Research On Ozone Feature Analysis And Prediction Algorithm Based On Spatio-temporal Data

Ozone is one of the main causes affecting the atmospheric environment,and has become the focus of attention in many fields such as the environment.The atmospheric environment monitoring system provides strong support for ozone related research.How to obtain relevant information of ozone time series and predict future ozone data through monitoring data,and provide technical support for effective environmental governance,is the current research challenge of interdisciplinary fields such as environment and computer.In response to this challenge,on the basis of systematically sorting out the environmental grid monitoring data,this paper conducts research on ozone feature analysis and prediction algorithms based on spatio-temporal data from the perspective of machine learning:(1)To solve the problem of insufficient utilization of ozone time series information,the research proposes a time domain-oriented ozone monitoring data prediction model.First,it uses the moving average filter to captive the time-domain features of the preprocessed ozone historical data,filters and then classifies the monitoring information.After that,the classified information is analyzed as integrated with the processed time series to establish the time-domain prediction model,which is based on the method of the Moving Average-Long Short Term Memory Network(MA-LSTM).The results show this method can solve the issue of insufficient utilization of time series.Furthermore,the use of mixed models improves the accuracy of prediction to a certain extent.(2)In response to the insufficient use of ozone airspace information,this research proposes an ozone monitoring data prediction model for spatio-temporal heterogeneous and multi-granularity.First,the spatially relevant research area is judged based on the space characteristics of the research site.The site partition,which mainly uses on the K-means method and the first geography law,forms the zoning strategy integration method based on the similarity of the spatial ozone sequence.It can train multiple different models,dynamically plans the forecast range as well as determine the relevant areas.Then the multi-site prediction results obtained by MA-LSTM are used to learn the spatial relationship through machine learning method.Finally,a spatial weighting system based on the traditional geography inverse distance weighting method(IDW)is used to establish an unknown point prediction model.The fusion of spatio-temporalinformation to obtain the final prediction results of unknown points show that the method effectively combines geographic information and reduce the impact of sudden changes.