A Strategic Delay Prediction Method for Airport Flight Scheduling Based on Delay Combination and Delay Splitting

Abstract

This invention discloses a strategic delay prediction method for airport flight schedules based on delay decomposition and combination, relating to the field of air traffic control and planning technology. It enables timely and effective prediction of arrival and departure delays for airport flights. The invention includes: a strategic delay prediction method for airport flight schedules based on delay decomposition and combination, used to obtain an airport capacity probability map using the airport flight schedule and historical flight operation data, and to obtain predicted values ​​for airport flight arrival and departure delays based on the airport flight schedule, airport operation mode modeling, and airport capacity probability map. This invention is applicable to the evaluation of the rationality and effectiveness of airport flight time / schedule arrangement during the strategic phase.

Description

Technical Field

[0001] This invention relates to the field of air traffic control and planning technology, and in particular to a method for predicting airport flight schedule delays based on delay splitting and combining. Background Technology

[0002] With the increase in total civil aviation traffic, flight delays have once again become a major constraint on the development of the civil aviation industry. As the demand in the aviation market continues to grow, airports are showing signs of insufficient capacity, and flight delays have become unavoidable due to the combined effects of various factors.

[0003] Globally, flight delays will be a persistent problem and a major obstacle to the development of the civil aviation industry. Therefore, predicting flight delay levels, adjusting flight schedules accordingly, and making strategic and tactical preparations and early warnings for flight delays have become crucial issues for the development of the civil aviation industry. This is of great significance for optimizing flight schedules and improving operational efficiency, enabling the civil aviation industry to optimize and modify existing flight plans earlier and propose corresponding strategic plans and countermeasures.

[0004] Among these factors, how to predict delays in airport flight schedules in a timely and effective manner has become a topic that needs to be studied. Summary of the Invention

[0005] The embodiments of the present invention provide a method for predicting airport flight schedule delays based on delay splitting and combining, which can predict the delays of airport flight schedules for arrival and departure in a timely and effective manner.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0007] Airport flight scheduling strategy delay prediction methods based on delay splitting and combining include:

[0008] S1. Utilize airport flight operation data to obtain clustering results for airport arrival and departure flights. The flight operation data includes: calculated off-block time data and actual landing time data. Calculated off-block time (SOBT) is a technical term referring to the time after an aircraft has passed its planned off-block time (SOBT), the target off-block time (TOBT) calculated based on the airport's minimum capacity, and the COBT calculated by the system after inputting TOBT into the traffic flow management system. In practical applications, K-means clustering is used to cluster the off-block traffic overview and the actual landing traffic overview for airport arrival and departure flight operation data, obtaining cluster centers under various K values ​​as the clustering results.

[0009] S2. Using the clustering results of the airport's arriving and departing flights, and through clustering result evaluation indicators and confidence interval judgment methods, obtain the airport's arrival and departure capacity probability map. The capacity probability map is represented as a set of capacity periodic curves, used to generate random airport departure and arrival capacities at different time windows. Specifically, the capacity probability map is a set of fused and separated broken lines in the same coordinate system. The fused broken lines represent fused scenarios formed by the fusion of different scenarios, and the independent broken lines represent independent scenarios. Each scenario has a corresponding formation probability, the probability of which is equal to the proportion of its corresponding cluster in the number of observations. The probability of a fused scenario is the difference between 1 and the probabilities of the other independent scenarios. Within a specific time window, when an independent scenario is selected, the arrival and departure capacity is the traffic volume value corresponding to the independent scenario within that specific time window. When a fused scenario is selected, the arrival and departure capacity is the weighted value of the traffic volume values ​​corresponding to each independent scenario within the fused scenario within that specific time window, where the weight is the quotient of the formation probability of each independent scenario in the fused scenario and the sum of their formation probabilities.

[0010] S3. Using the aforementioned arrival and departure capacity probability map, establish a prediction model for the airport's inherent arrival and departure delays. This model can break down airport flight schedule delays into inherent delays and flow management delays for analysis. Inherent delays are caused by the combined effects of the airport's own capacity, airport facility structure, and operational mode. Flow management delays are delays caused by human error or unforeseen circumstances during airport flow management. Both types of delays ultimately lead to delays and lags in airport flight schedules, and therefore are recorded and archived during the airport's daily management process.

[0011] S4. Utilize airport traffic management data to establish an airport flow control delay prediction model. The airport traffic management data includes: flight departure and destination airports, flight target wheel chock removal time data, and flight wheel chock calculation time data. Specifically, airport traffic generally refers to the actual flight departure traffic. The actual flight departure traffic overview is expressed as follows: dividing the statistical period into 15-minute intervals, counting the actual number of flights taking off within each time window, as the actual flight departure traffic overview for that time window. The actual wheel chock traffic overview is expressed as follows: dividing the statistical period into 15-minute intervals, counting the actual number of flights using wheel chocks within each time window, as the actual wheel chock traffic overview for that time window.

[0012] S5. Based on the arrival and departure original delay prediction model and the airport flow control delay prediction model, and using the airport flight delay propagation judgment formula, establish an airport flight schedule delay prediction model.

[0013] S6. Based on the airport flight schedule delay prediction model, perform prediction processing on the airport flight schedule to be processed, and obtain the arrival and departure delay prediction results of the airport flight schedule to be processed.

[0014] Specifically, S3 also includes: before establishing the airport's native arrival and departure delay prediction model using the arrival and departure capacity probability map, loading a queuing network model corresponding to the airport's operation mode, wherein the airport operation mode includes: runway service mode, airport departure operation mode, and airport arrival operation mode; specifically, the queuing network model corresponding to the airport operation mode can be divided into various types according to function, such as: flight flow characteristic model, runway takeoff time interval distribution model, runway landing time interval distribution model, parking stand wheel chock removal time interval distribution model, aircraft taxiing mode model, etc.

[0015] Furthermore, the arrival / departure native delay prediction model includes: a departure native delay prediction model and an arrival native delay prediction model; in the departure native delay prediction model, the airport departure capacity is used to predict the aircraft stand waiting time and the queuing network model is used to predict the aircraft runway head waiting time; in the arrival native delay prediction model, the airport arrival capacity is used to predict the aircraft arrival taxiing delay and the queuing network model is used to predict the aircraft terminal area delay time. The departure native delay prediction model is defined as follows: when the aircraft is at the parking stand, the key parameter limiting the aircraft's push-out is the departure capacity, which originates from the random capacity scenario generated by the departure capacity probability map. After the aircraft has completed push-out and taxied to the runway head, the key parameter limiting the aircraft's takeoff is the runway head service time interval; the departure native delay prediction model is a simulation system for each departing aircraft from the calculation of the wheel chock removal time to the actual takeoff time. The arrival native delay prediction model is expressed as follows: when an aircraft enters the terminal area earliest, the terminal area delay is predicted by simplifying the queuing network model. After the aircraft lands, the key parameter limiting the aircraft's entry into the runway is the arrival capacity, which is derived from the random capacity scenario generated by the arrival capacity probability map. The arrival native delay prediction model is a simulation system that calculates the terminal area delay for each arriving aircraft by shifting the time forward from the planned landing time to the actual wheel chute time.

[0016] Specifically, S4 also includes: before establishing the airport flow control delay prediction model, using airport flow management data to perform airport flow control delay correlation analysis; and establishing the flow control delay prediction model based on the results of the airport flow control delay correlation analysis.

[0017] The airport flow control delay correlation analysis includes: obtaining quantified values ​​of the correlation between the magnitude of flow control delays and the volume of traffic in different cities over different time periods. If the quantified correlation value is higher than a threshold, the flow control delay prediction model is set as a positive correlation function with the volume of traffic. If the quantified correlation value is less than or equal to the threshold, the flow control delay prediction model is obtained by statistically analyzing the magnitude of flow control delays and the probability of flow control delays occurring. The probability of flow control delays includes the ratio of the number of flights subject to flow control in a specific city to the total number of flights within a specific time period. For example, using airport traffic management data, an airport flow control delay prediction model is established through airport flow control delay correlation analysis. Based on the results of the airport flow control delay correlation analysis, the direction of the airport traffic flow and the flow control delay model selection is determined. Specifically, the airport flow control delay correlation analysis involves testing the correlation between the magnitude of flow control delays and the volume of flights in different cities over different time periods. If the correlation is high, the flow control delay prediction model is a positive correlation function with the volume of flights; otherwise, the flow control delay prediction model is obtained by statistically analyzing the magnitude of flow control delays and the probability of flow control delays occurring. The probability of flow control delays occurring is expressed as the ratio of the number of flights subject to flow control in a specific city to the total number of flights within a specific time period. For example, for the correlation test between the magnitude of flow control delays and the volume of flights in different cities over different time periods, if the correlation is higher than a threshold, the flow control delay prediction model is a positive correlation function with the volume of flights; otherwise, the flow control delay prediction model is obtained by statistically analyzing the average magnitude of flow control delays and the probability of flow control delays occurring. The threshold is generally 0.5, and the probability of flow control delays occurring is expressed as the ratio of the number of flights subject to flow control in a specific city to the total number of flights within a specific time period. Specifically, the magnitude of the flow control delay is the sum of the flow control delays for a specific city within a specific time period; the average magnitude of the flow control delay is the sum of the magnitude of the flow control delay and the number of flights affected by flow control.

[0018] Specifically, S5 includes: generating arrival delay prediction values ​​for the airport flight schedule using the arrival native delay prediction model, and then converting these arrival delay prediction values ​​into departure propagation delay prediction values; generating flight flow control delays using the flow control delay prediction model, and then generating departure delay prediction values ​​for the flight schedule using the departure native delay prediction model. For example, the flight schedule includes at least: time, aircraft registration number, planned wheel chock removal time, planned wheel chock placement time, departure airport, and destination airport. The airport flight delay propagation determination formula includes: aircraft registration number, flight number, aircraft type, actual wheel chock placement time and planned wheel chock placement time of the preceding arriving flight, planned wheel chock removal time of the subsequent departing flight, and the aircraft's shortest turnaround time.

[0019] Furthermore, in S5, it also includes: comparing the difference between the turnaround time and the minimum turnaround time of a specific flight at the airport with the arrival delay of the specific flight; if the former is greater, it is determined that the arrival delay of the specific flight will not propagate to subsequent flights; otherwise, the subsequent flights of the specific flight will experience delay propagation. The magnitude of the propagated delay is equal to the difference between the difference and the arrival delay of the flight. For example, the airport delay propagation determination formula includes: comparing the difference between the turnaround time and the minimum turnaround time of a specific flight at the airport with the arrival delay of the flight; if the former is greater, the arrival delay of the flight will not propagate to subsequent flights; otherwise, the subsequent flights of the flight will experience delay propagation, the magnitude of which is equal to the difference between the difference and the arrival delay of the flight. Specifically, the arrival delay of the flight is represented as AD. i =AIBT i –SIBT i AIBT i This refers to the actual turnaround time of the flight, SIBT. i This refers to the scheduled turnaround time for a flight; the difference between the turnaround time of a specific flight at the airport and the minimum turnaround time is the RDCT. i =SOBT i+1 –SIBT i –MTTT i SOBT i+1 It is the scheduled wheel-shifting time for subsequent flights, MTTT i It is the shortest turnaround time for the preceding flight; the magnitude of the delay propagation is DP. i =AD i -RDCT i If DP i If the value is less than 0, there is no delay propagation; otherwise, the target wheel-off time for subsequent flights is TOBT. i+1 =DP i +SOBT i+1 .

[0020] Specifically, in S6, the airport flight schedules to be processed include: the date of the airport flight to be processed, the departure airport and destination airport of the airport flight to be processed, the time of the flight schedule's shift change and the time of the flight schedule's shift change. The predicted arrival and departure delays of the airport flight schedules to be processed include: average arrival delay, average flow control delay of departure flights, average departure delay, and average total departure delay.

[0021] The average delay of inbound flights is the ratio of total inbound delay to total number of inbound flights; the average flow control delay of departing flights is the ratio of total flow control delay to total number of departing flights; the average delay of departing flights is the ratio of total departing flight delay to total number of departing flights; the average total delay of departing flights is the sum of the average flow control delay of departing flights and the average delay of departing flights.

[0022] Specifically, in S1, the airport arrival and departure flight clustering results include: flight departure traffic clustering results and flight arrival traffic clustering results. The flight departure traffic clustering results are: cluster center results of the airport's actual flight departure traffic using K-means clustering; the flight arrival traffic clustering results are: cluster center results of the airport's actual flight arrival traffic using K-means clustering. For example, the airport arrival and departure flight clustering results include: flight departure traffic clustering results and flight arrival traffic clustering results. The clustering method is K-means clustering, a data partitioning algorithm based on Euclidean distance, which is expressed as: dividing the dataset X into k clusters, where k is a moderately sized positive integer. For a specific cluster, the clustering between observations in the cluster should be less than the distance between that value and observations in other clusters. For each cluster, the center point of all observations in the cluster... k = 1, 2…K, also known as The centroid of cluster k, denoted as: in Let x be the q-th observation in cluster k, and n be the number of observations in cluster k. The K-means algorithm performs Euclidean distance detection between each observation x and each centroid, and assigns x to the cluster corresponding to the nearest centroid. The algorithm terminates when the sum of the distances between each observation and its corresponding cluster is minimized.

[0023] In this embodiment, the clustering object is the airport flight traffic overview. Specifically, the departure traffic overview is represented as follows: taking the actual departure time of each day's departing flights, and counting the number of departing flights in 15-minute time windows, we obtain the traffic overview dataset X = {x1, x2, ..., x...}. n}, where x i The number of departing flights within the i-th time window is represented by: The arrival traffic overview is represented as follows: taking the actual arrival times of daily arrival flights, and statistically analyzing the arrival flight volume in 15-minute time windows to obtain the traffic overview dataset. The clustering result is represented as: in the same coordinate system, a continuous broken line formed by different cluster centers with the time window as the x-axis and the traffic volume within a unit time window as the y-axis.

[0024] The clustering results include clusters ranging from 3 to 16, and these results need to be evaluated and filtered. The evaluation metrics for the clustering results include: silhouette coefficient, variance ratio criterion, and pruning method. For example, the silhouette coefficient is the similarity between a single observation and other observations in its assigned cluster, expressed as: s(x i ) is the silhouette coefficient for each observation. The silhouette coefficient for an observation x is expressed as: Where a(x i b(x) is the average distance from the observation x to all other data points in its cluster. i ) is the average distance between the observation x and the observations in the neighboring clusters; where, Where d(x) j ,x i ) is the distance between observations i and j, and |Ci| is the number of observations within the cluster. Based on all the obtained observations, the silhouette coefficients s(x) are calculated. i ), average profile coefficient An SI value close to 1 indicates that the clusters are compact and well separated, close to 0 indicates overlapping clusters, and close to -1 indicates that there are too many or too few clusters.

[0025] The variance ratio criterion is the ratio of the inter-cluster error variance to the intra-cluster error variance, expressed as: The inter-cluster variance is expressed as: Where n i c is the number of data points in cluster i. i Let c be the centroid of cluster i, and c be the global centroid of all data points. The intra-cluster error variance is expressed as: x represents all observations within cluster i, and c i The centroid of cluster i is denoted by k. A higher CHI value indicates better clustering performance. The pruning method removes unimportant weights from the model, reducing the number of model parameters and computational load without affecting model accuracy. Specifically, if the proportion of a single cluster in the airport's inbound and outbound flight clustering results is less than a certain threshold, that clustering result is not adopted. The pruning method removes unimportant weights from the model, reducing the number of model parameters and computational load without affecting model accuracy. Specifically, it calculates the proportion of each cluster in the overall clustering results. If a cluster with a proportion less than a threshold appears, it indicates that pruning involves clusters with smaller features and less impact, and the value of k is not further increased. The threshold is generally set to 5%-10%.

[0026] Specifically, S2 includes: using the airport arrival and departure flight clustering results, and through clustering result evaluation indicators and confidence interval judgment methods, obtaining an airport arrival and departure capacity probability map; the confidence interval judgment method includes: within a given time window, comparing the cluster center results with the confidence interval boundaries obtained using the airport arrival and departure flight clustering results; if the cluster center results do not exceed the confidence interval boundaries, then it is determined that the cluster centers and the airport arrival and departure flight clustering results are merged. Specifically, using the historical dataset as the overall sample, the clustering results are used for classification, and the sample center, standard deviation, and error limit are calculated for the independent samples in each cluster to obtain the upper and lower limits of the confidence interval. The upper and lower limits of the confidence interval are compared with the cluster centers in the same coordinate system; if the flow rate of the cluster center exceeds the upper and lower limits of the confidence interval within a specific time window, then the cluster and the cluster corresponding to the confidence interval are separated within the specific time window; otherwise, they are merged. Here, the sample center is represented as... |S q | represents the number of observations in the independent sample q. It is the j-th observation in the independent sample q; the sample standard deviation is The sample error limit is Where z_critical(0.995) is the Z-critical value with a confidence level of 99%.

[0027] This invention provides a method for predicting airport flight schedule delays based on delay splitting and combining. It utilizes airport flight operation data to obtain airport arrival and departure flight clustering results. Using these clustering results, and through evaluation indicators and confidence interval judgment methods, it obtains an airport arrival and departure capacity probability map. For the airport operation mode, it analyzes the airport operation queuing network model and uses the airport arrival and departure capacity probability map to establish a native arrival and departure delay prediction model. Using airport flow management data, it conducts airport flow control delay correlation analysis to establish an airport flow control delay prediction model. Based on the native arrival and departure delay prediction model and the airport flow control delay prediction model, it uses an airport flight delay propagation judgment formula to establish an airport flight schedule delay prediction model. Based on the airport flight schedule delay prediction model, it uses airport flight schedules for simulation to obtain predicted values ​​for airport flight schedule arrival and departure delays. This achieves timely and effective prediction of airport flight schedule arrival and departure delays. Attached Figure Description

[0028] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the method flow provided in an embodiment of the present invention;

[0030] Figure 2 This is a schematic diagram of the port arrival and departure capacity probability diagram provided in an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of the original departure delay queuing network model provided in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of the native delay queuing network model for arrivals provided in an embodiment of the present invention;

[0033] Figure 5 This is a schematic diagram of the airport flight delay propagation determination formula provided in an embodiment of the present invention;

[0034] Figure 6 This is a schematic diagram of an airport flight schedule delay prediction system provided in an embodiment of the present invention. Detailed Implementation

[0035] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Embodiments of the present invention will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of the present invention means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0036] This invention provides a method for strategic delay prediction of airport flight schedules based on delay decomposition and combination. The main design idea is as follows: This method utilizes the airport flight schedule and historical flight operation data to obtain an airport capacity probability map, and based on the airport flight schedule, airport operation mode modeling, and airport capacity probability map, obtains predicted values ​​for airport flight arrival and departure delays. This embodiment is applicable to the evaluation of the rationality and effectiveness of airport flight time / schedule arrangement during the strategic phase, such as... Figure 1 As shown, the main process of this method includes:

[0037] This invention provides a method for predicting airport flight scheduling strategic delays based on delay splitting and combining, such as... Figure 1 As shown, it includes:

[0038] Using airport flight operation data, clustering results of airport arrival and departure flights are obtained. The flight operation data includes: flight wheel chock calculation time data and flight actual landing time data.

[0039] Using the clustering results of the airport's arriving and departing flights, obtain the airport's arrival and departure capacity probability map;

[0040] Using the aforementioned arrival and departure capacity probability map, an airport's native arrival and departure delay prediction model is established;

[0041] An airport flow control delay prediction model is established using airport flow management data, wherein the airport flow management data includes: flight departure airport and destination airport, flight target wheel chute time data, and flight calculation wheel chute time data;

[0042] Based on the arrival and departure native delay prediction model and the airport flow control delay prediction model, an airport flight schedule delay prediction model is established.

[0043] Based on the airport flight schedule delay prediction model, the airport flight schedule to be processed is predicted, and the predicted values ​​of airport flight schedule arrival and departure delays are obtained.

[0044] Specifically, before constructing an airport flight schedule delay prediction model, it is necessary to first conduct an airport capacity assessment and statistically analyze the airport's arrival and departure traffic based on historical flight operation data. Secondly, the K-means clustering method is used to perform cluster analysis on the arrival and departure traffic, and the clustering results are evaluated and screened based on three methods: silhouette coefficient, variance ratio criterion, and pruning method. Finally, the confidence interval method is used to obtain key time nodes, and combined with the screened cluster center results, an airport arrival and departure capacity probability map is generated.

[0045] The airport queuing network model is analyzed to obtain the airport arrival and departure queuing network model, and then a simulation model of the original delay of airport arrival and departure is constructed. At the same time, historical flight flow control data is statistically analyzed to determine the correlation between the magnitude of flow control delay and the magnitude of traffic flow. Based on the correlation test results, a flow control delay prediction model is constructed. Finally, based on the airport flight delay propagation judgment formula, the original delay simulation model of arrival and departure and the flow control delay prediction model are coupled to construct an airport flight schedule delay prediction model based on delay decomposition and simulation.

[0046] Airport capacity assessment includes: statistical traffic overview, traffic data clustering, evaluation of clustering results, confidence interval method, and capacity probability plot.

[0047] The statistical traffic overview includes: departure traffic overview and arrival traffic overview. The departure traffic overview is obtained by taking the actual departure time of departing flights each day, and counting the number of departing flights in 15-minute time windows, resulting in a traffic overview dataset X = {x1, x2, ..., x...}. n}, where x i The number of departing flights within the i-th time window is represented by: taking the actual arrival times of the daily arriving flights, and counting the number of arriving flights in 15-minute time windows to obtain the traffic overview dataset. The data set is statistically analyzed as described in the departure traffic overview.

[0048] After obtaining the overview of inbound and outbound traffic, the K-means clustering method was used to cluster the inbound and outbound traffic. K-means clustering is a data partitioning algorithm based on Euclidean distance, which is expressed as: dividing the dataset X into k clusters, where k is a moderately sized positive integer. For a specific cluster, the clustering of observations within the cluster should be smaller than the distance between that value and observations in other clusters. For each cluster, the centroid of all observations in the cluster is... k = 1, 2…K, also known as The centroid of cluster k is represented as:

[0049]

[0050] In the formula Let x be the q-th observation in cluster k, and n be the number of observations in cluster k. The K-means algorithm performs Euclidean distance detection between each observation x and each centroid, and assigns x to the cluster corresponding to the nearest centroid. The algorithm terminates when the sum of the distances between each observation and its corresponding cluster is minimized.

[0051] K = 3-16 was chosen as the initial number of clusters. All cluster centers were plotted as a line graph, and the proportion of each cluster in the total observations was calculated. The line graph is a graph with the time window as the x-axis and the flow rate as the y-axis, containing K lines representing cluster centers.

[0052] The silhouette coefficient, variance ratio criterion, and pruning method were used to evaluate and filter all clustering results. Specifically:

[0053] Silhouette coefficient: A metric that measures the degree of separation between clusters by comparing the similarity of each observation to other observations in its assigned cluster. The silhouette coefficient of an observation x is:

[0054]

[0055] In the formula a(x i ) is the average distance from the observation x to all other data points in its cluster, calculated using the following formula:

[0056]

[0057] b(x i ) is the average distance between the observation x and the observations in the neighboring clusters, calculated using the following formula:

[0058]

[0059] In equation (3-4), d(x) j ,x i ) is the distance between observations i and j, and |Ci| is the number of observations within the cluster.

[0060] Based on the silhouette coefficients s(x) of all the observed values ​​obtained i The average profile coefficient is:

[0061]

[0062] An SI value close to 1 indicates that the clusters are compact and well separated, close to 0 indicates overlapping clusters, and close to -1 indicates that there are too many or too few clusters.

[0063] Variance ratio criterion: It is the ratio of the inter-cluster error variance to the intra-cluster error variance. The calculation formula is:

[0064]

[0065] In the formula, n is the total number of observations, and the formula for calculating the inter-cluster error variance is:

[0066]

[0067] In the formula, n i c is the number of observations in cluster i. i Let c be the centroid of cluster i, and c be the global centroid of all observations; the formula for calculating the intra-cluster error variance is:

[0068]

[0069] In the formula, x represents all observations within cluster i, and c i The CHI value is the centroid of cluster i. The higher the CHI value, the better the clustering effect.

[0070] Pruning is a method to remove unimportant weights from the model, reducing the number of model parameters and computational load without affecting the model's accuracy. Specifically, it involves calculating the proportion of each cluster in the overall clustering results. If a cluster with a proportion less than a threshold is found, it indicates that pruning involves clusters with smaller features and less impact, and the value of k is no longer increased. The threshold is generally set to 5%-10%.

[0071] Based on the above clustering evaluation results, the optimal clustering result is obtained, and the key time nodes are determined using the confidence interval method. The confidence interval determination method includes: within a given time window, comparing the cluster center results with the confidence interval boundaries obtained using the airport arrival and departure flight clustering results. If the cluster center results do not exceed the confidence interval boundaries, then the cluster centers are determined to be merged with the airport arrival and departure flight clustering results. Specifically, using the historical dataset as the overall sample, the clustering results are used for classification. For each cluster, the sample center, standard deviation, and error limit are calculated for the independent samples to obtain the upper and lower limits of the confidence interval. The upper and lower limits of the confidence interval are compared with the cluster centers in the same coordinate system. If the flow rate of the cluster center exceeds the upper and lower limits of the confidence interval within a specific time window, then the cluster is separated from the cluster corresponding to the confidence interval within that specific time window; otherwise, the two are merged.

[0072] Wherein, the sample center is represented as

[0073]

[0074] In the formula, |S q | represents the number of observations in the independent sample q. It is the j-th observation in the independent sample q; the sample standard deviation is:

[0075]

[0076] The formula for calculating sample error is:

[0077]

[0078] In the formula, z_critical(0.995) is the Z-critical value with a confidence level of 99%.

[0079] Once the optimal clustering result is obtained, the capacity probabilistic graph method is used to merge similar scenes and determine the separated scenes. The algorithm is as follows:

[0080] Using the confidence interval method to calculate the cluster centers, we obtain the following set of confidence intervals:

[0081]

[0082] In the formula, For scene i within the 99% confidence interval of time window t, take cluster center j:

[0083] {x j For confidence interval i and cluster center j, within the time interval [t)|t=1,2……,T}, the expression is used to express the expression for confidence interval i and cluster center j. x ,t y If the following equation is satisfied:

[0084]

[0085] Then clusters i and j in the time period [t] x , t y The area within [t] represents the fusion scenario; otherwise, the key node is specified within the time period [t]. x , t y At the left endpoint of ], define this key node as t. ij .

[0086] Let set g represent the set of all scenarios. If the following equation is satisfied:

[0087]

[0088] Then, for scene i, its separation time from other scenes is given by Equation 15:

[0089] t ij (g)=min(t ij ), j∈g, j≠i (15)

[0090] Let G be the set of all groups. The probability graph is constructed as follows: If t ij =1, then for a unique i∈g, scenario i is an independent scenario; otherwise The earliest branch time in scene i is t. ij (g)

[0091] The capacity probability map is obtained using the confidence interval method and the capacity probability map method, as shown in the attached figure. Figure 2As shown, the capacity probability map represents a set of capacity probability curves that merge and separate over time. Its usage is as follows: After selecting a specific time window, a capacity scenario within that window is randomly selected. If an independent scenario is selected, the selection probability is the proportion of the cluster corresponding to that scenario to the total observed values, and the capacity size is the traffic volume of the corresponding cluster center in the current time window. If a merged scenario is selected, the selection probability is the difference between 1 and the remaining independent scenarios, and the capacity size is the weighted average of the traffic volume values ​​of each independent scenario within the merged scenario within the specific time window. The weight represents the quotient of the selection probability of each independent scenario within the merged scenario and the sum of their selection probabilities.

[0092] Construct native delay prediction models for arrivals and departures, including: native delay prediction model for departures and native delay prediction model for arrivals.

[0093] The departure delay prediction model is a simulation model of the queuing network of parking stands and runway heads. The simulation process is attached. Figure 3 As shown, the initial arrival sequence and initial capacity sequence are obtained using flight schedule data and departure capacity probability map, where the initial arrival sequence includes: flight calculation and wheel chock removal time.

[0094] The flight schedule is iteratively traversed. When an aircraft is in a parking position, if the number of currently operating flights is less than the departure capacity, the aircraft can remove its wheel chocks and begin taxiing. Once the aircraft reaches the runway threshold, the waiting time is the difference between the actual departure time of the last aircraft that has not yet taken off and the time the current aircraft reaches the runway threshold. The iteration ends when all aircraft for the day have been processed or when the iteration time exceeds 24:00. The departure delay is the average departure delay of all flights for the day.

[0095] The arrival native delay prediction model is a simulation model of the queuing network in the terminal area and runway. The simulation process is attached. Figure 4 As shown, the initial arrival sequence and initial capacity sequence are obtained using flight schedule data and arrival capacity probability map, where the initial arrival sequence includes the scheduled flight landing time.

[0096] The flight schedule is iteratively traversed. When an aircraft is in the terminal area, the in-flight queuing delay is calculated using a steady-state queuing model. After an aircraft lands, its taxiing delay is the difference between the entry time of the last aircraft that has not yet entered the taxiing system and the actual landing time of the current aircraft. The iteration ends when all aircraft for the day have been processed or when the iteration time exceeds 24:00. The arrival delay is the average arrival delay of all flights for the day.

[0097] Airport flow control delay correlation analysis examines the correlation between the magnitude of flow control delays and flight volume in different cities over different time periods. If the correlation is higher than a threshold, the flow control delay prediction model is considered a positive correlation function with flight volume; otherwise, the model is derived by statistically analyzing the average flow control delay magnitude and the probability of flow control delays occurring. The threshold is typically 0.5. The probability of flow control delays is represented as the ratio of the number of flights subject to flow control in a specific city to the total number of flights within a specific time period. Specifically, the magnitude of flow control delays is represented as the sum of flow control delays in a specific city within a specific time period; the average flow control delay magnitude is represented as the sum of the flow control delay magnitude and the number of flights affected by flow control.

[0098] Airport delay propagation determination formula, as shown in the appendix. Figure 5 As shown, this includes comparing the difference between a specific flight's turnaround time and its minimum turnaround time at the airport with the flight's arrival delay. If the difference is greater, the arrival delay will not propagate to subsequent flights; otherwise, the delay will propagate to subsequent flights, and the magnitude of this propagation will be equal to the difference between the difference and the flight's arrival delay. The arrival delay is represented as AD. i =AIBT i –SIBT i AIBT i This refers to the actual turnaround time of the flight, SIBT. i This refers to the scheduled turnaround time for a flight; the difference between the specific flight's turnaround time at the airport and the minimum turnaround time is...

[0099] RDCT i =SOBT i+1 –SIBT i –MTTT i (16)

[0100] In the formula, SOBT i+1 It is the scheduled wheel-shifting time for subsequent flights, MTTT i It is the shortest turnaround time for the preceding flight; the magnitude of the delay propagation is DP. i =AD i -RDCT i If DP i If the value is less than 0, there is no delay propagation; otherwise, the target wheel-off time for subsequent flights is TOBT. i+1 =DP i +SOBT i+1 .

[0101] Based on the original arrival and departure delay prediction model and the airport flow control delay prediction model, an airport flight schedule delay prediction model is established using the airport flight delay propagation formula, as shown in the appendix. Figure 6As shown, the arrival flight schedule is simulated using the original arrival delay prediction model, which includes non-originating flights. The simulation yields the average delay of arrival flights. Departure flight schedules are generated using a delay propagation decision formula. Departure flights include both originating and non-originating flights. Originating flights do not have delay propagation, while non-originating flights have delay propagation based on their flight registration numbers and the delay propagation decision formula. Flow control delay prediction models are used to generate flow control delays for departure flight schedules. Departure flight schedules, delay propagation, and flow control delays are combined to generate a departure flight time series for calculating wheel chute clearance. Finally, the departure flight time series for calculating wheel chute clearance is simulated using the original departure delay prediction model to obtain the average delay of departure flights.

[0102] The airport flight schedule delay prediction method provided in this invention assesses airport capacity by statistically analyzing airport arrival and departure traffic based on historical flight operation data. It then uses K-means clustering to perform cluster analysis on the arrival and departure traffic data. The clustering results are evaluated and filtered using silhouette coefficient, variance ratio criterion, and pruning method. Key time nodes are obtained using the confidence interval method, and combined with the filtered cluster center results, an airport arrival and departure capacity probability map is generated. The airport queuing network model is analyzed to obtain the airport arrival and departure queuing network model, and a simulation model of the original airport arrival and departure delays is constructed. Historical flight flow control data is statistically analyzed to determine the correlation between flow control delay magnitude and traffic volume. Based on the correlation test results, a flow control delay prediction model is constructed. Finally, based on the airport flight delay propagation judgment formula, the original arrival and departure delay simulation model and the flow control delay prediction model are coupled to construct an airport flight schedule delay prediction model based on delay decomposition and simulation.

[0103] This invention provides a method for predicting airport flight schedule delays based on delay decomposition and simulation modeling. The method utilizes airport flight operation data to obtain airport arrival and departure flight clustering results. Using these clustering results, and employing clustering result evaluation indicators and confidence interval judgment methods, an airport arrival and departure capacity probability map is obtained. For the airport operation mode, by analyzing the airport operation queuing network model, and using the airport arrival and departure capacity probability map, a native arrival and departure delay prediction model is established. Using airport flow management data, and through airport flow control delay correlation analysis, an airport flow control delay prediction model is established. Based on the native arrival and departure delay prediction model and the airport flow control delay prediction model, and using the airport flight delay propagation judgment formula, an airport flight schedule delay prediction model is established. Based on the airport flight schedule delay prediction model, simulation is performed using the airport flight schedule to obtain the predicted values ​​of airport flight schedule arrival and departure delays. This achieves the prediction of airport flight schedule delays.

[0104] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting airport flight scheduling strategic delays based on delay splitting and combining, characterized in that, include: S1. Using airport flight operation data, obtain the clustering results of airport arrival and departure flights, wherein the flight operation data includes: flight wheel chock calculation time data and flight actual landing time data; S2. Using the clustering results of the airport's arriving and departing flights, obtain the airport's arrival and departure capacity probability map; wherein, the capacity probability map is represented as a set of capacity periodic curves, used to generate random airport departure capacity and airport arrival capacity at different time windows. S3. Using the arrival and departure capacity probability map, establish an airport arrival and departure native delay prediction model; wherein, native delay is flight delay caused by the combined effect of the airport's own capacity, airport facility structure and operation mode; S4. Using airport traffic management data, establish an airport traffic control delay prediction model, wherein the airport traffic management data includes: flight departure airport and destination airport, flight target wheel chock time data and flight calculated wheel chock time data; S5. Based on the arrival and departure native delay prediction model and the airport flow control delay prediction model, establish an airport flight schedule delay prediction model; S6. Based on the airport flight schedule delay prediction model, perform prediction processing on the airport flight schedule to be processed, and obtain the arrival and departure delay prediction results of the airport flight schedule to be processed. S3 also includes: Before establishing the airport's native arrival and departure delay prediction model using the arrival and departure capacity probability map, a queuing network model corresponding to the airport's operation mode is loaded. The airport operation mode includes: runway service mode, airport departure operation mode, and airport arrival operation mode. The port arrival / departure native delay prediction model includes: a departure native delay prediction model and an arrival native delay prediction model; In the original departure delay prediction model, the airport departure capacity is used to predict the aircraft gate waiting time and the queuing network model is used to predict the runway head waiting time. In the original arrival delay prediction model, the airport arrival capacity is used to predict flight arrival taxiing delays and the queuing network model is used to predict flight terminal area delay times. S4 also includes: Before establishing an airport flow control delay prediction model, airport flow management data is used to conduct a correlation analysis of airport flow control delays. Based on the results of the correlation analysis of airport flow control delays, an airport flow control delay prediction model is established. In S5, it includes: The arrival delay prediction model is used to generate the arrival delay prediction value of the airport flight plan. Then, the arrival delay prediction value of the airport flight plan is converted into the departure propagation delay prediction value. In this process, the delay propagation is generated for the departure flight plan using the delay propagation decision formula. The departure flights include originating flights and non-originating flights. Originating flights do not have delay propagation, while non-originating flights obtain delay propagation based on the flight registration number and the delay propagation decision formula. The flow control delay prediction model generates flight flow control delays for departing flights, merges the departing flight schedule with delay propagation and flow control delays, generates a time series for calculating the wheel chute withdrawal for departing flights, and then uses the original departure delay prediction model to generate the departure delay prediction value of the flight schedule.

2. The method according to claim 1, characterized in that, The airport flow control delay correlation analysis includes: The correlation between the magnitude of flow control delays and the volume of traffic in different cities over different time periods is obtained. If the quantified correlation value is higher than a threshold, the flow control delay prediction model is set as a positive correlation function with respect to the volume of traffic. If the quantified value of the correlation is less than or equal to the threshold, the flow control delay prediction model is obtained by statistically analyzing the magnitude of flow control delays and the probability of flow control delays occurring. The probability of flow control delays occurring includes the ratio of the number of flights subject to flow control in a specific city to the total number of flights within a specific time period.

3. The method according to claim 1, characterized in that, S5 also includes: The difference between the flight's turnaround time at the airport and the minimum turnaround time is compared with the flight's arrival delay. If the former is greater, it is determined that the flight's arrival delay will not propagate to subsequent flights; otherwise, the delay will propagate to subsequent flights.

4. The method according to claim 1, characterized in that, In S6, the airport flight plan to be processed includes: the date of the airport flight to be processed, the departure airport and destination airport of the airport flight to be processed, the time of the flight plan to be withdrawn from the schedule of the airport flight to be processed, and the time of the flight plan to be put into the schedule of the airport flight to be processed. The airport flight schedule arrival and departure delay prediction results to be processed include: average arrival flight delay, average departure flight flow control delay, average departure flight delay, and average total departure flight delay.

5. The method according to claim 4, characterized in that, The average delay of arriving flights is the ratio of total arriving delays to the total number of arriving flights; The average flow control delay for departing flights is the ratio of the total flow control delay to the total number of departing flights; The average delay of departing flights is the ratio of total departing flight delays to the total number of departing flights; The average total delay of departing flights is the sum of the average flow control delay of departing flights and the average delay of departing flights.

6. The method according to claim 4, characterized in that, In S1, the airport arrival and departure flight clustering results include: flight departure traffic clustering results and flight arrival traffic clustering results; The clustering results of the flight departure traffic are: the cluster center results of the airport's actual flight departure traffic profile using the Kmeans clustering method; The clustering results of the flight arrival traffic are: the cluster center results of the actual flight arrival traffic at the airport using the Kmeans clustering method.

Images