基于VMD-GRU的大型灌区干旱预测模型研究

Research on the Drought Prediction Model of Large Irrigation Areas Based on VMD-GRU

灌区干旱灾害的频繁发生严重制约了当地经济的发展,为了实现灌区内水资源的合理调配和利用,需对灌区未来的干旱情况进行有效预测,从而有针对性的采取预防措施降低干旱事件所造成的影响。基于“先分解,后重构”的思想,将能够对非线性、非平稳时间序列进行预处理的变分模态分解VMD(Variational Mode Decomposition)与门控循环单元网络GRU(Gated Recurrent Unit)相结合,建立VMD-GRU干旱预测模型,针对赵口大型灌区4个气象站点1981-2017年的逐月降水数据构建标准化降水指数SPI时间序列并采用所建干旱预测模型和3个单一预测模型对SPI值进行预测,研究采用3个月的时间尺度计算标准化降水指数SPI进行干旱事件识别和等级划分。结果表明:4个气象站点VMD-GRU干旱预测模型SPI值预测结果的平均相对误差在18.7%~20.6%之间,预测误差较为稳定。VMD-GRU干旱预测模型的平均绝对误差MAE、均方误差MSE、均方根误差RMSE均为4种模型中最小,不同气象站点的3个评价指标值均分别接近,所建模型不存在过拟合或者欠拟合问题。利用游程理论识别干旱历时和干旱强度,以开封气象站为例,1981-2017年间干旱历时的预测准确率达到81.1%,实际干旱强度的整体波动趋势与预测干旱强度大体一致,两序列的皮尔逊相关系数达到0.961,表明所建VMD-GRU干旱预测模型能够较为准确地对灌区干旱进行预测,为大型灌区干旱预警提供一种可靠的研究方法。

The frequent occurrence of drought disasters in irrigation areas seriously restricts the development of local economy.In order to realize the rational allocation and utilization of water resources in irrigation areas,it is necessary for this paper to effectively predict the future drought situation in irrigation areas so as to take targeted preventive measures to reduce the impact of drought events.Based on the idea of decomposing and reconstructing,the VMD-GRU drought prediction model is established by combining the variational modal decomposition VMD(Variational Mode Decomposition)which can preprocess nonlinear and non-stationary time series with GRU(Gated Recurrent Unit),and the standardized precipitation index SPI time series is established according to the monthly precipitation data of four meteorological stations in large irrigation areas in Zhaokou from 1981 to 2017,and the SPI value is predicted by using the established drought prediction model and three single prediction models.The paper adopts a three-month time scale.The results show that the average relative error of SPI value predicted by VMD-GRU drought prediction model in four meteorological stations is between 18.7%and 20.6%,and the prediction error is relatively stable.The mean absolute error MAE,mean square error MSE and root mean square error RMSE of VMD-GRU drought prediction model are the smallest among the four models,and the values of the three evaluation indexes of different meteorological stations are close to each other,so there is no over-fitting or under-fitting problem in the established model.According to the theory of run to identify drought duration and drought intensity and Kaifeng Meteorological Station,the prediction accuracy of drought duration from 1981 to 2017 reached 81.1%,and the overall fluctuation trend of actual drought intensity is basically consistent with the predicted drought intensity.Pearson correlation coefficient of the two series reaches 0.961,indicating that the established VMD-GRU drought prediction model can accurately predict the drought in irrigation areas and provide a reliable research method for droughts’early warning in large irrigation areas.