基于改进的Seq2Seq-LSTM模型的空气质量指数预测研究模型

The Air Quality Index Prediction Research Model Based on Improved Seq2Seq-LSTM Model

针对长短期记忆网络(Long Short-Term Memory,LSTM)模型输入输出时间步长度相等、处理长序列遗忘多、无法按重要程度分配权重等不足,构建了一种基于注意力机制(Attention Mechanism,Attention)改进的Seq2Seq-LSTM组合模型。该模型将序列到序列(Sequence to Sequence,Seq2Seq)模型中编码器、解码器设置为三层LSTM结构,并在解码器输出序列前引入注意力机制对模型进一步优化。为验证改进后的Seq2Seq-LSTM模型的有效性,本研究以山东省青岛市为研究区域,基于历史数据对未来1~7 h的空气质量指数进行模拟预测;预测结果与传统的机器学习模型支持向量机(Support Vector Machines,SVM)以及单一的LSTM模型预测结果进行了对比。结果表明:改进后的Seq2Seq-LSTM模型在中长期空气质量指数预测中的预测效果更突出。说明改进后的Seq2Seq-LSTM模型较单一模型具备更强的预测力,可作为山东省青岛市中长期空气质量指数预测模拟的可靠工具。

Since the reform and opening-up,our country is faced with the complex and changeable environmental pollution and governance problems at the same time of economic development,it is of practical value to develop an air quality index(AQI)prediction system to assist the Environmental monitoring work.The AQI prediction model based on the attention mechanism improved Seq2Seq-LSTM model in this paper aims to provide more reliable prediction results for workers,detect pollution ahead of time and reduce the cost of pollution control,then promote our country’s ecological civilization construction,promote people’s Life Happiness Index.In recent years,data-driven models,represented by deep learning(a classification of machine learning)algorithm,are widely used in AQI prediction due to their advantages of not considering complex parameters,simulating prediction by mining the latent law of data itself,and high simulation accuracy.The commonly used machine learning algorithm has achieved good results in air quality index prediction.Compared with machine learning model,deep learning model can learn the inherent laws and levels of sample data in a faster and more effective way,which greatly improves the prediction accuracy of the model.Long Short-Term Memory(LSTM)and Attention Mechanism(Attention)are commonly used models of deep learning.Among them,LSTM model is widely used in air quality index prediction research because of its simplicity,flexibility,stability and long-term memory Seq2Seq model is widely used in multivariable prediction task because it can handle the non-uniform sequence of input and output step.Attention mechanism shows its excellent performance in the field of natural language processing.Although the LSTM model has been proved to have good performance in air quality index prediction,it still has some defects,it is very important for AQI prediction research to break through the limitation of single model.In order to solve the problems of Long Short-Term Memory(LSTM)model,such as equal length of time step between input and output,Long sequence forgetting and weight distribution in importance,this paper constructs an improved Seq2Seq-LSTM combination model based on Attention Mechanism(Attention).In this model,the encoder and decoder in the Sequence to Sequence(Seq2Seq)model are set as three-layer LSTM structure,and the attention mechanism is introduced before the output Sequence of decoder to further optimize the model.In order to verify the validity of the improved Seq2Seq-LSTM model,this study takes Qingdao City of Shandong province as the research area,and based on the historical data,carries on the simulation forecast to the air quality index of the next 1~7 hours;the predicted results were compared with those of traditional Support Vector Machines(SVM)and single LSTM Support Vector machine.The results show that the improved Seq2Seq-LSTM model is more effective in the medium and long term air quality index prediction.The results show that the improved Seq2Seq-LSTM model is more powerful than the single model,and can be used as a reliable tool for the medium and long-term air quality index prediction simulation in Qingdao City,Shandong province.