基于梯度下降优化的LSTM对空气质量预测研究

Study on the prediction of air quality based on LSTM withgradient descent optimization

文章以2018年~2019年的太原市空气污染物监测数据为基础,建立基于自适应调节学习率的随机梯度下降算法(Adagrad、AdaDelta、Adam)优化的LSTM循环神经网络预测模型,对太原市的空气质量指数(AQI)进行仿真预测,通过对比可得:基于Adam优化的LSTM循环神经网络不仅具备更高的预测精度,而且收敛速度也较快,为城市大气污染治理工作提供了科学合理的理论研究,具有更远的发展前景.

Based on the monitoring data of air pollutants in Taiyuan city from 2018 to 2019,this paper establishes the LSTM cycle neural network prediction model optimized by the stochastic gradient descent algorithm(adagrad,adadelta,Adam)based on adaptive adjustment learning rate,and simulates and forecasts the air quality index(AQI)of Taiyuan city.Through comparison,it can be concluded that the LSTM recurrent neural network based on Adam optimization not only has the advantages of Higher prediction accuracy and faster convergence speed provide scientific and reasonable theoretical research for urban air pollution control,and have a further development prospect.