| In recent years,air quality issues have attracted more and more attention.Effective real-time and long-term air quality level prediction has important guiding significance for guiding people’s travel and urban construction planning.The air quality is affected by multiple factors such as weather,traffic conditions,related pollution source emissions and use of surrounding land.The non-linear relationship involved is very complex,but some data cannot be directly obtained due to privacy protection,which brings difficulties for forecasting.This paper proposes a forecasting system based on correlation analysis of multiple classes of data.The main contribution includes:1.Based on the protection of data privacy,we explore the associations within and between multiple classes of urban perception data,and design an end-to-end multi-task learning prediction model using the inter-class association relationship.This method achieves a good prediction and at the same time solves the problem of directly using multiple classes of data and association analysis is not obvious and difficult to promote and apply.2.This thesis discusses numbers’objectivity contained in the embedded expression of features in air quality prediction system,and designs a model combining TransE and mathematical expressions to learn number expressions of any magnitude.The performance of number expression learning is verified by effectiveness experiment of number reasoning and performance of air quality single time series prediction task.3.The air quality prediction system is established based on the multiple data correlation analysis model,which realizes the interface of model training,loading,multi-task series prediction,single-task series prediction,data distribution showing and data correlation analysis.At the same time,this thesis completed a number of experiments on the datasets of KDD Cup 2018 and AI challenge which meets user needs and has good accuracy. |
PDF Full Text Request