基于多特征融合和牛顿-拉夫逊优化算法的LSTM日径流预测

王铮,张梦君,姜楠,王万良,屠杭垚

Daily runoff prediction using LSTM based on multi-feature fusion and Newton-Raphson-based optimizer

Zheng WANG,Mengjun ZHANG,Nan JIANG,Wanliang WANG,Hangyao TU

表 3 各模型在测试集上的2 和 3 d预见期性能评价结果

Tab.3 Performance evalution results of multiple models with two- and three- day lead time on test set

模型 2 d 3 d 花园口 利津 花园口 利津 RMSE NSE RMSE NSE RMSE NSE RMSE NSE RF 437.774 4 0.805 4 386.129 4 0.854 9 535.451 5 0.709 2 481.175 6 0.774 8 CNN 398.110 3 0.839 1 340.881 0 0.886 9 487.641 7 0.758 8 433.588 5 0.817 1 TCN 411.112 9 0.828 4 378.179 3 0.860 8 505.366 6 0.741 0 437.845 3 0.813 5 TCN-LSTM 403.074 9 0.835 1 344.661 5 0.884 4 492.761 4 0.753 7 437.448 9 0.813 9 LSTM 409.039 9 0.830 1 335.820 8 0.830 1 500.632 2 0.745 8 433.363 7 0.817 3 MFF-LSTM 136.870 8 0.972 8 121.582 1 0.972 8 215.398 6 0.947 7 217.145 9 0.950 2 MFF-SSA-LSTM 154.234 2 0.978 9 121.582 1 0.978 9 227.501 3 0.941 6 170.716 3 0.969 2 MFF-BKA-LSTM 155.409 7 0.973 2 126.920 3 0.973 2 221.533 3 0.944 7 196.019 8 0.959 4 MFF-NRBO-LSTM 116.784 9 0.984 6 109.921 0 0.971 4 196.854 0 0.956 3 164.402 4 0.971 4