基于粒子群算法和注意力机制的LSTM的PM_(2.5)预测研究

PM_(2.5)Prediction of Long Short-Term Memory Network(LSTM)Based on Particle Swarm Optimization Algorithmand Attention Mechanism

PM_(2.5)是空气质量的重要影响因素之一,更加准确地预测PM_(2.5)的含量,对于预报空气质量变化、空气治理和促进科学绿色发展都有着重要的作用。本文提出一种基于粒子群算法和注意力机制的长短期记忆网络(LSTM)模型,该模型既具备了LSTM可以轻松提取数据的时间维度信息的能力,又具备了注意力机制可以完美解决特征权重分配的能力,可以较为准确地对空气中PM_(2.5)含量进行预测。通过与K近邻回归、支持向量回归、循环神经网络和未进行寻优处理的基于注意力机制的LSTM等模型进行对比试验,证明了基于粒子群算法和注意力机制的LSTM在预测空气中PM_(2.5)含量时具有更佳的性能,且模型的均方误差(MSE)、平均绝对误差(MAE)在保证相同相关系数(R^(2))的情况下,降低了50%以上。

PM_(2.5)is one of the important factors affecting air quality.More accurate prediction of the content of PM_(2.5)plays an important role in forecasting air quality changes,doing air governance and promoting the scientific and green development.This paper proposes a Long Short-Term Memory Network(LSTM)model based on particle swarm optimization algorithm and attention mechanism.This model has both the ability of LSTM to easily extract the time dimension information of data,and the ability of attention mechanism to perfectly solve the feature weight distribution,which can more accurately predict the content of PM_(2.5)in the air.Through comparative experiments with K nearest neighbor regression,support vector regression,recurrent neural network and LSTM based on attention mechanism without optimization processing,it is proved that the LSTM based on particle swarm optimization algorithm and attention mechanism has better performance in predicting PM_(2.5)content in the air,and the Mean Square Error(MSE)and Mean Absolute Error(MAE)of the model are reduced by more than 50%under the same correlation coefficient(R^(2)).