model reaches 90.00%, the average relative error is7.62%, and the root mean square error is 137.52×10~9m³. At the Cuntan station, the qualified rate of the model reaches 83.33%, the average relative error is 9.24%, and the root mean square error is 374.58×10~9m³.