CN115689073B - Flight delay prediction method and system based on high-order network - Google Patents
Flight delay prediction method and system based on high-order network Download PDFInfo
- Publication number
- CN115689073B CN115689073B CN202310009763.1A CN202310009763A CN115689073B CN 115689073 B CN115689073 B CN 115689073B CN 202310009763 A CN202310009763 A CN 202310009763A CN 115689073 B CN115689073 B CN 115689073B
- Authority
- CN
- China
- Prior art keywords
- delay
- order
- network
- edge
- composite structure
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a flight delay prediction method and a system based on a high-order network, belongs to the technical field of delay prediction, and solves the problem that inaccurate flight delay prediction is caused by the sweep effect among multiple airports in the prior art. Constructing a high-order delay network according to historical flight data, and converting the high-order delay network into a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels; converting the high-order delay network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order delay network and the delay projection network; calculating the weight of each composite structure based on the edge weight to serve as a high-order characteristic value, taking the high-order characteristic value as input, and taking the delay tag as output training classification model to obtain a delay prediction model; and calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into a delay prediction model to obtain a prediction result. Accurate prediction of flight delay among multiple airports is achieved.
Description
Technical Field
The invention relates to the technical field of delay prediction, in particular to a flight delay prediction method and a flight delay prediction system based on a high-order network.
Background
With the rapid development of the civil aviation industry, the problem of flight delay is becoming a worldwide challenge. There are many reasons for flight delays, but the inherent propagation effects of air traffic networks are key factors that continuously worsen flight delays. In an aviation network composed of a plurality of airports, flight delays among the airports can have influence, such as different flights share the airplane, departure flight delays of departure airports can possibly cause arrival delays of destination airports, and any initial delays can be amplified continuously to influence the whole aviation traffic network. Thus, research into the flight delay propagation mechanism is important and challenging.
Researchers at home and abroad have conducted systematic studies on the problem of flight delay spread. The economic statistical type method is initially applied to analyze delay spread problems. Then, with the rise of the research on the complex network at the end of the twentieth century, new understanding of the problem of delay has also been induced from the viewpoint of the complex network, and delay spread can be regarded as a collective behavior of nodes in an aviation network emerging through interaction. With the rise of the machine learning method around 2010, a practical flight delay prediction model can be established through the deep learning method.
Despite the progress made in the current research on delay spread, there are shortcomings. Traditional methods such as operation study, economy statistics and the like used in the mainstream, such as regression analysis and statistical inference methods, often obtain delay propagation results directly through analysis, such as whether an airport has a significant influence on other airports, but neglect influence caused by connection of a plurality of nodes in a system, such as influence of route delay between two airports on a plurality of follow-up routes, especially time sequence change condition. Some existing flight delay deducing methods based on statistical learning only consider the delay relation between pairs of airports, but do not consider the high-order delay propagation effect between multiple airports.
Disclosure of Invention
In view of the above analysis, the present invention aims to provide a method and a system for predicting flight delay based on a higher-order network, which are used for solving the problem that the existing method and system do not consider the inaccurate prediction of flight delay caused by sweep effect among multiple airports.
In one aspect, the embodiment of the invention provides a flight delay prediction method based on a high-order network, which comprises the following steps:
constructing a high-order delay network according to historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels;
converting the high-order delay network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order delay network and the delay projection network;
based on the edge weights, calculating weights of all composite structures as high-order characteristic values, taking the high-order characteristic values as input, and taking delay labels as output training classification models to obtain delay prediction models;
and taking the airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into a delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted.
Based on further improvement of the method, the construction of the high-order delay network according to the historical flight data comprises the following steps:
sequencing the historical flight data according to the sequence of the flight time, identifying whether the landing delay time of each piece of historical flight data exceeds a delay threshold value in one day, and recording corresponding planes, dates and airport sets as the delay data if the landing delay time exceeds the delay threshold value; continuously identifying whether the same aircraft has delay between different airports for 2 or more times continuously in one day according to the current delay data, and adding the different airports into an airport set of the corresponding aircraft if the delay data exist;
taking out airport sets, neglecting the airport sequence in the airport sets, and putting different airport sets into a data set;
the airport set corresponds to an edge in a high-order delay network, and airports in the airport set are used as nodes contained by the edge, wherein if 2 airports exist in the airport set, the airport set corresponds to a common edge, and the order is 2; otherwise, the airport set corresponds to a higher-order side, and the order of the higher-order side is the number of airports in the airport set.
Based on the further improvement of the method, the high-order edge in the high-order deferred network is converted into a common edge to obtain a deferred projection network, which comprises the following steps:
taking out each higher-order edge in the higher-order delay network in turn, and constructing a common edge of the delay projection network between any two nodes contained in the current higher-order edge; and directly mapping each common edge in the high-order deferred network into a common edge of the deferred projection network.
Based on a further improvement of the method, extracting the composite structure based on the deferred projection network comprises:
according to the nodes of the delay projection network, traversing and taking out the combinations with the same node number, wherein the node number is more than 2;
and sequentially taking out each combination, identifying whether edges exist between any two nodes in the current combination in the deferred projection network, and if so, forming the current combination into a composite structure.
Based on a further improvement of the above method, labeling the composite structure with a delay tag includes: if nodes in the composite structure all belong to the same high-order edge of the high-order delay network, the composite structure is a positive sample, and a label with delay is marked; otherwise, the composite structure is a negative sample, and a label without delay is marked.
Based on a further improvement of the above method, converting the high-order deferred network into a line graph includes:
acquiring edges which are smaller than or equal to an order threshold value in a high-order delay network as edges to be converted;
mapping the edges to be converted into nodes in the line graph, and if the same nodes exist in the two edges to be converted, constructing an edge between the nodes in the line graph corresponding to the two edges to be converted.
Based on a further improvement of the method, the index value of each node in the calculation line graph is calculated based on a centrality index, wherein the centrality index comprises: center of gravity, near center, feature vector center, cluster coefficients, web page rank, or core radius.
Based on a further improvement of the above method, mapping the index value to edge weights in the high order deferred network and the deferred projection network includes: according to the one-to-one correspondence between the nodes in the line graph and the edges of the high-order delay network, directly taking the index values of the nodes in the line graph as the edge weights of the corresponding high-order delay network;
according to the one-to-many conversion relation between the edges in the high-order deferred network and the edges in the deferred projection network, the index value is used as the edge weight of each edge in the converted deferred projection network.
Based on the further improvement of the method, based on the edge weight, the weight of each composite structure is calculated as a high-order characteristic value, and the method comprises the following steps: according to the edge weight between any two nodes in the composite structure, the weight of the composite structure is calculated by a geometric mean, a harmonic mean or an arithmetic mean method to be used as a high-order characteristic value.
On the other hand, the embodiment of the invention provides a flight delay prediction system based on a high-order network, which comprises the following steps:
the composite structure extraction module is used for constructing a high-order delay network according to the historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels;
the weight acquisition module is used for converting the high-order deferred network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order deferred network and the deferred projection network;
the model training module is used for calculating the weight of each composite structure based on the edge weight to serve as a high-order characteristic value, taking the high-order characteristic value as input, and taking the delay label as output training classification model to obtain a delay prediction model;
the delay prediction module is used for taking a plurality of airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into the delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted.
Compared with the prior art, the invention has at least one of the following beneficial effects:
1. aiming at the situation of the problem of delay propagation of an aviation network, the situation of delay of a plurality of subsequent flights of the same aircraft in one day caused by delay of the preceding flights is characterized by a high-order structure, and the defect of a delay data processing method only aiming at paired flights is overcome. Meanwhile, the high-order network is converted into the line graph so as to calculate the importance of any high-order side, and the weight is further given to each side in the projection graph, so that the operation is easy and the expansibility is strong.
2. The high-order network structure index and the high-order line graph index are synthesized, various high-order characteristics of the flight delay network are utilized efficiently through integrated learning, accuracy of flight delay prediction is improved greatly, and particularly, a local structure which is easy to generate continuous delay in an actual flight is predicted. The method has important practical significance for the principle of delay propagation, delay control and stable operation of flights.
In the invention, the technical schemes can be mutually combined to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, like reference numerals being used to designate like parts throughout the drawings;
FIG. 1 is a flow chart of a method for predicting flight delay based on a higher-order network in embodiment 1 of the present invention;
FIG. 2 is a schematic diagram of a high-order deferred network converted into a deferred projection network according to embodiment 1 of the present invention;
FIG. 3 is a diagram illustrating the conversion of the higher order delay network according to embodiment 1 of the present invention;
FIG. 4 is a graph showing the prediction accuracy improvement rate in various combinations in embodiment 1 of the present invention.
Detailed Description
Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and together with the description serve to explain the principles of the invention, and are not intended to limit the scope of the invention.
Example 1
The invention discloses a flight delay prediction method based on a high-order network, which is shown in fig. 1 and comprises the following steps:
s11: constructing a high-order delay network according to historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels.
It should be noted that, the historical flight data is flight data within a historical period of time, including: flight time, aircraft, departure airport, landing airport, and landing delay time.
Specifically, constructing a higher order deferred network from historical flight data includes:
sequencing the historical flight data according to the sequence of the flight time, identifying whether the landing delay time of each piece of historical flight data exceeds a delay threshold value in one day, and recording corresponding planes, dates and airport sets as the delay data if the landing delay time exceeds the delay threshold value; continuously identifying whether the same aircraft has delay between different airports for 2 or more times continuously in one day according to the current delay data, and adding the different airports into an airport set of the corresponding aircraft if the delay data exist;
taking out airport sets, neglecting the airport sequence in the airport sets, and putting different airport sets into a data set;
the airport set corresponds to the edge in the high-order delay network, and the airports in the airport set are used as nodes contained by the edge, wherein if 2 airports exist in the airport set, the airport set corresponds to the common edge, and the order is 2; otherwise, the airport set corresponds to a higher-order side, and the order of the higher-order side is the number of airports in the airport set.
The airport set includes a take-off airport and a landing airport. Illustratively, the delay threshold is 15 minutes, and landing delay time exceeds 15 minutes in all the days is taken as delay data, and when the aircraft continuously runs from Beijing to Shanghai, shanghai to Guangzhou in the same aircraft in the same period in one day, the airport collection records three airport name identifications of Beijing, shanghai and Guangzhou.
The important consideration of the embodiment is whether delay occurs between airports, the high-order delay network constructed in the step is an unoriented high-order network, and whether delay occurs between airports is indicated by the edges of the high-order network and the nodes on the edges.
Further, the high-order deferred network is converted into a deferred projection network, namely, the common edges of the high-order deferred network are reserved, the high-order edges in the high-order deferred network are converted into full-communication subgraphs among corresponding nodes, the full-communication subgraphs are used for integrally interacting a plurality of nodes in the high-order deferred network, and dimension reduction is characterized as the common edges between every two nodes in the deferred projection network.
Specifically, converting a higher-order edge in a higher-order deferred network into a common edge to obtain a deferred projection network includes:
each higher-order edge in the higher-order delay network is sequentially taken out, and a common edge of the delay projection network is constructed between any two nodes contained in the current higher-order edge, namely one node containsnHigher-order edges of individual nodes converting correspondence into a deferred projection network
A strip common edge; and directly mapping each common edge in the high-order deferred network into a common edge of the deferred projection network.
As shown in fig. 2, the left graph is an exemplary schematic diagram of a high-order deferred network, in which there are 4 normal edges including 2 nodes, 1 high-order edge including 3 levels of nodes, 1 high-order edge including 4 levels of nodes, and the right graph is a schematic diagram of a deferred projection network converted from the left graph, in which normal edges are directly reserved, mapped as edges between nodes, and the high-order edges of the three levels are converted into 3 normal edges, and the high-order edges of the four levels are converted into 6 normal edges.
Next, based on the obtained deferred projection network, extracting a combination of the same number of nodes, and identifying a composite structure from the combination, including:
according to the nodes of the delay projection network, traversing and taking out the combinations with the same node number, wherein the node number is more than 2;
and sequentially taking out each combination, identifying whether edges exist between any two nodes in the current combination in the deferred projection network, and if so, forming the current combination into a composite structure.
It should be noted that, the number of nodes in the composite structure indicates the number of airports for identifying whether delay occurs, preferably, the number of nodes is set to 3, and 3 nodes extracted by traversing obtain a triple, but not all triples are composite structures, and only when edges exist between every two nodes in the delay projection network in the triple, the composite structure is obtained.
Further, marking the composite structure with a delay tag according to the higher-order delay network, including: if nodes in the composite structure all belong to the same high-order edge of the high-order delay network, the composite structure is a positive sample, and a label with delay is marked; otherwise, the composite structure is a negative sample, and a label without delay is marked.
Illustratively, in the right diagram of fig. 2, triplet 1 results from the higher order edge transitions of the third order in the left diagram, so triplet 1 is a positive sample; whereas triplet 2 is generated by the 3 common edge transitions in the left graph, so triplet 2 is a negative sample.
Compared with the prior art, the method and the device have the advantages that whether delay occurs among a plurality of airports or not is accurately identified through the high-order network, and the prediction accuracy is improved.
S12: and converting the high-order deferred network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order deferred network and the deferred projection network.
It should be noted that, in this embodiment, the high-order deferred network is converted into a line graph, that is, the edge is converted into a node, and the importance of the edge of the high-order deferred network is represented by the importance of the node.
Specifically, converting a high-order deferred network into a line graph includes:
acquiring edges which are smaller than or equal to an order threshold value in a high-order delay network as edges to be converted;
mapping the edges to be converted into nodes in the line graph, and if the same nodes exist in the two edges to be converted, constructing an edge between the nodes in the line graph corresponding to the two edges to be converted.
It should be noted that, the higher-order side with the higher order number contains more nodes, which results in that the side owned by the higher-order side in the line graph is very large, and the number of the higher-order side in the higher-order delay network is smaller, so that the influence is smaller, and therefore, the embodiment can be converted into the side of the line graph through the order threshold limit. Preferably, the order threshold is set to 6, i.e. the common edges in the high order delinquent network and the higher order edges below six are converted to nodes in the line graph.
Illustratively, as shown in fig. 3, the left diagram is a schematic diagram of a high-order delay network, and the right diagram is a schematic diagram of a transition. The edge A, the edge B and the edge C in the left graph are respectively converted into the node A, the node B and the node C in the right graph, and because the edge A, the edge B and the edge C have the same node, an edge exists between every two nodes in the line graph.
Calculating the index value of each node in the line graph is based on the centrality index, wherein the centrality index comprises: center of gravity DC (Degree Centrality), center of proximity CC (Closeness Centrality), feature vector center EC (Eigenvector Centrality), cluster coefficient C (Clustering Coefficient), web page rank PR (PageRank), or core radius CO (coreness).
The calculation formula of each centering index is shown in table 1.
Table 1 calculation formulas for center indexes
It should be noted that, these centrality indexes may represent the importance of the nodes in the line graph, and correspondingly represent the importance of the edges in the higher-order network. The conversion mode can equally process the high-order side and the common side, so that the importance of the high-order side and the common side can be reasonably and effectively calculated.
Further, mapping the index value calculated according to the centrality index into the edge weights in the high-order deferred network and the deferred projection network includes: according to the one-to-one correspondence between the nodes in the line graph and the edges of the high-order delay network, directly taking the index values of the nodes in the line graph as the edge weights of the corresponding high-order delay network; according to the one-to-many conversion relation between the edge in the high-order deferred network and the edge in the deferred projection network, the index value is used as the edge weight of each edge in the deferred projection network after conversion, namely the original undirected edge is set as the calculated centrality index of the edge or the high-order edge to which the edge belongs, and the deferred projection network with undirected right is obtained.
Compared with the prior art, the method and the device convert the high-order network into the line graph so as to calculate the importance of any high-order side, further give weight to each side in the projection graph, and are easy to operate and strong in expansibility.
S13: based on the edge weights, the weights of the composite structures are calculated to serve as high-order characteristic values, the high-order characteristic values serve as inputs, and the delay labels serve as output training classification models, so that a delay prediction model is obtained.
It should be noted that, according to the edge weight and average number method, the weight of each composite structure is calculated as a higher-order feature value, including: according to the edge weight between any two nodes in the composite structure, the weight of the composite structure is calculated as a high-order characteristic value through a Geometric mean (Geometric mean), a Harmonic mean (Harmonic mean) or an Arithmetic mean (arithmetical mean) method.
As shown in table 2, the calculation formulas of geometric mean, harmonic mean or arithmetic mean are listed taking the third-order composite structure as an example.
Table 2 formula for mean method calculation
The classification model selects Random Forest (Random Forest) or Na ve Bayes (Na ve Bayes), takes the high-order eigenvalue as input, takes the delay tag as output to train the classification model, and takes the trained classification model as the delay prediction model.
It should be noted that, in machine learning, the network model is trained by the training set, and the performance test is performed on the network model by the test set, so as to prevent the network model from over fitting and under training of the training set, which are conventional methods, and therefore, are not separately described in the above step S11 to step S13. If described in terms of both training and testing for this embodiment, it includes:
in step S11, the historical flight data are sorted according to the sequence of the flight time and then are divided into a training set and a testing set according to a proportion; respectively constructing a high-order delay network and a delay projection network for the training set and the testing set, identifying a composite structure in the training set and the testing set, and marking delay labels;
in step S12, the high-order deferred network of the training set is converted into a line graph, and an index value is calculated to obtain a deferred projection network of the training set with right and no direction.
In step S13, according to the weighted undirected deferred projection network of the training set, calculating the weight of the composite structure of the training set as a training sample, and training a classification model; and then according to the weighted undirected delay projection network of the training set, calculating the weight of the composite structure of the testing set as a test sample, inputting the test sample into the trained classification model, using the obtained prediction result as a prediction label, comparing the prediction result with the actual delay label of the composite structure in the testing set, and calculating AUC (Area Under Curve) area to represent prediction accuracy after drawing a ROC (Receiver Operating Characteristic Curve) curve.
For example, when the method of the present embodiment is not adopted to perform random prediction, auc=0.5, in the present embodiment, the acquired national flight data of 2018 and 7 months are divided into a training set and a test set according to a ratio of 4:1, the above 6 centrality indexes and 3 average methods are respectively subjected to combined verification on a random forest and a bayesian regression model, the respectively obtained AUC values are compared with the randomly predicted AUC values, the effect of improving the prediction accuracy is obvious, and the prediction accuracy improvement rate in each case is shown in fig. 4.
In fig. 4, the higher-order eigenvalue of the composite structure calculated by the geometric mean method is trained and tested by using the page rank PR as the centrality index, and the accuracy of the AUC value obtained by training and testing the naive bayes classification model is improved by 25% compared with that of the random prediction. Then this optimal combination may be selected for prediction of flight delays.
S14: and taking the airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into a delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted.
The number of the airports to be predicted is used as the composite structure to be predicted, and is consistent with the number of nodes in the composite structure during training.
When calculating the weight of the composite structure to be predicted based on the weighted undirected deferred projection network corresponding to the trained prediction model, if two airports in the composite structure to be predicted have no edges in the deferred projection network, the corresponding edge weight can be preset to be a value close to 0, such as 0.1. And transmitting the calculated weight into a delay prediction model to obtain a predicted classification result, wherein the classification result is used for indicating whether flight delays are likely to occur among a plurality of airports to be predicted.
Compared with the prior art, the flight delay prediction method based on the high-order network provided by the embodiment aims at the situation of the flight delay propagation problem of the aviation network, the situation of the follow-up multiple flight delays caused by the front flight delay of the same aircraft in one day is characterized by a high-order structure, and the defect of the delay data processing method only aiming at the paired flights is overcome. Meanwhile, the high-order network is converted into the line graph so as to calculate the importance of any high-order side, and the weight is further given to each side in the projection graph, so that the operation is easy and the expansibility is strong. According to the embodiment, the high-order network structure index and the high-order line graph index are integrated, various high-order characteristics of the flight delay network are utilized efficiently through integrated learning, accuracy of flight delay prediction is improved greatly, and particularly, a local structure which is easy to generate continuous delay in actual flights is predicted. The method has important practical significance for the principle of delay propagation, delay control and stable operation of flights.
Example 2
In another embodiment of the present invention, a flight delay prediction system based on a higher order network is disclosed, so as to implement the flight delay prediction method based on the higher order network in embodiment 1. The specific implementation of each module is described with reference to the corresponding description in embodiment 1. The system comprises:
the composite structure extraction module is used for constructing a high-order delay network according to the historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; extracting the composite structure based on the delay projection network and labeling the composite structure with delay labels;
the weight acquisition module is used for converting the high-order deferred network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order deferred network and the deferred projection network;
the model training module is used for calculating the weight of each composite structure based on the edge weight to serve as a high-order characteristic value, taking the high-order characteristic value as input, and taking the delay label as output training classification model to obtain a delay prediction model;
the delay prediction module is used for taking a plurality of airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into the delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted.
Because the related parts of the flight delay prediction system based on the higher-order network and the flight delay prediction method based on the higher-order network provided in this embodiment can be referred to each other, the description is repeated here, and therefore, the description is not repeated here. The principle of the system embodiment is the same as that of the method embodiment, so the system embodiment also has the corresponding technical effects of the method embodiment.
Those skilled in the art will appreciate that all or part of the flow of the methods of the embodiments described above may be accomplished by way of a computer program to instruct associated hardware, where the program may be stored on a computer readable storage medium. Wherein the computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (7)
1. A flight delay prediction method based on a high-order network is characterized by comprising the following steps:
constructing a high-order delay network according to historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels;
converting the high-order delay network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order delay network and the delay projection network;
based on the edge weights, calculating weights of all composite structures as high-order characteristic values, taking the high-order characteristic values as input, and taking delay labels as output training classification models to obtain delay prediction models;
taking a plurality of airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into a delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted;
the construction of the high-order delay network according to the historical flight data comprises the following steps:
sequencing the historical flight data according to the sequence of the flight time, identifying whether the landing delay time of each piece of historical flight data exceeds a delay threshold value in one day, and recording corresponding planes, dates and airport sets as the delay data if the landing delay time exceeds the delay threshold value; continuously identifying whether the same aircraft has delay between different airports for 2 or more times continuously in one day according to the current delay data, and adding the different airports into an airport set of the corresponding aircraft if the delay data exist;
taking out airport sets, neglecting the airport sequence in the airport sets, and putting different airport sets into a data set;
the airport set corresponds to an edge in a high-order delay network, and airports in the airport set are used as nodes contained by the edge, wherein if 2 airports exist in the airport set, the airport set corresponds to a common edge, and the order is 2; otherwise, the airport set corresponds to a higher-order side, and the order of the higher-order side is the number of airports in the airport set;
the step of converting the higher-order edge in the higher-order deferred network into a common edge to obtain a deferred projection network comprises the following steps:
taking out each higher-order edge in the higher-order delay network in turn, and constructing a common edge of the delay projection network between any two nodes contained in the current higher-order edge; each common edge in the high-order delay network is directly mapped into a common edge of the delay projection network;
the delay projection network-based extraction of the composite structure comprises the following steps:
according to the nodes of the delay projection network, traversing and taking out the combinations with the same node number, wherein the node number is more than 2;
and sequentially taking out each combination, identifying whether edges exist between any two nodes in the current combination in the deferred projection network, and if so, forming the current combination into a composite structure.
2. The method for high-order network-based flight delay prediction of claim 1, wherein labeling the composite structure comprises: if nodes in the composite structure all belong to the same high-order edge of the high-order delay network, the composite structure is a positive sample, and a label with delay is marked; otherwise, the composite structure is a negative sample, and a label without delay is marked.
3. The high-order network-based flight delay prediction method of claim 1, wherein the converting the high-order delay network into a line graph comprises:
acquiring edges which are smaller than or equal to an order threshold value in a high-order delay network as edges to be converted;
mapping the edges to be converted into nodes in the line graph, and if the same nodes exist in the two edges to be converted, constructing an edge between the nodes in the line graph corresponding to the two edges to be converted.
4. The higher order network-based flight delay prediction method of claim 1, wherein the calculating the index value of each node in the line graph is calculated based on a centrality index, the centrality index comprising: center of gravity, near center, feature vector center, cluster coefficients, web page rank, or core radius.
5. A higher order network-based flight delay prediction method as recited in claim 3, wherein said mapping the index value to edge weights in the higher order delay network and the delay projection network comprises: according to the one-to-one correspondence between the nodes in the line graph and the edges of the high-order delay network, directly taking the index values of the nodes in the line graph as the edge weights of the corresponding high-order delay network;
according to the one-to-many conversion relation between the edges in the high-order deferred network and the edges in the deferred projection network, the index value is used as the edge weight of each edge in the converted deferred projection network.
6. The method for predicting flight delay based on higher-order network as recited in claim 1, wherein calculating weights of the respective composite structures as higher-order eigenvalues based on edge weights comprises: according to the edge weight between any two nodes in the composite structure, the weight of the composite structure is calculated by a geometric mean, a harmonic mean or an arithmetic mean method to be used as a high-order characteristic value.
7. A high-order network-based flight delay prediction system, comprising:
the composite structure extraction module is used for constructing a high-order delay network according to the historical flight data; converting a high-order edge in the high-order delay network into a common edge to obtain a delay projection network; based on the delay projection network, extracting the composite structure and labeling the composite structure with delay labels;
the weight acquisition module is used for converting the high-order deferred network into a line graph, calculating index values of all nodes in the line graph, and mapping the index values into edge weights in the high-order deferred network and the deferred projection network;
the model training module is used for calculating the weight of each composite structure based on the edge weight to serve as a high-order characteristic value, taking the high-order characteristic value as input, and taking the delay label as output training classification model to obtain a delay prediction model;
the delay prediction module is used for taking a plurality of airports to be predicted as a composite structure to be predicted, calculating the weight of the composite structure to be predicted according to the edge weight in the delay projection network, and transmitting the weight into the delay prediction model to obtain a prediction result, wherein the prediction result is used for indicating whether flight delay occurs among the airports to be predicted;
the construction of the high-order delay network according to the historical flight data comprises the following steps:
sequencing the historical flight data according to the sequence of the flight time, identifying whether the landing delay time of each piece of historical flight data exceeds a delay threshold value in one day, and recording corresponding planes, dates and airport sets as the delay data if the landing delay time exceeds the delay threshold value; continuously identifying whether the same aircraft has delay between different airports for 2 or more times continuously in one day according to the current delay data, and adding the different airports into an airport set of the corresponding aircraft if the delay data exist;
taking out airport sets, neglecting the airport sequence in the airport sets, and putting different airport sets into a data set;
the airport set corresponds to an edge in a high-order delay network, and airports in the airport set are used as nodes contained by the edge, wherein if 2 airports exist in the airport set, the airport set corresponds to a common edge, and the order is 2; otherwise, the airport set corresponds to a higher-order side, and the order of the higher-order side is the number of airports in the airport set;
the step of converting the higher-order edge in the higher-order deferred network into a common edge to obtain a deferred projection network comprises the following steps:
taking out each higher-order edge in the higher-order delay network in turn, and constructing a common edge of the delay projection network between any two nodes contained in the current higher-order edge; each common edge in the high-order delay network is directly mapped into a common edge of the delay projection network;
the delay projection network-based extraction of the composite structure comprises the following steps:
according to the nodes of the delay projection network, traversing and taking out the combinations with the same node number, wherein the node number is more than 2;
and sequentially taking out each combination, identifying whether edges exist between any two nodes in the current combination in the deferred projection network, and if so, forming the current combination into a composite structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310009763.1A CN115689073B (en) | 2023-01-04 | 2023-01-04 | Flight delay prediction method and system based on high-order network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310009763.1A CN115689073B (en) | 2023-01-04 | 2023-01-04 | Flight delay prediction method and system based on high-order network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115689073A CN115689073A (en) | 2023-02-03 |
| CN115689073B true CN115689073B (en) | 2023-05-23 |
Family
ID=85057628
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310009763.1A Active CN115689073B (en) | 2023-01-04 | 2023-01-04 | Flight delay prediction method and system based on high-order network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115689073B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116362430B (en) * | 2023-06-02 | 2023-08-01 | 中国民航大学 | Flight delay prediction method and system based on online increment MHHA-SRU |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110675007A (en) * | 2019-12-06 | 2020-01-10 | 南京萨利智能科技有限公司 | Flight delay prediction method and device and computer storage medium |
| CN111950791A (en) * | 2020-08-14 | 2020-11-17 | 中国民航信息网络股份有限公司 | Flight delay prediction method, device, server and storage medium |
| CN112182059B (en) * | 2020-08-20 | 2022-05-27 | 北京航空航天大学 | High-order analysis method for flight delay characteristics |
-
2023
- 2023-01-04 CN CN202310009763.1A patent/CN115689073B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115689073A (en) | 2023-02-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107918830B (en) | Power distribution network running state evaluation method based on big data technology | |
| CN115412947B (en) | Fault simulation method and system based on digital twin and AI algorithm | |
| CN105335752A (en) | Principal component analysis multivariable decision-making tree-based connection manner identification method | |
| CN110990718B (en) | Social network model building module of company image lifting system | |
| CN115689073B (en) | Flight delay prediction method and system based on high-order network | |
| CN109299160B (en) | Power CPS security analysis method based on monitoring big data mining | |
| CN105471647A (en) | Power communication network fault positioning method | |
| CN111814528B (en) | Connectivity analysis noctilucent image city grade classification method | |
| He et al. | Community Detection in Aviation Network Based on K-means and Complex Network. | |
| CN107729939A (en) | A kind of CIM extended method and device towards newly-increased power network resources | |
| CN115734274A (en) | Cellular network fault diagnosis method based on deep learning and knowledge graph | |
| CN113344408A (en) | Processing method for multi-scale situation perception process of civil aviation traffic control operation | |
| Li et al. | A delay prediction model for high-speed railway: an extreme learning machine tuned via particle swarm optimization | |
| CN117009518A (en) | Similar event judging method integrating basic attribute and text content and application thereof | |
| Tian et al. | Assessment of flight block time reliability under different delay time windows: A case study | |
| Ji et al. | Cost Prediction of Distribution Network Project Based on DART Model | |
| CN114511905A (en) | Face clustering method based on graph convolution neural network | |
| CN116522213A (en) | Service state level classification and classification model training method and electronic equipment | |
| CN114116921A (en) | Simplification method for mining network asset graph based on equivalent structure | |
| Du et al. | Unstructured log oriented fault diagnosis for operation and maintenance management | |
| CN111080110A (en) | Airport congestion risk analysis system | |
| Hu et al. | A multi-attribute decision analysis method based on rough sets dealing with uncertain information | |
| CN116541527B (en) | Document classification method based on model integration and data expansion | |
| Cao et al. | Research on deep mining of MOOC multimodal resources based on improved Eclat algorithm | |
| Xiong | Initial clustering based on the swarm intelligence algorithm for computing a data density parameter |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |