基于季节ARIMA模型和LSTM模型的西宁市降雨量预测研究

Rainfall Forecasting in Xining City Based on Seasonal ARIMA Model and LSTM Model

摘要: 西宁市是我国降雨量变化较大的城市。为了掌握该市降雨量趋势并为有关部门提供科学有效的降雨量信息，本文采用季节ARIMA (差分自回归移动平均)模型和LSTM模型对西宁市2009年~2021年月降雨量趋势进行拟合分析。西宁市年降雨量总体趋势呈现出明显的季节性，夏季降雨量最大。通过对模型参数的选择和设置，建立了西宁市降雨量的季节ARIMA预测模型ARIMA(2,1,1)(0,1,1)和LSTM模型，最后对2022年12个月的降雨量进行预测。预测结果如下：季节ARIMA模型对2022年前4个月的预测值较为准确，但随着预测步长的扩大，模型的预测精度会降低。LSTM模型则凭借其处理长期依赖性和复杂非线性关系的能力，在降雨量预测中表现出了不错的性能。使用季节ARIMA模型和LSTM模型能够较好地预测西宁市未来短期的降雨量变化。

Abstract: The city of Xining is the city with larger rainfall variation in our country. In order to grasp the trend of rainfall in the city and provide scientific and effective rainfall information for relevant departments, this paper uses seasonal ARIMA (differential autoregressive moving average) model and LSTM model to fit and analyze the monthly rainfall trend of Xining City from 2009 to 2021. The overall trend of annual rainfall in Xining city shows obvious seasonality, with the highest rainfall in summer. Through the selection and setting of model parameters, the seasonal ARIMA prediction model of rainfall in Xining City ARIMA(2,1,1)(0,1,1) and LSTM model are established, and finally the 12-month rainfall in 2022 is predicted. The prediction results are as follows: the seasonal ARIMA model predicts relatively accurate values in the first four months of 2022, but with the expansion of the prediction step, the prediction accuracy of the model will decrease. The LSTM model has shown good performance in rainfall prediction by virtue of its ability to deal with long-term dependence and complex nonlinear relationships. The seasonal ARIMA model and LSTM model can better predict the short-term rainfall change in Xining city in the future.