Air Quality Prediction Based On Neural Network

With the rapid development of economy and the acceleration of industrialization and urbanization in China,the problem of air pollution is becoming more and more serious.In the field of environmental science,air pollutant concentration prediction is a very important research topic.With the advancement of detection instruments and observation techniques,large-scale,multi-dimensional and complex air quality data have been obtained.The rapid development of computer technology has brought technological innovation to the field of air quality monitoring and opened the era of information and intelligence.To improve the prediction accuracy of PM_2.5 concentration,other air pollutant concentrations and meteorological conditions are considered.This context proposes prediction models for air quality forecast which are based on BP neural network and deep belief network respectively.Because the performance of BP neural network model is affected by initial weights and thresholds and this model is easy to fall into local minima,PSO optimization algorithm is selected to optimize the model parameter selection,and the inertia weight is improved.The crossover and mutation operation in genetic algorithm is introduced to establish the improved PSO-GA-BP prediction model,so as to improve the prediction accuracy and convergence performance of the model.In view of the large data set of hourly forecasting,DBN-SVR forecasting model is established.This model combines deep belief network with support vector regression.The grey relational analysis is used for feature selection.According to the past air quality data and meteorological data,a prediction model is established to predict the future PM_2.5 concentration.The experimental results of one hour time step show that compared with BP neural network model and PSO-GA-BP model,DBN-SVR model has the best prediction accuracy,shorter running time and higher efficiency when the sample size is large.To explore the performance of the model in larger prediction time step,the models of 1-6 hour prediction time step are trained respectively and compare with support vector regression and BP neural network.The results show that DBN-SVR prediction model has better performance than BP neural network model and support vector regression model.