An Airport Capacity Prediction Method Based on Causal Forest

Abstract

This invention belongs to the field of airport capacity prediction technology, specifically relating to an airport capacity prediction method based on causal forests. The steps include: collecting historical operational data of the target airport to form basic data and preprocessing it; classifying the meteorological types corresponding to the airport meteorological conditions into favorable and unfavorable meteorological conditions; training a causal forest model; obtaining the individual treatment effect for each unfavorable meteorological type and calculating the average treatment effect under each unfavorable meteorological type; obtaining the airport's theoretical capacity; given airport meteorological condition data for a certain prediction period, using a classification algorithm to obtain the meteorological type corresponding to that prediction period, and estimating the airport capacity for that prediction period based on the airport's theoretical capacity and the average treatment effect corresponding to the meteorological type in that prediction period. This invention can estimate the causal effect of treatment variables on the target variable under multiple covariate disturbances, thereby providing more accurate and stable capacity prediction results.

Description

Technical Field

[0001] This invention belongs to the field of airport capacity prediction technology, specifically relating to an airport capacity prediction method based on causal forest. Background Technology

[0002] Airport capacity forecasting plays a crucial role in air traffic flow management. Accurate forecasting of airport capacity under different weather conditions can provide decision support for air traffic control departments, airports, and airlines, enabling more precise flow allocation and resource scheduling. Accurate capacity forecasting results can improve airport resource utilization while ensuring air traffic safety and efficient operation, ultimately enhancing passenger experience and aviation operational efficiency.

[0003] Existing research typically employs airport capacity prediction methods based on traditional machine learning, estimating airport capacity using actual takeoffs and landings from historical data. These methods suffer from three main problems, leading to low accuracy in capacity prediction: First, under the same weather conditions, airport capacity may be the same in two time periods, but actual takeoffs and landings are affected by the number of planned flights; when the number of planned flights differs, changes in actual takeoffs and landings result in inaccurate capacity estimates. Second, airport capacity is affected not only by its own operations but also by interference from surrounding airports. The busyness of neighboring airports or flight adjustments directly impacts an airport's actual capacity, but traditional methods often fail to consider these external factors, leading to instability and bias in the prediction results. Third, while traditional machine learning methods can effectively handle large amounts of data, they typically rely on correlation for prediction, making it difficult to reveal causal relationships between variables. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide an airport capacity prediction method based on causal forest. By using causal forest, it is possible to estimate the causal effect of the treatment variable (meteorological conditions) on the target variable (airport capacity) under multiple covariate disturbances, thereby providing more accurate and stable capacity prediction results.

[0005] To achieve the above objectives, this invention provides an airport capacity prediction method based on causal forests, comprising the following steps:

[0006] S1. Collect historical operational data of the target airport, including the actual number of takeoffs and landings, airport weather conditions, number of planned flights, and delay status of related airports for each time period, to form basic data;

[0007] S2. Preprocess the basic data, including filling in missing values ​​and standardization.

[0008] S3. For airport meteorological condition data, use clustering methods to discretize the data and divide the meteorological types corresponding to the airport meteorological condition data into favorable meteorological conditions and unfavorable meteorological types.

[0009] S4. Using the discretized meteorological conditions as the processing variable, the actual number of takeoffs and landings at the airport as the target variable, and the number of planned flights at the airport and the delay status of related airports as covariates, train a causal forest model.

[0010] S5. Using the causal forest model, obtain the individual treatment effect for each adverse weather type and calculate the average treatment effect for each adverse weather type.

[0011] S6. Obtain the airport's theoretical capacity based on historical airport operation data;

[0012] S7. Given airport meteorological condition data for a certain period to be predicted, use a classification algorithm to obtain the meteorological type corresponding to the period to be predicted. Based on the airport's theoretical capacity and the average treatment effect corresponding to the meteorological type in the period to be predicted, estimate the airport capacity for the period to be predicted.

[0013] As a preferred embodiment of the present invention, in S1, a time range is set, the time period length is set to 1 hour, and historical operational data of the target airport in each time period within the time range is collected. The airport meteorological condition data includes weather conditions, visibility, cloud base height, temperature, air pressure, precipitation, wind direction, and wind speed. In the delay status of related airports, related airports include other airports in the same terminal area as the target airport, other airports within 250km of the target airport, and airports whose planned flights to and from the target airport account for more than 5% of the total planned flights of the target airport. The delay status is the average delay time.

[0014] Data from each time period forms a record, and all records together form the basic data.

[0015] As a preferred embodiment of the present invention, in step S2, the preprocessing involves filling in missing fields in the basic data using any of the following methods:

[0016] Method 1: Fill using the average of the preceding and following values;

[0017] Method 2: For missing fields in airport actual takeoffs and landings, airport meteorological conditions, and airport planned flight numbers, an LSTM-GRU hybrid model is used to predict missing values. Historical data from 12 hours before and after the missing period is used as input, and the predicted value of the missing field is output. For missing data in the delay status of related airports, the delay data of related airports that are not missing are weighted and fused using the proportion of flights between the target airport and related airports and geographical distance. The weighted fused value is used as the imputation value for the missing field.

[0018] After filling, the airport meteorological condition data is standardized using the Z-score method.

[0019] As a preferred embodiment of the present invention, the discretization process using clustering methods in S3 is as follows:

[0020] S3.1. Determine the number of clusters K by combining the silhouette coefficient and empirical methods;

[0021] S3.2 Execute the k-means clustering process to divide the airport meteorological condition data into K clusters. ;

[0022] S3.3 Discretize each data point in the airport meteorological condition data. If a data point belongs to a cluster... The meteorological conditions of this data point are then labeled as meteorological type k. Where k=0 corresponds to the first cluster This indicates favorable weather conditions, meaning no intervention or treatment is required. Corresponding to the 2nd to the Kth clusters This indicates an unfavorable weather type.

[0023] As a preferred embodiment of the present invention, the process of training the causal forest model in S4 is as follows:

[0024] S4.1 Replace the meteorological condition data in the preprocessed basic data with the corresponding meteorological type k to form the airport capacity prediction dataset D;

[0025] S4.2 From the airport capacity prediction dataset D, select weather types k=0 and The data, among which This is an index for unfavorable weather types, with a value of ;

[0026] S4.3, another That is, selecting data with weather types k=0 and k=1 from the airport capacity prediction dataset D to form a subset dataset. ,from 60% of the data is randomly selected to build the tree, and this is denoted as the training dataset. The remaining 40% of the data is used for estimation, denoted as the estimation dataset. ,definition As the processing variable in the causal forest model, the target variable Y is defined as... The actual number of takeoffs and landings at the airport during the corresponding time period is defined as covariate X. The number of scheduled flights and the average delay time at relevant airports for the corresponding time period;

[0027] S4.4, Call the Python causal machine learning library causalml, to... Using the fitting function of the causal random forest regressor as input, the causal forest model is trained to obtain... The causal forest model of time is denoted as ;

[0028] S4.5, will Set to the following order Repeat the data selection and training process in S4.3-S4.4 to form... The causal forest model of time is denoted as .

[0029] As a preferred embodiment of the present invention, in S5, the individual treatment effect for each adverse weather type is obtained as follows: ,use calculate Individual treatment effect of the i-th sample , ;against Similarly, the individual treatment effect can be obtained. ;

[0030] Calculate the average treatment effect under each adverse weather type:

[0031] ;

[0032] In the formula, For the corresponding The average treatment effect of the values; For the corresponding The individual treatment effect of the i-th sample; For the corresponding An estimated dataset of values.

[0033] As a preferred embodiment of the present invention, in S6, an evaluation method based on historical statistical data is adopted, with the envelope interval set to 98%. Based on the airport's historical operation data, the historical peak service flights are obtained and used as the airport's theoretical capacity.

[0034] As a preferred embodiment of the present invention, in step S7, for a given airport meteorological condition data for a certain period to be predicted, the XGBoost classification algorithm is used to obtain the meteorological type k corresponding to the period to be predicted, and the airport capacity for the period to be predicted is estimated.

[0035] ;

[0036] In the formula, This is the estimated airport capacity for the period to be predicted; This represents the airport's theoretical capacity. This represents the average treatment effect corresponding to the weather type for the forecast period. When k=0, The value is 0.

[0037] The beneficial effects of this invention are:

[0038] This invention comprehensively reflects various factors affecting airport capacity by collecting and processing historical airport operational data, including actual takeoffs and landings, weather conditions, planned flight numbers, and related airport delay information. The clustering and discretization processing of weather condition data categorizes different weather conditions into multiple types, thereby enabling refined capacity prediction.

[0039] This invention utilizes a causal forest model to reveal the causal effect of meteorological conditions on airport capacity and accurately assess the changing patterns of airport capacity under different meteorological conditions. It considers multiple covariates (such as the number of planned flights and the delay status of other airports) and effectively controls these interfering factors through the causal forest model. This feature enables the model to provide stable and accurate prediction results even in complex air traffic flow environments with multiple factors and interferences. Thus, it can accurately and stably predict airport capacity under the influence of multiple factors and variables, providing more reliable technical support for air traffic flow management. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating the principle of this invention;

[0041] Figure 2 This is a schematic diagram of the data partitioning and training process in Embodiment 1 of the present invention;

[0042] Figure 3 This is a schematic diagram of the individual treatment effect calculation process in Embodiment 1 of the present invention. Detailed Implementation

[0043] The embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0044] Example 1: As Figure 1 As shown, an airport capacity prediction method based on causal forests includes the following steps:

[0045] S1. Collect historical operational data of the target airport, including the actual number of takeoffs and landings, airport weather conditions, number of planned flights, and delay status of related airports for each time period, to form basic data;

[0046] S2. Preprocess the basic data, including filling in missing values ​​and standardization.

[0047] S3. For airport meteorological condition data, use clustering methods to discretize the data and divide the meteorological types corresponding to the airport meteorological condition data into favorable meteorological conditions and unfavorable meteorological types.

[0048] S4. Using the discretized meteorological conditions as the processing variable, the actual number of takeoffs and landings at the airport as the target variable, and the number of planned flights at the airport and the delay status of related airports as covariates, train a causal forest model.

[0049] S5. Using the causal forest model, obtain the individual treatment effect for each adverse weather type and calculate the average treatment effect for each adverse weather type.

[0050] S6. Obtain the airport's theoretical capacity based on historical airport operation data;

[0051] S7. Given airport meteorological condition data for a certain period to be predicted, use a classification algorithm to obtain the meteorological type corresponding to the period to be predicted. Based on the airport's theoretical capacity and the average treatment effect corresponding to the meteorological type in the period to be predicted, estimate the airport capacity for the period to be predicted.

[0052] In S1, a time range is set, and the time period length is set to 1 hour. Historical operational data of the target airport in each time period within the time range is collected. The airport meteorological condition data includes weather conditions, visibility, cloud base height, temperature, air pressure, precipitation, wind direction, and wind speed. The delay status of related airports includes other airports in the same terminal area as the target airport, other airports within 250km of the target airport, and airports whose planned flights to and from the target airport account for more than 5% of the total planned flights of the target airport. The delay status is the average delay time.

[0053] Data from each time period forms a record, and all records together form the basic data.

[0054] For example, for a certain day from 09:00 to 10:00, the corresponding record contains the following data:

[0055] Actual number of takeoffs and landings at the airport: 38 flights / hour;

[0056] Airport meteorological conditions: visibility 8km, cloud base 1200m, no precipitation, wind speed 3m / s;

[0057] Planned number of flights at the airport: 42;

[0058] Average delay time at relevant airports: 8 minutes.

[0059] In S2, the preprocessing involves filling in missing fields in the basic data using any of the following methods:

[0060] Method 1: Fill using the average of the preceding and following values;

[0061] Method 2: For missing fields in airport actual takeoffs and landings, airport meteorological conditions, and airport planned flight numbers, an LSTM-GRU hybrid model is used to predict missing values. Historical data from 12 hours before and after the missing period is used as input, and the predicted value of the missing field is output. For missing data in the delay status of related airports, the delay data of related airports that are not missing are weighted and fused using the proportion of flights between the target airport and related airports and geographical distance. The weighted fused value is used as the imputation value for the missing field.

[0062] After filling, the airport meteorological condition data is standardized using the Z-score method.

[0063] In S3, the process of discretization using clustering methods is as follows:

[0064] S3.1. Determine the number of clusters K by combining the silhouette coefficient and empirical methods;

[0065] S3.2 Execute the k-means clustering process to divide the airport meteorological condition data into K clusters. ;

[0066] S3.3 Discretize each data point in the airport meteorological condition data. If a data point belongs to a cluster... The meteorological conditions of this data point are then labeled as meteorological type k. Where k=0 corresponds to the first cluster This indicates favorable weather conditions, meaning no intervention or treatment is required. Corresponding to the 2nd to the Kth clusters This indicates an unfavorable weather type.

[0067] Favorable weather conditions will not negatively suppress airport capacity, allowing it to operate at its theoretical capacity. Unfavorable weather types, including heavy fog, heavy rain, strong winds, and thunderstorms, will lead to a decrease in airport capacity.

[0068] In S3.1, the process for determining the number of clusters K using a combination of silhouette coefficient and empirical methods is as follows:

[0069] First, based on the aviation industry's experience and understanding of the impact of meteorological conditions on airport operations, a reasonable range for K is defined, specifying that it must include one type of favorable weather (k=0, no intervention, no capacity suppression, such as clear skies, high visibility, light winds, etc.). Then, considering common unfavorable weather types encountered in actual operations (such as heavy fog, heavy rain, strong winds, thunderstorms, blizzards, etc.), the candidate range for K is initially set to 3-8 (avoiding K less than 3 leading to insufficient differentiation of unfavorable weather, and greater than 8 leading to over-segmentation without operational significance). Next, within this candidate range, each potential K value is iterated, and k-means clustering is performed on the preprocessed meteorological data (standardized visibility, wind speed, precipitation, etc.). For each K value, the overall average profile coefficient s is calculated according to the formula... Where a is the average distance between a sample and other samples in the same cluster (cohesion), and b is the average distance between a sample and the nearest sample from a different cluster (separation). The closer the coefficient is to 1, the better the clustering effect. The K with the largest coefficient is selected as the initial optimal solution. Finally, the clustering results corresponding to K are verified again through experience to see if they conform to business logic. For example, check whether heavy fog (visibility < 1km) and heavy rain (precipitation > 20mm / h) are divided into different clusters, and whether favorable weather is in a separate cluster without mixed unfavorable weather data. If the verification is successful, the value of K is finally determined. If there is data within a cluster that does not conform to business understanding (such as light fog and clear weather being mistakenly classified into the same cluster), the candidate range of K is fine-tuned (such as expanding it to 4~9), and the above contour coefficient calculation and empirical verification steps are repeated until the number of clusters K that both meets the quantitative clustering quality (high contour coefficient) and adapts to the actual operation of the airport is obtained.

[0070] In S4, the process of training the causal forest model is as follows:

[0071] S4.1 Replace the meteorological condition data in the preprocessed basic data with the corresponding meteorological type k to form the airport capacity prediction dataset D;

[0072] S4.2 From the airport capacity prediction dataset D, select weather types k=0 and The data, among which This is an index for unfavorable weather types, with a value of ;

[0073] S4.3, another That is, selecting data with weather types k=0 and k=1 from the airport capacity prediction dataset D to form a subset dataset. ,from 60% of the data is randomly selected to build the tree, and this is denoted as the training dataset. The remaining 40% of the data is used for estimation, denoted as the estimation dataset. ,definition As the processing variable in the causal forest model, the target variable Y is defined as... The actual number of takeoffs and landings at the airport during the corresponding time period is defined as covariate X. The number of scheduled flights and the average delay time at relevant airports for the corresponding time period;

[0074] S4.4, Call the Python causal machine learning library causalml, to... Using the fitting function of the Causal Random Forest Regressor as input, the causal forest model is trained to obtain... The causal forest model of time is denoted as ;

[0075] S4.5, will Set to the following order Repeat the data selection and training process in S4.3-S4.4 to form... The causal forest model of time is denoted as The process of obtaining the complete causal forest model is as follows: Figure 2 As shown (including the data partitioning and training process).

[0076] In S5, such as Figure 3 As shown, the individual treatment effect for each adverse weather type is obtained, targeting ,use calculate Individual treatment effect of the i-th sample , ;against Similarly, the individual treatment effect can be obtained. ;

[0077] Calculate the average treatment effect under each adverse weather type:

[0078] ;

[0079] In the formula, For the corresponding The average treatment effect of the values; For the corresponding The individual treatment effect of the i-th sample; For the corresponding An estimated dataset of values.

[0080] In S6, an evaluation method based on historical statistical data is adopted, with the envelope interval (also known as the envelope line or capacity envelope) set at 98%. Based on the airport's historical operation data, the historical peak service flights are obtained and used as the airport's theoretical capacity.

[0081] In S7, for a given airport meteorological condition data for a certain period to be predicted, the XGBoost classification algorithm is used to obtain the meteorological type k corresponding to that period, and the airport capacity for that period is estimated.

[0082] ;

[0083] In the formula, This is the estimated airport capacity for the period to be predicted; This represents the airport's theoretical capacity. This represents the average treatment effect corresponding to the weather type for the forecast period. When k=0, The value is 0.

[0084] Example 2: Based on Example 1, Pearson correlation coefficient is introduced into Method 2 of preprocessing to form an improved Method 2:

[0085] First, the correlation between each feature and the missing field is calculated using the Pearson correlation coefficient. Features with an absolute correlation value greater than 0.6 are selected as input variables. Then, the filtered feature data from 12 hours before and after the missing period are input into the LSTM-GRU hybrid model, and the predicted value of the missing field is output. For the missing data in the delay status of relevant airports, an airport operation priority coefficient is introduced on the basis of the original weighted fusion logic. The priority coefficient is set according to the airport level and flight throughput. The priority coefficient, together with the proportion of flights and geographical distance, constitutes a weighting factor. The weighted fusion value is calculated as the imputation value of the missing field.

[0086] Furthermore, the segmented Z-score method is used to standardize the airport meteorological data. The meteorological data is divided into distance (visibility, cloud base height), intensity (precipitation, wind speed), and numerical (temperature, air pressure) categories according to feature type. The Z-score value is calculated separately for each category of features to avoid mutual interference between features of different magnitudes.

[0087] By filtering features based on relevance, the interference of redundant inputs on the LSTM-GRU model is reduced, improving the accuracy of missing value prediction. Introducing airport operation priority coefficients makes the weighted fusion of relevant airport delay data more closely reflect actual operational scenarios, resulting in more accurate completion of missing values. During Z-score standardization, segmentation based on feature type addresses the distortion problem after standardization of meteorological features of different magnitudes, improving the effectiveness of subsequent clustering and model training.

[0088] Example 3: The difference between this example and Example 1 is that in S3.1, the number of clusters K is determined by integrating the profile coefficient, elbow rule, and airport operational meteorological threshold constraints. Specifically:

[0089] Set meteorological threshold constraints for airport operations, such as visibility below 800 meters as the critical value for low visibility and precipitation greater than 25 mm as the critical value for heavy rain. Filter the number of clusters that meet the meteorological thresholds, and then combine the maximum profile coefficient with the elbow rule inflection point to determine the optimal K value.

[0090] The weighted k-means++ clustering process is executed, assigning differentiated weights to different features in the airport meteorological condition data. Weight coefficients are set for visibility, wind speed, precipitation, and other meteorological features. For example, the weight coefficients for visibility, wind speed, and precipitation are set to 0.3, 0.25, and 0.2, respectively, while the remaining weight of 0.25 is evenly distributed among the weight coefficients of other meteorological features. k-means++ clustering avoids getting trapped in local optima by optimizing the selection of initial cluster centers.

[0091] Each data point in the airport meteorological condition data is discretized. If a data point belongs to a cluster, the meteorological condition of that data point is marked as meteorological type k. K=0 corresponds to the first cluster, which represents favorable meteorological conditions. K=2 to Kth clusters correspond to unfavorable meteorological types. The data within each unfavorable meteorological type cluster meets the preset airport operation meteorological threshold constraints.

[0092] By introducing meteorological threshold constraints for airport operations, the selection of cluster size is made to better reflect the actual operational scenarios of airports, avoiding biases from empirical values. Weighted k-means++ is used instead of ordinary k-means, which not only solves the local optima problem caused by the random initial centers of k-means, but also highlights meteorological factors that have a greater impact on airport capacity through feature weighting, thus improving the accuracy of clustering results.

[0093] Example 4: The difference between this example and Example 1 is that in S6, a hierarchical statistical evaluation method combined with dynamic correction is used to obtain the theoretical capacity of the airport, specifically:

[0094] S6.1. The airport's historical operation data is divided into two layers according to time period type (morning peak, midday off-peak, evening peak, nighttime off-peak) and season type (Spring Festival peak season, summer peak season, shoulder season, off-season) to obtain multiple layered subset datasets;

[0095] S6.2 Set a 98% envelope interval for each hierarchical subset, calculate the historical peak service flights within each hierarchical subset, and record them as hierarchical peak flights;

[0096] S6.3. Based on the proportion of each hierarchical subset in the total annual runtime, set the hierarchical weight coefficient and calculate the initial theoretical capacity using weighted average.

[0097] S6.4. An adverse weather correction coefficient is introduced. The adverse weather correction coefficient is determined based on the ratio of the airport's actual capacity under historical adverse weather conditions (such as blizzards and strong typhoons) to the number of flights during normal peak periods. This coefficient corrects the initial theoretical capacity and ultimately yields the airport's theoretical capacity.

[0098] Stratifying data by time period and season allows theoretical capacity to better reflect actual service capabilities in different operational scenarios, avoiding interference from off-peak data in peak capacity assessments. Incorporating adverse weather correction factors enhances the applicability of theoretical capacity under specific conditions, thereby optimizing the accuracy of final capacity forecasts.

[0099] Example 5: An airport capacity prediction device based on causal forest, comprising:

[0100] One or more processors;

[0101] Memory, used to store one or more computer programs;

[0102] When one or more programs are executed by one or more processors, the one or more processors perform the method of any one of embodiments 1-4.

[0103] Example 6: A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method of any one of Examples 1-4.

Claims

1. An airport capacity prediction method based on causal forest, characterized in that... Includes the following steps: S1. Collect historical operational data of the target airport, including the actual number of takeoffs and landings, airport weather conditions, number of planned flights, and delay status of related airports for each time period, to form basic data; S2. Preprocess the basic data, including filling in missing values ​​and standardization. S3. For airport meteorological condition data, use clustering methods to discretize the data and divide the meteorological types corresponding to the airport meteorological condition data into favorable meteorological conditions and unfavorable meteorological types. S4. Using the discretized meteorological conditions as the processing variable, the actual number of takeoffs and landings at the airport as the target variable, and the number of planned flights at the airport and the delay status of related airports as covariates, train a causal forest model. S5. Using the causal forest model, obtain the individual treatment effect for each adverse weather type and calculate the average treatment effect for each adverse weather type. S6. Obtain the airport's theoretical capacity based on historical airport operation data; S7. Given airport meteorological condition data for a certain period to be predicted, use a classification algorithm to obtain the meteorological type corresponding to the period to be predicted, and estimate the airport capacity for the period to be predicted based on the airport's theoretical capacity and the average treatment effect corresponding to the meteorological type in the period to be predicted. In S4, the process of training the causal forest model is as follows: S4.1 Replace the meteorological condition data in the preprocessed basic data with the corresponding meteorological type k to form the airport capacity prediction dataset D; S4.2 From the airport capacity prediction dataset D, select weather types k=0 and The data, among which This is an index for unfavorable weather types, with a value of K is the number of clusters; S4.3, another That is, selecting data with weather types k=0 and k=1 from the airport capacity prediction dataset D to form a subset dataset. ,from 60% of the data is randomly selected to build the tree, and this is denoted as the training dataset. The remaining 40% of the data is used for estimation, denoted as the estimation dataset. ,definition As the processing variable in the causal forest model, the target variable Y is defined as... The actual number of takeoffs and landings at the airport during the corresponding time period is defined as covariate X. The number of scheduled flights and the average delay time at relevant airports for the corresponding time period; S4.4, Call the Python causal machine learning library causalml, to... Using the fitting function of the causal random forest regressor as input, the causal forest model is trained to obtain... The causal forest model of time is denoted as ; S4.5, will Set to the following order Repeat the data selection and training process in S4.3-S4.4 to form... The causal forest model of time is denoted as .

2. The airport capacity prediction method based on causal forest according to claim 1, characterized in that: In S1, a time range is set, and the time period length is set to 1 hour. Historical operational data of the target airport in each time period within the time range is collected. The airport meteorological condition data includes weather conditions, visibility, cloud base height, temperature, air pressure, precipitation, wind direction, and wind speed. The delay status of related airports includes other airports in the same terminal area as the target airport, other airports within 250km of the target airport, and airports whose planned flights to and from the target airport account for more than 5% of the total planned flights of the target airport. The delay status is the average delay time. Data from each time period forms a record, and all records together form the basic data.

3. The airport capacity prediction method based on causal forest according to claim 2, characterized in that: In S2, the preprocessing involves filling in missing fields in the basic data using any of the following methods: Method 1: Fill using the average of the preceding and following values; Method 2: For missing fields in airport actual takeoffs and landings, airport meteorological conditions, and airport planned flight numbers, an LSTM-GRU hybrid model is used to predict missing values. Historical data from 12 hours before and after the missing period is used as input, and the predicted value of the missing field is output. For missing data in the delay status of related airports, the delay data of related airports that are not missing are weighted and fused using the proportion of flights between the target airport and related airports and geographical distance. The weighted fused value is used as the imputation value for the missing field. After filling, the airport meteorological condition data is standardized using the Z-score method.

4. The airport capacity prediction method based on causal forest according to claim 1, characterized in that, In S3, the process of discretization using clustering methods is as follows: S3.

1. Determine the number of clusters K by combining the silhouette coefficient and empirical methods; S3.2 Execute the k-means clustering process to divide the airport meteorological condition data into K clusters. ; S3.3 Discretize each data point in the airport meteorological condition data. If a data point belongs to a cluster... The meteorological conditions of this data point are then labeled as meteorological type k. Where k=0 corresponds to the first cluster This indicates favorable weather conditions, meaning no intervention or treatment is required. Corresponding to the 2nd to the Kth clusters This indicates an unfavorable weather type.

5. The airport capacity prediction method based on causal forest according to claim 1, characterized in that: In S5, the individual treatment effect for each adverse weather type is obtained as follows: ,use calculate Individual treatment effect of the i-th sample , ;against Similarly, the individual treatment effect can be obtained. ; Calculate the average treatment effect under each adverse weather type: ； In the formula, For the corresponding The average treatment effect of the values; For the corresponding The individual treatment effect of the i-th sample; For the corresponding An estimated dataset of values.

6. The airport capacity prediction method based on causal forest according to claim 4, characterized in that, In S6, an evaluation method based on historical statistical data is adopted, with the envelope interval set to 98%. Based on the airport's historical operation data, the historical peak service flights are obtained and used as the airport's theoretical capacity.

7. The airport capacity prediction method based on causal forest according to claim 4, characterized in that: In step S7, for a given airport meteorological condition data for a specific time period to be predicted, the XGBoost classification algorithm is used to obtain the meteorological type k corresponding to that time period, and the airport capacity for that time period is estimated. ； In the formula, This is the estimated airport capacity for the period to be predicted; This represents the airport's theoretical capacity. This represents the average treatment effect corresponding to the weather type for the forecast period. When k=0, The value is 0.

Images