An airport delay level prediction method based on association rules and LSTM
By combining association rules and LSTM models, an airport delay level prediction system was constructed, which solved the problems of spatiotemporal propagation and insufficient interpretability in airport delay prediction, and achieved fast and accurate delay prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CIVIL AVIATION UNIV OF CHINA
- Filing Date
- 2022-10-31
- Publication Date
- 2026-05-29
AI Technical Summary
Existing airport delay prediction methods mainly consider the propagation characteristics of flight delays in the time dimension, lacking consideration of spatiotemporal propagation. Deep learning models also suffer from insufficient interpretability and slow training speed.
An airport delay level prediction method based on association rules and LSTM is adopted. By constructing an airport delay index calculation module, a level identification module, a spatiotemporal association rule mining module, and a prediction module, and combining FP-Growth and fuzzy C-means clustering algorithms, spatiotemporal association rules are mined and prediction is performed using an LSTM model.
It improves the accuracy and interpretability of airport delay prediction, while accelerating model training and providing reliable prediction basis.
Smart Images
Figure CN115660181B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of air traffic management technology, and specifically relates to a method for predicting airport delay levels based on Association Rules (AR) and LSTM (Long Short-Term Memory) models. Background Technology
[0002] Currently, compared to 1950, my country's civil aviation industry has experienced exponential growth in both annual airport takeoffs and landings and total passenger and cargo traffic. Simultaneously, the number of aircraft is increasing rapidly, while existing route design, planning, and traffic management systems are unable to respond in a timely manner. This results in a large number of flights being unable to take off or land at airports as scheduled, leading to severe airport delays. Therefore, timely and accurate prediction of airport delays is crucial, enabling air traffic management departments to take appropriate measures based on the predictions.
[0003] Currently, methods for predicting flight delays generally fall into three categories: prediction using modeling and simulation, and prediction using data-driven methods. Air traffic systems are composed of many different entities, and these elements interact with each other, forming a nonlinear, aggregated dynamic characteristic, which poses a challenge to accurate delay prediction. Modeling and simulation-based prediction methods primarily establish mathematical models, requiring sophisticated use of many model parameters and incorporating unrealistic assumptions. Data-driven methods often directly utilize past flight operation data from a specific airport to predict future delays, mainly considering the temporal propagation of delays and rarely considering the spatial propagation characteristics of delays between different airports.
[0004] With the rise of computer and artificial intelligence technologies, many scholars have proposed deep learning-based prediction methods for airport flight delays. Deep learning-based multilayer prediction models, especially LSTM models, have a strong ability to extract nonlinear features from time series data, and their performance in airport flight delay prediction surpasses traditional machine learning methods. While LSTM models perform well in prediction results, a significant limitation is the difficulty in observing the model's learning process; the training process is like a "black box," making the output difficult to interpret, thus affecting its credibility and acceptability. Furthermore, achieving satisfactory prediction results requires a large amount of data for training, resulting in a slow optimization process.
[0005] In summary, the current research has the following main problems: (1) Current airport delay prediction mainly considers the propagation characteristics of flight delays in the time dimension, but lacks consideration of the propagation of delays in time and space at the same time; (2) Simple deep learning prediction often lacks certain explanations for the results, thus affecting the credibility and acceptability of the results; (3) Deep learning models require a large amount of training data input, and the optimization speed is slow. Summary of the Invention
[0006] To address the aforementioned problems, the present invention aims to provide a method for predicting airport delay levels based on association rules and LSTM.
[0007] To achieve the above objectives, the airport delay level prediction method based on association rules and LSTM provided by this invention includes the following steps performed in sequence:
[0008] 1) Construct an airport delay level prediction system based on association rules and LSTM, consisting of an airport delay index calculation module, an airport delay level identification module, an airport delay spatiotemporal association rule mining module, and an AR-LSTM-based airport delay level prediction module. The airport delay index calculation module is mainly used to calculate airport delay index data; the airport delay level identification module is mainly used to identify the delay levels of each airport at different times; the airport delay spatiotemporal association rule mining module is mainly used to mine the spatiotemporal association rules of delay propagation between airports; and the AR-LSTM-based airport delay level prediction module is mainly used to predict the delay levels of airports in future time periods.
[0009] 2) In the above airport delay index calculation module, the delay time of each flight's takeoff and landing is calculated based on flight plan data and actual operation data; then, based on the delay time of each flight, the average delay time and flight delay rate of each airport at different times are calculated.
[0010] 3) In the airport delay level identification module, the average delay time and flight delay rate of each airport at different time periods obtained in step 2) are used to classify the delay levels of each airport at different time periods using the fuzzy C-means clustering algorithm;
[0011] 4) In the airport delay spatiotemporal association rule mining module, based on the delay level classification results of each airport at different time periods obtained in step 3), an airport delay transaction set Y is constructed. The airport delay transaction set Y contains multiple airport delay itemsets I, and the airport delay itemsets I contain the time period of the delay, the airport where the delay occurred, and the delay level label.
[0012] 5) In the airport delay spatiotemporal association rule mining module, based on the above-mentioned airport delay transaction set Y, the spatiotemporal association rule mining method based on FP-Growth is applied to mine the spatiotemporal association rules of airport delays.
[0013] 6) In the AR-LSTM-based airport delay level prediction module, based on the strong association rules mined in step 5), strong association rules associated with the airport to be predicted are extracted for the airport and the prediction period. Then, the LSTM method is applied to predict the future delay level of the airport and calculate the prediction accuracy.
[0014] In step 2), the delay times for takeoff and landing of each flight are calculated based on flight schedule data and actual operation data; then, the specific method for calculating the average delay time and flight delay rate of each airport at different times based on the delay times of each flight is as follows:
[0015] 2.1) Calculate the takeoff / landing delay time for each flight based on the actual takeoff / landing time and the estimated takeoff / landing time:
[0016]
[0017] Where, d ijk This represents the takeoff / landing delay time of the k-th flight in the j-th time period at airport i. and Let represent the actual takeoff / landing time and the estimated takeoff / landing time of the k-th flight in the j-th time period at airport i;
[0018] 2.2) Based on the takeoff / landing delay times of the above flights, the average delay time and flight delay rate of each airport in different time periods are calculated, with one hour as a time period. The average delay time refers to the average delay time of all takeoff / landing flights within the study period. The flight delay rate refers to the ratio of the number of delayed flights to the total number of takeoff / landing flights within the study period.
[0019]
[0020]
[0021] Among them, D ij Let N be the average delay time of airport i during time period j. ij R represents the number of flights taking off and landing at airport i during time period j. ij γ represents the flight delay rate at airport i during time period j. ijk The delay discrimination value γ is the delay discrimination value for the k-th flight in the j-th time period at airport i. When the delay time of this flight is greater than or equal to 15 minutes, the delay discrimination value γ is...ijk A value of 1 indicates the flight is considered delayed; otherwise, a value of 0 indicates the flight is considered non-delayed.
[0022] In step 3), the specific method for classifying the delay levels of each airport at different times using the average delay times and flight delay rates obtained in step 2) is as follows:
[0023] 3.1) Standardize the average delay time and flight delay rate obtained in step 2), as shown in the following formula:
[0024]
[0025] Where: Q is the average delay time or flight delay rate to be standardized, μ is the mean of all average delay times or flight delay rates, and σ is the standard deviation of all average delay times or flight delay rates.
[0026] 3.2) Using the average delay time and flight delay rate of each airport at different times after standardization as samples, the delay level of each airport at different times is divided into four levels: normal, mild delay, moderate delay and severe delay, and labeled with the numbers 1, 2, 3 and 4 respectively.
[0027] In step 5), the specific method for mining the spatiotemporal association rules of airport delays based on the above-mentioned airport delay transaction set Y is as follows:
[0028] 5.1) The relevant definitions of spatiotemporal association rules are as follows:
[0029] Definition 1:
[0030] Items and Itemsets: Define a set I = {I1, I2, ..., I...} m}, where I is an itemset, and element I i For an itemset I to contain K items, it is called a K-itemset.
[0031] Definition 2:
[0032] Transaction set: Let transaction set D = {T1, T2, ..., T} n}, each transaction Y i (i = 1, 2, ..., n) each represent an event, and each transaction is a subset of itemset I;
[0033] Definition 3:
[0034] Support of association rules: For an association rule X→Y, X∈I, Y∈I, and The support sum(X→Y) is defined as the frequency of transactions containing both X and Y in transaction set D. The calculation formula is as follows:
[0035]
[0036] Definition 4:
[0037] The confidence level of an association rule is the probability that itemset Y will occur given that itemset X occurs within the same transaction. Essentially, it is a conditional probability, P(Y|X), calculated using the following formula:
[0038]
[0039] Definition 5:
[0040] Minimum Support of Itemsets and Frequent Itemsets: Minimum Support min This is a threshold set before the mining rules are implemented. As long as the support of itemset I is greater than this threshold, it is a frequent itemset. If the itemset is a k-itemset, then the frequent itemset is called a frequent k-itemset.
[0041] Definition 6:
[0042] Minimum confidence and strong association rules: Minimum confidence conf min It is also a pre-set threshold; if the conditional pattern composed of elements in the frequent itemset meets the minimum confidence requirement, the conditional pattern can be extracted and recorded as a strong association rule to characterize the effective intrinsic relationship between transactions.
[0043] 5.2) Set minimum support (sup) min and minimum confidence conf min The value range is 0-1;
[0044] 5.3) Scan the airport delay transaction set Y, sort the transaction-related data in descending order of itemset frequency, and delete items with less than the minimum support to obtain frequent itemsets;
[0045] 5.4) Construct an FP-tree. Scan again to create a frequent item header table and construct an FP-tree from the sorted and deleted transactions in the airport delay transaction set Y in descending order. An FP-tree is a tree structure consisting of a root node marked as empty and an item prefix subtree. Each node in the item prefix subtree consists of two fields: item name and support count. The item name indicates which item the node represents, and the support count indicates the number of transactions in the path up to the current node.
[0046] 5.5) Combining all frequent itemsets to form a conditional pattern will satisfy the minimum confidence level (conf). min The conditional pattern output is a spatiotemporal association rule;
[0047] 5.6) Retain the two-items and three-items from the spatiotemporal association rules mined in step 5.5) as strong association rules.
[0048] In step 6), based on the strong correlation rules mined in step 5), for the airport to be predicted and the prediction period, strong correlation rules associated with the predicted airport are extracted. Then, the LSTM method is applied to predict the future delay level of the airport and calculate the prediction accuracy. The specific method is as follows:
[0049] 6.1) Constructing the sample set: The sample set mainly consists of two parts. One part is a rule sample that is consistent with the strong correlation rules mined based on daily airport delay data. The other part is a time series sample composed of the predicted airport delay index sequence.
[0050] 6.2) Use 80% of the above sample set as the training dataset and the remaining 20% as the test dataset;
[0051] 6.3) Input the samples from the training dataset into the LSTM model for training. The LSTM model architecture is LSTM1→LSTM2→ReLU→Softmax, where LSTM1 and LSTM2 are training layers used to select features, learn and train on the samples, and fit the feature factors of airport delays input in the previous step; the ReLU layer is the activation function layer, which ensures the effectiveness of the training results and improves learning efficiency; finally, the training results are classified and output through the Softmax layer.
[0052] 6.4) Backpropagation is performed on the above LSTM model using the mean squared error loss function. During this process, the relevant parameters in the neurons are updated, and the training iteration termination condition is determined by calculating whether the loss result reaches the preset value.
[0053] 6.5) When the iteration termination condition is met, the LSTM model training is completed, thus obtaining the airport delay level prediction model based on AR-LSTM. Then, the samples in the test dataset are input into the airport delay level prediction model based on AR-LSTM to obtain the delay level prediction value.
[0054] 6.6) Compare the predicted delay level obtained in the previous step with the actual delay level to calculate the prediction accuracy, as shown in the following formula:
[0055]
[0056] Where N is the total number of samples in the sample set. Let be the predicted delay level for the i-th sample. Let be the actual value of the delay level for the i-th sample.
[0057] Advantages of this invention:
[0058] The airport delay level prediction model constructed in this invention applies association rules with spatiotemporal information to the LSTM model, taking into account the impact of other airport delays on the predicted airport, as well as the impact of different time periods on the predicted time period, while improving the optimization speed of the LSTM model.
[0059] The data used in this invention is easy to collect, and through big data mining methods, the patterns of delay propagation can be obtained quickly and conveniently, and the future delay situation can be predicted, which has practical operational value.
[0060] The airport delay level prediction model constructed in this invention provides a new modeling approach for the research field of airport delay prediction in China, enriches the research methods in the field of delay prediction, and the delay prediction results can provide decision-making basis for airports, airlines, and air traffic control. Attached Figure Description
[0061] Figure 1 This is a flowchart of the delay prediction method based on association rules and LSTM model provided by the present invention.
[0062] Figure 2 This is a graph showing the delay level classification results in an embodiment of the present invention. Detailed Implementation
[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are merely for explaining the invention and are not intended to limit the invention.
[0064] like Figure 1 As shown, the airport delay level prediction method based on association rules and LSTM provided by this invention includes the following steps performed in sequence:
[0065] 1) Construct an airport delay level prediction system based on association rules and LSTM, consisting of an airport delay index calculation module, an airport delay level identification module, an airport delay spatiotemporal association rule mining module, and an AR-LSTM-based airport delay level prediction module. The airport delay index calculation module is mainly used to calculate airport delay index data; the airport delay level identification module is mainly used to identify the delay levels of each airport at different times; the airport delay spatiotemporal association rule mining module is mainly used to mine the spatiotemporal association rules of delay propagation between airports; and the AR-LSTM-based airport delay level prediction module is mainly used to predict the delay levels of airports in future time periods.
[0066] 2) In the above airport delay index calculation module, the delay time of each flight's takeoff and landing is calculated based on flight plan data and actual operation data; then, based on the delay time of each flight, the average delay time and flight delay rate of each airport at different times are calculated.
[0067] The specific method is as follows:
[0068] 2.1) Calculate the takeoff / landing delay time for each flight based on the actual takeoff / landing time and the estimated takeoff / landing time:
[0069]
[0070] Where, d ijk This represents the takeoff / landing delay time of the k-th flight in the j-th time period at airport i. and Let represent the actual takeoff / landing time and the estimated takeoff / landing time of the k-th flight in the j-th time period at airport i;
[0071] 2.2) Based on the takeoff / landing delay times of the above flights, the average delay time and flight delay rate of each airport in different time periods are calculated, with one hour as a time period. The average delay time refers to the average delay time of all takeoff / landing flights within the study period. The flight delay rate refers to the ratio of the number of delayed flights to the total number of takeoff / landing flights within the study period.
[0072]
[0073]
[0074] Among them, D ij Let N be the average delay time of airport i during time period j. ij R represents the number of flights taking off and landing at airport i during time period j. ij γ represents the flight delay rate at airport i during time period j. ijk The delay discrimination value γ is the delay discrimination value for the k-th flight in the j-th time period at airport i. When the delay time of this flight is greater than or equal to 15 minutes, the delay discrimination value γ is... ijk A value of 1 indicates the flight is considered delayed; otherwise, a value of 0 indicates the flight is not delayed.
[0075] 3) In the airport delay level identification module, the average delay time and flight delay rate of each airport at different time periods obtained in step 2) are used to classify the delay levels of each airport at different time periods using the fuzzy C-means (FCM) clustering algorithm;
[0076] The specific method is as follows:
[0077] 3.1) Standardize the average delay time and flight delay rate obtained in step 2), as shown in the following formula:
[0078]
[0079] Where: Q is the average delay time or flight delay rate to be standardized, μ is the mean of all average delay times or flight delay rates, and σ is the standard deviation of all average delay times or flight delay rates.
[0080] 3.2) Using the average delay time and flight delay rate of each airport at different times after standardization as samples, the delay level of each airport at different times is divided into four levels: normal, slight delay, moderate delay and severe delay, and labeled with the numbers 1, 2, 3 and 4 respectively.
[0081] 4) In the airport delay spatiotemporal association rule mining module, based on the delay level classification results of each airport at different time periods obtained in step 3), an airport delay transaction set Y is constructed. The airport delay transaction set Y contains multiple airport delay itemsets I, and the airport delay itemsets I contain the time period of the delay, the airport where the delay occurred, and the delay level label.
[0082] A sample table of airport delay itemsets is shown in Table 1:
[0083] Table 1 Sample table of airport delay items
[0084]
[0085] 5) In the airport delay spatiotemporal association rule mining module, based on the above-mentioned airport delay transaction set Y, the spatiotemporal association rule mining method based on FP-Growth is applied to mine the spatiotemporal association rules of airport delays.
[0086] The specific method is as follows:
[0087] 5.1) The relevant definitions of spatiotemporal association rules are as follows:
[0088] Definition 1:
[0089] Items and Itemsets: Define a set I = {I1, I2, ..., I...} m Let I be an itemset, and let element I be an item set. i For an itemset I, if there are K items, then the itemset I is called a K-itemset; for example, the itemset I = {Airport A is delayed, Airport B is not delayed, Airport C is delayed, Airport D is not delayed} is a 4-itemset.
[0090] Definition 2:
[0091] Transaction set: Let transaction set D = {T1, T2, ..., T} n}, each transaction Y i (i = 1, 2, ..., n) represent an event, and each transaction is a subset of itemset I; for example, if the transaction set is the delay status of each airport every hour throughout the year, then the delay status of these airports in a certain hour can constitute a transaction.
[0092] Definition 3:
[0093] Support of association rules: For an association rule X→Y, X∈I, Y∈I, and The support sum(X→Y) is defined as the frequency of occurrence of transactions containing both X and Y in a transaction set D. Support reflects the importance of this transaction in the transaction set, and the calculation formula is as follows:
[0094]
[0095] Definition 4:
[0096] The confidence level of an association rule is the probability that itemset Y will occur given that itemset X occurs within the same transaction. Essentially, it is a conditional probability, P(Y|X), calculated using the following formula:
[0097]
[0098] Definition 5:
[0099] Minimum Support of Itemsets and Frequent Itemsets: Minimum Support min This is a threshold set before rule mining. If the support of itemset I is greater than this threshold, it is considered a frequent itemset (FP). The smaller the minimum support is set, the more association rules will be mined. Furthermore, if an itemset is a k-itemset, it is called a frequent k-itemset.
[0100] Definition 6:
[0101] Minimum confidence and strong association rules: and minimum support min Similarly, minimum confidence level (conf) min It is also a pre-set threshold; if the conditional pattern composed of elements in the frequent itemset meets the minimum confidence requirement, the conditional pattern can be extracted and recorded as a strong association rule to characterize the effective intrinsic relationship between transactions.
[0102] 5.2) Set minimum support (sup) min and minimum confidence conf min The value range is 0-1;
[0103] 5.3) Scan the airport delay transaction set Y, sort the transaction-related data in descending order of itemset frequency, and delete items with less than the minimum support to obtain frequent itemsets;
[0104] 5.4) Construct an FP-tree. Scan again to create a frequent item header table and construct an FP-tree from the sorted and deleted transactions in the airport delay transaction set Y in descending order. An FP-tree is a tree structure consisting of a root node marked as NULL and an item prefix subtree. Each node in the item prefix subtree consists of two fields: item name and support count. The item name indicates which item the node represents, and the support count indicates the number of transactions in the path up to the current node.
[0105] 5.5) Combining all frequent itemsets to form a conditional pattern will satisfy the minimum confidence level (conf). min The conditional pattern output is a spatiotemporal association rule, thereby realizing the mining of all spatiotemporal association rules;
[0106] 5.6) To improve the calculation speed of subsequent delay prediction, the two-items and three-items in the spatiotemporal association rules mined in step 5.5) are retained as strong association rules for subsequent prediction.
[0107] 6) In the AR-LSTM-based airport delay level prediction module, based on the strong association rules mined in step 5), strong association rules associated with the airport to be predicted are extracted for the airport and the prediction period. Then, the LSTM method is applied to predict the future delay level of the airport and calculate the prediction accuracy.
[0108] The specific method is as follows:
[0109] 6.1) Constructing the sample set: The sample set mainly consists of two parts. One part is a rule sample that is consistent with the strong correlation rules mined based on daily airport delay data. The other part is a time series sample composed of the predicted airport delay index sequence.
[0110] 6.2) Use 80% of the above sample set as the training dataset and the remaining 20% as the test dataset;
[0111] 6.3) Input the samples from the training dataset into the LSTM model for training. The LSTM model architecture is LSTM1→LSTM2→ReLU→Softmax, where LSTM1 and LSTM2 are training layers used to select features, learn and train on the samples, and fit the feature factors of airport delays input in the previous step; the ReLU layer is the activation function layer, which ensures the effectiveness of the training results and improves learning efficiency; finally, the training results are classified and output through the Softmax layer.
[0112] 6.4) The mean squared error (MSE) loss function is used to backpropagate the above LSTM model. During this process, the relevant parameters in the neurons are updated, and the training iteration termination condition is determined by calculating whether the loss result reaches the preset value.
[0113] 6.5) When the iteration termination condition is met, the LSTM model training is completed, thus obtaining the airport delay level prediction model based on AR-LSTM. Then, the samples in the test dataset are input into the airport delay level prediction model based on AR-LSTM to obtain the delay level prediction value.
[0114] 6.6) Compare the predicted delay level obtained in the previous step with the actual delay level to calculate the prediction accuracy. This accuracy is used to evaluate the prediction performance of the AR-LSTM airport delay level prediction model. The formula is as follows:
[0115]
[0116] Where N is the total number of samples in the sample set. Let be the predicted delay level for the i-th sample. Let be the actual value of the delay level for the i-th sample.
[0117] Here are some application examples of the airport delay level prediction method based on association rules and LSTM provided by this invention:
[0118] (1) Construct an airport delay level prediction system based on association rules and LSTM, consisting of an airport delay index calculation module, an airport delay level identification module, an airport delay spatiotemporal association rule mining module, and an airport delay level prediction module based on AR-LSTM.
[0119] (2) In the airport delay index calculation module, the actual flight operation data and planned data of the United States throughout 2019 are used to calculate the average delay time and flight delay rate of each airport in each hour.
[0120] (3) In the airport delay level identification module, the average delay time and flight delay rate calculated in the previous step are standardized. Then, the standardized average delay time and flight delay rate of each airport at different times are used as samples and divided into 1-4 levels using the FCM clustering algorithm. Level 1 represents normal delay, level 2 represents mild delay, level 3 represents moderate delay, and level 4 represents severe delay. The delay level of each airport at different times is then identified. The classification results are as follows: Figure 2 As shown.
[0121] (4) In the airport delay spatiotemporal association rule mining module, based on the delay level identification results of each airport at different time periods obtained in step 3), the airport delay transaction set Y is constructed.
[0122] (5) In the airport delay spatiotemporal association rule mining module, based on the airport delay transaction set Y, the spatiotemporal association rule mining method based on FP-Growth is applied to mine the spatiotemporal association rules of airport delays. The minimum support is set to 0.3 and the minimum confidence is set to 0.75. The binary set and triplet are mined as the strong association rules of airport delays mined by this module. Some rules are shown in Table 2. In the table, the first two digits of each item represent the time period, the middle three letters represent the airport's three-letter code, and the last digit represents the delay level. Rule 5 means: when the delay level of DTW airport from 10:00 to 11:00 is 1 (normal), and the delay level of LAX airport from 12:00 to 13:00 is 2 (mild delay), then the delay level of ORD airport from 11:00 to 12:00 is 3 (moderate delay). The confidence of this rule is 91.3%. This rule illustrates that the delays of other airports affect the target airport through propagation.
[0123] Table 2 Examples of Strong Association Rules for Airport Delays
[0124]
[0125] (6) Based on the strong correlation rules obtained in the previous step, extract the strong correlation rules associated with the predicted airport. For example, to predict the delay level of ORD airport from 11:00 to 12:00, rules 1 and 5 in Table 2 need to be extracted. Further, based on the extracted strong correlation rules, average delay time, and flight delay rate, a sample set is constructed. 80% of the samples are input into the LSTM model for training, and the remaining 20% are used for testing. Taking ORD airport as an example again, its prediction accuracy for different time periods between 11:00 and 20:00 is shown in Table 3.
[0126] Table 3 Analysis of Forecast Results for Different Time Periods
[0127]
[0128] Note: Contents not described in detail in this specification are prior art known to those skilled in the art. The solution of this invention can be implemented by combining the methods described in the specification with the art itself.
Claims
1. A method for predicting airport delay levels based on association rules and LSTM, characterized in that: The airport delay level prediction method based on association rules and LSTM includes the following steps performed in sequence: 1) Construct an airport delay level prediction system based on association rules and LSTM, consisting of an airport delay index calculation module, an airport delay level identification module, an airport delay spatiotemporal association rule mining module, and an AR-LSTM-based airport delay level prediction module. The airport delay index calculation module is mainly used to calculate airport delay index data; the airport delay level identification module is mainly used to identify the delay levels of each airport at different times; the airport delay spatiotemporal association rule mining module is mainly used to mine the spatiotemporal association rules of delay propagation between airports; and the AR-LSTM-based airport delay level prediction module is mainly used to predict the delay levels of airports in future time periods. 2) In the above airport delay index calculation module, the delay time of each flight's takeoff and landing is calculated based on flight plan data and actual operation data; then, based on the delay time of each flight, the average delay time and flight delay rate of each airport at different times are calculated. 3) In the airport delay level identification module, the average delay time and flight delay rate of each airport at different time periods obtained in step 2) are used to classify the delay levels of each airport at different time periods using the fuzzy C-means clustering algorithm; 4) In the airport delay spatiotemporal association rule mining module, based on the delay level classification results of each airport at different time periods obtained in step 3), an airport delay transaction set Y is constructed. The airport delay transaction set Y contains multiple airport delay itemsets I, and the airport delay itemsets I contain the time period of the delay, the airport where the delay occurred, and the delay level label. 5) In the airport delay spatiotemporal association rule mining module, based on the above-mentioned airport delay transaction set Y, the spatiotemporal association rule mining method based on FP-Growth is applied to mine the spatiotemporal association rules of airport delays. 6) In the AR-LSTM-based airport delay level prediction module, based on the strong association rules mined in step 5), strong association rules associated with the airport to be predicted are extracted for the airport and the prediction period. Then, the LSTM method is applied to predict the future delay level of the airport and calculate the prediction accuracy.
2. The airport delay level prediction method based on association rules and LSTM according to claim 1, characterized in that: In step 2), the delay times for takeoff and landing of each flight are calculated based on flight schedule data and actual operation data; then, the specific method for calculating the average delay time and flight delay rate of each airport at different times based on the delay times of each flight is as follows: 2.1) Calculate the takeoff / landing delay time for each flight based on the actual takeoff / landing time and the estimated takeoff / landing time: Where, d ijk This represents the takeoff / landing delay time of the k-th flight in the j-th time period at airport i. and Let represent the actual takeoff / landing time and the estimated takeoff / landing time of the k-th flight in the j-th time period at airport i; 2.2) Based on the takeoff / landing delay times of the above flights, the average delay time and flight delay rate of each airport in different time periods are calculated, with one hour as a time period. The average delay time refers to the average delay time of all takeoff / landing flights within the study period. The flight delay rate refers to the ratio of the number of delayed flights to the total number of takeoff / landing flights within the study period. Among them, D ij Let N be the average delay time of airport i during time period j. ij R represents the number of flights taking off and landing at airport i during time period j. ij γ represents the flight delay rate at airport i during time period j. ijk The delay discrimination value γ is the delay discrimination value for the k-th flight in the j-th time period at airport i. When the delay time of this flight is greater than or equal to 15 minutes, the delay discrimination value γ is... ijk A value of 1 indicates the flight is considered delayed; otherwise, a value of 0 indicates the flight is considered non-delayed.
3. The airport delay level prediction method based on association rules and LSTM according to claim 1, characterized in that: In step 3), the specific method for classifying the delay levels of each airport at different times using the average delay times and flight delay rates obtained in step 2) is as follows: 3.1) Standardize the average delay time and flight delay rate obtained in step 2), as shown in the following formula: Where: Q is the average delay time or flight delay rate to be standardized, μ is the mean of all average delay times or flight delay rates, and σ is the standard deviation of all average delay times or flight delay rates. 3.2) Using the average delay time and flight delay rate of each airport at different times after standardization as samples, the delay level of each airport at different times is divided into four levels: normal, mild delay, moderate delay and severe delay, and labeled with the numbers 1, 2, 3 and 4 respectively.
4. The airport delay level prediction method based on association rules and LSTM according to claim 1, characterized in that: In step 5), the specific method for mining the spatiotemporal association rules of airport delays based on the above-mentioned airport delay transaction set Y is as follows: 5.1) The relevant definitions of spatiotemporal association rules are as follows: Definition 1: Items and Itemsets: Define a set I = {I1, I2, ..., I...} m }, where I is an itemset, and element I i For an itemset I to contain K items, it is called a K-itemset. Definition 2: Transaction set: Let transaction set D = {T1, T2, ..., T} n }, each transaction Y i (i = 1, 2, ..., n) each represent an event, and each transaction is a subset of itemset I; Definition 3: Support of association rules: For an association rule X→Y, X∈I, Y∈I, and The support sum(X→Y) is defined as the frequency of transactions containing both X and Y in transaction set D. The calculation formula is as follows: Definition 4: The confidence level of an association rule is the probability that itemset Y will occur given that itemset X occurs within the same transaction. Essentially, it is a conditional probability, P(Y|X), calculated using the following formula: Definition 5: Minimum Support of Itemsets and Frequent Itemsets: Minimum Support min This is a threshold set before the mining rules are implemented. As long as the support of itemset I is greater than this threshold, it is a frequent itemset. If the itemset is a k-itemset, then the frequent itemset is called a frequent k-itemset. Definition 6: Minimum confidence and strong association rules: Minimum confidence conf min It is also a pre-set threshold; If a conditional pattern consisting of elements within a frequent itemset satisfies the minimum confidence requirement, the conditional pattern can be extracted and recorded as a strong association rule to characterize the effective intrinsic relationship between transactions. 5.2) Set minimum support (sup) min and minimum confidence conf min The value range is 0-1; 5.3) Scan the airport delay transaction set Y, sort the transaction-related data in descending order of itemset frequency, and delete items with less than the minimum support to obtain frequent itemsets; 5.4) Construct an FP-tree, scan again, and create a frequent item header table and construct an FP-tree by sorting and deleting the transactions in the airport delay transaction set Y in descending order from top to bottom; An FP-tree is a tree structure consisting of a root node marked as empty and a project prefix subtree. Each node in the project prefix subtree consists of two fields: project name and support count. The project name indicates which project the node represents, and the support count indicates the number of transactions in the path up to the current node. 5.5) Combining all frequent itemsets to form a conditional pattern will satisfy the minimum confidence level (conf). min The conditional pattern output is a spatiotemporal association rule; 5.6) Retain the two-items and three-items from the spatiotemporal association rules mined in step 5.5) as strong association rules.
5. The airport delay level prediction method based on association rules and LSTM according to claim 1, characterized in that: In step 6), based on the strong correlation rules mined in step 5), for the airport to be predicted and the prediction period, strong correlation rules associated with the predicted airport are extracted. Then, the LSTM method is applied to predict the future delay level of the airport and calculate the prediction accuracy. The specific method is as follows: 6.1) Constructing the sample set: The sample set mainly consists of two parts. One part is a rule sample that is consistent with the strong correlation rules mined based on daily airport delay data. The other part is a time series sample composed of the predicted airport delay index sequence. 6.2) Use 80% of the above sample set as the training dataset and the remaining 20% as the test dataset; 6.3) Input the samples in the training dataset into the LSTM model for training. The LSTM model architecture is LSTM1→LSTM2→ReLU→Softmax, where LSTM1 and LSTM2 are training layers used to select features, learn and train the samples, and fit the feature factors of airport delays input in the previous step. The ReLU layer is the activation function layer, which ensures the effectiveness of the training results and improves learning efficiency; finally, the Softmax layer classifies and outputs the training results. 6.4) Backpropagation is performed on the above LSTM model using the mean squared error loss function. During this process, the relevant parameters in the neurons are updated, and the training iteration termination condition is determined by calculating whether the loss result reaches the preset value. 6.5) When the iteration termination condition is met, the LSTM model training is completed, thus obtaining the airport delay level prediction model based on AR-LSTM. Then, the samples in the test dataset are input into the airport delay level prediction model based on AR-LSTM to obtain the delay level prediction value. 6.6) Compare the predicted delay level obtained in the previous step with the actual delay level to calculate the prediction accuracy, as shown in the following formula: Where N is the total number of samples in the sample set. Let be the predicted delay level for the i-th sample. Let be the actual value of the delay level for the i-th sample.