Research On Air Quality Prediction Based On Neural Network And Kriging Model

Air pollution has become the barrier for the development of China,which has caused much inconvenience in production and life and serious harm to people’s health.For environmental protection departments,the accurate forecasts for air quality enable them to take preventive measure in advance.Meanwhile,Air pollution control work can be completed efficiently and the harm from harmful weather for people can be reduced.Therefore,it is significant to forecast air quality with high accurate.The main research contents are as follows:Firstly,the meteorological factors closely related to air pollution concentration are analyzed and explored.We carried out the correlation analysis among temperature,air pressure,humidity,precipitation and wind speed of Hebei Province in recent years.Meanwhile,we compared and analyzed the meteorological data and air quality index over the years to verify the correlation between each meteorological factor and AQI.the meteorological factors,which have a significant impact on AQI,are selected as the input of the air quality prediction model.Secondly,BP neural network model is selected for the prediction model in time dimension,and an improved particle swarm optimization algorithm is proposed to improve its accuracy.The improved particle swarm optimization algorithm change the inertia weight by two nonlinear decreasing curves,the learning factor changes dynamically with tangent function,and keeps the adaptive variation,which can avoid the particle falling into the local extreme value and ensure the fast convergence of the particle.By analyzing the experimental results,it is shown that the BP neural network model optimized,based on the improved particle swarm optimization,has the advantages of high accuracy and strong stability.Finally,kriging interpolation model was selected as the spatial prediction model,and the improved genetic simulated annealing algorithm was used to optimize the interpolation accuracy.The sigmoid crossover mutation curve and elite retention strategy were adopted in the algorithm to maximize the retention of excellent individuals,while the Metropolis criterion carried out gene reversal operation to increase individual diversity in the face of poor new individuals.The experimental results show that the kriging model,which is optimized by the improved genetic simulated annealing algorithm,has faster convergence speed and higher interpolation accuracy.In addition,the AQI concentration distribution can be forecast based on the previous air quality information and meteorological information,and the prediction results are intuitive and accurate.