基于互信息的湖泊日水位预测——以西洞庭湖为例

鉴于传统的湖泊水位预测在输入因子选择时具有一定的盲目性,以西洞庭湖为例,利用基于互信息的输入因子选择法建立了日水位预测模型。按河流生态功能将水文年划分为枯水期、汛前涨水期、汛期、汛后退水期4个时期,然后分期计算影响湖泊日水位的自变量与日水位的互信息,并引入广义相关系数将互信息归一化,选出各时期互信息最大的自变量因子作为模型的输入变量。经过模型计算与数据分析可得:F检验结果显著,回归值与实测值的相关度高,剩余标准差小。由此证明用互信息筛选出的因子作为模型的输入变量能取得较好的精度并在实际中易于操作。

Lake daily level forecast by using mutual information: case of west Dongting Lake

For some blindness in selection of input factors for lake daily water level forecasting by traditional methods, we established a lake daily level forecast model for west Dongting Lake by using predictor identification approach based on the mutual information. Firstly, hydrological year is divided into low water period, pre-flood period, flood period, after-flood period by the ecological functions of the Yangtze River. Then the mutual information between the independent variables affecting the lake level and the water level is calculated by stages and the generalized correlation coefficient is introduced to normalize the mutual information. The independent variables that have the max mutual information in each period are selected as the input variables of the model. The analysis results indicate that the results of F test are significant, and the correlation between the regression and the measured values is high, and the residual standard deviation is small, which demonstrates that the variables selected by mutual information can be used as input variables of the model. The proposed method can be applied to actual operations and the results perform a better precision.