Offline And Online Air Quality Forecast With Improved Integrated Learning Algorithm

Air quality predictions is related to air quality and human disease.With many influencing factors that are of complex nonlinear relationships,the air environment is complex and variable.Air quality data has the general characteristics of real-time data streams which needs combination of off-line and online prediction.Therefore,this thesis has important application value for the study of air quality analysis and prediction.This thesis proposes an integrated learning method that combines with PI system to analyze and predict air quality.In the offline section,we propose a new combination OPGBoost algorithm based on XGBoost,which is mainly optimized by custom XGBoost loss function and Bagging integration.According to the characteristics of air quality data,the data is processed by LaGrange interpolation method,constructing the properties of the air data over a period of time by using the staggered method to expand the original data characteristics and choose out important features.Then,predictive models with XGBoost,RF,OPGBoost and BP algorithms,OPGBoost algorithm is selected as the final air quality prediction model.In addition,through secondary development for achieving online forecasting,the PI system introduces machine learning modules and adds data mining capabilities.The PI data fusion function is used to integrate the feature subsets,the sliding window is used to cache the data set and the attenuation function is used to control the weight of the generated model,finally,combining the previously generated model to achieve online real-time prediction using OPGBoost algorithm on PI system.Experiments show that the air quality prediction for PM_2.5 and AQI has achieved better off-line modeling and online prediction effects,it has good practical significance.