Atmospheric Electric Field Data Analysis And Application Research Based On Autoregressive Model And Machine Learning

The atmospheric electric field is the basic parameter of atmospheric electricity.The ground has a vertical downward atmospheric electric field in sunny days.On thunderstorm days,the ground atmospheric electric field is significantly enhanced.The observation and research of the atmospheric electric field is of great significance to reduce the lightning disaster,research on atmospheric electricity and guarantee the aerospace activities.Atmospheric electric field mill is the basic instrument to measure atmospheric electric field.The effective calibration,data correction and outlier detection of the atmospheric electric field mill are important means to improve the quality of observation data.Atmospheric electric field has the characteristics of non-stationary signal.Using non-stationary signal processing method,combined with machine learning algorithm,the atmospheric electric field under different weather conditions is analyzed,which provides ideas for lightning warning research.The main research contents and conclusions of this paper include:(1)In order to improve the accuracy and consistency of electric field observation data,research the calibration methods of atmospheric electric field sensors.Establish the calibration model of electric field sensor,and the key parameters such as the size and structure of the calibration device are simulated and analyzed based on the finite element method.Considering that the building has a certain distortion effect on the atmospheric electric field.By changing the building height,width,the distance between the electric field meter and the building,and thr elevation angle range of electric field meter,study the influence of the actual installation environment of the atmospheric electric field on the electric field distortion.Distortion coefficient,the actual measurement value of the atmospheric electric field instrument is revised.(2)The cleaning of the atmospheric electric field sequence is the key to preprocessing,which is of great significance to the subsequent mining research.Aiming at the shortcomings of traditional anomaly detection algorithms that need to specify the corresponding parameters and fail to use the relevant information between time series,a detection and correction method for atmospheric electric field anomalies based on isolated forest combined with Chen-Liu algorithm is proposed.This method uses the autoregressive integrated moving average model(ARIMA)model to fit the atmospheric electric field time series and obtain the fitting residuals.Based on the residual series,an isolated forest model is constructed to determine the location of abnormal points,and finally corrected by the Chen-Liu algorithm..The reliability of the proposed method is verified through experiments with simulated sequence and measured atmospheric electric field data.Compared with the original sequence prediction,the root mean square error and average percentage error of the atmospheric electric field sequence after cleaning are improved by 27.8% and 34.98%,respectively.(3)The traditional methods of lightning warning ignore the oscillation scale characteristics of atmospheric electric field,which lead to low probability of detection,A lightning warning method based on Ensemble Empirical Mode Decomposition(EEMD)and extreme Gradient Boosting(XGBoost)is proposed.This method uses EEMD to decompose the electric field which observed by the atmospheric electric field mill,then calculates the sample entropy of original data and each Intrinsic Mode Function(IMF),reconstruct them according to random component,detail component and trend component,and extracts the statistical and autoencoder features of reconstructed components respectively,The XGBoost algorithm is used to establish a lightning warning model,and fuse the classifiers of each component.The experimental research is carried out by using the observation data of atmospheric electric field mill and lightning positioning system,and performance of the algorithm is analyzed.Compared with the general voting decision-making method,the probability of detection is increased by 4.8% at the highest,and the false alarm rate is reduced by 5.2% ? 6.4%.(4)Build a high availability and reliability atmospheric electric field monitoring platform based on the big data.Flume is used to achieve real-time electric field data acquisition,Spark is used to achieve offline analysis and modeling,Kafka is used to achieve data caching,and spark-Streaming is used to read and process the data in real time.Through the experimental analysis,the advantages and reliability of the atmospheric electric field monitoring platform based on big data in offline modeling performance and real-time analysis are verified.The analysis of the structural parameters of the electric field calibration device and the experimental conclusions of the influence of buildings on electric field distortion,the detection and correction of atmospheric electric field anomalies,and the lightning warning methods have laid a theoretical foundation for manufacturers to develop better performance lightning warning products based on atmospheric electric fields.