An airport automatic observation data anomaly detection method

Abstract

This application provides a method for detecting anomalies in automatic airport observation data. The method includes: acquiring real-time meteorological observation data of an airport runway; wherein the real-time meteorological observation data includes real-time observation values ​​of multiple meteorological variables collected by multiple meteorological sensors at different physical locations; inputting the real-time meteorological observation data into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network; calculating reconstruction residual data based on the real-time meteorological observation data and the meteorological reconstruction data; using a pre-constructed independent physical knowledge base, applying physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data; and determining whether there are any anomalous sensors among the multiple sensors at different physical locations based on the reconstruction residual data and the physical residual data.

Description

Technical Field

[0001] This application relates to the field of aviation meteorological observation quality control technology, and more specifically, to a method for detecting anomalies in automatic airport observation data. Background Technology

[0002] The quality of airport aviation meteorological observation data directly affects flight safety and operational efficiency. Failure to promptly identify anomalies in sensor data can lead to erroneous weather reports, thereby impacting aircraft takeoff and landing decisions. Therefore, efficient and accurate anomaly detection of multi-source meteorological sensor data in airport areas is a crucial technical aspect of aviation meteorological operations.

[0003] Existing sensor anomaly detection methods are based on statistical models or machine learning approaches, such as neural network models like autoencoders, and have been gradually applied to meteorological data anomaly detection. However, this purely data-driven approach is prone to learning spurious variable associations caused by sampling bias. For example, when using meteorological variables from a specific season for model training, the visibility, temperature, and humidity of airport runways vary across seasons, leading to false alarms or missed alarms in the trained anomaly detection method. Summary of the Invention

[0004] The purpose of this application is to provide a method for detecting anomalies in airport automatic observation data, in order to solve the problem that the current method for detecting anomalies in airport automatic observation data using a pure data-driven learning model has a high false alarm rate due to different seasonal meteorological variables.

[0005] In a first aspect, this application provides a method for detecting anomalies in automatic airport observation data. The method includes: acquiring real-time meteorological observation data of an airport runway; wherein the real-time meteorological observation data includes real-time observation values ​​of multiple meteorological variables collected by meteorological sensors at multiple different physical locations; inputting the real-time meteorological observation data into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network, wherein the meteorological reconstruction data includes the reconstruction value of each meteorological variable corresponding to each physical location; calculating reconstruction residual data based on the real-time meteorological observation data and the meteorological reconstruction data, wherein the reconstruction residual data includes the reconstruction residual value corresponding to each meteorological variable at each physical location; using a pre-constructed independent physical knowledge base, applying physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data, wherein the physical residual data includes the physical residual value corresponding to each target meteorological variable; and determining whether there are any anomalous sensors among the sensors at multiple different physical locations based on the reconstruction residual data and the physical residual data.

[0006] The airport automatic observation data anomaly detection method provided in this solution learns the normal joint statistical distribution of multi-location, multi-variable meteorological data from the runway through an autoencoder network. It identifies anomalies deviating from historical patterns by reconstructing residuals and, through an independent physical knowledge base, outputs physical residuals based on deterministic atmospheric physical laws and spatial correlation laws to identify anomalies that violate physical laws. Finally, sensor anomalies are determined from both statistical patterns and physical laws by reconstructing residual data and physical residual data. Therefore, the method designed in this solution integrates statistical reconstruction and physical constraints for dual-dimensional detection. It can identify minute sensor drifts that traditional standard methods cannot detect and can filter statistical false alarms from purely data-driven methods through physical rules. In addition, the physical knowledge base designed in this solution is decoupled from the autoencoder network. When changing airports or adjusting physical rules, only the configuration file needs to be modified, without retraining the neural network.

[0007] In an optional implementation of the first aspect, determining whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstructed residual data and the physical residual data includes: for each meteorological variable at each physical location, determining whether the corresponding reconstructed residual value exceeds a first preset threshold, or determining whether the physical residual value corresponding to each target meteorological variable exceeds a second preset threshold; if the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, then determining that the sensor at the physical location corresponding to the meteorological variable that exceeds the threshold is abnormal.

[0008] The above implementation method performs threshold verification on the two residuals of the reconstructed residual data and the physical residual data for each meteorological variable at each physical location. If any residual exceeds the corresponding threshold, it is determined that the sensor of the corresponding variable at that location is abnormal, thus ensuring the sensitivity of anomaly detection and avoiding missed detections in a single detection dimension.

[0009] In an optional implementation of the first aspect, determining whether there are anomalous sensors among sensors at multiple different physical locations based on reconstructed residual data and physical residual data includes: acquiring time prediction data for real-time meteorological observation data, wherein the time prediction data is used to reflect the temporal consistency of the current real-time observation value with respect to the observation value at the previous time step; calculating prediction residual data based on the time prediction data and real-time meteorological observation data, wherein the prediction residual data includes the prediction residual value corresponding to each meteorological variable at each physical location; and determining whether there are anomalous sensors among sensors at multiple different physical locations based on the reconstructed residual data, physical residual data, and prediction residual data.

[0010] The above implementation method adds a time dimension consistency check to the dual-channel detection of reconstruction residual and physical residual, forming a three-dimensional verification system. The added time consistency dimension can effectively identify sensor jump-type faults and supplement the detection blind spots of reconstruction residual and physical residual.

[0011] In an optional implementation of the first aspect, determining whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstructed residual data, physical residual data, and predicted residual data includes: for each meteorological variable at each physical location, determining whether the corresponding reconstructed residual value exceeds a first preset threshold, or determining whether the physical residual value corresponding to each target meteorological variable exceeds a second preset threshold, or determining whether the predicted residual value corresponding to each meteorological variable at each physical location exceeds a third preset threshold; if the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, or the predicted residual value corresponding to any meteorological variable exceeds the third preset threshold, then it is determined that the sensor at the physical location corresponding to the meteorological variable exceeding the threshold is abnormal.

[0012] In the above implementation method, the three residuals—reconstruction residual data, physical residual data, and prediction residual data—are each set with independent thresholds, corresponding to three different types of normality constraints: spatial variable correlation, physical law constraints, and temporal continuity. Anomaly screening is performed from three orthogonal dimensions. The three detection paths correspond to different anomaly modes, covering almost all common sensor anomaly types. Furthermore, each path's threshold is set and judged independently, allowing operators to adjust the detection sensitivity of a specific path according to the airport's actual needs without affecting other detection logic, thus providing high maintenance flexibility.

[0013] In an optional implementation of the first aspect, obtaining time prediction data for real-time meteorological observation data includes: obtaining meteorological observation data from the previous time step, wherein the meteorological observation data from the previous time step includes observation values ​​of multiple meteorological variables collected by multiple meteorological sensors at different physical locations in the previous time step; inputting the meteorological observation data from the previous time step into a pre-trained autoencoder network to obtain the predicted values ​​of each meteorological variable at the current time output by the autoencoder network, thereby obtaining time prediction data, wherein the parallel feedforward branch of the pre-trained autoencoder network is the time prediction branch.

[0014] In the above implementation, the time prediction branch and the autoencoder reconstruction path share the underlying input features, and learn the temporal evolution of meteorological data through an independent feedforward network. Because they share the input layer and some feature representations, the time prediction branch does not require a separately trained feature extractor, and can achieve time-series prediction without significantly increasing the number of network parameters.

[0015] In an optional implementation of the first aspect, the training process of the autoencoder network is as follows: acquiring historical meteorological observation data of the airport runway; constructing a three-dimensional dataset based on the historical meteorological observation data, wherein the three-dimensional dataset includes multiple three-dimensional data samples, and the three-dimensional dimensions corresponding to each three-dimensional data sample include time dimension, location dimension, and observation value dimension; dividing the three-dimensional dataset into multiple training sample pairs, wherein the sampling time difference between the two samples in each training sample pair is 1; training the reconstruction path and time prediction branch of the autoencoder network based on the multiple training sample pairs until the reconstruction loss of the reconstruction path and the prediction loss of the time prediction branch meet the convergence conditions respectively.

[0016] The above implementation optimizes the reconstruction loss and prediction loss during the training phase, enabling simultaneous acquisition of both reconstruction and prediction capabilities in a single training session, eliminating the need to train two independent models and significantly shortening the training cycle. Furthermore, training is based on a massive amount of minute-level samples throughout the year, fully learning normal meteorological patterns under different seasons and weather conditions. Combined with prior constraints from the adjacency matrix, it can effectively avoid spurious associations caused by seasonal sample bias and improve the model's generalization ability.

[0017] In an optional implementation of the first aspect, a pre-constructed independent physical knowledge base is used to apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data. This includes: parsing the pre-constructed physical knowledge base to obtain atmospheric physical constraints and spatial consistency constraints; applying atmospheric physical constraints to at least one first target meteorological variable in the real-time meteorological observation data according to the atmospheric physical constraints to obtain a physical prediction value corresponding to each first target meteorological variable; calculating the difference between the real-time observed value and the physical prediction value corresponding to each first target meteorological variable to obtain a physical residual value corresponding to each first target meteorological variable; and performing runway spatial consistency verification on at least one second target meteorological variable in the real-time meteorological observation data according to the spatial consistency constraints to obtain a physical residual value corresponding to each second target meteorological variable.

[0018] The above implementation divides the independent physical knowledge base into two levels: atmospheric physical constraints and spatial consistency constraints. Physical residuals are calculated from two dimensions: variable physical properties and multi-location spatial distribution patterns. This allows physical rules to be classified and maintained, universal atmospheric physical rules to be reused across airports, and spatial consistency rules to be configured individually for specific runways, enabling the identification of even smaller sensor deviations.

[0019] In an optional embodiment of the first aspect, the atmospheric physical constraints include target atmospheric physical boundary conditions and target atmospheric physical prediction formulas. Based on the atmospheric physical constraints, at least one first target meteorological variable in the real-time meteorological observation data is subject to atmospheric physical constraints to obtain a physical prediction value corresponding to each first target meteorological variable. This includes: determining whether at least one first target meteorological variable in the real-time meteorological observation data is within the target atmospheric physical boundary conditions; if it is determined that at least one first target meteorological variable in the real-time meteorological observation data is within the target atmospheric physical boundary conditions, then the at least one first target meteorological variable is predicted and calculated using the target atmospheric physical prediction formula to obtain a physical prediction value corresponding to each first target meteorological variable.

[0020] The above implementation method first uses boundary conditions for rapid screening to directly identify outliers that clearly cross the boundary. Then, for the first target meteorological variable within the boundary, the theoretical value is calculated using physical prediction formulas to obtain continuous residuals and identify minor deviations within the boundary. This improves the overall detection speed through the pre-screening step of boundary conditions, and the two-level mechanism simultaneously covers serious faults and slight drifts, achieving full-range anomaly detection.

[0021] In an optional implementation of the first aspect, at least one target meteorological variable is predicted and calculated using a target atmospheric physics prediction formula to obtain a physical prediction value corresponding to each target meteorological variable, including: when there are multiple target atmospheric physics prediction formulas corresponding to the first target meteorological variable, obtaining an initial physical prediction value calculated by each target atmospheric physics prediction formula; and taking a weighted average of all initial physical prediction values ​​to obtain a physical prediction value corresponding to the first target meteorological variable.

[0022] In the above implementation method, when multiple applicable physical prediction formulas exist for the same target variable, weights are assigned based on the historical accuracy and applicable scenarios of each formula, and the multiple calculation results are weighted and fused to obtain a more robust physical prediction value. Multi-formula fusion can combine the advantages of different physical formulas, maintaining high prediction accuracy under different weather conditions and avoiding the bias of a single formula in specific scenarios.

[0023] In an alternative embodiment of the first aspect, multiple different physical locations include the first end of the runway, the second end of the runway, and the mid-field location along the runway of the same airport runway.

[0024] The above implementation method adopts a three-point sensor layout at both ends of the runway and the middle of the runway, covering the entire length of the runway. It utilizes the physical characteristics of the three points being located in the same micro-meteorological air mass within the same airport air mass range to establish spatial consistency constraints. The three-point layout forms multiple sets of spatial comparison pairs, which can accurately locate individual abnormal sensors through pairwise comparison and three-point consistency analysis, avoiding mutual confusion of single-point faults. At the same time, the sensor deployment position is aligned with the key nodes of runway take-off and landing, and the detection results can directly correspond to different stages of aircraft take-off and landing, with strong business adaptability.

[0025] Secondly, this application provides an airport automatic observation data anomaly detection device, which includes an acquisition module, an autoencoder network module, a reconstruction residual calculation module, a physical residual calculation module, and a determination module. The acquisition module is used to acquire real-time meteorological observation data of the airport runway; wherein, the real-time meteorological observation data includes real-time observation values ​​of multiple meteorological variables collected by meteorological sensors at multiple different physical locations; the autoencoder network module is used to input multi-dimensional meteorological data vectors into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network, wherein the meteorological reconstruction data includes each meteorological variable corresponding to each physical location. The module calculates the reconstructed values ​​of variables; the reconstruction residual calculation module is used to calculate the reconstruction residual data based on real-time meteorological observation data and meteorological reconstruction data, wherein the reconstruction residual data includes the reconstruction residual value corresponding to each meteorological variable at each physical location; the physical residual calculation module is used to apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints using a pre-built independent physical knowledge base, thereby obtaining physical residual data, wherein the physical residual data includes the physical residual value corresponding to each target meteorological variable; the determination module is used to determine whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstruction residual data and the physical residual data.

[0026] The airport automatic observation data anomaly detection device provided in this solution learns the normal joint statistical distribution of multi-location, multi-variable meteorological data from the runway through an autoencoder network. It identifies anomalies that deviate from historical patterns by reconstructing residuals and, through an independent physical knowledge base, outputs physical residuals based on deterministic atmospheric physical laws and spatial correlation laws to identify anomalies that violate physical laws. Finally, it determines sensor anomalies from both statistical patterns and physical laws by using reconstructed residual data and physical residual data. Therefore, the method designed in this solution integrates statistical reconstruction and physical constraint dual-dimensional detection, which can identify minute sensor drifts that traditional standard methods cannot detect, and can also filter statistical false alarms from purely data-driven methods through physical rules. In addition, the physical knowledge base designed in this solution is decoupled from the autoencoder network. When changing airports or adjusting physical rules, only the configuration file needs to be modified, without retraining the neural network.

[0027] Thirdly, the present invention provides an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the method described in the first aspect and any optional embodiment of the first aspect.

[0028] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the method described in the first aspect or any optional embodiment of the first aspect.

[0029] Fifthly, the present invention provides a computer program product, including a computer program / instructions, which, when executed by a processor, perform the method described in the first aspect and any optional embodiment of the first aspect.

[0030] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a first flowchart of the airport automatic observation data anomaly detection method provided in the embodiments of this application; Figure 2 This is a second flowchart of the airport automatic observation data anomaly detection method provided in the embodiments of this application; Figure 3 A third flowchart of the airport automatic observation data anomaly detection method provided in the embodiments of this application; Figure 4 The fourth flowchart of the airport automatic observation data anomaly detection method provided in the embodiments of this application; Figure 5 The fifth flowchart of the airport automatic observation data anomaly detection method provided in the embodiments of this application; Figure 6 A structural diagram of the airport automatic observation data anomaly detection device provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0033] Icons: 600 - Acquisition Module; 610 - Autoencoder Network Module; 620 - Reconstruction Residual Calculation Module; 630 - Physical Residual Calculation Module; 640 - Determination Module; 7 - Electronic Equipment; 701 - Processor; 702 - Memory; 703 - Communication Bus. Detailed Implementation

[0034] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0035] 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 this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0036] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0037] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0038] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0039] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0040] In the description of the embodiments of this application, the technical terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of this application and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0041] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0042] The quality of airport aviation meteorological observation data directly affects flight safety and operational efficiency. Failure to promptly identify anomalies in sensor data can lead to erroneous weather reports, thereby impacting aircraft takeoff and landing decisions. Therefore, efficient and accurate anomaly detection of multi-source meteorological sensor data in airport areas is a crucial technical aspect of aviation meteorological operations.

[0043] Existing sensor anomaly detection methods are based on statistical models or machine learning approaches, such as neural network models like autoencoders, and have been gradually applied to meteorological data anomaly detection. However, this purely data-driven approach is prone to learning spurious variable associations caused by sampling bias. For example, when using meteorological variables from a specific season for model training, the visibility, temperature, and humidity of airport runways vary across seasons, leading to false alarms or missed alarms in the trained anomaly detection method.

[0044] To address the aforementioned issues, this application provides a method for detecting anomalies in airport automatic observation data. It learns the normal joint statistical distribution of multi-location, multi-variable meteorological data from the runway through an autoencoder network, identifies anomalies deviating from historical patterns by reconstructing residuals, and uses an independent physical knowledge base to output physical residuals based on deterministic atmospheric physical laws and spatial correlation patterns to identify anomalies violating physical laws. Finally, it determines sensor anomalies from both statistical patterns and physical laws using reconstructed residual data and physical residual data. Therefore, this method integrates statistical reconstruction and physical constraints for dual-dimensional detection, enabling it to identify minute sensor drifts that traditional standard methods cannot detect, and filtering statistical false alarms from purely data-driven methods through physical rules. Furthermore, the physical knowledge base is decoupled from the autoencoder network, so when changing airports or adjusting physical rules, only the configuration file needs to be modified, without retraining the neural network.

[0045] Based on the above ideas, this application first provides a method for detecting anomalies in airport automatic observation data. This method can be applied to computing devices, including but not limited to computers, servers, control chips, and host computers, etc. Figure 1 As shown, the automatic observation data anomaly detection method for this airport can be implemented in the following ways, including: Step S100: Obtain real-time meteorological observation data of the airport runway.

[0046] Step S110: Input real-time meteorological observation data into the pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network.

[0047] Step S120: Calculate the reconstruction residual data based on real-time meteorological observation data and meteorological reconstruction data.

[0048] Step S130: Using a pre-built independent physical knowledge base, apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data.

[0049] Step S140: Based on the reconstructed residual data and the physical residual data, determine whether there are any abnormal sensors among the sensors at multiple different physical locations.

[0050] In the above implementation, real-time meteorological observation data refers to real-time observations of various meteorological variables collected by meteorological sensors at multiple physical locations on the airport runway at a fixed time resolution, covering meteorological elements such as temperature, humidity, air pressure, wind speed, wind direction, visibility, cloud base height, and precipitation. This fixed time resolution can be adaptively adjusted according to the actual required data volume and frequency; for example, it can be 1 minute.

[0051] An independent physics knowledge base represents a set of atmospheric physics rules stored in a structured configuration file. It is completely decoupled from the autoencoder network and can be independently edited, maintained, and migrated. Specifically, it can contain multiple physical constraints, which may include hard constraint boundary conditions, soft constraint prediction formulas, and spatial consistency checks.

[0052] Both the reconstruction residual and the physical residual can retain the original physical units, such as ℃, %, m / s, hPa, etc., which respectively characterize the degree to which the observed data deviates from the historical normal statistical pattern and the laws of atmospheric physics.

[0053] The above implementation method of this scheme first acquires real-time meteorological observation data of the airport runway, and then inputs the real-time meteorological observation data into a pre-trained autoencoder network. The core mechanism of the autoencoder is that normal data patterns can be completely reconstructed through a low-dimensional bottleneck layer, and the reconstruction residual approaches 0; abnormal data loses information during compression due to deviation from the normal joint distribution, and the reconstruction residual increases significantly. Therefore, the autoencoder network outputs the reconstruction value of each meteorological variable corresponding to each physical location, and obtains meteorological reconstruction data. Then, based on the real-time meteorological observation data and the meteorological reconstruction data, the reconstruction residual data is calculated.

[0054] Specifically, as a specific implementation method, this scheme first constructs a multi-dimensional observation vector from real-time meteorological observation data. Then, the meteorological reconstruction data output by the autoencoder network is a meteorological reconstruction vector with the same dimension as the multi-dimensional observation vector. This scheme then performs element-wise subtraction between the multi-dimensional observation vector and the meteorological reconstruction vector. Taking a typical configuration of 3 physical locations and 12 types of meteorological variables as an example, following a fixed sequence of "Location 1, Variable 1; Location 1, Variable 2; ... Location 1, Variable 12; Location 2, Variable 1; ... Location 3, Variable 12", the residuals for each type of meteorological variable at each location are calculated independently. The calculation formula is as follows: ; in, This is the physical location index, with values ​​of 1, 2, and 3, corresponding to the three different physical locations of the runway. This is a meteorological variable index, with values ​​from 1 to 12, corresponding to meteorological elements such as temperature, relative humidity, surface air pressure, wind speed, wind direction, visibility, and cloud base height. For the first The position Real-time observations of class variables, This refers to the reconstructed value of the corresponding variable at the corresponding position in the output of the autoencoder network.

[0055] The calculated reconstruction residuals have the same physical units as the observed values. For example, the temperature residual is in °C, the humidity residual is in % and the wind speed residual is in m / s and the air pressure residual is in hPa. The sign of the residuals indicates the direction of deviation of the observed values ​​from the normal statistical pattern, and the absolute value indicates the degree of deviation. It can be directly used for quantitative assessment of the magnitude of anomalies and root cause tracing without the need for additional unit conversion or normalization.

[0056] In the case where the reconstructed residual data is obtained through the above method, this scheme can execute step S130 to use a pre-built independent physical knowledge base to apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints, thereby obtaining physical residual data. The physical residual data includes the physical residual value corresponding to each target meteorological variable. In this way, the rationality of the observation values ​​is verified from the physical essence level by using the independent physical knowledge base based on deterministic atmospheric physical laws.

[0057] Specifically, as one implementation method, this scheme can parse a pre-constructed physical knowledge base to obtain atmospheric physical constraints and spatial consistency constraints. Then, based on the atmospheric physical constraints, atmospheric physical constraints are applied to at least one first target meteorological variable in the real-time meteorological observation data to obtain the physical prediction value corresponding to each first target meteorological variable. The difference between the real-time observed value and the physical prediction value corresponding to each first target meteorological variable is calculated to obtain the physical residual value corresponding to each first target meteorological variable. Based on the spatial consistency constraints, runway spatial consistency verification is performed on at least one second target meteorological variable in the real-time meteorological observation data to obtain the physical residual value corresponding to each second target meteorological variable.

[0058] In the above implementation scheme, the independent physical knowledge base designed in this scheme is divided into two levels: the first level is atmospheric physical constraints, which are universal physical rules that are not affected by the location of the airport and can be reused across airports; the second level is spatial consistency constraints, which are exclusive spatial rules at the runway scale and can be configured separately for specific runways.

[0059] Specifically, based on the two levels of the independent physical knowledge base, the physical residual data obtained by this scheme are divided into two categories: atmospheric physical constraint residuals and spatial consistency constraint residuals, which are calculated in the following ways: First, the atmospheric physical constraint residual calculation can begin by parsing the atmospheric physical constraint conditions in the independent physics knowledge base to obtain the target atmospheric physical boundary conditions and the target atmospheric physical prediction formula. This scheme determines whether the real-time observed value of the first target meteorological variable is within the range specified by the target atmospheric physical boundary conditions, such as relative humidity ranging from [0, 100]%, wind speed ≥ 0 m / s, temperature ranging from [-20℃, 45℃], and surface pressure ranging from [980 hPa, 1040 hPa]. If the observed value exceeds the boundary range, it is directly determined that the variable violates the hard physical constraint, and its atmospheric physical residual value is recorded as the amplitude value exceeding the boundary, marked as an anomaly, and does not need to be included in subsequent calculations.

[0060] If the observed values ​​fall within the range specified by the target atmospheric physical boundary conditions, this scheme calls the target atmospheric physical prediction formula, using the real-time observed values ​​of the associated meteorological variables as input, to calculate the physical prediction value of the first target meteorological variable. For example, the theoretical prediction value of relative humidity is calculated based on the observed values ​​of air temperature and dew point temperature using the Magnus saturated vapor pressure formula; the rationality of the relative humidity at the time of precipitation is verified through precipitation-humidity logical constraints.

[0061] In this approach, when there are multiple target atmospheric physics prediction formulas corresponding to the first target meteorological variable, the initial physical prediction value calculated by each formula is obtained. A weighted average of all initial physical prediction values ​​is then taken to obtain the physical prediction value corresponding to the first target meteorological variable. This multi-formula fusion combines the advantages of different physical formulas, maintaining high prediction accuracy under various weather conditions and avoiding the bias of a single formula in specific scenarios. Furthermore, the weight configuration of different atmospheric physics prediction formulas can be dynamically adjusted based on season and weather type, further improving the scenario adaptability of physical predictions. When adding a new physical formula, only the corresponding formula and weight configuration need to be added to integrate it into the existing prediction system without reconstructing the original calculation logic, resulting in strong knowledge base scalability.

[0062] The calculation of spatial consistency constraint residuals involves: parsing the spatial consistency constraints in the independent physical knowledge base; performing runway-scale spatial cross-validation on the second target meteorological variables (such as temperature, surface pressure, relative humidity, and other spatially highly correlated variables); calculating the deviation between the observed values ​​at each location and the reference values ​​at the remaining locations; and obtaining the spatial consistency residuals. Specifically, this scheme can determine the spatial consistency residuals by comparing single-point and multi-location means, or by using a benchmark location comparison method, taking the calibrated benchmark sensor at one end of the runway as the benchmark, and calculating the observation differences between the remaining locations and the benchmark location as the spatial residuals.

[0063] The two types of physical residuals mentioned above together constitute the physical residual data output in this step. Each residual value corresponds to a clear physical constraint source and physical unit, which can be directly used for subsequent anomaly detection and root cause tracing.

[0064] With the reconstruction residual data and physical residual data obtained through the above methods, this solution can determine whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstruction residual data and physical residual data.

[0065] Specifically, as a specific implementation method, such as Figure 2 As shown, this solution can achieve anomaly identification through independent threshold determination of dual residuals, including: Step S200: For each meteorological variable at each physical location, determine whether the corresponding reconstruction residual value exceeds the first preset threshold, or determine whether the physical residual value corresponding to each target meteorological variable exceeds the second preset threshold.

[0066] Step S210: If the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, then it is determined that the sensor at the physical location corresponding to the meteorological variable that exceeds the threshold is abnormal.

[0067] In the above implementation, the first preset threshold is an anomaly judgment threshold set for the reconstruction residual, which is adaptively determined based on the statistical distribution of historical normal data. Different meteorological variables and different physical locations may correspond to different thresholds. The second preset threshold is an anomaly judgment threshold set for the physical residual, which is determined based on the allowable deviation range of physical laws and the statistical characteristics of historical observations.

[0068] The implementation scheme described above employs an independent threshold determination mechanism. If any residual exceeds the corresponding threshold, the sensor for that variable at that location is determined to be faulty, ensuring high sensitivity in anomaly detection and preventing missed detections from a single detection dimension. Simultaneously, the threshold determination logic requires minimal computation, enabling rapid anomaly determination for all variables at a single moment, thus meeting real-time detection needs.

[0069] In an optional implementation of this embodiment, the solution can also introduce time consistency verification to form a three-way residual detection system, such as... Figure 3 As shown, it includes: Step S300: Obtain time-prediction data based on real-time meteorological observation data.

[0070] Step S310: Calculate the prediction residual data based on the time prediction data and real-time meteorological observation data.

[0071] Step S320: Based on the reconstructed residual data, physical residual data, and predicted residual data, determine whether there are any abnormal sensors among the sensors at multiple different physical locations.

[0072] In the above implementation, the time prediction data is used to reflect the temporal consistency of the real-time observation at the current moment with respect to the observation at the previous time step.

[0073] The prediction residual data includes the prediction residual value for each meteorological variable at each physical location, which can detect sudden jumps in the sensor reading frequency that exceed the natural rate of change.

[0074] The implementation scheme described above adds a time dimension consistency check to the dual-path detection of reconstructed residual data and physical residual data. Under normal circumstances, the variation range of meteorological variables within a short period of time is within the natural variability range, while sensor failures can cause numerical jumps exceeding the natural variability range. The added time consistency dimension can effectively identify sensor jump-type faults and fill the detection blind spots of reconstructed residuals and physical residuals.

[0075] In an optional implementation of this embodiment, for step S320, as follows: Figure 4 As shown, this solution can implement the threshold determination logic for the three residuals in the following ways: Step S400: For each meteorological variable at each physical location, determine whether the corresponding reconstruction residual value exceeds the first preset threshold, or determine whether the physical residual value corresponding to each target meteorological variable exceeds the second preset threshold, or determine whether the prediction residual value corresponding to each meteorological variable at each physical location exceeds the third preset threshold.

[0076] Step S410: If the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, or the prediction residual value of any meteorological variable exceeds the third preset threshold, then it is determined that the sensor at the physical location corresponding to the meteorological variable that exceeds the threshold is abnormal.

[0077] The above-described implementation scheme features three independent threshold settings for the three residuals, corresponding to three different types of normality constraints: spatial variable correlation, physical law constraints, and temporal continuity. Anomaly screening is performed from three orthogonal dimensions, covering all common sensor anomaly types such as drift, jumps, jamming, and logic errors. Furthermore, each threshold is set and judged independently, allowing operators to adjust the detection sensitivity of a specific path according to the airport's actual needs without affecting other detection logic, thus offering high maintenance flexibility.

[0078] Additionally, it should be noted that, in order to improve the reliability of sensor anomaly detection, this solution can also be configured to determine anomalies only when multiple thresholds are exceeded. That is, if the reconstructed residual value of any meteorological variable exceeds the first preset threshold, the physical residual value exceeds the second preset threshold, and the prediction residual value exceeds the third preset threshold, this solution will determine that the sensor at the physical location corresponding to the meteorological variable whose three residuals all exceed the thresholds is abnormal.

[0079] In an optional implementation of this embodiment, the time prediction data can be obtained from the output of the time prediction branch of the autoencoder network, such as... Figure 5 As shown, it includes: Step S500: Obtain meteorological observation data from the previous time step.

[0080] Step S510: Input the meteorological observation data from the previous time step into the pre-trained autoencoder network to obtain the predicted values ​​of each meteorological variable at the current time output by the autoencoder network, and obtain the time prediction data.

[0081] In the above implementation method, the meteorological observation data of the previous time step includes the observation values ​​of multiple meteorological variables collected by multiple meteorological sensors at different physical locations in the previous time step.

[0082] The parallel feedforward branch of the pre-trained autoencoder network is a time prediction branch. The time prediction branch shares the underlying input features with the autoencoder reconstruction path. It learns the temporal evolution of meteorological data through an independent feedforward network, without the need to train a feature extractor separately.

[0083] Specifically, the aforementioned autoencoder network containing the time prediction branch can be trained as follows: acquire historical meteorological observation data of the airport runway, and then construct a three-dimensional dataset based on the historical meteorological observation data. The three-dimensional dataset includes multiple three-dimensional data samples, and the three-dimensional dimensions corresponding to each three-dimensional data sample include the time dimension, the location dimension, and the observation value dimension. Then, divide the three-dimensional dataset into multiple training sample pairs, wherein the sampling time step difference between the two samples in each training sample pair is 1. Based on the multiple training sample pairs, train the reconstruction path and the time prediction branch of the autoencoder network until the reconstruction loss of the reconstruction path and the prediction loss of the time prediction branch meet the convergence conditions respectively.

[0084] The autoencoder network includes a fully connected structure with an input layer, an encoder, and a decoder. For example, the input layer can take in a 36-dimensional multidimensional observation vector, i.e., a 3-position × 12-variable multidimensional observation vector. The encoder compresses the multidimensional observation vector to a low-dimensional bottleneck (64→32→16-dimensional bottleneck) through the fully connected layer, learns the joint probability density structure of the variables, and then the decoder reconstructs the original dimensional observation vector from the bottleneck layer (32→64→36-dimensional output). Then, it calculates the loss value between the reconstructed observation vector and the original input multidimensional observation vector to obtain the reconstruction loss. At the same time, it obtains the time prediction value based on the training sample pairs from the time prediction branch, and then calculates the prediction loss value based on the time prediction value and the observation value of the next time step in the sample pair.

[0085] Based on the above, this solution adopts a joint loss function to optimize both reconstruction loss and prediction loss. Both reconstruction and prediction capabilities can be obtained in one training session, eliminating the need to train two independent models separately and significantly shortening the training cycle.

[0086] Furthermore, during training, this scheme can introduce an adjacency matrix constructed based on variable correlation coefficients as a prior constraint. This retains only physically relevant variable connection paths and masks variable associations without physical meaning, thus preventing the network from learning spurious statistical correlations at the training level and improving the model's generalization ability. This adjacency matrix can be constructed based on the Pearson correlation coefficients between variables in the year's observation data, establishing connection paths when |r|>0.3, where r represents the Pearson correlation coefficient. This adjacency matrix is ​​only used as a weight mask during the training phase and is not involved in the calculation during the inference phase, thus not affecting inference efficiency.

[0087] In the implementation scheme described above, the time prediction branch and the autoencoder reconstruction path share the underlying input features. They learn the temporal evolution patterns of meteorological data through an independent feedforward network, eliminating the need for separate feature extractor training. This allows for time-series prediction without significantly increasing the number of network parameters. Simultaneously, the shared input layer ensures consistency in the feature space, giving the prediction residual and reconstruction residual a unified physical dimension and semantics, facilitating subsequent fusion judgment and result interpretation. During the inference phase, the reconstruction path and time prediction branch can be computed in parallel, with no significant increase in inference latency at a single time step, thus not affecting overall detection efficiency.

[0088] In an optional embodiment of this scheme, the sensors are positioned at multiple different physical locations, including the first end of the runway, the second end of the runway, and the mid-runway along the same airport runway. This three-point sensor layout, combining the two ends and the mid-runway, covers the entire length of the runway. Spatial consistency constraints are established by utilizing the physical characteristics of these three points being located within the same micrometeorological air mass within the same airport air mass range. Simultaneously, the three-point layout forms multiple sets of spatial comparison pairs. Through pairwise comparisons and three-point consistency analysis, individual abnormal sensors can be accurately located, avoiding confusion caused by single-point faults. Furthermore, the sensor deployment locations are aligned with critical nodes of runway takeoff and landing, allowing detection results to directly correspond to different stages of aircraft takeoff and landing, thus improving the precision of aviation meteorological services.

[0089] The airport automatic observation data anomaly detection method provided in this solution learns the normal joint statistical distribution of multi-location, multi-variable meteorological data from the runway through an autoencoder network. It identifies anomalies deviating from historical patterns by reconstructing residuals and, through an independent physical knowledge base, outputs physical residuals based on deterministic atmospheric physical laws and spatial correlation laws to identify anomalies that violate physical laws. Finally, sensor anomalies are determined from both statistical patterns and physical laws by reconstructing residual data and physical residual data. Therefore, the method designed in this solution integrates statistical reconstruction and physical constraints for dual-dimensional detection. It can identify minute sensor drifts that traditional standard methods cannot detect and can filter statistical false alarms from purely data-driven methods through physical rules. In addition, the physical knowledge base designed in this solution is decoupled from the autoencoder network. When changing airports or adjusting physical rules, only the configuration file needs to be modified, without retraining the neural network.

[0090] Figure 6 A schematic structural block diagram of an airport automatic observation data anomaly detection device provided in this application is presented. It should be understood that this device is applied to the computing device described above, and this device is used in conjunction with... Figures 1 to 5The method implementation in this paper corresponds to the steps involved in the aforementioned method. The specific functions of this device can be found in the description above; to avoid repetition, detailed descriptions are omitted here. This device includes at least one software function module that can be stored in a memory or embedded in the device's operating system (OS) in the form of software or firmware. Specifically, the device includes: an acquisition module 600, an autoencoder network module 610, a reconstruction residual calculation module 620, a physical residual calculation module 630, and a determination module 640. The acquisition module 600 is used to acquire real-time meteorological observation data of the airport runway; wherein, the real-time meteorological observation data includes real-time observation values ​​of multiple meteorological variables collected by meteorological sensors at multiple different physical locations; the autoencoder network module 610 is used to input multi-dimensional meteorological data vectors into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network, wherein the meteorological reconstruction data includes the reconstruction value of each meteorological variable corresponding to each physical location; the reconstruction residual... The difference calculation module 620 is used to calculate the reconstruction residual data based on real-time meteorological observation data and meteorological reconstruction data, wherein the reconstruction residual data includes the reconstruction residual value corresponding to each meteorological variable at each physical location; the physical residual calculation module 630 is used to use a pre-built independent physical knowledge base to apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data, wherein the physical residual data includes the physical residual value corresponding to each target meteorological variable; the determination module 640 is used to determine whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstruction residual data and the physical residual data.

[0091] The airport automatic observation data anomaly detection device provided in this solution learns the normal joint statistical distribution of multi-location, multi-variable meteorological data from the runway through an autoencoder network. It identifies anomalies that deviate from historical patterns by reconstructing residuals and, through an independent physical knowledge base, outputs physical residuals based on deterministic atmospheric physical laws and spatial correlation laws to identify anomalies that violate physical laws. Finally, it determines sensor anomalies from both statistical patterns and physical laws by using reconstructed residual data and physical residual data. Therefore, the method designed in this solution integrates statistical reconstruction and physical constraint dual-dimensional detection, which can identify minute sensor drifts that traditional standard methods cannot detect, and can also filter statistical false alarms from purely data-driven methods through physical rules. In addition, the physical knowledge base designed in this solution is decoupled from the autoencoder network. When changing airports or adjusting physical rules, only the configuration file needs to be modified, without retraining the neural network.

[0092] According to some embodiments of this application, such as Figure 7As shown, this application provides an electronic device 7, including: a processor 701 and a memory 702. The processor 701 and the memory 702 are interconnected and communicate with each other through a communication bus 703 and / or other forms of connection mechanism (not shown). The memory 702 stores a computer program executable by the processor 701. When the computing device is running, the processor 701 executes the computer program to perform any optional implementation method, such as steps S100 to S140: acquiring real-time meteorological observation data of an airport runway; inputting the real-time meteorological observation data into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network; calculating reconstruction residual data based on the real-time meteorological observation data and the meteorological reconstruction data; using a pre-built independent physical knowledge base, applying physical constraints to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data; and determining whether there are abnormal sensors among sensors at multiple different physical locations based on the reconstruction residual data and the physical residual data.

[0093] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the method in any of the aforementioned optional implementations.

[0094] The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0095] This application provides a computer program product that, when run on a computer, causes the computer to perform a method in any of the optional implementations.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and they should all be covered within the scope of the claims and specification of this application. In particular, as long as there is no structural conflict, the various technical features mentioned in the embodiments can be combined in any way. This application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for detecting anomalies in automatic airport observation data, characterized in that, The method includes: Acquire real-time meteorological observation data of the airport runway; wherein, the real-time meteorological observation data includes real-time observation values ​​of multiple meteorological variables collected by multiple meteorological sensors at different physical locations; The real-time meteorological observation data is input into a pre-trained autoencoder network to obtain meteorological reconstruction data output by the autoencoder network, wherein the meteorological reconstruction data includes the reconstructed value of each meteorological variable corresponding to each physical location; Based on the real-time meteorological observation data and the meteorological reconstruction data, the reconstruction residual data is calculated, wherein the reconstruction residual data includes the reconstruction residual value corresponding to each meteorological variable for each physical location; Using a pre-built independent physical knowledge base, physical constraints are applied to at least one target meteorological variable in the real-time meteorological observation data based on physical constraints to obtain physical residual data, wherein the physical residual data includes the physical residual value corresponding to each target meteorological variable; Based on the reconstructed residual data and the physical residual data, it is determined whether there are any abnormal sensors among the sensors at the multiple different physical locations.

2. The method according to claim 1, characterized in that, The step of determining whether there are abnormal sensors among the multiple sensors at different physical locations based on the reconstructed residual data and the physical residual data includes: For each meteorological variable at each physical location, determine whether the corresponding reconstructed residual value exceeds the first preset threshold, or determine whether the physical residual value corresponding to each target meteorological variable exceeds the second preset threshold. If the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, then it is determined that the sensor at the physical location corresponding to the meteorological variable that exceeds the threshold is abnormal.

3. The method according to claim 1, characterized in that, The step of determining whether there are abnormal sensors among the multiple sensors at different physical locations based on the reconstructed residual data and the physical residual data includes: Obtain time prediction data for the real-time meteorological observation data, wherein the time prediction data is used to reflect the time consistency of the real-time observation value at the current moment with the observation value at the previous time step. Based on the time-prediction data and the real-time meteorological observation data, prediction residual data is calculated, wherein the prediction residual data includes the prediction residual value corresponding to each meteorological variable for each physical location; Based on the reconstructed residual data, the physical residual data, and the predicted residual data, it is determined whether there are any abnormal sensors among the sensors at the multiple different physical locations.

4. The method according to claim 3, characterized in that, The step of determining whether there are abnormal sensors among the multiple sensors at different physical locations based on the reconstructed residual data, the physical residual data, and the predicted residual data includes: For each meteorological variable at each physical location, determine whether the corresponding reconstruction residual value exceeds the first preset threshold, or determine whether the physical residual value corresponding to each target meteorological variable exceeds the second preset threshold, or determine whether the prediction residual value corresponding to each meteorological variable at each physical location exceeds the third preset threshold. If the reconstructed residual value of any meteorological variable exceeds the first preset threshold, or the physical residual value corresponding to any target meteorological variable exceeds the second preset threshold, or the prediction residual value corresponding to any meteorological variable exceeds the third preset threshold, then it is determined that the sensor at the physical location corresponding to the meteorological variable that exceeds the threshold is abnormal.

5. The method according to claim 3, characterized in that, The acquisition of time-prediction data for the real-time meteorological observation data includes: Acquire meteorological observation data from the previous time step, wherein the meteorological observation data from the previous time step includes observation values ​​of multiple meteorological variables collected by multiple meteorological sensors at different physical locations in the previous time step. The meteorological observation data of the previous time step is input into the pre-trained autoencoder network to obtain the predicted values ​​of each meteorological variable at the current time output by the autoencoder network, thereby obtaining the time prediction data. The parallel feedforward branch of the pre-trained autoencoder network is the time prediction branch.

6. The method according to claim 5, characterized in that, The training process of the autoencoder network is as follows: Obtain historical meteorological observation data for airport runways; A three-dimensional dataset is constructed based on the historical meteorological observation data. The three-dimensional dataset includes multiple three-dimensional data samples, and the three-dimensional dimensions corresponding to each three-dimensional data sample include time dimension, location dimension and observation value dimension. The three-dimensional dataset is divided into multiple training sample pairs, wherein the sampling step difference between the two samples in each training sample pair is 1. Based on the multiple training sample pairs, the reconstruction path and the temporal prediction branch of the autoencoder network are trained until the reconstruction loss of the reconstruction path and the prediction loss of the temporal prediction branch respectively meet the convergence conditions.

7. The method according to claim 1, characterized in that, The process of using a pre-built independent physical knowledge base to apply physical constraints to at least one target meteorological variable in the real-time meteorological observation data, based on physical constraints, to obtain physical residual data includes: The pre-constructed physical knowledge base is analyzed to obtain atmospheric physical constraints and spatial consistency constraints. Based on the atmospheric physical constraints, at least one first target meteorological variable in the real-time meteorological observation data is subjected to atmospheric physical constraints to obtain the physical prediction value corresponding to each first target meteorological variable. Calculate the difference between the real-time observed value and the physical prediction value corresponding to each first target meteorological variable to obtain the physical residual value corresponding to each first target meteorological variable; Based on the spatial consistency constraints, runway spatial consistency verification is performed on at least one second target meteorological variable in the real-time meteorological observation data to obtain the physical residual value corresponding to each second target meteorological variable.

8. The method according to claim 7, characterized in that, The atmospheric physical constraints include target atmospheric physical boundary conditions and target atmospheric physical prediction formulas. The step of applying atmospheric physical constraints to at least one first target meteorological variable in the real-time meteorological observation data, based on the atmospheric physical constraints, to obtain the physical prediction value corresponding to each first target meteorological variable, includes: Determine whether at least one first target meteorological variable in the real-time meteorological observation data is within the target atmospheric physical boundary conditions; If it is determined that at least one first target meteorological variable in the real-time meteorological observation data is within the target atmospheric physical boundary conditions, then the at least one first target meteorological variable is predicted and calculated using the target atmospheric physical prediction formula to obtain the physical prediction value corresponding to each first target meteorological variable.

9. The method according to claim 8, characterized in that, The step of predicting and calculating at least one first target meteorological variable using the target atmospheric physics prediction formula to obtain the physical prediction value corresponding to each first target meteorological variable includes: When there are multiple target atmospheric physics prediction formulas corresponding to the first target meteorological variable, obtain the initial physical prediction value calculated by each target atmospheric physics prediction formula. The weighted average of all the initial physical prediction values ​​is used to obtain the physical prediction value corresponding to the first target meteorological variable.

10. The method according to claim 1, characterized in that, The multiple different physical locations include the first end of the runway, the second end of the runway, and the mid-field location along the runway of the same airport runway.

Images