基于SSA-LSTM的日径流预测

Daily Runoff Prediction Based on SSA-LSTM

摘要: 关于径流的中长期预测始终是水文预报所研究的重难点问题。为提高日径流的预测精度，提出一种基于麻雀搜索算法(SSA)和长短期记忆神经网络(LSTM)的日径流预测模型，即利用SSA算法优化LSTM模型的超参数后，对日径流进行预测。结果显示：对于特拉华河在霍巴特的西分支、纽约霍巴特东南的小镇小溪、特拉华河西分支从德里上游和德里附近的小特拉华河的日径流，基于SSA-LSTM模型的预测准确率分别为99.9994%、99.9977%、99.9991%、99.9997%，相较于LSTM模型，分别提升了0.014%、0.004%、0.011%、0.008%。该模型其他指标平方绝对百分比误差(MAPE)、根均方误差(RMSE)、平均绝对误差(MAE)相较于对照模型也有明显下降。研究表明，利用SSA-LSTM模型预测日径流量具有良好的准确性，可以为日径流量的预测提供依据。

Abstract: The mid-long term prediction of runoff is always a key and difficult problem in hydrological forecasting. In order to improve the prediction accuracy of daily runoff, a daily runoff prediction model based on Sparrow search algorithm (SSA) and Long Short-Term Memory model (LSTM) was proposed. The daily runoff is forecasted. The results show that for the western branch of Delaware River in Hobart, the Small Town Creek in southeastern new Hobart, and the western branch of Delaware River, daily runoff from the Little Delaware River upstream and near Delhi, the prediction accuracy of SSA-LSTM model was 99.9994%, 99.9977%, 99.9991%, 99.9997% respectively, which was 0.014%, 0.004%, 0.011%, 0.008% higher than that of LSTM model. Compared with the control model, the square absolute percentage error (Mape), root mean square error (RMSE) and mean absolute error (MAE) of other indexes of the model also decreased significantly. The study shows that the SSA-LSTM model has good accuracy in predicting daily runoff, which can provide a basis for the prediction of daily runoff.