基于双向长短期记忆的广州市PM_(2.5)浓度预测研究

Prediction of PM_(2.5)Concentration in Guangzhou Based on Bidirectional Long Short-term Memory Network

提出一种基于双向长短期记忆网络的PM_(2.5)浓度预测方法,其可以利用深度学习模型准确预测1 h后的PM_(2.5)浓度。方法基于随机森林计算的特征重要性来选择预测模型的输入变量,并对数据进行权重分配以减少预测误差,构建双向长短期记忆网络模型最终实现对PM_(2.5)浓度的精准预测。借助广州市番禺区和南沙区2021年~2023年的监控数据进行验证分析,并与传统利用所有输入变量的方法进行对比,所提方案均方根误差减少了4.92%,平均绝对误差减小了7.57%,相对均方根误差减小了4.92%,所提方案能够获得更高的预测精度。

This paper proposes a prediction method of PM_(2.5)concentration based on bidirectional long short-term memory network,which can accurately predict PM_(2.5)concentration after one hour by using deep learning model.This method selects the input variables of the prediction model based on the feature importance of random forest calculation,assigns the weight of the data to reduce the prediction error,and builds a bidirectional long and short term memory network model to achieve the accurate prediction of PM_(2.5)concentration.Using the monitoring data of Panyu District and Nansha District of Guangzhou from 2021 to 2023 for verification and analysis,and compared with the traditional method using all input variables,the minimum root-mean-square error of the proposed scheme is reduced by 4.92%,the average absolute error is reduced by 7.57%,and the relative root-mean-square error is reduced by 4.92%.The proposed scheme can obtain higher prediction accuracy.