基于机器学习模型的兰州市月降水量预测研究

Research on Monthly Precipitation Prediction in Lanzhou City Based on Machine Learning Model

月降水量的精确预测对于国民生产、防灾减灾有重大意义,然而单独的模型难以完成准确预测降水的任务。本文把自适应噪声完备集合经验模态分解(CEEMDAN)分别与误差反向传播模型(BP)和长短期记忆神经网络(LSTM)结合起来,以兰州市降水数据为例,与单一的LSTM模型、差分整合移动平均自回归模型(ARIMA)和BP模型的性能进行比较。结果表明,两个复合模型有效提高了观测值和预测值的拟合度,克服了峰值预测精度不高的问题,显著优于对比模型。

Accurate forecasting of monthly precipitation is of great significance for national production as well as disaster prevention and mitigation.However,it is difficult for a single model to complete the task of accurate precipitation prediction.We combine CEEMDAN with the error reverse communication model(BP)and with the long short-term memory neural network(LSTM),respectively.And we compare Lanzhou's precipitation data with the precipitation predictions with a single LSTM model,ARIMA model and BP model.The research results show that the two composite models effectively improve the fitting of observation values and prediction values.Thus,the problem of low accuracy in peak prediction is overcomed,showing significantly higher performance than the comparative models.