基于EMD和CatBoost算法的改进时间序列模型——以大连市PM_(2.5)预测为例

Improved time series models based on EMD and CatBoost algorithms:taking PM_(2.5) prediction of Dalian City as an example

【目的】解决传统大气PM_(2.5)浓度时序预测时精度较低问题,减少PM_(2.5)时间序列的非线性、高噪声、不平稳与波动性对预测的影响,从而更精确地预测PM_(2.5)浓度。【方法】以2014年1月1日至2022年1月31日大连市雾霾天气时PM_(2.5)数据为例,提出了经验模态分解(EMD)、分类提升(CatBoost)、自回归综合移动平均模型(ARIMA)组合的混合机器学习时间序列模型,并与传统自回归模型(AR)、ARIMA,以及只加入EMD方法后的混合模型进行比较。【结果】混合模型EMD-CatBoost-ARIMA较原始序列均方根误差(RMSE)改进20.76%,平均绝对值误差(MAE)改进17.40%,希尔不等系数(TIC)改进29.17%。【结论】对于高熵值的重构序列,EMD分解方法和CatBoost算法能够显著提升PM_(2.5)时间序列模型的预测性能。相比较传统时间序列模型,EMD-CatBoost-ARIMA模型对大气PM2.5浓度预测性能较高。

【Objective】The study aims to address the problem of low accuracy in traditional PM_(2.5) concentration time series prediction,and to reduce the impact of nonlinearity,high noise,instability and volatility on the prediction of PM_(2.5) time series,to predict PM_(2.5) concentration more accurately.【Method】The haze PM_(2.5) data of Dalian City from January 1,2014 to January 31,2022 was used as an example.In this study,a hybrid machine learning time series model with the combination of empirical modal decomposition(EMD),classification boosting(CatBoost)and autoregressive integrated moving average model(ARIMA)was proposed.It was compared with the traditional autoregressive model(AR),ARIMA and the hybrid model with only the EMD method.【Result】The hybrid model EMD-CatBoost-ARIMA improved the root mean square error(RMSE)of the original sequence by 20.76%,the mean absolute error(MAE)by 17.40%,and the theil inequality coefficient(TIC)by 29.17%.【Conclusion】For reconstructed sequences with high entropy values,the EMD decomposition method and CatBoost algorithm can significantly improve the prediction performance of PM_(2.5) time series models.Compared with the traditional time series models,the EMD-CatBoost-ARIMA model has higher performance in PM_(2.5) concentration prediction.