基于LCD-SSA-BiLSTM模型的月径流预测研究

Research on Monthly Runoff Prediction Method Based on LCD-SSA-BiLSTM Model

径流预测在水资源优化配置和防汛抗旱方面发挥着重要作用。但径流序列非平稳会导致预测误差及峰值预测误差较大,因此提出了基于局部特征尺度分解(LCD)、麻雀搜索算法(SSA)和双向长短期记忆神经网络(BiLSTM)的组合预测模型(LCD-SSA-BiLSTM),以提高非平稳径流序列的预测精度。以汾河上游4个站点(汾河水库站、上静游站、兰村站和寨上站)为研究对象开展月径流序列预测研究,采用纳什效率系数、平均绝对误差、均方根误差、合格率4个评价指标对预测结果进行定量评价。结果表明,LCD-SSA-BiLSTM模型的平均绝对误差为10.346×10^(4)~124.629×10^(4)m^(3),均方根误差为19.416×10^(4)~191.284×10^(4)m^(3),纳什效率系数为0.975~0.988,4个水文站的合格率均在90%及以上,预测精度为甲级,与单一BiLSTM、EMD-BiLSTM、LCD-BiLSTM及EMD-SSA-BiLSTM模型相比预测效果更好,因此LCD-SSA-BiLSTM模型是预测非平稳月径流序列的有效方法。

Runoff prediction plays an important role in the optimal allocation of water resources and flood control and drought relief.To solve the problem of large prediction errors caused by non-smoothness and extreme values of runoff series and improve the prediction accuracy,this paper proposes a combined prediction model(LCD-SSA-BiLSTM)based on local characteristic-scale decomposition(LCD),sparrow search algorithm(SSA)and bi-directional long short-term memory(BiLSTM)to study the monthly runoff series of four stations in the upper reaches of Fenhe River(Fenhe Reservoir Station,Shangjingyou Station,Lancun Station and Zhaishang Station).Nash efficiency coefficient(N NSE),mean absolute error(M MAE),root mean square error(RRMSE),and qualification rate(QQR)are used to quantitatively evaluate the prediction results.Compared with the single BiLSTM model,EMD-BiLSTM model,LCD-BiLSTM model and EMD-SSA-BiLSTM model,the results show that the LCD-SSA-BiLSTM model has higher prediction accuracy with M MAE of 10.346×10^(4)-124.629×10^(4)m^(3),RRMSE of 19.416×10^(4)-191.284×10^(4)m^(3),N NSE of 0.975-0.988,and the QQR of all four hydrological stations were 90%and above,and the prediction accuracy was grade A.Thus,the LCD-SSA-BiLSTM model is an effective method to predict non-stationary monthly runoff series.