基于GEO-RVM模型的枯水期月径流预报研究

Research on Monthly Runoff Forecast in Dry Seasons Based on GEO-RVM Model

为提高枯水期月径流预测精度,研究提出金雕优化(GEO)算法与相关向量机(RVM)相融合的预测方法。基于云南省某水文站67 a的径流资料,选取预报月之前具有较好相关性的月径流作为预报影响因子,通过主成分分析(PCA)对影响因子进行降维处理,利用GEO算法优化RVM核宽度因子和超参数,建立GEO-RVM模型对该站枯水期11月、12月和次年1—4月月径流进行预报,预报结果与基于GEO算法优化的支持向量机(SVM)模型(GEO-SVM)作对比。结果表明:GEO-RVM模型对实例11月、12月和次年1—4月月径流预报的平均相对误差分别为8.59%、7.34%、5.97%、6.07%、5.99%、5.04%,预报精度优于GEO-SVM模型。GEO算法能有效优化RVM核宽度因子和超参数,GEO-RVM模型具有较好的预报精度,将其用于枯水期月径流预报是可行的。

To improve the accuracy of monthly runoff forecasts during dry seasons,this study proposes a forecasting method that combines the golden eagle optimization(GEO)algorithm and the relevance vector machine(RVM).On the basis of the runoff data of 67 a from a hydrological station in Yunnan Province,the monthly runoff with good correlation before the forecast month is selected as the influencing factor of forecasts,and the influencing factor is reduced in dimension by principal component analysis(PCA).The kernel width factor and hyperparameters of RVM are optimized by the GEO algorithm,and the GEO-RVM model is built to forecast the monthly runoff of the station during the dry season from November to April of the following year.Moreover,the forecast results are compared with those of the GEO-based support vector machine(SVM)model(GEO-SVM).The results demonstrate that the average relative errors of the GEO-RVM model for the monthly runoff forecasts from November to April of the following year are 8.59%,7.34%,5.97%,6.07%,5.99%,and 5.04%,respectively,which means the accuracy is better than that of the GEO-SVM model.The GEO algorithm can effectively optimize the kernel width factor and hyperparameters of RVM,and the GEO-RVM model has better forecast accuracy,which can be used for monthly runoff forecasting during dry seasons.