Flight path data anomaly analysis method and monitoring source evaluation method

Abstract

The invention discloses a track data anomaly analysis method and a monitoring source evaluation method. The track data anomaly analysis method comprises the following steps: acquiring historical track data; inputting historical track data into the trained LSTM prediction network model to obtain predicted track data; acquiring real track data; calculating a vector difference value between the real track data and the predicted track data, judging whether the conformity of the vector difference value is greater than a set threshold value or not, and if the conformity of the vector difference value is less than or equal to the set threshold value, judging that the real track data is abnormal. According to the monitoring source evaluation method, the monitoring source track credibility evaluation model is constructed based on the real track data and the predicted track data, multiple factors such as height difference and track vector difference are comprehensively considered, and the reliability of the monitoring source is flexibly judged according to different conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of track detection, and particularly relates to a track data anomaly analysis method and a surveillance source evaluation method. BACKGROUND

[0002] With the rapid development of surveillance source technology and computer storage and computing capacity, track data such as radar track, ADS-B track and automatic integrated track are growing explosively. These track data contain rich and valuable target activity information. Mining and extracting valuable information from the data and applying the information to realize digital processing, intelligent response and intelligent support are a new chapter in the development of intelligent air traffic control.

[0003] However, the existing track detection technology mainly depends on the quality indicators of surveillance data issued by surveillance sources, such as the quality factor in ADS-B data and the north sector information in radar data. Surveillance data that does not meet these quality indicators can be easily identified and discarded by the system. However, data that meets the quality indicators but is not accurate in position often has a significant impact. Therefore, a new track detection technology is needed to more accurately process such data. SUMMARY

[0004] To solve the problems in the prior art, the present application provides a track data anomaly analysis method and a surveillance source evaluation method.

[0005] The technical scheme adopted by the present application is as follows:

[0006] The first aspect of the present application provides a track data anomaly analysis method, comprising the following contents:

[0007] acquiring historical track data;

[0008] inputting the historical track data into a trained LSTM prediction network model to obtain predicted track data;

[0009] acquiring real track data;

[0010] calculating the vector difference between the real track data and the predicted track data, and determining whether the degree of coincidence of the vector difference is greater than a set threshold value, if the degree of coincidence of the vector difference is less than or equal to the set threshold value, determining that the real track data is abnormal.

[0011] Preferably, the training process of the LSTM prediction network model comprises:

[0012] acquiring track sample data;

[0013] preprocessing the track sample data to generate second track sample data;

[0014] feature extraction is performed on the second track sample data to generate track sample feature data;

[0015] The track sample feature data is divided into a training data set and a validation data set;

[0016] The training data set is input into a neural network model for training to obtain a trained LSTM network model;

[0017] The validation data is input into the trained LSTM network model to obtain feature data of a track at a next time;

[0018] The feature data of the track at the next time is compared with real track feature data to obtain a comparison result;

[0019] The trained LSTM network model is adjusted based on the comparison result to obtain an LSTM prediction network model.

[0020] Preferably, the historical track data is input into the trained LSTM prediction network model to obtain predicted track data, including the following formula:

[0021]

[0022] wherein P n is an arbitrary track point weight; x n , y n , and z n are current position coordinates; x n-1 , y n-1 , and z n-1 are previous position coordinates.

[0023] When L≦|W(P n -P n-1 )| is satisfied, the output data is track feature data.

[0024] wherein L is a monitoring source constant, and W is a monitoring source weight parameter.

[0025] Preferably, the track sample data is preprocessed to generate second track sample data, including:

[0026] The track sample data is sequentially subjected to data cleaning, data conversion, data difference, and data completion to generate second track sample data.

[0027] Preferably, the feature extraction is performed on the second track sample data to generate track sample feature data, including:

[0028] Sample identifiers, track time sequence features, sample position features, and sample height features are extracted from the second track sample data to generate sample feature data.

[0029] The sample feature data is classified, deduplicated, and error-removed to generate track sample feature data.

[0030] Preferably, the formula for calculating the vector difference between the actual trajectory data and the predicted trajectory data is as follows:

[0031]

[0032] Where μ t x represents the vector difference between the actual trajectory data and the predicted trajectory data, where x, y, and z are the coordinates of the actual trajectory data. t y t z t Predict the coordinates of the trajectory data.

[0033] A second aspect of this application provides a method for evaluating surveillance sources, including the following:

[0034] A credibility assessment model for surveillance source tracks is established based on the real track data and predicted track data.

[0035] The reliability of the surveillance source is determined based on the surveillance source track reliability assessment model.

[0036] Preferably, the surveillance source track reliability assessment model includes:

[0037] H≤|h-h1|;

[0038] Where H is the altitude difference constant, h is the crossing altitude corresponding to the planned crossing time, and h1 is the aircraft altitude data;

[0039]

[0040] Where μ t x represents the vector difference between the actual trajectory data and the predicted trajectory data, where x, y, and z are the coordinates of the actual trajectory data. t y t z t Predict the coordinates of the trajectory data.

[0041]

[0042] Where N is a weighting constant, 2∑ t The first variance;

[0043]

[0044] Where M is the weighting constant of the planned trajectory, 2∑ p This represents the second variance.

[0045] Preferably, determining the reliability of a surveillance source based on the surveillance source track reliability assessment model includes the following:

[0046] When H ≤ |h-h1| and At that time, the monitoring source is determined to be reliable;

[0047] When H≤|h-h1| is not satisfied, it is necessary to calculate the vector difference μ between the actual trajectory data and the predicted trajectory data within a preset time period before the current time. t , and μ t The monitoring source is deemed reliable when the compliance rate exceeds the set threshold.

[0048] When not satisfied At this time, it is necessary to calculate the vector difference μ between the planned trajectory data and the actual trajectory data. p When H ≤ |h - h1| and satisfies At that time, the monitoring source was determined to be reliable.

[0049] The beneficial effects of this invention are: the track data anomaly analysis method of this application breaks through the limitations of traditional methods that rely on monitoring source quality indicators, and can identify data with qualified quality indicators but inaccurate positions. By calculating the vector difference between the actual track and the predicted track and comparing it with a set threshold, it is possible to determine whether the actual track data is abnormal, thereby improving the accuracy of track detection.

[0050] The surveillance source assessment method in this application constructs a surveillance source track reliability assessment model based on real track data and predicted track data. It comprehensively considers multiple factors such as altitude difference and track vector difference, and flexibly determines the reliability of the surveillance source according to different conditions. It determines reliability when specific conditions are met; when some conditions are not met, it further assesses reliability by calculating the vector difference over a preset time period or the vector difference between the planned track and the actual track. This multi-method assessment mechanism ensures the reliability of the assessment results. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the method flow of Embodiment 1 of this application;

[0052] Figure 2 This is a schematic diagram illustrating the effect of the predicted trajectory in Embodiment 2 of this application;

[0053] Figure 3 This is a schematic diagram showing the comparison between monitoring source data and predicted trajectory in Embodiment 2 of this application;

[0054] Figure 4 This is a schematic diagram showing the difference between automated monitoring data and predicted data in Embodiment 2 of this application. Detailed Implementation

[0055] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0056] This application utilizes existing track data parsing tools to analyze radar data, ADS-B data, and integrated track data accessed by the automated system. It fuses and processes this low-level air traffic control data to obtain a preliminary sample set. The sample set is then filtered, and feature values ​​are extracted to form feature samples. An LSTM model is used for algorithm training to develop a prediction algorithm. Existing track detection technologies primarily rely on the quality indicators of surveillance data from the monitoring source (e.g., quality factor in ADS-B data, north sector information in radar data). Monitoring data with substandard quality indicators is easily identified and discarded by the system. Therefore, data with acceptable quality indicators but inaccurate location often has a significant impact. This application can bypass the limitations of the quality factor in the monitoring source data itself and perform detection and analysis entirely based on the track location itself. The prediction algorithm detects track anomalies, thereby guiding actual monitoring source interference troubleshooting, monitoring source blind spot correction, and station site selection, improving the automation and intelligence level of monitoring trajectory data analysis.

[0057] Example 1:

[0058] like Figure 1 As shown, a method for anomaly analysis of flight track data includes the following steps:

[0059] S1, acquire historical flight track data.

[0060] Existing civil aviation system flight track data mainly includes the following categories:

[0061] Radar tracks. Radar tracks can be divided into primary air traffic control radar data and secondary air traffic control radar data. Radar track information mainly includes the aircraft's azimuth, range, speed, barometric altitude, secondary code (identification code), and extended S-mode information. In addition to data information, radar track information also includes quality indicators such as: north information loss rate, sector information loss rate, etc.

[0062] ADS-B tracks. ADS-B track information mainly includes aircraft identification information (ID), position, altitude, speed, direction, and rate of climb. ADS-B track information also includes quality factor information such as NICP, NUC, and SIC, which vary slightly depending on the data version.

[0063] Integrated track data. Integrated track data is mainly composed of information from various surveillance sources, including but not limited to radar track information and ADS-B track information, which are weighted and processed, and output by the air traffic control automation system to integrate track information from various surveillance sources.

[0064] Based on a clear understanding of the existing surveillance trajectory data categories used, this application primarily focuses on the analysis and research of radar tracks, ADSB, and integrated track data. The raw track data is cleaned, removing dirty and problematic data, and a data feature sample pool is established to ensure the effectiveness of the training data.

[0065] S2, input the historical trajectory data into the trained LSTM prediction network model to obtain predicted trajectory data.

[0066] The formulas used to obtain the predicted trajectory data include the following:

[0067]

[0068] Among them, P n x represents the weight of any waypoint; n y n , z n x represents the current position coordinates; n-1 y n-1 , z n-1 The coordinates of the previous position;

[0069] When L≦|W(P) is satisfied n -P n-1 The output data is track feature data;

[0070] Where L is the surveillance source constant and W is the surveillance source weight parameter, where W = 1.2 for secondary radar surveillance source and W = 1.1 for ADS-B surveillance source.

[0071] The training process of the LSTM prediction network model includes:

[0072] S2.1, Obtain track sample data.

[0073] It should be noted that the track sample data needs to be collected from relevant air traffic control units, mainly including radar data, ADSB data, airspace structure data, etc.

[0074] In this embodiment, the system track data from the European TRAFFIC CONTROLE CAT062 standard is used as sample data (partial). See Table 1 below:

[0075]

[0076] Table 1S2.2 shows the preprocessing of the track sample data to generate second track sample data.

[0077] Specifically, the trajectory sample data is sequentially cleaned, transformed, and interpolated and completed to generate the second trajectory sample data.

[0078] For example, the original acquired track sample data undergoes a comprehensive inspection to remove dirty and problematic data, such as incomplete data records, obviously erroneous or abnormal data points, ensuring data quality and reliability. The cleaned data is then formatted to standardize its representation, ensuring compatibility and matching between data from different sources or in different formats during subsequent processing. For missing values ​​in the data, appropriate interpolation methods are used to complete them, ensuring data continuity and integrity. After processing, second track sample data is generated.

[0079] S2..3, Perform feature extraction on the second track sample data to generate track sample feature data.

[0080] Representative and discriminative features are extracted from the second track sample data, specifically including sample identifiers, track temporal features, sample location features, and sample altitude features. Sample identifiers uniquely identify each sample, facilitating data location and management; track temporal features reflect the time sequence and relative positional relationships of the data; sample location features, composed of longitude and latitude data, determine the geographical location of the track; and sample altitude features reflect the sample's true altitude information. The extracted sample feature data undergoes classification, deduplication, and error correction. Classification groups the data according to certain rules for easier subsequent analysis and processing; deduplication removes duplicate data records to avoid data redundancy; and error correction corrects errors in the data to improve accuracy. After processing, track sample feature data is generated.

[0081] For example, in this data processing phase, considering issues such as training load, the obtained raw track data is preprocessed and features are extracted. Four types of features are extracted from the obtained raw data as the primary features of the samples.

[0082] Sample identifier. Primarily used for extracting samples and locating their positions.

[0083] Temporal characteristics of flight paths. This mainly obtains the relative positional characteristics of the data and the positional relationship between samples.

[0084] Sample location characteristics. This mainly includes two types of data: the longitude and latitude of the sample data.

[0085] Sample height characteristics. Primarily reflects the true height data of the sample data.

[0086] After confirming the sample characteristics, the data was classified, deduplicated, and error-corrected, resulting in the sample data pool shown in Table 2.

[0087] No. TRACK NUMBER LON [deg] LAT [deg] Geometric Altitude [feet] 351 458 101.4212322 37.28425562 35062.5 1942 458 101.4340425 37.27878392 35081.25 3486 458 101.4470458 37.27319956 35075 5080 458 101.4599741 37.26774931 35118.75 6681 458 101.473074 37.26211131 35081.25 8247 458 101.4863455 37.25645185 35012.5

[0088] Table 2

[0089] S2.4, the trajectory sample feature data is divided into a training dataset and a validation dataset.

[0090] The trajectory sample feature data is divided into training and validation datasets according to a certain ratio (such as the common 7:3 or 8:2). The training dataset is used to train the neural network model, allowing the model to learn the patterns and rules in the data; the validation dataset is used to evaluate the model's performance during the training process and adjust the model's parameters to prevent overfitting.

[0091] S2.5 Input the training dataset into the neural network model for training to obtain the trained LSTM network model.

[0092] For example, the pre-defined training dataset is input into a pre-built neural network model, and a suitable training algorithm (such as stochastic gradient descent) is used to train the model. During training, the model continuously adjusts its parameters to minimize the error between the predicted and actual results. After multiple iterations of training, the trained LSTM network model is obtained.

[0093] S2.6, the validation data and the input to the trained LSTM network model are used to obtain the feature data of the trajectory at the next time step.

[0094] For example, the validation dataset is input into a trained LSTM network model, which then makes predictions based on the input data and outputs feature data of the trajectory at the next time step. These predictions will be used for subsequent comparison with the actual trajectory feature data.

[0095] S2.7, compare the feature data of the trajectory at the next moment with the feature data of the actual trajectory to obtain the comparison result.

[0096] For example, the feature data of the trajectory predicted by the model at the next time step is compared in detail with the actual feature data of the trajectory. Various comparison methods can be used, such as calculating the error between the two (e.g., mean squared error, mean absolute error, etc.), to evaluate the model's prediction accuracy. Through comparison, a result reflecting the model's prediction performance is obtained.

[0097] S2.8, Based on the comparison results, the parameters of the trained LSTM network model are tuned to obtain the LSTM prediction network model.

[0098] For example, based on the comparison results, analyze the problems and shortcomings of the model. If the model's prediction error is large, it indicates that the model's parameters may need further adjustment. Optimize the model's parameters using appropriate parameter tuning methods (such as grid search, random search, etc.), such as adjusting the number of neurons in the LSTM layer and the learning rate. After multiple parameter tuning and validations, until the model's performance reaches a satisfactory level, the final LSTM prediction network model is obtained.

[0099] It should be noted that currently, flight track data mainly comes from various surveillance sources, such as automated systems, radar, ADS-B, multi-point positioning, etc. From a practical application perspective, the main research objective for establishing the algorithm model is to collect comprehensive track information output from air traffic management automation systems. This comprehensive track information mainly includes basic information such as flight track identifier data, sector information, status information, longitude, latitude, and altitude. Based on the established sample pool and features, sample track data is input into a neural network system for training. After training, test samples are input into the neural network system, and the feature data of the track at the next moment is used as the system's output. By comparing this output with the actual values, the relationship between the actual and predicted track data is established, thus achieving both future track prediction and verification of the actual track.

[0100] The feature Y(t) of a certain trajectory at time t can be expressed as Y(t) = {lng,lat,h}, where lng,lat,h represent the longitude, latitude, and altitude of the trajectory at time t, respectively. When predicting the trajectory, the feature data of n consecutive future flight trajectories Y(t-n+1), ..., Y(t-1), and Y(t) are used as network inputs, and the data Y(t+1) at time t+1 is used as the output. From this, a prediction data expression can be established:

[0101] Y(t+1)=f({Y(t-n+1),…,,Y(t-1),Y(t)}

[0102] S3, acquire real flight path data.

[0103] S4, calculate the vector difference between the actual trajectory data and the predicted trajectory data, and determine whether the conformity of the vector difference is greater than a set threshold. If the conformity of the vector difference is less than or equal to the set threshold, then the actual trajectory data is determined to be abnormal.

[0104] The formula for calculating the vector difference between the actual trajectory data and the predicted trajectory data is as follows:

[0105]

[0106] Where μ tx represents the vector difference between the actual trajectory data and the predicted trajectory data, where x, y, and z are the coordinates of the actual trajectory data. t y t z t Predict the coordinates of the trajectory data.

[0107] It's important to note that the conformity of vector differences is determined by the distribution of vector differences over a period of time. For example, by statistically analyzing vector differences at multiple moments and calculating the frequency of these differences within a certain range, this frequency can be used as the conformity. Suppose that vector differences are statistically analyzed at 100 moments over a period of time, and the vector differences at 80 of those moments fall within a specific range, then the conformity is 80%. When the conformity is greater than a set threshold, it indicates that the track data is normal; when the conformity is less than or equal to the set threshold, the actual track data is considered abnormal.

[0108] Example 2 provides a monitoring source evaluation method, including the following:

[0109] A credibility assessment model for surveillance source tracks is established based on the real track data and predicted track data.

[0110] The reliability of the surveillance source is determined based on the surveillance source track reliability assessment model.

[0111] Specifically, the credibility assessment model for surveillance source tracks includes:

[0112] Based on the system's flight plan data, and according to the planned transit times and corresponding altitudes, using the planned waypoints as detection points, the aircraft altitude data at the corresponding time points is detected. The difference between the planned transit times and the aircraft altitude data is calculated, using the following formula:

[0113] H≤|h-h1|;

[0114] Where H is the altitude difference constant, h is the altitude corresponding to the planned time of passing the point, and h1 is the aircraft altitude data.

[0115] Calculate the vector difference between the actual trajectory data and the predicted trajectory data; the calculation formula is as follows:

[0116]

[0117] Where μ t x represents the vector difference between the actual trajectory data and the predicted trajectory data, where x, y, and z are the coordinates of the actual trajectory data. t y t z t Predict the coordinates of the trajectory data;

[0118]

[0119] Where N is a weighting constant, 2∑ t This represents the first variance.

[0120] Calculate the vector difference μ between the planned trajectory data and the actual trajectory data. p :

[0121]

[0122] Where M is the weighting constant of the planned trajectory, 2∑ p This represents the second variance.

[0123] Specifically, determining whether a surveillance source is reliable based on the surveillance source track reliability assessment model includes the following:

[0124] When H ≤ |h-h1| and At that time, the monitoring source is determined to be reliable.

[0125] When H≤|h-h1| is not satisfied, it is necessary to calculate the vector difference μ between the actual trajectory data and the predicted trajectory data within a preset time period before the current time. t , and μ t The monitoring source is deemed reliable when the compliance exceeds a set threshold; for example, if H≤|h-h1| is not satisfied, the vector difference μ between the actual trajectory data and the predicted trajectory data for the previous 60 seconds needs to be calculated. t , and μ t The monitoring source is considered reliable when the compliance exceeds the set threshold.

[0126] When not satisfied At this time, it is necessary to calculate the vector difference μ between the planned trajectory data and the actual trajectory data. p When H ≤ |h - h1| and satisfies At that time, the monitoring source was determined to be reliable.

[0127] It should be noted that the surveillance source track reliability assessment model is an evaluation system for the quality of surveillance sources, derived through comparative analysis of actual field data, supported by the track prediction model. Three evaluation systems will be established for different application scenarios.

[0128] Single surveillance source trajectory evaluation system. In areas covered by only a single surveillance source, an algorithm model is built by collecting its own trajectory data, and self-verification is performed to determine the stability of the surveillance source.

[0129] For example, before a monitoring source is put into operation, the real data from the monitoring source can be combined with this algorithm for self-testing. According to quality standards, when the error rate of the monitoring source's output data is... A time percentage less than 20% indicates that the monitoring source data is generally reliable. Where p a p represents the probability of an anomaly.b To meet the requirements of track point count, Q represents the total number of detected tracks.

[0130] A multi-source surveillance evaluation system is implemented. In areas covered by multiple surveillance sources, an algorithmic model is built by collecting multi-dimensional surveillance source information. The system's correction weights are then evaluated using actual data, predicted data, and relevant flight plan data. By incorporating predicted flight paths into the correction weights, the reliability of each surveillance source within a given area is assessed.

[0131] For example, in an environment with coverage from multiple surveillance sources, the accuracy probability p of the predicted trajectory can be calculated separately by combining the actual data from each surveillance source with this algorithm. n Using a reliable monitoring source as a reference baseline probability p q Compare the detection accuracy Wp of various monitoring sources n, Where W is the monitoring source weight parameter (constant), A value less than 10% (including time) indicates that the data from this monitoring source is generally reliable. Its p t This represents the anomaly rate.

[0132] Evaluation framework for integrated flight path display systems. In display systems (air traffic control automation systems) that provide integrated flight path services for air traffic control, the deviations between the flight paths displayed by various air traffic control automation systems and the actual flight path data are identified based on the established algorithm model predicting the trajectory, thus providing a reference for adjusting the parameters of the air traffic control automation system.

[0133] In the evaluation of integrated display systems, this algorithm can be used to calculate the probability p of the output track conformity of each integrated display system. n Using a reliable monitoring source as a reference baseline probability p q, , According to p t Determine the system contrast.

[0134] The establishment of a trajectory analysis system is the foundation for theoretical data to guide practical work. By using the obtained prediction algorithms, we can discover and summarize the patterns of abnormal data occurrence and the significance of abnormal indicators, thereby guiding practical work such as interference investigation of surveillance sources, blind spot correction of surveillance sources, and station site selection.

[0135] Existing automated systems primarily derive their predicted trajectories from flight plans, while actual trajectory prediction directly stems from trajectory data. Therefore, predicted trajectory data is of greater research significance. Algorithms trained using sample data demonstrate that... Figure 1 The accuracy of the predicted trajectory with the actual trajectory reaches over 85%, which can be used as a basis for theoretical analysis of abnormal flight path situations.

[0136] By comparing with the predicted trajectory, it is possible to Figure 2It can be seen that ADS-B data is more stable and smoother than radar data in this area. The accuracy of the flight path can be improved by adjusting the weight parameters of the automated surveillance data or by correcting the radar data, depending on the actual situation. It can also be used as a theoretical tool for investigating interference from surveillance sources and selecting blind spot stations.

[0137] By comparing with the predicted trajectory, it is possible to Figure 3 It can be seen that there is a difference in the flight path between Automation A and Automation B in this area. The difference in flight path between automation can be reduced by adjusting the parameters of the automation monitoring data according to the actual situation.

[0138] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for analyzing track data anomalies, the method comprising: The method comprises the following steps: Obtaining historical track data; Inputting the historical track data into a trained LSTM prediction network model to obtain predicted track data; Obtaining real track data; Calculating the vector difference between the real track data and the predicted track data, and determining whether the degree of coincidence of the vector difference is greater than a set threshold value, if the degree of coincidence of the vector difference is less than or equal to the set threshold value, it is determined that the real track data is abnormal.

2. The track data anomaly analysis method according to claim 1, characterized in that: The training process of the LSTM prediction network model comprises: Obtaining track sample data; Preprocessing the track sample data to generate second track sample data; Extracting features from the second track sample data to generate track sample feature data; Dividing the track sample feature data into a training data set and a validation data set; Inputting the training data set into a neural network model for training to obtain a trained LSTM network model; Inputting the validation data into the trained LSTM network model to obtain feature data of the track at the next moment; Comparing the feature data of the track at the next moment with real track feature data to obtain a comparison result; Based on the comparison result, the trained LSTM network model is adjusted to obtain an LSTM prediction network model.

3. The track data anomaly analysis method according to claim 1, characterized in that: Inputting the historical track data into the trained LSTM prediction network model to obtain predicted track data comprises the following formula: wherein P n is an arbitrary waypoint weight; x n , y n , z n are current position coordinates; x n-1 , y n-1 , z n-1 are previous position coordinates; When L≦|W(P n -P n-1 )| is satisfied, the output data is track feature data; Where L is a monitoring source constant, and W is a monitoring source weight parameter.

4. The track data anomaly analysis method according to claim 2, characterized in that: The preprocessing of the track sample data to generate the second track sample data comprises: Sequentially performing data cleaning, data conversion, data difference and completion on the track sample data to generate the second track sample data.

5. The track data anomaly analysis method according to claim 2, characterized in that: The feature extraction of the second track sample data to generate the track sample feature data comprises: Extracting sample identifiers, track time sequence features, sample position features and sample height features from the second track sample data to generate sample feature data; Classifying, deduplicating and correcting the sample feature data to generate track sample feature data.

6. The track data anomaly analysis method according to claim 1, characterized in that: The formula for calculating the vector difference between the real track data and the predicted track data is: where μ t is the vector difference between the true track data and the predicted track data, x, y, z are the coordinates of the true track data, and x t , y t , z t are the coordinates of the predicted track data.

7. A method of monitoring source evaluation, characterized by: The method comprises the following steps: Based on the real track data and the predicted track data, a monitoring source track credibility evaluation model is established; Based on the monitoring source track credibility evaluation model, it is determined whether the monitoring source is reliable.

8. The method of monitoring source evaluation of claim 7, wherein: The monitoring source track credibility evaluation model comprises: H≤|h-h1|; Where H is a height difference constant, h is the overpass height corresponding to the planned overpass time, and h1 is the aircraft height data; where μ t is the vector difference between the true track data and the predicted track data, x, y, z are the coordinates of the true track data, x t , y t , z t are the coordinates of the predicted track data. where N is a weighting constant, 2∑ t is the first variance; where M is a planned track weighting constant, 2∑ p is the second variance.

9. The method of monitoring source evaluation of claim 8, wherein: Based on the monitoring source track credibility evaluation model, it is determined whether the monitoring source is reliable, which comprises the following steps: When H ≤ |h - h1| and the monitoring source is determined to be reliable. When H≤|h-h1| is not satisfied, the vector difference μ between the real track data and the predicted track data within the preset time before the current time is calculated t , and when the conformity of μ t exceeds the set threshold, it is determined that the monitoring source is reliable. When H > | h - hi | is not satisfied , the vector difference μ between the planned track data and the real track data needs to be calculated p When H ≤ | h - hi | is satisfied and , it is determined that the monitoring source is reliable.

Images