Prediction Of Ozone Concentration Based On C-PSODE Algorithm And BP Neural Network

Ozone pollution has a negative impact on human health,climate and vegetation.In order to comprehensive understand and Grasp trend of ozone concentration,the research on tropospheric ozone concentration prediction is crucial.It's necessary to establish an ozone concentration prediction model to describe the complex relationship between changes in ozone concentration and related variables that cause or hinder ozone production.An effective ozone concentration prediction model has theoretical and practical significance for providing effective early warning and improving prediction accuracy.This paper takes the meteorological monitoring and ozone premature historical data of the Banqiao monitoring station in Taipei in 2015 as the research object.The literature research showed that the ozone concentration in Taipei has seasonal fluctuations.According to the season,the historical data is divided into four data sets,which are respectively handled using factor analysis of SPSS.Then eliminate factors which are weakly related to ozone,and use PCA to reduce network complexity.This paper raises a differential evolution algorithm based on chaotic algorithm optimization and particle-group hybrid evolution(C-PSODE)algorithm.The algorithm is compared with PSO algorithm,DE algorithm and PSODE algorithm by multi-extremum function verification.It is concluded that the algorithm has strong adaptability,high stability and high accuracy.The optimal initial weight and threshold of BP neural network are obtained by this algorithm.The BP neural network ozone hourly concentration prediction model(C-PSODE-BP)based on the algorithm is designed and implemented.The model can effectively improve the prediction accuracy of ozone and effectively avoid the shortcomings of the network falling into local optimum.In this study,the ozone concentration of spring,summer,autumn and winter in Taipei was predicted by the model.The average prediction accuracy was 91.04%,85.25%,90.22%and 91.15%,respectively.The following conclusions were found:the model is significantly better than the PSO-BP model,the DE-BP model,and the PSODE-BP model.The ozone precursors were found to be the most influential factors for ozone concentration prediction:CO,C02,N02,NOx,CH4,THC and NMHC;the ozone concentration in spring and autumn in Taipei is significantly higher than that in summer and winter.The research in this paper believes that using BP neural network to compensate BP neural network is a feasible and accurate method for modeling air pollution.