基于混合神经网络模型的空气污染物预测研究

Research on air pollutant prediction based on hybrid neural network modelling

为了更好地预测未来一段时间内空气中污染物的含量,提高预测精度,减小误差,提出了一种基于一维卷积神经网络(CNN),和双向长短时记忆网络(BiLSTM)结合了注意力机制(AT)和粒子群优化(PSO)的CNN-BiLSTM-ATPSO空气污染物预测模型。一维卷积用于学习局部特征趋势,BiLSTM用于捕获时间序列之间的依赖关系,据此设计CNN-BiLSTM时间预测模型,并结合注意力机制和粒子群两种深度学习算法,进一步对模型进行改进。通过与LSTM-Attention模型和BiLSTM-Attention模型进行对比,通过比较评估指标平均绝对误差(MAE)和均方根误差(RMSE),得出结论,CNN-BiLSTM-ATPSO模型在预测精度方面具有显著优势。

s:In order to better predict the air pollutant content in the future period,improve the prediction accuracy and reduce the error,an air pollutant content prediction model based on a one-dimensional convolutional neural network(CNN),and a bidirection-al long and short-term memory network(BiLSTM)combining the attention mechanism(AT)and particle swarm optimisation(PSO)is proposed.One-dimensional convolution is used to learn local feature trends and BiLSTM is used to capture the dependencies between time series,according to which the CNN-BiLSTM temporal prediction model is designed and further improved by combining two deep learning algorithms,namely Attention Mechanism(AT)and Particle Swarm(PSO).By comparing with the LSTM-Attention model and BiLSTM-Attention model,it is concluded that the CNN-BiLSTM-ATPSO model has a significant advantage in terms of prediction ac-curacy by comparing the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)of the evaluation metrics.