Ionospheric irregularity identification method and system based on machine learning

Abstract

The application discloses a method and system for ionospheric irregularity identification based on machine learning. The method comprises the following steps: receiving a GNSS observation file and an ephemeris file, and preprocessing the GNSS observation file and the ephemeris file; extracting an electron total content rate index feature, a time integral and total electron content index feature, and a Doppler coefficient feature according to the preprocessed GNSS observation file and ephemeris file; identifying ionospheric irregularities according to the extracted electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature, and outputting an identification result. The method can avoid setting a single threshold to determine whether an ionospheric irregularity event occurs during the process of identifying ionospheric irregularities, and effectively avoids data loss caused by cycle slip detection, thereby effectively identifying EPB events and TID events.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of GNSS and ionosphere, and particularly relates to an ionospheric irregularity identification method and system based on machine learning. BACKGROUND

[0002] The ionosphere is a crucial region of the Earth's atmosphere, located at altitudes of approximately 60 to 1000 kilometers. The atmosphere in this region is ionized by solar radiation (mainly ultraviolet and X-rays) and cosmic rays, resulting in a large number of free electrons and positive ions, thus making the atmosphere in this region conductive. The existence of the ionosphere has a significant impact on the propagation of radio waves, as it can refract, reflect, scatter, and absorb radio waves, thereby affecting the performance of radio information systems such as satellite navigation, communication, and radar.

[0003] Equatorial Plasma Bubbles (EPBs) are large-scale ionospheric irregular structures that occur in the magnetic equatorial and low-latitude regions, usually at night. These structures are generally believed to be seeded by disturbances in the bottom of the ionospheric F layer over the magnetic equator, which are then amplified by Rayleigh-Taylor (R-T) instability, and gradually rise to higher altitudes and map along the magnetic field lines to low-latitude regions. The existence of equatorial plasma bubbles not only affects radio communication and navigation systems, but also has important significance for space weather research. They can cause ionospheric scintillation phenomena and affect the stability of communication and navigation systems. Therefore, the study of the occurrence regularity, morphological characteristics, and evolution process of equatorial plasma bubbles has important scientific and application value for understanding and predicting space weather phenomena.

[0004] Traveling Ionospheric Disturbances (TIDs) are periodic electron density fluctuations in the ionosphere, closely related to the motion of acoustic-gravity waves in the upper atmosphere. TIDs affect the electron density distribution and signal propagation characteristics of the ionosphere, and have various impacts on satellite communication, including changes in signal propagation path, Doppler effect, ionospheric second-order effects, and signal interference. These impacts can reduce communication quality, increase signal processing complexity, and in some cases, cause interference to satellite navigation. The study of TIDs is of great significance for understanding the dynamic changes of the ionosphere and space weather phenomena.

[0005] Currently, in the ionospheric irregularity identification method, there are mainly two problems: one is that in the process of detecting equatorial plasma bubble, the disturbance is large and the cycle slip is easy to occur, and if a single threshold is set, it is easy to lead to the difficulty in identifying EPB (equatorial plasma bubble); the other is that it is difficult to identify equatorial plasma bubble and traveling ionospheric disturbance by using ROTI coefficient alone, thereby leading to low identification efficiency of traveling ionospheric disturbance. SUMMARY

[0006] In order to overcome the problems that the existing ionospheric irregularity identification technology cannot identify EPB event due to cycle slip detection, and the distinction between EPB event and TID event is not high enough, the purpose of the present application is to provide an ionospheric irregularity identification method and system based on machine learning, which avoids using a single threshold to determine whether an EPB event occurs, can effectively distinguish EPB event, and improves the accuracy of EPB event and TID event detection.

[0007] In order to achieve the purpose of the present application, the present application adopts the following technical solutions:

[0008] An ionospheric irregularity identification method based on machine learning, the method comprising the following steps:

[0009] Receiving a GNSS observation file and an ephemeris file, and preprocessing the GNSS observation file and the ephemeris file;

[0010] Extracting an electron total content rate index feature, a time integral and total electron content index feature, and a Doppler coefficient feature according to the preprocessed GNSS observation file and ephemeris file;

[0011] Identifying ionospheric irregularities according to the extracted electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature, and outputting the identification result.

[0012] In the above technical solution, by preprocessing the received GNSS observation file and ephemeris file, the reliability and practicability of the data can be improved, and the efficiency of subsequent data processing can be improved; the extracted electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature can avoid using a single threshold to determine whether an EPB event occurs in the process of identifying ionospheric irregularities, and effectively avoid data loss caused by cycle slip detection, thereby effectively identifying EPB event, and improving the accuracy of EPB event and TID event detection.

[0013] Further, the process of preprocessing the GNSS observation file and ephemeris file comprises:

[0014] The received GNSS observation file is subjected to rough error detection and abnormal data elimination processing;

[0015] The received ephemeris file is subjected to data integrity check, data error analysis and data correction processing.

[0016] In the above technical solution, the GNSS observation file is subjected to rough error detection and abnormal data elimination processing, and the ephemeris file is subjected to data integrity check, data error analysis and data correction processing, and finally the GNSS observation file and the ephemeris file are subjected to noise reduction processing, which can improve the reliability and practicality of the data and improve the efficiency of subsequent data processing.

[0017] Further, the process of extracting the total electron content change rate index feature according to the preprocessed GNSS observation file includes:

[0018] Extracting total electron content data according to the preprocessed GNSS observation file;

[0019] Extracting the total electron content change rate feature according to the extracted total electron content data;

[0020] Extracting the total electron content change rate index feature according to the total electron content change rate feature.

[0021] Further, the time derivative between two consecutive time periods is calculated according to the extracted total electron content data to extract the total electron content change rate feature, and the expression is:

[0022]

[0023] The expression for extracting the total electron content change rate index feature according to the total electron content change rate feature is:

[0024]

[0025] Where, t i represents the time stamp, k represents the serial number of the satellite, STEC represents the original calculated total electron content data, and < > represents the average value in a time interval.

[0026] In the above technical solution, the total electron content data is extracted according to the preprocessed GNSS observation file, which can use the total electron content data to monitor the typical indicators of ionospheric irregularities in a short time window, so as to extract the time derivative between two consecutive time periods to extract the total electron content change rate feature.

[0027] Further, the process of extracting the time integral and total electron content index index feature according to the preprocessed GNSS observation file includes:

[0028] Extracting the total electron content without trend from the pretreated GNSS observation file;

[0029] Extracting the integral of the total electron content without trend data in a fixed time window from the total electron content without trend;

[0030] Combining the extracted total electron content rate of change index feature with the integral of the total electron content without trend data to obtain the time integral and total electron content index feature.

[0031] Further, the moving average method is used to extract the total electron content without trend, and the expression is:

[0032]

[0033] The background value of the total electron content data is extracted by using the SG filter, and the expression is:

[0034] DTEC = MF × (STEC - SGTEC)

[0035] Based on the extracted total electron content without trend and the background value of the total electron content data, the integral of the total electron content without trend data in a fixed time window is extracted, and the expression is:

[0036]

[0037] Combining the total electron content rate of change index feature with the integral of the total electron content without trend data to obtain the time integral and total electron content index feature, and the expression is:

[0038] IROTI = IDTEC · ROTI

[0039] Wherein, N represents a natural number, MF represents the satellite elevation angle, STEC represents the original calculated total electron content data, SGTEC represents the background total electron content data obtained after SG filter fitting, |DTEC(i)| represents the absolute value of DTEC data at time point t, Δt represents the time interval of DTEC data, n represents the number of DTEC data in the time window, the unit of IDTEC value is TECU, and the unit of IROTI is generally TECU 2 .

[0040] In the technical solution, based on the background value of the total electron content data after removing the trend extracted, the integral of the total electron content data after removing the trend in a fixed time window is extracted, and then the electron total content change rate index feature is combined with the integral of the total electron content data after removing the trend, to obtain the time integral and total electron content index feature, which can effectively maintain the communication quality during the ionospheric traveling disturbance detection process, reduce the signal processing complexity, reduce the interference to satellite navigation, and improve the accuracy of TID detection.

[0041] Further, the process of extracting the Doppler coefficient feature according to the preprocessed ephemeris file includes:

[0042] The Doppler measurement values at two frequencies are calculated, and the expression is:

[0043]

[0044] According to the Doppler measurement values at two frequencies, the differential frequency AD of the Doppler measurement values is calculated, and the expression is:

[0045]

[0046] Based on the differential frequency AD of the Doppler measurement values, the Doppler coefficient feature of AD in a specific time period is calculated by using a sliding window method, and the expression is:

[0047]

[0048] Where i represents the number of frequencies, D i represents the Doppler measurement value, λ i represents the wavelength, represents the geometric distance rate between the satellite and the receiver, and c represents the speed of light, and represents the clock error of the satellite and the receiver, represents the ionospheric delay rate, represents the tropospheric delay rate, and ε i represents noise.

[0049] Further, the process of identifying ionospheric irregularities according to the extracted electron total content change rate index feature, time integral and total electron content index feature, and Doppler coefficient feature includes:

[0050] A machine learning model is set, and the electron total content change rate index feature, the time integral and total electron content index feature, and the Doppler coefficient feature are used to iteratively optimize and train the machine learning model, and the process is:

[0051] Based on the electronic total content change rate index characteristic, time integral and total electron content index index characteristic and Doppler coefficient characteristic, a decision tree is iteratively constructed, and the expression is:

[0052]

[0053] A target function is constructed by using a regularization method to iteratively optimize the machine learning model, and the expression is:

[0054]

[0055] A loss function is set to optimize the parameters of the machine learning model in the iterative optimization process. When the iteration optimization training round ends or the loss function converges, a trained machine learning model is obtained.

[0056] The trained machine learning model is used to identify ionospheric irregularities in the GNSS observation file and ephemeris file to be identified, and the expression is:

[0057]

[0058] A binary classifier is designed to classify the recognition results and convert the logical function output into a probability, and the expression is:

[0059]

[0060] Among them, the recognition results include equatorial plasmasphere events, ionospheric traveling disturbance events and quiet periods, K t represents the number of leaf nodes in the tth tree, represents the weight of the kth leaf node, represents an indicator function, x represents sample data, represents the area of the kth leaf node in the tth tree n represents the number of samples, y i represents the true label of the ith sample, represents the prediction value of the model for the ith sample, Ω(f t ) represents a regularization term.

[0061] Further, in the process of iteratively optimizing the machine learning model by constructing a target function using a regularization method, the regularization term is used to calculate the sum of squares of the number of leaf nodes and the weight of the leaf nodes of the tree, and the expression is:

[0062]

[0063] where γ and λ are regularization parameters.

[0064] In the technical solution, the threshold is avoided to determine whether the ionospheric irregularity event occurs, the total electron content change rate index feature, the time integral and total electron content index feature and the Doppler coefficient feature are used to perform parameter iterative optimization training on the set machine learning model, so that the machine learning model can effectively master the data change in the GNSS observation file and the ephemeris file, thereby avoiding using a single threshold to determine whether the EPB event occurs in the process of identifying the ionospheric irregularity, and effectively avoiding data loss caused by cycle slip detection, so as to effectively identify the EPB event, and improve the accuracy of EPB event and TID event detection.

[0065] An ionospheric irregularity identification system based on machine learning, the system comprises:

[0066] A data receiving module is configured to receive a GNSS observation file and an ephemeris file.

[0067] A data processing module is configured to preprocess the GNSS observation file and the ephemeris file.

[0068] A feature extraction module is configured to extract a total electron content change rate index feature, a time integral and total electron content index feature and a Doppler coefficient feature from the preprocessed GNSS observation file and ephemeris file.

[0069] An ionospheric irregularity identification module is configured to identify the ionospheric irregularity according to the extracted total electron content change rate index feature, time integral and total electron content index feature and Doppler coefficient feature, and output an identification result.

[0070] Compared with the prior art, the present application has the following advantages:

[0071] The present application provides an ionospheric irregularity identification method and system based on machine learning, which can improve the reliability and practicality of the data by preprocessing the received GNSS observation file and ephemeris file, and improve the efficiency of subsequent data processing. The extracted total electron content change rate index feature, time integral and total electron content index feature and Doppler coefficient feature avoid using a single threshold to determine whether the EPB event occurs, effectively avoid data loss caused by cycle slip detection, effectively identify the EPB event, and improve the accuracy of EPB event and TID event detection. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 A step flowchart of an ionospheric irregularity identification method based on machine learning provided by the present application embodiment;

[0073] Figure 2 A flowchart of machine learning model training provided by the present application embodiment;

[0074] Figure 3 A TEC file and a result predicted based on a machine learning model in an experiment provided for an embodiment of the present application;

[0075] Figure 4 A confusion matrix of a test set in a machine learning model training process in an experiment provided for an embodiment of the present application;

[0076] Figure 5 A TEC sequence and a ROTI sequence in an experiment provided for an embodiment of the present application;

[0077] Figure 6 A prediction sample and a confusion matrix two of a test set in a machine learning model training process in an experiment provided for an embodiment of the present application;

[0078] Figure 7 A structure diagram of a machine learning-based ionospheric irregularity identification system provided for an embodiment of the present application. DETAILED DESCRIPTION

[0079] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The preferred embodiments of the present application are shown in the drawings. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0080] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0081] Embodiment one:

[0082] The present embodiment provides a machine learning-based ionospheric irregularity identification method, referring to Figure 1 , the method comprises the following steps:

[0083] Step S1: The receiver receives a GNSS observation file and an ephemeris file, and pre-processes the GNSS observation file and the ephemeris file;

[0084] Step S2: Extracting the total electron content rate of change index feature, time integral and total electron content index feature, Doppler coefficient feature according to the pre-processed GNSS observation file and ephemeris file;

[0085] Step S3: Ionospheric irregularity identification is performed according to the extracted rate of TEC index, the time integral and rate of the total electron content index, and the Doppler index, and an identification result is output.

[0086] The rate of TEC index is ROTI, the time integral and rate of the total electron content index is IROTI, and the Doppler index is DI.

[0087] In step S1, the process of preprocessing the GNSS observation file and the ephemeris file includes:

[0088] The received GNSS observation file is subjected to gross error detection and abnormal data rejection processing.

[0089] The received ephemeris file is subjected to data integrity checking, data error analysis, and data correction processing.

[0090] It can be understood that the GNSS observation file is subjected to gross error detection and abnormal data rejection processing, and the ephemeris file is subjected to data integrity checking, data error analysis, and data correction processing, and finally the GNSS observation file and the ephemeris file are subjected to noise reduction processing, which can improve the reliability and practicality of the data and improve the efficiency of subsequent data processing.

[0091] In step S2, the process of extracting the rate of TEC index according to the preprocessed GNSS observation file includes:

[0092] The total electron content data (TEC data) is extracted from the preprocessed GNSS observation file.

[0093] The rate of change of the total electron content is extracted from the extracted total electron content data.

[0094] The rate of TEC index is extracted from the rate of change of the total electron content.

[0095] Specifically, the time derivative between two consecutive time periods is calculated according to the extracted total electron content data to extract the rate of change of the total electron content, and the expression is:

[0096]

[0097] The expression for extracting the rate of TEC index from the rate of change of the total electron content is:

[0098]

[0099] where t i denotes the time stamp, k denotes the satellite sequence number, STEC denotes the original calculated electron content data, and < > denotes the average value in a time interval.

[0100] It can be understood that, according to the pre-processed GNSS observation file, the electron content data is extracted, the electron content data can be used to monitor the typical indicators of ionospheric irregularities in a short time window, so as to extract the time derivative between two consecutive time periods, to extract the change rate characteristics of the electron content, and to further improve the accuracy of TID detection in the process of ionospheric traveling disturbance detection.

[0101] In step S2, the process of extracting the time integral and total electron content index characteristics according to the pre-processed GNSS observation file includes:

[0102] According to the pre-processed GNSS observation file, the total electron content data is extracted.

[0103] According to the total electron content data, the integral of the total electron content data in a fixed time window is extracted.

[0104] The extracted electron content change rate index characteristics are combined with the integral of the total electron content data to obtain the time integral and total electron content index characteristics.

[0105] Specifically, the moving average method is used to extract the total electron content data, and the expression is:

[0106]

[0107] The background value of the electron content data is extracted by using the SG filter, and the expression is:

[0108] DTEC = MF × (STEC - SGTEC)

[0109] Based on the extracted total electron content data and the background value of the total electron content data, the integral of the total electron content data in a fixed time window is extracted, and the expression is:

[0110]

[0111] The electron content change rate index characteristics are combined with the integral of the total electron content data to obtain the time integral and total electron content index characteristics, and the expression is:

[0112] IROTI = IDTEC · ROTI

[0113] wherein, N represents a natural number, MF represents a satellite elevation angle, STEC represents the original calculated total electron content data, SGTEC represents the background total electron content data obtained after SG filtering fitting, |DTEC(i)| represents the absolute value of DTEC data at time point t, Δt represents the time interval of DTEC data, n represents the number of DTEC data in the time window, the unit of IDTEC value is TECU, and the unit of IROTI is generally TECU 2 wherein, 1 TECU = 10 16 electrons per square meter.

[0114] It can be understood that, based on the background value of the total electron content data extracted from the removed trend, the integral of the total electron content data removed from the trend in a fixed time window is extracted, and then the electron total content change rate index feature is combined with the integral of the total electron content data removed from the trend, to obtain the time integral and total electron content index feature.

[0115] In step S2, the process of extracting the Doppler coefficient feature according to the preprocessed ephemeris file includes:

[0116] The Doppler measurement values at two frequencies are calculated, and the expression is:

[0117]

[0118] According to the Doppler measurement values at two frequencies, the differential frequency ΔD of the Doppler measurement values is calculated, and the expression is:

[0119]

[0120] Based on the differential frequency ΔD of the Doppler measurement values, the sliding window method is used to calculate the Doppler coefficient feature of ΔD in a specific time period, and the expression is:

[0121]

[0122] wherein, i represents the number of frequencies, D i represents the Doppler measurement value, λ i represents the wavelength, represents the geometric distance rate between the satellite and the receiver, c represents the speed of light, and respectively represent the clock error of the satellite and the receiver, represents the ionospheric delay rate, represents the tropospheric delay rate, ε i represents noise.

[0123] In step S3, referring to Figure 2, avoid using threshold to determine whether ionospheric irregularity event occurs, the process of ionospheric irregularity identification according to the extracted electron total content rate index characteristic, time integral and total electron content index index characteristic and Doppler coefficient characteristic includes:

[0124] A machine learning model is set, and the machine learning model is iteratively optimized and trained by using the electron total content rate index characteristic, the time integral and total electron content index index characteristic and the Doppler coefficient characteristic, and the process is:

[0125] Based on the electron total content rate index characteristic, the time integral and total electron content index index characteristic and the Doppler coefficient characteristic, a decision tree is iteratively constructed, and the expression is:

[0126]

[0127] A regularization method is used to construct a target function to iteratively optimize the machine learning model, and the expression is:

[0128]

[0129] A loss function is set to optimize the parameters of the machine learning model in the iterative optimization process, and when the iteration optimization training round ends or the loss function converges, a trained machine learning model is obtained;

[0130] The trained machine learning model is used to identify ionospheric irregularities in the GNSS observation file and ephemeris file to be identified, and the expression is:

[0131]

[0132] A binary classifier is designed to classify the identification results, and the logical function output is converted into probability, and the expression is:

[0133]

[0134] Among them, the identification results include equatorial plasmasphere event, ionospheric traveling disturbance event and calm period, K t The number of leaf nodes in the tth tree is represented by K The weight of the kth leaf node is represented by W An indicator function is represented by x, and x represents sample data. The area of the kth leaf node in the tth tree is represented by R n represents the number of samples, y i The true label of the ith sample is represented by y The prediction value of the model for the ith sample is represented by f t Ω(f

[0135] As a preferred embodiment, in the process of iteratively optimizing the machine learning model by constructing the objective function with the regularization method, the sum of squares of the number of leaf nodes and the weight of the leaf nodes of the tree is calculated by the regularization term, and the expression is:

[0136]

[0137] Wherein, γ and λ represent regularization parameters.

[0138] In this embodiment, the machine learning model uses an XGBoost (eXtreme Gradient Boosting) classification model, which improves the prediction performance of the model by constructing multiple decision trees.

[0139] It can be understood that the machine learning model is iteratively optimized and trained by using the total electron content rate of change index feature, the time integral and total electron content index feature, and the Doppler coefficient feature, so that the machine learning model can effectively master the data changes in the GNSS observation file and the ephemeris file, thereby avoiding the use of a single threshold to determine whether an EPB event occurs in the process of ionospheric irregularity identification, and effectively avoiding data loss caused by cycle slip detection, thereby effectively identifying EPB events, and improving the accuracy of EPB event and TID event detection.

[0140] In this embodiment, by preprocessing the received GNSS observation file and ephemeris file, the reliability and practicality of the data can be improved, and the efficiency of subsequent data processing can be improved; the total electron content rate of change index feature, the time integral and total electron content index feature, and the Doppler coefficient feature can be extracted, which can avoid the use of a single threshold to determine whether an EPB event occurs in the process of ionospheric irregularity identification, and effectively avoid data loss caused by cycle slip detection, thereby effectively identifying EPB events, and improving the accuracy of EPB event and TID event detection.

[0141] Embodiment Two:

[0142] This embodiment is based on the method described in Embodiment One, and gives relative experimental data as follows:

[0143] Case One:

[0144] The data receiving time is March 3, 2024, and the satellite is R22 satellite (GLONASS).

[0145] From Figure 3From the middle TEC image, it can be seen that a large amount of data from 12:30 to 13:30 was identified as a cycle slip and was excluded, so the data in that period was missing and could not be identified when calculating ROTI. Because multiple coefficients were introduced instead of a single coefficient, the XGBoost model can still identify the EPB event in this period without being affected by the accuracy of the ROTI data.

[0146] From Figure 3 , it can be seen that the identified time period corresponds to the TEC in the above figure, and the occurrence time period is about 20:00-22:00 (night) in local time, which is consistent with the occurrence time of the EPB event. Before and after the data disappears, it can be seen that it is consumed downward, which is consistent with the characteristics of the EPB event. In summary, this is an EPB event, which proves the reliability of the XGBoost identification model.

[0147] From Figure 4 , it can be seen from the confusion matrix that a total of 760 ephemeris data were identified, of which 565 ephemeris data were determined to be non-EPB events (quiet period), and 195 ephemeris data were EPB events. This is also consistent with the actual situation.

[0148] Case two:

[0149] The data receiving time is July 17, 2014, and the satellite is G05 (GPS).

[0150] From the TEC sequence in Figure 5 , it can be seen that a fluctuation occurred at about UTC 14:00, which is consistent with the performance of TID in the TEC sequence. From the occurrence time period and the occurrence time, it is consistent with the characteristics of MSTID (mesoscale traveling ionospheric disturbance). However, if only the ROTI sequence is used as a judgment, it is difficult to set a good threshold to distinguish TID, because the ROTI value of the quiet period may also reach about 0.2. Therefore, IROTI index is used in the XGBoost model, so that the identification model has the ability to distinguish TID.

[0151] Referring to Figure 6 , from the current identification effect, the XGBoost identification model of the present application has strong identification ability for equatorial plasma bubble (EPB) events, and can also well identify traveling ionospheric disturbance (TID) events.

[0152] Example three:

[0153] The embodiment provides an ionospheric irregularity identification system based on machine learning, referring to Figure 7 , the system comprises:

[0154] a data receiving module configured to receive a GNSS observation file and an ephemeris file;

[0155] a data processing module configured to preprocess the GNSS observation file and the ephemeris file;

[0156] a feature extraction module configured to extract an electron total content rate of change index feature, a time integral and total electron content index feature, and a Doppler coefficient feature according to the preprocessed GNSS observation file and the ephemeris file;

[0157] an ionospheric irregularity identification module configured to identify ionospheric irregularities according to the extracted electron total content rate of change index feature, the time integral and total electron content index feature, and the Doppler coefficient feature, and output an identification result.

[0158] The above merely describes the embodiments of the present application, and does not limit the patent scope of the present application, and any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, which are made according to the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A method for ionospheric irregularity identification based on machine learning, characterized by, The method comprises the following steps: Receiving a GNSS observation file and an ephemeris file, and preprocessing the GNSS observation file and the ephemeris file; Extracting an electron total content rate index feature, a time integral and total electron content index feature, and a Doppler coefficient feature from the preprocessed GNSS observation file and ephemeris file; Identifying an ionospheric irregularity according to the extracted electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature, and outputting an identification result; Setting up a machine learning model, and iteratively optimizing and training the machine learning model by using the electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature. 2.The machine learning based ionospheric irregularity identification method of claim 1, wherein, The preprocessing process of the GNSS observation file and the ephemeris file comprises: Performing rough error detection and abnormal data rejection processing on the received GNSS observation file; Performing data integrity checking, data error analysis, and data correction processing on the received ephemeris file. 3.The machine learning based ionospheric irregularity identification method of claim 1, wherein, The process of extracting the electron total content rate index feature from the preprocessed GNSS observation file comprises: Extracting electron total content data from the preprocessed GNSS observation file; Extracting an electron total content rate feature from the extracted electron total content data; Extracting an electron total content rate index feature from the electron total content rate feature. 4.The machine learning based ionospheric irregularity identification method of claim 3, wherein, The time derivative between two consecutive time periods is calculated according to the extracted electron total content data to extract the electron total content rate feature, and the expression is: The expression for extracting the electron total content rate index feature from the electron total content rate feature is: wherein, represents a time stamp, represents a serial number of the satellite, represents the originally calculated total electron content data, and < > represents an average value over a time interval. 5.The machine learning based ionospheric irregularity identification method of claim 4, wherein, The process of extracting the time integral and total electron content index feature from the preprocessed GNSS observation file comprises: Extracting trend-removed total electron content from the preprocessed GNSS observation file; Extracting the integral of trend-removed total electron content data in a fixed time window according to the trend-removed total electron content; Combining the extracted electron total content rate index feature and the integral of trend-removed total electron content data to obtain the time integral and total electron content index feature. 6.The machine learning based ionospheric irregularity identification method of claim 5, wherein, The trend-removed total electron content is extracted by using the moving average method, and the expression is: The background value of the electron total content data is extracted by using the SG filter, and the expression is: Based on the extracted trend-removed total electron content and the background value of the electron total content data, the integral of trend-removed total electron content data in a fixed time window is extracted, and the expression is: The electron total content rate index feature and the integral of trend-removed total electron content data are combined to obtain the time integral and total electron content index feature, and the expression is: Wherein, N represents a natural number, MF represents a satellite elevation angle, STEC represents original calculated total electron content data, SGTEC represents background total electron content data obtained after SG filtering fitting, represents the absolute value of the DTEC data at the time point t, represents the time interval of the DTEC data, n represents the number of DTEC data in the time window, the unit of the IDTEC value is TECU, and the unit of the IROTI represents . 7.The machine learning based ionospheric irregularity identification method of claim 1, wherein, The process of extracting the Doppler coefficient feature from the preprocessed ephemeris file comprises: Calculating the Doppler measurement value at two frequencies, and the expression is: According to the Doppler measurements at the two frequencies, a differential frequency of the Doppler measurements is calculated with the expression: Differential frequency based on doppler measurements Doppler coefficient features are calculated using a sliding window approach with the expression: where i denotes the number of the frequency, denotes the Doppler measurement, denotes the wavelength, denotes the geometric range rate between satellite and receiver, denotes the speed of light, and denote the clock errors of the satellite and the receiver, respectively, denotes the ionospheric delay rate, denotes the tropospheric delay rate, denotes the noise. 8.The machine learning based ionospheric irregularity identification method of claim 7, wherein, The process of identifying an ionospheric irregularity according to the extracted electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature comprises: Iteratively constructing a decision tree based on the electron total content rate index feature, time integral and total electron content index feature, and Doppler coefficient feature, and the expression is: The regularization method is used to construct a target function for iterative optimization of the machine learning model, and the expression is as follows: The loss function is set to optimize the parameters of the machine learning model in the iterative optimization process, and when the iteration optimization training round ends or the loss function converges, a trained machine learning model is obtained; The trained machine learning model is used to identify ionospheric irregularities in the GNSS observation file and ephemeris file to be identified, and the expression is as follows: A binary classifier is designed to classify the identification results and convert the logical function output into a probability, and the expression is as follows: wherein the recognition result includes equatorial plasma bubble events, ionospheric travelling disturbance events and quiet periods, represents the number of leaf nodes in the tree, represents the number of leaf nodes in the tree, represents the weight of the i-th leaf node, represents an indicator function, represents sample data, represents the number of leaf nodes in the tree, represents the number of leaf nodes in the tree, represents the area of the i-th leaf node in the tree, represents the area of the i-th leaf node in the tree, , represents the number of samples, represents the true label of the i-th sample, represents the prediction value of the model for the i-th sample, represents the prediction value of the model for the i-th sample, represents the prediction value of the model for the i-th sample, represents the regularization term. 9.The machine learning based ionospheric irregularity identification method of claim 8, wherein, In the process of using the regularization method to construct a target function for iterative optimization of the machine learning model, the regularization term is used to calculate the sum of squares of the number of leaf nodes and the weights of the leaf nodes of the tree, and the expression is as follows: wherein and both represent a regularization parameter.

10. A machine learning based ionospheric irregularity identification system, the system based on the method of any of claims 1-9, characterized in that, The system comprises: A data receiving module is configured to receive GNSS observation files and ephemeris files. A data processing module is configured to preprocess the GNSS observation files and ephemeris files. A feature extraction module is configured to extract the total electron content rate of change index feature, time integral, and total electron content index feature from the preprocessed GNSS observation files, and extract the Doppler coefficient feature from the preprocessed ephemeris files. An ionospheric irregularity identification module is configured to identify ionospheric irregularities according to the extracted total electron content rate of change index feature, time integral, and total electron content index feature, and the Doppler coefficient feature, and output the identification results.

Images