Font Size: a A A

Research On Urban Air Quality Prediction Based On Multi-source Data Fusion

Posted on:2024-07-13Degree:MasterType:Thesis
Country:ChinaCandidate:L BiFull Text:PDF
GTID:2531307160455394Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of industry and the increase of population,urban environmental problems have become increasingly prominent.At the same time,air pollution has also brought a series of challenges to urban safety management.Therefore,accurate air quality prediction is helpful to improve the scientific management level of urban environment and reduce the risk of environmental pollution to people.However,there is a spatial correlation between each air quality monitoring station and the surrounding stations,and the complex space-time interaction between air quality monitoring stations makes it difficult to improve the accuracy of air quality prediction,so the problem of air quality prediction needs to be solved urgently.The air monitoring data are collected by air monitoring stations distributed in different areas of the city.These data provide data basis for mining the relationship between data and establishing a scientific air quality prediction model.At present,there are many deep learning methods to predict the air quality.The existing prediction methods have the following shortcomings.Firstly,because the air is diffusive,the air quality in the space to be predicted is easily affected by the air quality in adjacent areas,and the existing research is difficult to accurately capture the complex time-space interaction relationship between air quality monitoring stations.Secondly,air quality prediction depends on short-term data modeling,ignoring the long-term characteristics of air quality data.In view of the above problems,the main research work and innovative achievements of this thesis are as follows:(1)Multi-source data preprocessing.There are some problems such as data missing in the urban air quality data acquired by monitoring stations.Firstly,the urban multi-source air quality data are preprocessed by means of cleaning and denoising,and then the data set needed for research is extracted from the preprocessed data according to the research area.Finally,the data are normalized by unified time granularity and converted into time series data.(2)Research on spatial dependence based on urban quality prediction model.Accurate and reliable air quality prediction is very important for human beings to avoid air pollution.In order to realize short-term prediction of urban air quality,this thesis puts forward a prediction model of urban air quality based on graph convolution neural network.Taking Beijing urban area as an example,according to the spatial and temporal characteristics of multi-source data in this city,the urban air quality is analyzed by combining convolution graph neural network.The proposed model uses the processed data to construct the spatial-temporal data set of urban air quality,and uses the graph volume network to extract the spatial features between stations.The temporal features are extracted by the gated convolution neural network.Finally,the prediction results of air quality are obtained by analyzing the whole connection layer.Experimental results show that compared with LSTM(Long and Short Term Memory Networks)prediction method,the proposed model has better prediction accuracy,and can capture the spatial correlation between different stations and the time dependence within the same stations.(3)Based on the improved LSTM model,the long-term prediction accuracy can be obtained.Aiming at the long-term prediction of air pollutants,the encoder-decoder model based on LSTM has become the mainstream method of time series prediction.LSTM network depends on various gate units,but many existing studies ignore the correlation between gate units,and the limited memory information makes it difficult for the constructed model to obtain more accurate long-term prediction results.In view of the fact that the existing prediction models tend to ignore the features that play a key role in the long-term prediction of air quality,and the traditional neural network can not accurately capture the long-term dependence of data,an improved LSTM model is proposed to realize the deep feature extraction of more long-term dependence in sequence data.The simulation results show that compared with the traditional model,the proposed model improves the prediction accuracy by fully extracting the data correlation,and overcomes the long-term dependence and other problems.
Keywords/Search Tags:multi-source data, data fusion, air quality prediction, LSTM neural network
PDF Full Text Request
Related items