Research And Application Of Deep Learning Based Intelligent Control Of Air Quality Data

In recent years,China’s infrastructure,real estate industry and construction industry development rapidly,our atmosphere environment has suffered serious damage.In order to effectively protect the atmospheric environment and further reduce the harm caused by air pollution,China has established automatic air quality monitoring stations in various regions to monitor and monitor air pollutants.Due to many reasons such as equipment failure and rapid deterioration of external environment,the data collected by the air quality automatic monitoring station are often abnormal.The abnormal detection system of air quality data built by manual and traditional threshold judgment has low accuracy and slow efficiency,which cannot meet the needs of today.In recent years,deep learning technology has made good achievements in image classification,speech synthesis,time series prediction and other fields.Some scholars also try to apply deep learning technology to anomaly detection of air quality data,but the current research in this area is not in-depth enough.This paper studies and innovates long short-term memory network with long term memory and attention mechanism that can obtain global information.Finally,we build an intelligent air quality data control system based on the anomaly detection algorithm.Firstly,we propose an anomaly detection framework for air quality data,which is divided into prediction stage and anomaly detection stage.In this framework,we propose a short-term attention-LSTM prediction model based on long short-term memory network and multi-head Attention mechanism.We use the two-layer LSTM network to capture the complex time series correlation features among air quality data,and use the multi-head attention mechanism to extract the time series correlation features of different subspaces,which can obtain more complete and effective feature information.The model can accurately predict PM_2.5 concentration through the experiment of a station in Shenyang.In the stage of anomaly detection,we compare three anomaly detection algorithms including Quartile,One Class SVM and 3σrule.Finally,it is proved that Quartile can identify abnormal data efficiently and reduce false positives of normal data.Secondly,considering the influence of multiple factors on PM2.5,we construct a long-term prediction model of attention-Gaussian based on multi-head attention mechanism and Gaussian density estimation.In order to capture the long-term dependence and global information between air quality data,we use the information of different time stamps（hour,day,week）as input.In order to obtain more robust prediction results,we use gaussian density estimation to sample the predicted values.In the stage of anomaly detection,the 3σrule achieves optimal results,we finally determine the attention-Gaussian-3σanomaly detection model.Finally,we design and implement an intelligent air quality data control system based on the attention-Gaussian-3σanomaly detection model.The system realizes the functions of data monitoring,statistical analysis,abnormal notification,abnormal cause reporting and so on.By testing each functional module of the system,the system finally achieves the expected effect and meets the needs of users.