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 2 各模型在测试集上的1 d预见期性能评价结果

Tab.2 Performance evalution results of multiple models with one-day lead time on test set

模型 花园口 利津 RMSE MAE MAPE NSE RMSE MAE MAPE NSE RF 301.368 6 152.467 2 0.112 5 0.907 9 261.708 0 135.698 8 0.122 9 0.933 4 CNN 281.343 0 141.966 0 0.108 3 0.919 7 201.703 6 106.062 8 0.106 1 0.960 4 TCN 267.640 8 131.018 9 0.108 6 0.927 3 204.422 5 105.637 6 0.120 7 0.959 4 TCN-LSTM 269.727 5 131.594 5 0.103 3 0.926 2 212.500 0 114.145 7 0.111 2 0.956 1 LSTM 267.388 6 131.654 0 0.110 2 0.927 5 200.944 3 104.585 3 0.105 7 0.960 7 MFF-LSTM 134.740 5 77.958 8 0.066 0 0.979 5 116.979 6 53.791 3 0.068 8 0.985 5 MFF-SSA-LSTM 80.723 3 37.219 4 0.029 4 0.992 6 78.314 8 33.306 1 0.041 4 0.993 5 MFF-BKA-LSTM 84.636 9 38.968 8 0.033 3 0.991 9 86.887 8 34.992 7 0.029 4 0.992 0 MFF-NRBO-LSTM 61.289 6 31.822 4 0.028 3 0.995 8 55.684 4 27.875 0 0.034 2 0.996 7