基于耦合机器学习模型的洪水预报研究  

Research on the Flood Forecasting Based on Coupled Machine Learning Model

作　　者：阚光远 洪阳[1] 梁珂[3] KAN Guang-yuan;HONG Yang;LIANG Ke(Department of Hydraulic Engineering,Tsinghua University,Beijing 100084,China;Beijing IWHR Corporation,China Institute of Water Resources and Hydropower Research,Beijing 100048,China;Research Center on Flood and Drought Disaster Reduction of the Ministry of Water Resources,China Institute of Water Resources and Hydropower Research,Beijing 100038,China)

机构地区：[1]清华大学水利系,北京100084 [2]中国水利水电科学研究院水利部防洪抗旱减灾工程技术研究中心,北京100038 [3]中国水利水电科学研究院北京中水科工程总公司,北京100048

出　　处：《中国农村水利水电》2018年第10期165-169,176,共6页China Rural Water and Hydropower

基　　金：北京市自然科学基金资助项目(8184094);中国水科院科研专项资助项目(JZ0145B022018,JZ0145B022017);中国博士后科学基金资助项目(2016M600096)

摘　　要：近年来,以人工神经网络(ANN)为代表的机器学习模型在很多领域取得了突破性进展,例如:用于图像识别的深度学习模型、用于围棋对弈软件Alpha Go的强化学习模型等。本文提出了一种耦合机器学习模型,并用于流域洪水预报。该模型通过独特的建模方式将ANN与K最近邻方法相耦合,利用多目标遗传算法和Levenberg-Marquardt算法进行训练,较好地解决了传统ANN模型预见期仅为一个计算时段长、ANN拓扑结构和参数难以同时优化、ANN训练局部极小、单个ANN预报能力不佳等问题。在屯溪流域洪水预报中的应用表明,耦合机器学习模型的精度和可靠性较好,具有较好的应用前景。In recent years,machine learning models,such as artificial neural network（ ANN）,have made great progresses in many fields,such as deep learning model for image recognition and reinforcement learning model for go software Alpha Go. In this paper,a coupled machine learning（ CML） model for flood forecasting is proposed. The CML model couples the ANN with the K nearest neighbour method by a specially designed modelling approach and is trained by multi-objective genetic and Levenberg-Marquardt algorithms. The model resolves the insufficient foreseeable period（ only one time-step ahead）,not able to simultaneously optimize the ANN topology structure and parameter,the local minimum,and poor performance of single ANN problems concerning the traditional ANN model applications. Real-world application of the CML model in the Tunxi watershed flood forecasting indicates its satisfactory performance and reliability,which enlightens the possibility of further applications of the CML model in flood forecasting.

关 键 词：耦合机器学习模型 水文模型 洪水预报 人工神经网络 K最近邻方法 

分 类 号：TV11[水利工程—水文学及水资源]