【摘要】 对于具有随机性和突变型的复杂水文时间序列数据，使用单一模型进行水文预测结果准确度有待提高，因此组合模型更为科学。提出一种基于LSTM和GPR的水文时间序列预测方法。利用0-1规划和遗传算法筛选水位的最优特征组合，重构数据集后初步用LSTM模型进行预测，将LSTM预测结果与真实值的残差和LSTM预测结果作为GPR模型的输入，同时进行最终预测。以屯溪流域屯溪水文站的水位数据为例，对水位进行预测。实验结果表明，基于LSTM和GPR的水文时间序列预测方法能有效提升预测的准确度。更多还原

【Abstract】 For complex hydrological time series data with randomness and mutation, the accuracy of Hydrological Prediction results using a single model needs to be improved, so the combined model is more scientific. A method of hydrological time series prediction based on LSTM and GPR is proposed. 0-1 programming and genetic algorithm were used to screen the optimal feature combination of water level. After reconstruction of the data set, the LSTM model was preliminarily used for prediction. The residual difference between the LSTM prediction result and the real value and the LSTM prediction result were taken as the input of THE GPR model, and the final prediction was made at the same time. Taking the water level data of Tunxi Hydrological Station in Tunxi basin as an example, the water level was predicted. The experimental results show that the hydrological time series prediction method based on LSTM and GPR can improve the accuracy of the predictioneffectively.更多还原