Design Of Air Quality Prediction System In Chengdu Based On Data Mining

In recent years,haze weather has appeared more and more frequently.People pay more and more attention to air quality,and expect to use relevant technologies to excavate useful information in the air quality data to predict air quality,thus preventing the harm caused by air pollution.However,the air quality is greatly affected by meteorological factors,and it has characteristics such as non-linearity,time-variation,and high degree of instability.How to establish an effective air quality prediction model and prediction system has always been a problem that people urgently need to solve.The main contents of the thesis are as follows:(1)According to the prediction system requirement analysis,the overall design and parts design of the system are carried out.The entire system includes four layers: data acquisition layer,data service layer,data processing layer,and user interface layer.It is divided into three parts: data acquisition and storage,background management,data analysis and prediction.(2)A prediction system development environment was set up,and the function modules and interfaces of the system are programmed.Furthermore,the realization of system function is implemented with instance code and implementation renderings,including web crawler,web-side,data analysis and prediction and so on.After the system was implemented,its functions and performance were tested accordingly.(3)To solve the problem of outliers and missing values of Chengdu’s urban meteorological data and air quality data,this paper preprocessed the acquired data and selected two independent variables PM10 and PM2.5 that affect the AQI using the Adaptive-Lasso algorithm.When exploring the main factors affecting PM10 and PM2.5,the Pearson coefficient and Spearman coefficient obtained from the experiment were used to analyze their correlation with 5 meteorological factors,and the input parameters of the predictive model were found.(4)For the problem that the standard differential evolution algorithm(DE)has low search efficiency,this paper improves the searching efficiency of the DE algorithm by changing the variation factor and cross factor of the DE algorithm,and obtains an improved differential evolution algorithm(IDE).Tested with 5 multi-peak functions,the test results showed that IDE improved the DE’s efficiency.The influence of step size on the cuckoo search algorithm(CS)cannot be ignored.This paper improves the CS algorithm by improving the step size control and the improved cuckoo algorithm(ICS)is obtained.(5)For the traditional BP neural network model is too dependent on the initial value,the convergence speed is slow,easy to fall into the local minimum value and other issues,this paper uses the improved differential evolution algorithm(IDE)and the improved cuckoo search algorithm(ICS)to optimize the weights and thresholds of the BP neural network model,and obtained CS-BPNN and IDE-BPNN two models.The two improved models are compared with the traditional BP neural network model and the four optimized BP neural network models.The prediction results of PM10 and PM2.5 concentrations shows that the root mean square error(RMSE),the mean absolute error(MAE)and the mean absolute percentage error(MAPE)of IDE-BPNN are smaller than those of other models,and the index of agreement(IA)of IDE-BPNN is the highest,The mean bias error(MBE)of IDE-BPNN tends to be zero.From the result analysis,the IDE-BPNN model has better prediction performance.In this paper,through a comparative analysis of various air quality prediction models,an effective air quality prediction model is obtained,which improves the prediction accuracy of air quality.Finally,an air quality prediction system integrating data acquisition,storage,preprocessing and predictive analysis was designed and implemented.After many tests,the system test results reached the expected design goals.