【摘要】 为解决灌区渠道泄水调度决策流程过于复杂的问题，本研究以淠史杭灌区灌口集泄水闸为例，基于历史调度数据，利用SHAP(Shapley additive explanations)法挑选聚类因子，采用自组织映射(Self-organizing map, SOM)神经网络和K均值(K-means)耦合的聚类算法SOM-K-means对已有调度数据进行分类，通过寻找历史相似场景预测其调度流量。结果表明：SHAP法确定的SOM-K-means算法聚类因子为过去6 h降雨量、过去9 h降雨量、未来6 h降雨量、灌口集泄水闸闸上水位。SOM-K-means聚类的5种聚类标签(0、1、2、3、4)分别可归类为洪峰期、无调度期、起始期、消退期、衰减期。5种标签中最大均方根误差(RMSE)为0.201,最大平均绝对误差(MAE)为0.146,最大均方误差(MSE)为0.040,最小决定系数(R~2)为0.793。研究结果可为灌区水资源优化配置提供重要应用参考。更多还原

【Abstract】 Channel discharge scheduling is an important measure to ensure the safe operation of irrigation districts. Taking Guankouji drainage gate in irrigation area of Pishihang as example, the spillway dispatch flow rate was used as the target variable, and the historical and future rainfall at different time periods, the real-time water level on the spillway gate and its change amount were used as the characterization factors. The Shapley additive explanations(SHAP) method was used to analyze the importance of the feature factors, select the combination of clustering factors, and use SOM-K-means, a clustering algorithm coupled with self-organizing map(SOM) neural network and K-means, to classify the existing scheduling data. Classification was done to find historically similar scheduling scenarios. The results showed that the SHAP method determined the clustering factors for the SOM-K-means algorithm as the past 6 h of rainfall, the past 9 h of rainfall, the next 6 h of rainfall, and the water level on the gates of the irrigation catchment sluice. The five clustering labels(0, 1, 2, 3, 4) of SOM-K-means clustering can be categorized as flood peak period, no scheduling period, onset period, fading period, and decay period, respectively. The root mean square error of the five labels was below 0.3, the average absolute error was below 0.2, the mean square error was below 0.1, and the coefficients of determination were above 0.75, and it was feasible to search for historical similar scheduling scenarios by using the SOM-K-means clustering algorithm was feasible for searching historically similar scheduling scenarios. The results of the study had important reference value for flood control and scheduling decisions in irrigation areas.更多还原