基于LSTM的PM2.5浓度预测模型

Prediction of PM2.5 concentration based on LSTM

摘要:

随着近年雾霾天气的频繁出现,空气质量开始越来越受到公众关注。PM2.5浓度指数是判断空气质量的重要指标,如何根据历史数据有效地预测空气中PM2.5浓度,具有很高的应用价值。分析以往空气质量数据表明,PM2.5浓度有明显的非线性和不确定性波动,很难用传统机器学习算法有效地预测。本文基于LSTM循环神经网络,依据过去20小时采集的空气数据,预测未来5小时的PM2.5浓度指数。实验结果表明,LSTM可以有效地捕获空气质量的时序特征,较准确预测出未来时刻的PM2.5浓度指数。

Abstract:

With the frequent occurrence of smoggy weather in recent years, air quality has begun to receive more and more public attention. The PM2.5 index is an important indicator for judging air quality. How to effectively predict the PM2.5 concentration in the air based on historical data, which has application value. And analysis of previous air quality data show that PM2.5 concentration has obvious nonlinearity and uncertain volatility, which is difficult to predict effectively with traditional machine learning algorithms. Based on the LSTM (Long Short-Term Memory) Recurrent Neural Network (RNN), this paper predicts the PM2.5 concentration for the next 5 hours based on the air data collected over the past 20 hours. Experiments show that LSTM can effectively capture the feature of time series for air quality and accurately predict the PM2.5 concentration in the future.