疏勒河中游梯级水库库容和库水位关系模型优化

Relationship Model Optimization Between Storage Capacity and Water Level of Cascade Reservoirs in Middle Reaches of Shule River

为了研究梯级水库库容与库水位之间的关系,解决水库由原来承担单一灌溉或发电任务,到现在变为一库多用的多目标决策问题。以疏勒河流域中游地区的3座水库(昌马水库、双塔水库和赤金峡水库)为研究对象,利用高斯函数和两种神经网络建立3大水库的库水位和库容之间的关系模型。通过对比不同模型预测的均方根误差(RMSE),相对均方根误差(RRMSE)和平均绝对百分比误差(MAPE)来判断模型的表现。结果表明:3种模型均可较好地拟合3大水库库水位和库容之间的关系,对于昌马水库和双塔水库来说,径向基函数神经网络(RBF)的表现优于非线性回归模型和前馈反向传播神经网络(FBNN)模型;对于赤金峡水库而言,FBNN神经网络的表现优于非线性回归和RBF神经网络模型。与经典回归模型相比,人工神经网络更适合于水库库容的确定。

In order to study relationship between storage capacity and water level of cascade reservoir,multi-objective decisionmaking problem that reservoir was originally responsible for a single irrigation or power generation task has now become a multipurpose reservoir.Three reservoirs(Changma Reservoir,Shuangta Reservoir and Chijinxia Reservoir)in the middle reaches of the Shule River Basin were selected as research objects.Relationship between reservoir water levels and reservoir capacity of three reservoirs was established using Gaussian function and two types of neural networks model.By comparing root mean square error(RMSE),relative root mean square error(RRMSE)and mean absolute percentage error(MAPE)predicted by different models,performance of the model was evaluated.Results showed that three models could better fit relationship between water level and capacity of three reservoirs.For Changma Reservoir and Shuangta Reservoir,performance of radial basis function neural network(RBF)was better than nonlinear regression model and feed forward back propagation neural network(FBNN)model.For Chijinxia Reservoir,performance of FBNN neural network was better than nonlinear regression and RBF neural network models.Compared with classical regression model,artificial neural network had advantages of high precision and scientific reliability in prediction of reservoir capacity.It was more suitable for determination of reservoir capacity.