针对现在PM2.5浓度预测模的预测精度不高和泛化能力差的问题,提出一种结合时间模式注意力机制和改进时间卷积网络(Temporal Pattern Attention and Temporal Convolutional Network,TPA-TCN)的 PM2.5 浓度预测模型.通过对气象数据和空气污染物监测站点数据进行时空分析,选择具有高相关性的邻近站点作为辅助变量.引入TPA机制,在PM2.5数据时间序列的每个时间步上计算注意力权重,改进TCN的残差结构,提高模型的训练速度和鲁棒性.使用自回归(Au-toregressive,AR)算法优化模型的线性提取能力.实验结果表明,该模型在PM2.5预测对比实验任务中表现优异,具备更高的预测精度和更强的泛化能力.
PM2.5 Concentration Prediction Method Based on Temporal Pattern Attention Mechanism and Improved TCN
To address the current issues of low predictive accuracy and poor generalization ability in PM2.5 concentration prediction models,a PM2.5 concentration prediction model is proposed that combines a Temporal Pattern Attention mechanism with an improved Temporal Convolutional Network(TPA-TCN).Firstly,through spatiotemporal analysis of meteorological data and air pollution monitoring station data,neighboring stations with high correlation are selected as auxiliary variables.Secondly,the TPA mechanism is introduced to calculate attention weights at each time step of the PM2.5 data time series,and then the residual structure of the TCN is improved to improve the training speed and robustness of the model.Finally,the Autoregressive(AR)algorithm is used to optimize the linear extraction ability of the model.Experimental results show that the proposed model exhibits outstanding performance in PM2.5 prediction tasks.Compared with traditional time series prediction models,the proposed model has better prediction performance and generalization ability.
国家科技期刊平台
NSTL
万方数据
