基于信息融合与识别的洪水时段分类预报

Flood Period Classification Forecast Based on Information Fusion and Recognition

为解决流域历史洪水资料有限引起的洪水预报模型模拟精度不高的问题。以A水库为研究对象,利用K均值聚类方法对典型洪水进行聚类,分析降雨雨强、降雨中心和天气系统等水文影响因子,通过遗传优化算法计算汇流模型的各类参数,利用粗糙集方法挖掘影响因子与时段汇流模式间的关系,建立了基于信息融合与识别的洪水时段分类预报。结果表明:①选取的4场典型洪水通过分类预报方法计算得到的洪峰流量绝对误差与相对误差的绝对值分别为第一场9.01 m^(3)/s、2.95%,第二场116.46 m^(3)/s、6.78%,第三场30.92 m^(3)/s、17.55%,第四场6.12 m^(3)/s、1.86%;②洪水分类预报模型的模拟精度较传统预报方法更高,不同典型洪水的确定系数均在0.8以上。研究结果可为洪水资料较少的华北等地区的洪水时段分类预报提供参考和借鉴。

To solve the low simulation accuracy of flood forecast models caused by limited historical flood data in river basins,this paper employs the K-means clustering method to cluster typical floods by taking reservoir A as the research object.Meanwhile,it analyzes hydrological influencing factors such as rainfall intensity,rainfall center,and weather system,calculates various parameters of confluence models through a genetic optimization algorithm,and adopts a rough set method to explore the relationship between influencing factors and flood period confluence patterns.Finally,the flood period classification forecast based on information fusion and recognition is conducted.The results are as follows:①The absolute and relative errors of the four selected typical floods calculated by the classification forecast method are 9.01 m^(3)/s and 2.95%,116.46 m^(3)/s and 6.78%,30.92 m^(3)/s and 17.55%,and 6.12 m^(3)/s and 1.86%respectively;②The simulation accuracy of the flood classification forecast model built in this paper is higher than that of the traditional forecast methods,and the determination coefficients of different typical floods are all above 0.8.The results can provide references for the flood period classification and forecast in north China and other regions with less flood data.