LU501642B1 - Prediction method of departure flight delay based on timely information and deep learning - Google Patents
Prediction method of departure flight delay based on timely information and deep learning Download PDFInfo
- Publication number
- LU501642B1 LU501642B1 LU501642A LU501642A LU501642B1 LU 501642 B1 LU501642 B1 LU 501642B1 LU 501642 A LU501642 A LU 501642A LU 501642 A LU501642 A LU 501642A LU 501642 B1 LU501642 B1 LU 501642B1
- Authority
- LU
- Luxembourg
- Prior art keywords
- flight
- data
- information data
- deep learning
- model
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a departure flight delay prediction method based on timely information and deep learning, which relates to the technical field of computer prediction. by using a deep neural network model, combining aging flight information data and aging weather data, the departure flight delay prediction model is adopted to predict the flight delay situation, and numerical experiments are carried out by using real data. the results show that the constructed delay prediction model can obtain higher flight delay prediction accuracy in a short time, and has larger flight delay prediction aging. at the same time, with the increase of delay time threshold, the prediction accuracy is continuously improved and the loss value is continuously reduced. especially when 60 minutes is taken as the threshold, the prediction accuracy of the model can reach 91. 26%, which shows the effectiveness of the method of the invention.
Description
DESCRIPTION LU501642 Prediction method of departure flight delay based on timely information and deep learning
TECHNICAL FIELD The invention relates to the technical field of computer prediction, in particular to a departure flight delay prediction method based on timely information and deep learning.
BACKGROUND Due to weather, traffic and passengers, some flights are often delayed or cancelled. Under the influence of flight time-space network, it is easy to form a large number of flight delays. A large number of flight delays often cause conflicts between passengers, airports and airlines, which has become a potential problem affecting public social security. Therefore, early prediction of flight delays can give early warning to airlines, airports and related units, and win valuable time for formulating measures to flight delay, thus reducing economic losses caused by flight delays and improving passenger satisfaction. Therefore, flight delay prediction is of great practical significance to the civil aviation industry.
A large number of aviation workers have carried out research on flight delay prediction from different angles. At present, most of the researches use flight information data and meteorological data to predict flight delay, without considering the influence of the timeliness of input data on the timeliness of flight delay prediction, resulting in low timeliness of prediction results.
SUMMARY The embodiment of the invention provides a departure flight delay prediction method based on timely information and deep learning, which can solve the problems existing in the prior art.
The invention provides a departure flight delay prediction method based on timely information and deep learning, which comprises the following steps: Acquiring flight information data and meteorological information data; According to the flight information data and meteorological information data, the problem of departure flight delay prediction is converted into the form of g(X ,W), where X represents the flight information data, W represents the meteorological information data, when g(X ,W) = 1, the flight is delayed, when G (x, w) = 0, the flight is not delayed, and the deep learning model is used to find out the form of G (x, w). [0008] The departure flight delay prediction model baséd/501642 on deep learning is used to predict flight delay.
The departure flight delay prediction method based on timely information and deep learning in the invention can achieve higher prediction accuracy by using flight information data and meteorological data. By setting different flight delay time thresholds, it is found that with the increase of flight delay threshold, the prediction accuracy of the model increases continuously, and the prediction accuracy of the model can reach 91. 26% when the threshold is 60 minutes. By comparing and analyzing the prediction accuracy of the model on training set, verification set and test set, it can be seen that the model has good generalization performance. After analysis, due to the low real-time requirement of the model for input data, the model theoretically has a high time limit for flight delay prediction, which further enhances the practical application value of the method of the invention.
BRIEF DESCRIPTION OF THE FIGURES In order to more clearly explain the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only this.
According to some embodiments of the invention, for ordinary technicians in the field, other drawings can be obtained according to these drawings without paying creative labor.
Fig. 1 is a flow chart of the method of the present invention; Fig. 2 shows the change of prediction accuracy with the number of iterations when the delay time threshold is 15min.
Fig. 3 shows the change of loss value with the number of iterations when the delay time threshold is 15min.
Fig. 4 shows the change of prediction accuracy with the number of iterations when the delay time threshold is 30min.
Fig. 5 shows the change of loss value with the number of iterations when the delay time threshold is 30min.
Fig. 6 shows the change of prediction accuracy with the number of iterations when the delay time threshold is 60min.
Fig. 7 shows the change of loss value with the number of iterations when the delay tink&J501642 threshold is 60min.
DESCRIPTION OF THE INVENTION Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of the present invention.
Before explaining the method of the present invention, the definitions of some terms are explained: Definition 1: Flight delay refers to the fact that the actual flight arrival block time a; is more than 15 minutes later than the planned arrival time “.
Definition 2: Flight departure delay (flight departure delay) refers to the fact that the actual departure time d: of the flight is more than 15 minutes later than the planned departure time d, Definition 3: Flight delay prediction refers to giving whether the flight will be delayed or the time of flight delay before the flight delay occurs through certain technologies and methods.
Classification of flight delay prediction problems. According to the different stages of flight delay, the flight delay prediction problem can be divided into departure flight delay prediction, en route flight delay prediction, and inbound flight delay prediction.
Definition 4: The time limit of flight delay prediction refers to the length from the time when the flight delay prediction result is given by a certain technology and method to the scheduled departure time of the flight, that is, 7 7: >, where pr is the time when the flight delay prediction result is obtained, and #. is the scheduled departure time of the flight. Generally referred to as "pr delay report", there are common "2-hour delay report", "10-hour delay report", "24-hour delay report" and so on.
Definition 5. timely information means that a certain amount of time is given before the flight takes off through certain technologies and methods.
key information for the degree.
The usually given weather information of time limit is usually 40 days, 15 days, 7 days artd/501642 24 hours weather forecast; the actual time for removing the wheel block is generally not more than 1.5 hours; in order to use the more accurate information as much as possible, and give as long For the time-limited flight delay forecast, the present invention adopts 24-hour weather forecast instead of the actual wheel-off-block time.
From the perspective of passengers, the present invention studies the problem of delay prediction of departure flights. From the perspective of passengers, a flight delay of 15 minutes to 30 minutes is often acceptable. Therefore, different delay durations are selected as the threshold Th for judging whether the flight is delayed for research, namely: dd =T, Here Th is 15min, 30min, 60min respectively.
Definition 6: Flight information data, including airline, flight departure date, flight scheduled departure time, flight actual departure time, departure airport, destination airport, flight distance and other data; generally denoted as X.
Definition 7: Meteorological information data, including observation time, temperature, air pressure, visibility, cloud height, cloud amount, wind speed, wind direction, airport weather phenomena such as rain, snow, etc., and special weather thunderstorms that occur at the airport; generally denoted as W.
Referring to Fig. 1, the present invention provides an outbound flight delay prediction method based on timely information and deep learning, and the method includes the following steps: Step 1: Obtain flight information data and weather information data, and preprocess the data.
The meteorological information data obtained in the present invention is the meteorological forecast data 24 hours before the flight takes off, and the preprocessing of the flight information data and the meteorological information data includes: (1) Except for the selected flight information data and weather information data, delete other data in the obtained original data; (2) Delete the cancelled flight data in the flight information data;
(3) Convert the data of the weather phenomenon group in the meteorological informatid#/501642 data into numerical data, and the conversion rule is: if there is such a weather phenomenon, the data is set to 1, otherwise, it is set to 0; (4) Convert the discrete data such as airport and date in the flight information data into an n-bit binary string through one-hot encoding; (5) Standardize the numerical data in flight information data and weather information data by standard deviation standardization: a where - is the mean of the original data, and © is the standard deviation of the original data.
Step 2: According to the preprocessed flight information data and weather information data, the problem of delay prediction of departure flight is converted into the following form: i No Sight delays (X) Among them Aa is the flight information data, Xi, X> ,... , X14 are 14 items of data such as the departure airport, destination airport, planned departure time, flight actual arrival iw Wei 7: time, etc. Hs is meteorological data, and W1, Wo, ..., Wis are the weather conditions of the originating airport, including 18 data such as visibility, humidity, temperature, air pressure, wind speed, wind direction, the airport where the observation station is located and observation time.
The deep learning model is used to find the relationship between flight information data and meteorological information data in g(X ,W) form, and then the departure flight delay prediction model based on deep learning is established.
According to the invention, a deep learning model with a depth of 8 is selected to establish a departure flight delay prediction model, aiming at the over-fitting phenomenon existing in deep learning, a dropout regularization layer is added after each fully connected layer, and thé&J501642 selection of loss function and activation function is as follows.
The problem of flight delay prediction is essentially a two-category problem, and the cross information entropy function has a good performance in dealing with the two-category problem. Therefore, the cross information entropy function is chosen as the loss function of the evaluation model: J meee Flop) N°01 mt” Tr Regarding the selection of activation function, in the present invention, the hidden layer selects the tanh function as the activation function: Fxy=tanh{x)= € ne te While the problem of departure flight delay prediction is a kind of two-class problem, so the sigmod function is selected as the activation function of the output layer: Fly = i+e Step 3, use the departure flight delay prediction model based on deep learning to predict the flight delay.
Simulation and result analysis The data used are derived from the flight data of the US Department of Transportation in 2017, flight delay statistics and the historical data observed by weather stations at airports in 2017 by the US National Meteorological Data Center. Training data set, verification data set and test data set are 211,695 pieces of departure flight data from San Francisco Airport, including 1,711,472 pieces of training data set, 19,053 pieces of verification data set and 21,170 pieces of test data set. The departure flight delay prediction model based on deep learning specifically consists of 8 fully connected layers, in which the dropout regularization layer 1s added after each fully connected layer, and the tanh function is selected as the activation function between the input layer and the hidden layer. Because the departure flight delay prediction problem is a two-class problem, the sigmod function is selected as the activation function between the hidden layer and the output layer. The configuration of neurons in the input layer is determined according to the number of data items after fusion, and the number of neurons is 159 when the flight information data and meteorological data are fused. See Table 1 for the configuration 68501642 neurons in each hidden layer.
Table 1 Configuration of neurons in each layer of hidden layer Hidden layer serial number The first hidden layer The second hidden layer The third hidden layer The fifth hidden layer The sixth hidden layer | 0s | Flight delay prediction accuracy, also known as flight prediction accuracy Pacc, refers to the ratio between the number of correctly predicted flights and the total number of flights: “TPH TNA FN + FP The loss value refers to the distance evaluation value between the target value and the predicted value. The invention selects the cross information entropy as the loss function to calculate the loss value between the target value and the prediction. Accuracy and loss value are important indexes to evaluate the model, so the invention selects the prediction accuracy value and loss value to evaluate and analyze the model.
The symbol explanation of confusion matrix of flight delay prediction results is given below. Actual flight delays and predicted delays are true positive examples (TP), actual flight delays and predicted delays are true negative examples (TN), actual flight delays and predicted delays are false positive examples (FP), and actual flight delays and predicted delays are false negative examples (fn). See Table 2 for details.
Table 2 Confusion matrix of flight delay prediction results
PU i Fact isis 3 ı Delay | Withou delay : Pw TP FN FE 77 The prediction accuracy and loss value analysis under different thresholds are given below.
In the numerical experiment, 15 minutes, 30 minutes and 60 minutes were selected as flight/501642 delay time thresholds, and the model was simulated. The changes of the prediction accuracy and loss value of the model with the number of iterations in the training set and the verification set under the corresponding thresholds are shown in Figures 2 to 7.
From Figure 2, it can be seen that the prediction accuracy of the model on the training set and the verification set gradually improves to 80. 51% with the number of iterations, and the time limit of flight delay prediction is 24 hours. Fig. 3 shows that the loss value decreases steadily with the number of iterations on the training set and the verification set, thus illustrating the effectiveness of the model.
By comparing Figure 2 to Figure 7, we can see that with the increase of the threshold, the precision stability value of the model on the training set and the verification set is constantly improving, while the loss stability value is constantly decreasing. In addition, according to the change of the precision of the model on the training set and the verification set with the number of iterations, we can see that the over-fitting of the model 1s light, so the model 1s relatively stable. At the same time, the accuracy value of the model on the test set (see Table 3) is not much different from that on the training set and the verification set, which shows that the model has a good generalization effect.
Table 3 Model prediction accuracy under different thresholds ; Thesskold value | Precision of prediction . {= [San
EEE 3 ag ss. Re ë - i Sy € 3 EEE : à, = bln Cg Hey From Table 3, it can be seen that the prediction accuracy of the model is increasing with the increase of the threshold, which is due to the fact that the flight delay prediction is a partial classification problem, in which the flights without delay account for the majority. With the increase of the threshold, the proportion of flights without delay is further increased, and the prediction accuracy is further improved. At the same time, for passengers, whether the flight is delayed for 15 to 30 minutes often does not affect their travel experience. However, if the deldy’ 501642 is as long as 60 minutes or more, it will have a great impact on the travel experience of passengers, and the prediction accuracy of this model can reach 91. 26% when the threshold is 60 minutes, so the model has high practical value.
The flight information data in the model of the invention does not contain actual gear/turn-off time, so that the flight information data can be acquired in advance; The 24-hour weather forecast is relatively accurate, so theoretically, the departure flight delay forecast model based on timely information and deep learning proposed by the invention can realize "24-hour delay forecast", which also shows that the model has high practical application value.
Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concepts. Therefore, the appended claims are intended to be interpreted to include the preferred embodiments and all changes and modifications that fall within the scope of the present invention.
Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims (4)
1. A departure flight delay prediction method based on timely information and deep learning, characterized in that the method comprises the following steps: obtain flight information data and weather information data; according to the flight information data and the weather information data, the outbound flight delay prediction problem is converted into the form of g(X ,W), where x represents the flight information data, w represents the weather information data, when g(X ,W )=1, it means that the flight is delayed, and when g(X, W)=0, it means that the flight is not delayed; the deep learning model is used to find the relationship between the flight information data and the meteorological information data in the form of g(X, W), then establish a deep learning-based departure flight delay prediction model; a deep learning-based outbound flight delay prediction model is used to predict flight delays.
2. The outbound flight delay prediction method based on timely information and deep learning as claimed in claim 1, is characterized in that, after obtaining described flight information data and weather information data, also carry out preprocessing, and preprocessing comprises the following steps : except for the selected flight information data and weather information data, delete other data in the obtained original data; delete the canceled flight data in the flight information data; convert the data of the weather phenomenon group in the meteorological information data into numerical data; the conversion rule is: if there is such a weather phenomenon, set the data to 1, otherwise set it to 0; convert the airport, date and other discrete data in the flight information data into binary strings through one-hot encoding; standardize the data by standard deviation standardization on the numerical data in the flight information data and weather information data.
3. The method for predicting departure flight delay based on timely information and dedp}°01642 learning according to claim 1, wherein the obtained weather information data is the weather forecast data 24 hours before the flight takes off.
4. The outbound flight delay prediction method based on timely information and deep learning according to claim 1, is characterized in that, described deep learning based outbound flight delay prediction model adopts the deep learning model that the depth is 8 to establish, the each fully connected layer in the deep learning model has a dropout regularization layer; the cross information entropy function is used as the loss function of the evaluation model, the tanh function is selected as the activation function of the hidden layer, and the sigmod function is selected as the activation function of the output layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
LU501642A LU501642B1 (en) | 2022-03-10 | 2022-03-10 | Prediction method of departure flight delay based on timely information and deep learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
LU501642A LU501642B1 (en) | 2022-03-10 | 2022-03-10 | Prediction method of departure flight delay based on timely information and deep learning |
Publications (1)
Publication Number | Publication Date |
---|---|
LU501642B1 true LU501642B1 (en) | 2022-09-12 |
Family
ID=83231027
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
LU501642A LU501642B1 (en) | 2022-03-10 | 2022-03-10 | Prediction method of departure flight delay based on timely information and deep learning |
Country Status (1)
Country | Link |
---|---|
LU (1) | LU501642B1 (en) |
-
2022
- 2022-03-10 LU LU501642A patent/LU501642B1/en active IP Right Grant
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111144631B (en) | Real-time probability prediction method for flight delay based on Bayesian network algorithm | |
CN109448445A (en) | Flight based on shot and long term Memory Neural Networks is delayed grading forewarning system method | |
CN110796315B (en) | Departure flight delay prediction method based on aging information and deep learning | |
CN109409587A (en) | A kind of airport excavated based on weather data is into traffic flow forecasting method of leaving the theatre | |
Kern et al. | Data-driven aircraft estimated time of arrival prediction | |
Achenbach et al. | Prescriptive analytics in airline operations: Arrival time prediction and cost index optimization for short-haul flights | |
CN110443448B (en) | Bidirectional LSTM-based airplane position classification prediction method and system | |
Nan et al. | Prediction of Departure Aircraft Taxi Time Based on Deep Learning. | |
CN111160612A (en) | Off-site flight delay analysis and prediction method based on weather influence | |
Wang et al. | Modeling weather impact on ground delay programs | |
Grabbe et al. | Clustering days with similar airport weather conditions | |
CN113344408A (en) | Processing method for multi-scale situation perception process of civil aviation traffic control operation | |
Kim et al. | Airport and route classification by modelling flight delay propagation | |
CN113284369B (en) | Prediction method for actually measured airway data based on ADS-B | |
CN109145415B (en) | Air traffic flow time distance distribution analysis method | |
Khan et al. | A novel parallel series data-driven model for IATA-coded flight delays prediction and features analysis | |
CN114237110A (en) | Multi-mode data driving-based general aviation fleet reliability monitoring and early warning system | |
LU501642B1 (en) | Prediction method of departure flight delay based on timely information and deep learning | |
CN112182059B (en) | High-order analysis method for flight delay characteristics | |
Bardach et al. | Predicting flight delay risk using a random forest classifier based on air traffic scenarios and environmental conditions | |
CN117391653A (en) | Airport flight delay propagation prediction method and system | |
Dhanawade et al. | Analyzing factors influencing flight delay prediction | |
Jiang et al. | A multi-index prediction method for flight delay based on long short-term memory network model | |
Shao et al. | Air transportation delay propagation analysis with uncertainty in coloured–timed Petri nets | |
Hunter | Probabilistic forecasting of airport capacity |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FG | Patent granted |
Effective date: 20220912 |