基于回归支持向量机的水库防洪承载力预测模型研究

Forecasting Model of Reservoir's Flood Control Capacity Based on Support Vector Machine for Regression

防洪承载力,即水库目前剩余防洪库容条件下,不泄洪所能容纳的流域面降雨量。根据水量平衡原理分析概化出防洪承载力的预报因子为当前土壤含水量,当前入库流量以及水库剩余库容。利用2010—2020年青山水库55场历史洪水建立基于回归支持向量机的防洪承载力预测模型,利用贝叶斯优化进行超参数率定,通过分析预测值确定回归支持向量机核函数为线性,预测值与实测值相关系数为0.9527,平均绝对百分误差为24.2853%,预测偏小百分比为49.0909%,表明模型精度较高且计算结果较为稳定,可为水库防洪提供参考。

Flood control capacity means the area rainfall of the basin that could be accommodated without discharge under the current remaining flood control capacity of the reservoir. According to the principle of water balance, the paper analyzed and generalized the predictors of flood control carrying capacity, including current soil water content, current inflow flow and remaining reservoir capacity. Then a regression support vector machine-based flood control capacity prediction model was established. The data come from the 55 historical floods of Qingshan Reservoir from 2010 to 2020. Using Bayesian optimization method to calibrate the hyperparameters, the kernel function of the regression support vector machine is determined to be linear by analyzing the predicted value. The correlation coefficient between the predicted value and the measured value is 0.9527, the average absolute percentage error is 24.2853%, and the under-predicted percentage is 49.0909%, indicating that the model has high accuracy and the calculation result is relatively stable, which can provide a reference for reservoir flood control.