基于时间序列的季节性气温预测研究

Research on Seasonal Temperature Forecasting Based on Time Series

气温的变化受风速、湿度、日照时数等因素的影响,可以通过分析这些因素预测气温的变化情况。考虑到气温序列中存在季节特性,采用One-Hot编码方法提取气温序列中的季节性信息,并作为随机森林模型的输入特征,对月平均气温进行拟合与预测。由于模型构建时涉及众多超参数,文中利用随机搜索和网格搜索两种算法优化模型中的超参数。结果表明:考虑季节性的随机森林模型拟合效果优于简单随机森林模型,预测数据变化趋势与实际观测基本一致,拟合精度可以达到96.14%。经两种方法对超参数寻优之后,模型拟合精度可以达到96.45%。

The variation of temperature is influenced by factors including wind speed,humidity,sunshine duration,which can be analyzed to predict the temperature.Considering the seasonal feature in the temperature series,the seasonal information in the temperature series was extracted with the One-Hot encoding method and used as the input variable of the random forest model to fit and forecast the monthly average temperature.Since numerous hyperparameters are involved in the model,random search algorithm and grid search algorithm are used to optimize the hyperparameters.The results show that the fitting effect of the random forest model considering seasonality is better than that of the simple random forest model,and the predicted data is basically consistent with the actual situation,with the fitting accuracy reaching 96.14%.The fitting accuracy can be up to 96.45%after the hyperparameters search by both methods.