Study On Estimation And Prediction Of Near-Surface Ozone Based On Neural Network

Near-surface ozone is a very important trace gas in the atmosphere.It is not only the third largest greenhouse gas in the world,but also it has important effects on human health,regional air quality and ecosystems.The study of the long-term change characteristics of near-surface ozone and its influencing factors will be conducive to the scientific understanding and understanding of the impact of human activities and meteorological factors on the atmospheric environment and its internal relationship,and to the formulation of more effective air pollution control measures.Due to the complexity of the surface ozone photochemical synthesis process and the transmission characteristics,it is difficult to monitor the ozone concentration near the ground by traditional monitoring methods of air pollutants.Currently,the main monitoring means of near surface ozone in China is ground air quality monitoring stations.But the cost of site construction is too high,and most of them are distributed in urban areas,which can not be effectively monitored in the area without site.At the same time,near-surface ozone is limited to real-time monitoring and historical data analysis,so the prediction of near-surface ozone in the future has become one of the main research directions of ozone.In order to solve the above problems,this paper utilized the neural network which has the ability to fit and generalize complex nonlinear relations to model the ozone monitoring station area,realized the grid distribution of near-ground ozone estimation,and achieved the regional monitoring of ground ozone.Combined with the historical data and influencing factors of ozone,the neural network model was also used to predict the near surface ozone in the future.The main work of this paper was as follows:(1)The process of constructing and preprocessing data sets for estimation and prediction of near-surface ozone was proposed.Combined with the photochemical process and transmission of ozone and other factors,the data sets were constructed by resampling and longitude and latitude matching methods,and the data sets were preprocessed by the defined data pretreatment process.Combined with data characteristics and models,the estimation and prediction data sets were constructed.(2)A near-surface ozone estimation model based on CNN was proposed.Combining the residual network structure and the fully connected network,the feature extraction model was designed,and the 1-dimensional input feature was converted into the 2-dimensional feature as the input of CNN.According to the data characteristics,the corresponding CNN model was designed,and the multi-layer convolution kernel was designed to reduce the training parameters of the convolution kernel greatly.The method of L2 regularization and Dropout mechanism were used to optimize the model overfitting.Compared with BP neural network and support vector regression model,the experimental results showed that the estimation effect of the model designed in this paper had the best estimation effect.And the result of gridding was feasible.(3)A near-surface ozone prediction model based on LSTM-CNN was proposed.An improved R-LSTM model was proposed by combining the residual network structure and LSTM.The multi-layer LSTM model was used to extract the preliminary features of the time series data set,and the LSTM-CNN model was constructed by combining with the CNN network and the fully connected network to realize the feature fusion.Multiple experiments were set to select the appropriate optimizer and set learning,and the model training process was overfitted.The experimental results show that the determination coefficient(R2)of the test set of the prediction model designed in this paper is higher than that of the BP neural network,CNN and LSTM models.The estimation model proposed in this paper can obtain the grid distribution of near-surface ozone,solve the problem of non-site area can not be monitored,and provide data support for the treatment of near-surface ozone pollution.The high determination coefficient of the prediction model indicates that the model can be effectively applied to the prediction of near-surface ozone and provides a new way of thinking for the prediction of other pollutants.