Interpretable similarity measurement method and system for airport surface operation scenarios

Abstract

The present invention discloses an interpretable similarity measurement method and system for airport surface operation scenarios. The method comprises the following steps: step 1, collecting flight data and meteorological data from different data channels as scenario data, preprocessing the data, and constructing a scenario operation data set from the perspective of departing flights; step 2, processing the data in the scenario operation data set obtained in step 1 with secondary features, and dividing the data into two types of features, namely flight static attribute features and environment dynamic attribute features, according to the feature characteristics; step 3, constructing a data structure for interpretable similarity data using the two types of features obtained in step 2; and step 4, applying the data obtained in step 3 to a network for calculating interpretable similarity, and calculating the interpretable similarity of the airport surface operation scenarios based on the two types of data.

Description

Technical Field

[0001] The present invention belongs to the field of airport scene situation measurement and recognition, and specifically relates to an interpretable similarity measurement method and system, in particular to an interpretable similarity measurement method and system for airport scene operation scenarios. Background Art

[0002] In airport surface research, as predictive model design becomes increasingly complex, the accuracy of various indicators is also increasing. However, in actual surface control operations, the opacity of the prediction process prevents controllers from fully understanding the model's working principles. This leads them to rely more on their own experience to ensure operational safety, hindering the promotion of related technologies. To enhance controllers' trust in predictive models, one feasible approach is to utilize a prediction model based on similar scenarios. This involves comparing actual operational information differences between historical scenarios and the current scenario, providing historical statistical results as a reference for indicator prediction in the target scenario. In the field of indicator prediction, this approach uses multiple historical data sets that are most similar to the current scenario to improve the prediction accuracy of departing aircraft indicators in the target scenario. Given that the scenarios obtained by this method are based on actual historical operational data and are fully visible, the prediction results are more easily accepted by frontline operators, and therefore this process is considered explainable.

[0003] Previous research has partially examined similar scenarios, primarily categorized into two approaches: clustering-based "scenario segmentation, identification, and prediction" and improved similarity metrics-based indicator prediction. However, it is clear that clustering-based methods yield imprecise segmentation results. The resulting scenario prediction methods suffer from weak generalization and limited improvement in overall prediction performance, making them difficult to apply to refined scene operation management. Methods based on constructing similarity metrics, on the other hand, calculate deep features between scenarios and perform similarity measurements. The average of the prediction metrics from the closest scenarios is then taken as the prediction metric for the target scenario. This approach alleviates the difficulty of traditional similarity metrics in applying to airport scene operation environments with a high number of classification features, improving prediction performance to a certain extent. Therefore, current prediction research urgently needs a method that can effectively measure the similarity between the target scenario and similar scenarios, to screen similar scenarios and assist in subsequent prediction or classification. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to address the deficiencies of the existing technology and provide an interpretable similarity measurement method and system for airport surface operation scenarios, and to calculate the quantitative calculation results of scenario similarity under different application backgrounds.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for measuring the interpretable similarity of airport surface operation scenarios includes the following steps:

[0007] Step 1: Collect flight data and weather data from different data channels as scenario data, pre-process it, and construct a scenario operation data set from the perspective of departing flights;

[0008] Step 2: Process the data in the scenario operation data set obtained in step 1 by performing secondary feature processing, and divide it into two categories according to the characteristics of the features: flight static attribute features and environment dynamic attribute features;

[0009] Step 3: Use the two types of features obtained in step 2 to construct a data structure that can interpret similarity data;

[0010] Step 4: Apply the data obtained in step 3 to the network for calculating interpretable similarity, and calculate the interpretable similarity of the field operation scenarios based on the two types of data.

[0011] The step 1 specifically includes the following steps:

[0012] Step 1-1: Collect original flight and weather data from different data channels, including at least: airport collaborative management system data, airline operation control system data, and airport weather message data related to the target airport;

[0013] Step 1-2: preprocess the raw data collected in step 1-1, including screening out abnormal data and completing missing data.

[0014] The step 2 specifically includes the following steps:

[0015] Step 2-1: perform secondary feature extraction on the data processed in step 1, integrate all features, and perform normalization and encoding processing;

[0016] Step 2-2: Divide the features into two parts according to the characteristics of the extracted features, namely, flight static attribute features and environmental dynamic attribute features; store the processed and statistically processed data, and build a flight static attribute feature database and an environmental dynamic attribute feature database.

[0017] The step 3 specifically includes the following steps:

[0018] Step 3-1: Construct the static attribute input structure of the flight. Most of the static attribute features of the flight are categorical features and require additional encoding processing. The processed vector is called the static attribute vector θ i, all sample target scenes and their respective candidate similar scene sets are stacked into an input structure, and the final input data format is s×n×2×a cat , where s represents the number of sample scenes, α cat is the dimension of the processed static attribute vector, and n is the number of candidate scenes. Under this construction method, the static attributes of all sample target scenes and each candidate similar scene are compared.

[0019] Step 3-2: Construct the environment dynamic attribute input structure. Combine the environment dynamic attribute features of all scenes at the same time in the previous few days to form a multi-time scale environment dynamic attribute input vector for one day, called the dynamic attribute vector θ e , process the vectors of all scene data according to the above steps; combine all scenes with their respective candidate similar scene sets to finally form the input data.

[0020] The step 4 specifically includes the following steps:

[0021] Step 4-1: For the main supervision index T(n) under the nth candidate similar scene, use the supervision index T of the target scene ξ ξ The deviation δ between the nth candidate similar scene and the target scene n Therefore, the following method is used to model and analyze the composition of the departure supervision indicators of candidate similar scenarios:

[0022] T(n)=T ξ +δ n

[0023] Assume that the supervision index deviation in the nth candidate scenario obeys a normal distribution with a mean of 0 so that in The variance is generated under the condition of this scene due to the difference in key features between it and the target scene;

[0024] Step 4-2: Assume that there are only two different departure flight operation scenarios i and j in a scenario class, and rewrite the supervision indicator as Similarly, rewrite another sample as

[0025] Combine multiple samples with the same mean but different variances and apply weights to each sample to minimize the overall variance and obtain a more accurate T S Value:

[0026]

[0027] Among them, ω is the scene similarity, is the estimated value of the target scene supervision index, limiting the sum of similarities ω i +ωj =1;

[0028] Step 4-3: Get T through statistical knowledge S The variance value of

[0029]

[0030] Derivative the similarity in the formula:

[0031]

[0032]

[0033] Obviously, the second-order derivative is greater than zero, so the minimum value of the variance is obtained, and the corresponding similarity values ​​are:

[0034]

[0035] If the corresponding similarity of multiple scenes is determined, the expression is as follows:

[0036]

[0037] Where S is a scene set, which contains n different scenes, so the following results are obtained:

[0038]

[0039] The sum of all similarities is 1, that is,

[0040] Step 4-4: Taking into account the minimum value of the sample variance and combining the final prediction results, the supervision index of the target scene is approximately considered to be the weighted sum of the samples, which is expressed as:

[0041]

[0042] The same similarity sum is 1, that is Each weight ω i That is the explainable similarity of the scene;

[0043] Based on the characteristics of interpretable similarity, the two types of processed data are respectively input into two neural networks NET1 and NET2. The original data input shapes of the two neural networks are n×1×a cat and n×2×40, where n is the number of candidate similar scenes, α cat is the processed classification feature dimension. The calculation process is shown in steps 4-5 to 4-8:

[0044] Step 4-5: Downsample the data to form an image. Use a 1×2 convolution kernel to downsample the data. Expand the simplified data to form a two-dimensional tensor, which is similar in size and shape to common images.

[0045] Steps 4-6: Image-like convolution, using the LeNet-based network framework to perform convolution pooling on the tensor. At the end of the two neural networks, a fully connected layer is connected to obtain their respective similarity vectors, and the weights are combined to obtain an n×1 similarity fitting result, the number of which is consistent with the number of candidate similar scenes. The loss function is calculated as follows:

[0046]

[0047] Where ω is the similarity set of candidate similar scenes under all target scenes, ω mn is the similarity of the nth scene under the mth target scene, The meaning of is the mth target scene feature, x n The meaning is the nth historical scene feature, α is the target scene set, β is the candidate similar scene set, λ is the deviation importance parameter, K is the number of target operation scenes of all departing aircraft during the training process, T(x n ) is the supervision index of the nth candidate similar scene, For the target scene The actual supervision indicator under the target scenario; the loss function consists of two parts, which minimizes the prediction error of the supervision indicator on the training set while reducing the similarity value of high-deviation scenarios. The actual role of the neural network in this process is to calculate the similarity between different scenarios and the target scenario and use it to linearly generate the supervision indicator of the departure flight under the target scenario;

[0048] Steps 4-7: Priori processing. In order to ensure that the sum of similarities is 1, the following conditions are required: Based on this constraint, the similarities output by the two neural networks need to be processed a priori: first, all negative similarities need to be mapped to 0, that is, an additional ReLU layer is added to the end of the two neural networks. Second, all weights need to be normalized, that is, a normalization layer is added after the ReLU layer of the two neural networks. The similarities processed above are combined and weighted to produce the overall similarity between the target scene and the candidate similar scenes, which is calculated as follows:

[0049]

[0050] Where μ is the combination similarity distribution coefficient, and are the static and dynamic vector similarities of the candidate similar scenes output by NET1 and NET2 under the nth sample target scene;

[0051] Steps 4-8: Iterative training and similarity extraction. By linearly generating supervision indicators, the predicted combined similarity is linearly weighted summed with the historical supervision indicators in the candidate similar scenes, and iterative training is performed according to the loss function to reduce the error between the two.

[0052] An interpretable similarity measurement system for airport surface operation scenarios is characterized by being used in the above method and comprising:

[0053] Airport surface operation scenario data processing module, used to collect data from multiple data sources and perform pre-processing;

[0054] The surface operation scenario feature extraction and grouping module is used to extract secondary features according to data type and classify all features into flight static attribute features and environment dynamic attribute features according to their characteristics;

[0055] The scene dynamic interpretable similarity calculation module integrates the input data and calculates the dynamic interpretable similarity between the target scene and the historical operation scene.

[0056] Beneficial Effects: The present invention's method and system for measuring the interpretable similarity of airport surface operation scenarios employs a highly interpretable neural network model based on similar scenarios to implement the scenario similarity calculation process. This method proposes an interpretable similarity measurement method for airport surface operation scenarios, calculating interpretable similarity based on heterogeneous feature types using multi-timescale comparisons. This approach provides a novel approach for achieving precise surface control and can be widely used in the prediction and classification of key indicators in subsequent systems, filling a technological gap in methods for calculating dynamic interpretable similarity in airport surface operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A hierarchical partition diagram of the interpretable similarity measurement method for airport surface operation scenarios;

[0058] Figure 2 The core flow chart of the interpretable similarity measurement method for airport surface operation scenarios;

[0059] Figure 3 The flight static attribute feature data structure for interpretable similarity measurement method of airport surface operation scenarios;

[0060] Figure 4 An environmental dynamic attribute feature data structure for interpretable similarity measurement methods for airport surface operation scenarios;

[0061] Figure 5The weight comparison results of a day obtained by the interpretable similarity measurement method for airport surface operation scenarios;

[0062] Figure 6 The weight comparison results of different dates obtained by the interpretable similarity measurement method for airport surface operation scenarios;

[0063] Figure 7 Flowchart of the interpretable similarity measurement method for airport surface operation scenarios in an embodiment. DETAILED DESCRIPTION

[0064] In order to facilitate understanding of the design objectives, improvements, and combined advantages of the present invention, the present invention is further explained in detail in conjunction with the following figures and examples. It should also be noted that the scene classification samples used in this section are only used to illustrate the feasibility of the classification method and are not exclusively applicable to the processes included in the present invention.

[0065] Figure 1 This is a hierarchical division diagram of the interpretable similarity measurement method for airport surface operation scenarios, which describes the hierarchical division of the interpretable similarity measurement method for airport surface operation scenarios.

[0066] Figure 2 The flowchart of the method for interpretable similarity measurement of airport surface operation scenarios describes the process of this method.

[0067] We selected scenarios and accompanying data from departing flights at a hub airport for summary, setting the monitoring target as delay time. The analysis results are as follows:

[0068] Step 1: Collect and process multi-source data. The essence is to collect data from various available data sources such as airlines, airports and meteorological departments, integrate and pre-process the data, and construct a surface operation scenario feature system from the perspective of preliminary departure flights.

[0069] It includes the following sub-steps:

[0070] (1.1) Collect original flight and weather data from different data channels, including at least: airport collaborative management system data, airline operation control system data, and airport weather message data related to the target airport;

[0071] (1.2) Preprocessing the raw data from multiple data sources involves filtering out anomalous data and completing missing data. This preprocessing method involves deleting all missing data that cannot be completed and extracting flight operation scenario data for the target time period based on the selected features. Based on the Laida criterion, this method removes all flight entries whose taxi times exceed three standard deviations of the mean.

[0072] Statistics were collected for various flight-related data at this busy airport, including but not limited to airport collaborative decision-making data, flight plan data submitted by airlines, and airport weather report times. The raw data was processed by interpolation or deletion of abnormal data. The cleaned raw data was then integrated into a database of original departure flight scenarios.

[0073] Step 2: Feature extraction and grouping in the field operation scenario essentially involves processing secondary features based on the extracted scene data. Based on their characteristics, these features are divided into two categories: flight static attribute features and environmental dynamic attribute features, providing conditions for subsequent network input and data construction. The specific process is as follows:

[0074] (2.1) Perform secondary feature extraction on the processed data, integrate all features, and perform normalization and encoding processing;

[0075] (2.2) Based on the characteristics of the extracted features, the features are divided into two categories: static flight attributes and dynamic environmental attributes. The processed and statistically analyzed data is stored to construct a database of static flight attributes and a database of dynamic environmental attributes. Static flight attributes refer to features related to the flight's status or those specified in the flight plan. Dynamic environmental attributes refer to features related to dynamic attributes such as the airport's current traffic and weather conditions.

[0076] Secondary indicators are extracted based on the characteristics of the monitoring target, such as flight number, aircraft type, scene congestion evaluation index, airport weather conditions, etc. The categorical variables are processed with unique hot encoding. The processed similar scene data are classified according to their characteristics. Among them, the characteristics unique to the current flight are classified as flight static attribute characteristics, while the characteristics that change periodically over time are classified as environment dynamic attribute characteristics. The two characteristics are matched and stored separately. The construction result is as follows Figure 3 and Figure 4 shown.

[0077] Step 3: The data construction method for interpretable similarity data is essentially to compare the similarity of two types of data based on their characteristics. The two types of data are constructed separately according to the repeatability of the data. The specific construction process includes the following steps:

[0078] (3.1) Flight static attribute input structure construction. Most flight static attribute features are categorical features and require additional encoding processing. The processed vector is called the static attribute vector θ i , all sample target scenes and their respective candidate similar scene sets are stacked into an input structure, and the final input data format is s×n×2×a cat , where s represents the number of sample scenes, αcat is the dimension of the processed static attribute vector, and n is the number of candidate scenarios. In this construction method, the static attributes of all sample target scenarios and each candidate similar scenario are compared.

[0079] (3.2) The input structure of environmental dynamic attributes is constructed. The environmental dynamic attribute features of all scenes at the same time in the previous few days are spliced ​​to form a multi-time scale environmental dynamic attribute input vector for one day, which is called the dynamic attribute vector θ e , process all scene data according to the above steps; combine all scenes with their respective candidate similar scene sets to finally form the input data.

[0080] The data obtained in step 2 is further processed. For data without time-periodic characteristics, namely flight static attribute features, they are stacked with the corresponding data and aircraft models of similar candidate scenarios to form the flight static attribute input structure. For dynamic environment attribute features with time-periodic characteristics, several periodic data are extracted to construct the dynamic environment attribute input structure. Ultimately, two feature sets are formed for training dynamic similar scenarios.

[0081] Step 4: The data construction method of dynamic explainable similarity is essentially based on the dynamic explainable similarity calculation method of flight static attribute feature data and environment dynamic attribute data to evaluate similarity. Specifically, it includes the following steps:

[0082] (4.1) Scene index decomposition. For the main supervisory index T(n) under the nth candidate similar scene, the present invention uses the supervisory index T of the target scene ξ ξ The deviation δ between the nth candidate similar scene and the target scene n Therefore, the following method is used to model and analyze the composition of the candidate similar scene departure supervision indicators:

[0083] T(n)=T ξ +δ n

[0084] The various influencing factors and uncertainties involved in the scene will cause the supervision index of the current scene to be shortened or extended compared with the target scene. Therefore, for the convenience of analysis, it is assumed that the supervision index deviation in the nth candidate scene obeys a normal distribution with a mean of 0. in The variance is generated under the condition of this scene due to the difference in key features between it and the target scene.

[0085] (4.2) Assuming that there are only two different departure flight operation scenarios i and j in a scenario class, the supervision index can be rewritten as Another example can be rewritten as

[0086] Combine multiple samples with the same mean but different variances and apply weights to each sample to minimize the overall variance and obtain a more accurate T S Value:

[0087]

[0088] Among them, ω is the scene similarity, is the estimated value of the target scene supervision index, limiting the sum of similarities ω i +ω j =1.

[0089] (4.3) Through statistical knowledge, we can get T S The variance value of

[0090]

[0091] By taking the derivative of the similarity in the formula, we can get:

[0092]

[0093]

[0094] Obviously, the second-order derivative is greater than zero, so the minimum value of the variance can be obtained, and the corresponding similarity values ​​are:

[0095]

[0096] If the corresponding similarity of multiple scenes is determined, the expression is as follows:

[0097]

[0098] Where S is a scene set, which contains n different scenes. Therefore, we can get the following results:

[0099]

[0100] The sum of all similarities is 1, that is, (4.4) Taking into account the minimum value of the sample variance and combining it with the final prediction result, we can approximate that the supervision index of the target scene is the weighted sum of the samples. The experimental results show that this approximation method can meet the actual calculation requirements, and its expression is:

[0101]

[0102] The same similarity sum is 1, that is Each weight ω i This is the explainable similarity of the scene.

[0103] Based on the characteristics of interpretable similarity, the two types of processed data are respectively input into two neural networks NET1 and NET2. The original data input shapes of the two neural networks are n×1×a cat and n×2×40, where n is the number of candidate similar scenes, α cat is the processed classification feature dimension, and the calculation process is shown in steps (4-5) to (4-8):

[0104] (4.5) Data downsampling and visualization: downsample the data using a 1×2 convolution kernel; expand the simplified data to form a two-dimensional tensor, whose size and shape are similar to common images;

[0105] (4.6) Image convolution: convolution and pooling of tensors are performed using a LeNet-based network framework. The two neural networks are connected to a fully connected layer at the end to obtain their respective similarity vectors. The weights are then combined to obtain an n×1 similarity fitting result, which is consistent with the number of candidate similar scenes used in this invention. The loss function is calculated as follows:

[0106]

[0107] Where ω is the similarity set of candidate similar scenes under all target scenes, ω mn is the similarity of the nth scene under the mth target scene, The meaning of is the mth target scene feature, x n The meaning is the nth historical scene feature, α is the target scene set, β is the candidate similar scene set, λ is the deviation importance parameter, K is the number of target operation scenes of all departing aircraft during the training process, T(x n ) is the supervision index of the nth candidate similar scene, For the target scene The actual supervisory indicator under the target scenario. This loss function consists of two parts: minimizing the prediction error of the supervisory indicator on the training set while reducing the similarity value of high-deviation scenarios. The actual role of the neural network in this process is to calculate the similarity between different scenarios and the target scenario and use it to linearly generate the supervisory indicator of the departure flight under the target scenario;

[0108] (4.7) Priori processing, in order to ensure that the sum of similarities is 1, the following conditions are required: Based on this constraint, the similarities output by the two neural networks need to be processed a priori: first, all negative similarities need to be mapped to 0, that is, an additional ReLU layer is added to the end of the two neural networks. Second, all weights need to be normalized, that is, a normalization layer is added after the ReLU layer of the two neural networks. The combined weighted output of the similarities completed above can produce the overall similarity between the target scene and the candidate similar scenes, which is calculated as follows:

[0109]

[0110] Where μ is the combination similarity distribution coefficient, and The static and dynamic vector similarities of the candidate similar scenes output by NET1 and NET2 for the nth sample target scene, respectively, can be significantly reduced in the parameter search space and accelerated neural network convergence. (4.8) Iterative training and similarity extraction: By linearly generating supervision indicators, the predicted combined similarity is linearly weighted summed with the historical supervision indicators of the candidate similar scenes. Iterative training is then performed based on the loss function to reduce the error between the two.

[0111] The flight data extracted in step 3 is trained using the convolutional neural network proposed in step 4 of the present invention. The parameter settings are shown in Table 1. The scene similarities in the static attribute features and dynamic environment attribute features of the flight are compared respectively. Finally, the similarity between the target scene and all candidate similar scenes is evaluated based on the scene metric vector after the fully connected layer and prior processing. The similarity evaluation results of one day are shown in Figure 5 As shown in the figure, the similarity comparison results of multiple days are as follows Figure 6 The resulting scene dynamic similarity can be used for subsequent scene index prediction or scene category cluster analysis.

[0112] Table 1. Neural network parameter diagram in the interpretable similarity measurement method for airport surface operation scenarios

[0113]

[0114]

[0115] The present invention also provides an interpretable similarity measurement system for airport surface operation scenarios, which is used to implement the above method, specifically comprising:

[0116] Airport surface operation scenario data processing module, used to collect data from multiple data sources and perform pre-processing;

[0117] The surface operation scenario feature extraction and grouping module is used to extract secondary features according to data type and classify all features into flight static attribute features and environment dynamic attribute features according to their characteristics;

[0118] The scene dynamic interpretable similarity calculation module integrates the input data and calculates the dynamic interpretable similarity between the target scene and the historical operation scene.

[0119] The present invention provides a method and system for interpretable similarity measurement of airport surface operation scenarios. There are numerous methods and approaches for implementing this technical solution. The foregoing description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art may make improvements and modifications without departing from the principles of the present invention, and such improvements and modifications are also within the scope of protection of the present invention. Any components not specified in this embodiment may be implemented using existing technologies.

Claims

1. A method for measuring the interpretable similarity of airport surface operation scenarios, characterized by: The following steps are involved: Step 1: Collect flight data and weather data from different data channels as scenario data, pre-process it, and construct a scenario operation data set from the perspective of departing flights; Step 2: Process the data in the scenario operation data set obtained in step 1 by performing secondary feature processing, and divide it into two categories according to the characteristics of the features: flight static attribute features and environment dynamic attribute features; Step 3: Use the two types of features obtained in step 2 to construct a data structure that can interpret similarity data; Step 4, applying the data obtained in step 3 to the network for calculating interpretable similarity, and calculating the interpretable similarity of the field scene operation scenarios based on the two types of data respectively; specifically comprising the following steps: Step 4-1: For the main supervision index T(n) under the nth candidate similar scene, use the supervision index T of the target scene ξ ξ The deviation δ between the nth candidate similar scene and the target scene n Therefore, the following method is used to model and analyze the composition of the departure supervision indicators of candidate similar scenarios: T(n)=T ξ +δ n Assume that the supervision index deviation in the nth candidate scenario obeys a normal distribution with a mean of 0 so that in The variance is generated under the condition of this scene due to the difference in key features between it and the target scene; Step 4-2: Assume that there are only two different departure flight operation scenarios i and j in a scenario class, and rewrite the supervision indicator as Similarly, rewrite another sample as Combine multiple samples with the same mean but different variances and apply weights to each sample to minimize the overall variance and obtain a more accurate T S Value: Among them, ω is the scene similarity, is the estimated value of the target scene supervision index, limiting the sum of similarities ω i +ω j =1; Step 4-3: Get T through statistical knowledge S The variance value of Derivative the similarity in the formula: Obviously, the second-order derivative is greater than zero, so the minimum value of the variance is obtained, and the corresponding similarity values ​​are: If the corresponding similarity of multiple scenes is determined, the expression is as follows: Where S is a scene set, which contains n different scenes, so the following results are obtained: The sum of all similarities is 1, that is, Step 4-4: Taking into account the minimum value of the sample variance and combining the final prediction results, the supervision index of the target scene is approximately considered to be the weighted sum of the samples, which is expressed as: The same similarity sum is 1, that is Each weight ω i That is the explainable similarity of the scene; Based on the characteristics of interpretable similarity, the two types of processed data are respectively input into two neural networks NET1 and NET2. The original data input shapes of the two neural networks are n×1×α cat and n×2×40, where n is the number of candidate similar scenes, α cat is the processed classification feature dimension. The calculation process is shown in steps 4-5 to 4-8: Step 4-5: Downsample the data to form an image. Use a 1×2 convolution kernel to downsample the data. Expand the simplified data to form a two-dimensional tensor, which is similar in size and shape to common images. Steps 4-6: Image-like convolution, using the LeNet-based network framework to perform convolution pooling on the tensor. At the end of the two neural networks, a fully connected layer is connected to obtain their respective similarity vectors, and the weights are combined to obtain an n×1 similarity fitting result, the number of which is consistent with the number of candidate similar scenes. The loss function is calculated as follows: Where ω is the similarity set of candidate similar scenes under all target scenes, ω mn is the similarity of the nth scene under the mth target scene, The meaning of is the mth target scene feature, x n The meaning is the nth historical scene feature, α is the target scene set, β is the candidate similar scene set, λ is the deviation importance parameter, K is the number of target operation scenes of all departing aircraft during the training process, T(x n ) is the supervision index of the nth candidate similar scene, For the target scene The actual supervision indicator under the target scenario; the loss function consists of two parts, which minimizes the prediction error of the supervision indicator on the training set while reducing the similarity value of high-deviation scenarios. The actual role of the neural network in this process is to calculate the similarity between different scenarios and the target scenario and use it to linearly generate the supervision indicator of the departure flight under the target scenario; Steps 4-7: Priori processing. In order to ensure that the sum of similarities is 1, the following conditions are required: Based on this constraint, the similarities output by the two neural networks need to be processed a priori: first, all negative similarities need to be mapped to 0, that is, an additional ReLU layer is added to the end of the two neural networks. Second, all weights need to be normalized, that is, a normalization layer is added after the ReLU layer of the two neural networks. The similarities processed above are combined and weighted to produce the overall similarity between the target scene and the candidate similar scenes, which is calculated as follows: Where μ is the combination similarity distribution coefficient, and are the static and dynamic vector similarities of the candidate similar scenes output by NET1 and NET2 under the nth sample target scene; Steps 4-8: Iterative training and similarity extraction. By linearly generating supervision indicators, the predicted combined similarity is linearly weighted summed with the historical supervision indicators in the candidate similar scenes, and iterative training is performed according to the loss function to reduce the error between the two.

2. The method for measuring the interpretable similarity of airport surface operation scenarios according to claim 1 is characterized by: The step 1 specifically includes the following steps: Step 1-1: Collect original flight and weather data from different data channels, including at least: airport collaborative management system data, airline operation control system data, and airport weather message data related to the target airport; Step 1-2: preprocess the raw data collected in step 1-1, including screening out abnormal data and completing missing data.

3. The interpretable similarity measurement method for airport surface operation scenarios according to claim 1 is characterized by: The step 2 specifically includes the following steps: Step 2-1: perform secondary feature extraction on the data processed in step 1, integrate all features, and perform normalization and encoding processing; Step 2-2: Divide the features into two parts according to the characteristics of the extracted features, namely, flight static attribute features and environmental dynamic attribute features; store the processed and statistically processed data, and build a flight static attribute feature database and an environmental dynamic attribute feature database.

4. The method for measuring the interpretable similarity of airport surface operation scenarios according to claim 1 is characterized by: The step 3 specifically includes the following steps: Step 3-1: Construct the static attribute input structure of the flight. Most of the static attribute features of the flight are categorical features and require additional encoding processing. The processed vector is called the static attribute vector θ i , all sample target scenes and their respective candidate similar scene sets are stacked into an input structure, and the final input data format is s×n×2×α cat , where s represents the number of sample scenes, α cat is the dimension of the processed static attribute vector, and n is the number of candidate scenes. Under this construction method, the static attributes of all sample target scenes and each candidate similar scene are compared. Step 3-2: Construct the environment dynamic attribute input structure. Combine the environment dynamic attribute features of all scenes at the same time in the previous few days to form a multi-time scale environment dynamic attribute input vector for one day, called the dynamic attribute vector θ e , process the vectors of all scene data according to the above steps; combine all scenes with their respective candidate similar scene sets to finally form the input data.

5. An interpretable similarity measurement system for airport surface operation scenarios, characterized by: The method for implementing any one of claims 1 to 4 comprises: Airport surface operation scenario data processing module, used to collect data from multiple data sources and perform pre-processing; The surface operation scenario feature extraction and grouping module is used to extract secondary features according to data type and classify all features into flight static attribute features and environment dynamic attribute features according to their characteristics; The scene dynamic interpretable similarity calculation module integrates the input data and calculates the dynamic interpretable similarity between the target scene and the historical operation scene.

Images