Urban Air Quality Prediction Based On Integrated LSTM

With the continuous improvement of data acquisition and storage capabilities,spatio-temporal data like meteorological quality is becoming the research object of statistics and other disciplines.Spatio-temporal sequence data,as a spatial extension of time series data,has been widely used in many scenarios which has received more and more attention.This paper starts from the data preprocessing dealing with the missing values and outliers,solving the shortcomings such as incomplete data and abnormalities caused by equipment and other reasons,building a good foundation for modeling.By referring to the relevant literature,a novel multi-variable spatial weight matrix is proposed,which replaces the traditional weight matrix established by inverse distance,point connection and edge connection,and applied to the traditional STARMA model.In order to improve the forecasting effect furtherly,LSTM(longterm and short-term memory model)is used as the base model.The learning efficiency of this model is improved by the proposal of weighted minimum square error and the penalth of Group＿lasso,which treat as loss function.Besides,a series of base model are integrated by using bagging ideas.Multiple groups of comparative experiments based on examples are compared to verify the accuracy and feasibility of the model prediction and show the superiority of the model.Finally,based on the research conclusions,this paper initially obtains the validity of the model prediction and the availability of its prediction results in other fields.Spatio-temporal sequence prediction can be further applied in the field of meteorology,and can also play a role in prediction of crime rate,transportation flow,macroeconomics and other fields.It is worthy of further practical application and promotion.