Mixed Prediction Of PM2.5 Concentration On ARIMA-BP-SVM Model

As China enters the stage of rapid development,the urban population is growing,industry and agriculture continue to develop rapidly,and PM2.5 and many molecules and other substances have increased significantly.PM2.5 is a soluble molecule that absorbs large amounts of water,water vapor,and weather fog to form,reducing visibility and recurrent respiratory disease.The frequent occurrence of these events indicates that China's environmental problems are very serious,so the current air quality research attaches great importance to the monitoring and prediction of PM2.5.This paper studies the PM2.5 numerical prediction.The data of the first six months of2019 in Beijing are used for training,and the data for July are tested.The BPANN neural network,SVM support vector machine,and ARIMA model are used for prediction.The three are mixed to form The combined model compares the prediction accuracy of each prediction model by analyzing the error between the prediction result and the true value.First we use BP neural network to predict PM2.5,develop training and test samples according to the characteristics of automatic data connection to determine the optimal neural network structure,and then adjust parameters to improve the BP learning algorithm.Get the expectation,and finally get the calculation error of 24.9%,and the accuracy is more than70%.Next,we combine the advantages of the good prediction effect of support vector machines for lack of samples or small samples,to further predict large-scale sample sets with constraints,replace the relaxation variables with error squares,and change the inequality constraints to equality constraints the prediction result was revised again with an accuracy of about 75%.Immediately after we analyzed the characteristics of PM2.5 data,the data was stabilized,the graph analysis was automatically connected and partly automatically connected to determine a reasonable ARIMA model,and then the AIC and BIC values were calculated to determine the best model,and then the parameters were used Estimate the value to get the PM2.5 prediction model based on ARIMA.The error rate is 24% and the accuracy is above70%.Then we combine ARIMA model,support vector machine SVM and BP neural network model to obtain weights and integrated structure(ARIMA-BP-SVM).The prediction results show that the accuracy of the comprehensive model is higher than the accuracy of the individual model.On this basis,neurons are combined to obtain a comprehensive ARIMA-BP-SVM,and models and results are obtained with higher accuracy.Finally,the average error rate of the prediction results is 0.1864,and the accuracy is about 82%,which meets the expected prediction accuracy and the overall prediction accuracy is high.Finally,we extended and evaluated the model.The ARIMA-BP-SVM comprehensive prediction model not only solves the linear limitations of the ARIMA model and the nonlinear limitations of the BP model,but also combines the advantages of multiple models to obtain a highly accurate comprehensive prediction model.For the next PM2.5 and other multivariate predictors provide a reference.