广义相加模型在乌江夏季径流预报中的应用

基于前期冬季海温指数,构建具有4个非线性指数和9个线性指数的广义相加模型(GAM),对乌江流域洪家渡夏季径流进行了模拟与预测,利用5种评估指标,包括最小信息准则(AIC)、均方根误差(RMSE)、平均绝对误差(MAE)、概率空间线性误差(LEPS)和线性相关(r),评估GAM和广义线性模型(GLM)模拟效果.结果表明:...

Application of generalized additive model in summer runoff forecasting of Wujiang Basin

Based on the sea surface temperature(SST)indices in previous winters, a Generalized additive model(GAM)model was constructed with four non-linear and nine linear indices, to simulate and forecast the summer runoff of Hongjiadu hydropower station on the Wujiang Basin. Five evaluation indices were used to evaluate the simulation results of GAM and Generalized linear model(GLM), including the Akaike information criterion(AIC), the root mean square error(RMSE), the mean absolute error(MAE), the linear error in probability space(LEPS)and the linear correlation r. The results showed that the ratios of simulated data to real data by GAM were smaller than those by GLM in AIC, RMSE and LEPS, while the linear correlation coefficient of GAM was obviously larger than that of GLM, thus GAM performed better than GLM. Leave-one-out Cross validation was used to forecast the summer runoff of Hongjiadu hydropower station with GAM and GLM, respectively. The forecast results indicated that the correlation coefficient between the GAM result and observed data was improved to 0. 41 and predicated data with relative error less than 30% was more than 60%. The prediction error of GAM was around 10% in the typical flood year of Hongjiadu and less than 10% in the typical dry year. It can be concluded that the prediction accuracy was improved in GAM compared with that in GLM. Consequently, considering the the non-linear relationship between runoff and prediction factors, the use of GAM in runoff forecasting can effectively improve the simulation result and prediction accuracy than the linear regression model.