基于人工神经网络的莺落峡月径流模拟预测

莺落峡是黑河干流出山口径流量的重要控制站，莺落峡径流量的多少直接影响着该流域经济、社会的发展和生态环境保护，水资源分配和调度的管理和决策。论文基于人工神经网络，对莺落峡径流进行了模拟预测。将月径流分为汛期和非汛期，分别建立BP人工神经网络，通过对径流分类前后的模型进行比较，发现分类后的月径流BP模型的性能显然优于未分类的模型，故此设计了4种不同气候情景，采用分类后的模型对莺落峡2030年的径流量进行了预测。即，在降水量不变、气温增加0.5℃，2030年莺落峡年径流量将增加8.92%；气温增加1℃、降水量不变，年径流量将减少5.414%；气温不变、降水量增加10%，年径流量将增加9.905%；气温增加0.5℃、降水量增加10%，年径流量将增加8.98%。

The Simulation and Predication of Monthly Runoff in Yingluoxia Based on Artificial Neural Network

Yingluoxia is a key control station on the main stream of Heihe River basin, and the quantity of runoff at Yingluoxia directly affects the economic and social development and eco-environmental preservation of this region. The artificial neural network (ANN) is a data-driven model, and is used widely in solving the complicated non-liner problems. So this paper,using artificial neural network, simulates and predicates the monthly runoff of Yingluoxia. In order to avoid building a black-box model, firstly, we have researched on the trend of runoff and found that the runoff of Yingluoxia is mainly affected by natural factors, so we mainly consider the natural factor in building the model;then we respectively construct ANN model of classifying runoff and not classifying runoff and compare the performance of ANN models, finding that the ANN model considering the classifying runoff has a better performance, so we use the latter model to predicate the runoff of Yingluoxia in 2030. Under the background of global warming, in different scenarios, we find that it coincides with the real situation. When the precipitation is constant, the temperature increases 0.5℃, the runoff will increase 8.92%;when the temperature increases 1 ℃, the runoff will decrease 5.414%;when keeping the temperature constant, the precipitation increases 10%, the runoff will increase 9.905%;when the temperature increases 0.5 ℃, the precipitation increases 10%, the annual runoff will increase 8.98%.