基于LSTM多步预测模型的空气质量预测与预警

Air Quality Prediction and Early Warning Based on LSTM Multi-Step Prediction Model

本文首先通过斯皮尔曼相关性分析及灰色关联分析,以两方式相对比的方式筛选出与PM2.5浓度变化有关的因素,通过随机森林回归算法得出因素对PM2.5浓度的影响程度。然后将LSTM神经网络模型与多步预测模型相结合,构建LSTM多步预测模型,并设置步长用于预测PM2.5的值,根据均方根误差检验对模型效果进行评估。最后,将数据集带入LSTM多步预测模型并设置步长用以预测AQI的值。

In this paper, firstly, the factors related to the changes of PM2.5 concentration are screened out by Spearman correlation analysis and grey relation analysis through two relative comparisons, and the degree of influence of the factors on the concentration of PM2.5 is derived by random forest regres-sion. Then the LSTM neural network model was combined with the multi-step prediction model to construct an LSTM multi-step prediction model and the step size was set for predicting the value of PM2.5, and the model effect was evaluated according to the root mean square error test. Finally, the dataset was brought into the LSTM multi-step prediction model and the step size was set to predict the value of AQI.