基于SVM-CEEMDAN-BiLSTM模型的日降水量预测

Daily Precipitation Prediction Based on SVM-CEEMDAN-BiLSTM Model

针对日尺度降水序列中极大值与无雨日预测精度低的问题,提出基于支持向量机(SVM)、完全集合经验模态分解(CEEMDAN)和双向长短期记忆神经网络(BiLSTM)的降水预测耦合模型。将提出的模型应用于鄱阳湖流域景德镇站和赣县站2个典型站点的日降水量预测中,并与各种传统模型组合的预测结果进行了对比。结果表明:耦合模型的降水预测结果与实测结果基本一致且精度最高,为日尺度降水量预测中极大值与无雨日预测精度低的问题提供了一种解决参考。

In order to solve the problem of low prediction accuracy of the maximum value and rain-free days in daily precipitation series,a coupled model of precipitation prediction based on a support vector machine(SVM),complete ensemble empirical modal decomposition(CEEMDAN),and bi-directional long and short-term memory neural network(BiLSTM)was proposed.This paper applied the coupled model to predict the daily precipitation at Jingdezhen Station and Ganxian Station in the Poyang Lake Basin,and the results were compared with those of traditional model combinations.The results show that the precipitation prediction results of the coupled model are basically consistent with the measured ones,with the highest accuracy.It thus provides a reference for solving the problem of low prediction accuracy of the maximum value and rain-free days during daily precipitation prediction.