Research Of Air Quality Prediction Model Based On Ensemble Learning

With the development of industrial agricultural science and technology,the air quality of each city gradually begins to affect people's daily life and physical and mental health.The improvement of air quality has become an imperative,and air quality prediction can make an important contribution to the improvement of air quality.Ensemble learning is a nonlinear prediction method and it is an important branch of machine learning.Python is a concisely readable programming language.Its rich library of machine learning allows many problems to be solved.This thesis takes Xi'an AQI daily data from 2013 to 2017 as a sample,and realizes the air quality prediction on the Python machine learning platform.The main work of the thesis is as follows:The paper firstly studies the traditional air quality prediction methods,introduces the theoretical knowledge of machine learning and ensemble learning methods,and researches the commonly used machine learning algorithms:multiple linear regression algorithm,decision tree algorithm,Bagging algorithm,random forest algorithm and GBDT algorithms,and use air quality data for instance analysis of algorithms.In view of the deficiencies of the random forest algorithm in predicting and classifying unbalanced data,the classification effect of the decision tree in the random forest is selected and optimized.The results show that the error of the optimized random forest algorithm is reduced.Pre-processing of air quality data,visual analysis of the trends of various pollutant gases in the air,analysis of seasonal changes in the concentration of polluted gases,and recommendations for air treatment.Then,three predictive regression models were established based on the ensemble learning method:air quality regression prediction model based on Bagging algorithm,air quality regression prediction model based on random forest algorithm,and air quality regression prediction model based on GBDT algorithm.And three prediction models were used respectively,the air quality is predicted and the prediction error and the relationship between the decision tree size are analyzed and studied.PM₁₀ is the most serious air pollution.By adjusting the parameters of the three prediction models,the optimal parameters of each model are obtained.By comparing the forecasted mean square errors of the three optimal models,the random forest prediction model is determined as the final prediction model.Using random forest regression prediction model to predict air quality in Xi'an over the next five days,the expected results were obtained,and based on the forecast results,suggestions were made as to whether it is suitable for people to travel under this air quality level.