基于PCA-MDBN模型的开都河年径流预测

Annual runoff forecast of Kaidu River based on PCA-MDBN model

河川径流预测对于干旱区水库调度、防汛抗旱及水资源优化配置具有重要意义。由于河川径流变化影响因子众多,且各因子之间相互关联并呈现非线性变化,采用数学方法及传统神经网络很难精准预测,且在进行数据训练时存在局部收敛和计算效率差的问题。针对上述问题,本文首先通过主成分分析筛选影响径流变化的主要因子作为模型输入,利用小波阈值方法实现噪声消除。然后,提出基于粒子群优化算法并结合多变量深度信念网络(BP-MDBN),对河川径流进行预测,并与传统BP神经网络、DBN模型进行比较分析。结果表明:本文方法平均百分比误差为6.2,与BP、DBN方法进行对比,其MAPE分别降低0.077、0.10;BP-MDBN模型的RMSE和MAE值也明显小于其他两种方法。此方法具有较高的预测精度及泛化性能,研究成果可为河川径流精准预测提供理论支撑。

River runoff prediction is of great significance for reservoir operation,flood control and drought relief,and optimal allocation of water resources in arid areas.Because there are many influencing factors of river runoff change,and the factors are interrelated and present nonlinear changes,it is difficult to accurately predict using mathematical methods and traditional neural networks,and there are problems of local convergence and poor computational efficiency in data training.In order to solve the above problems,this paper first uses principal component analysis to screen the main factors that affect runoff changes as model inputs,and uses wavelet threshold method to eliminate noise.Then,based on particle swarm optimization algorithm and combined with multi-variable depth belief network(BP-MDBN),river runoff is predicted and compared with traditional BP neural network and DBN model.The results show that the average percentage error of the proposed method is 6.2.Compared with BP and DBN methods,its MAPE is reduced by 0.077 and 0.10 respectively.The RMSE and MAE values of the BP-MDBN model are also significantly smaller than those of the other two methods.This method has high prediction accuracy and generalization performance,and the research results can provide theoretical support for precise prediction of river runoff.