CN104156594B - Dynamic flight station-crossing time estimation method based on Bayes network - Google Patents

Abstract

A method for dynamic estimation of flight transit time based on Bayesian network. It uses the method of data mining in the estimation of flight transit time. First, several factors that have a significant impact on flight transit time are extracted, and the Bayesian network is used to obtain the transit time estimation model, and then different conditions are obtained. Download the transit time estimate. When estimating the departure time of a flight, it is only necessary to know the probability distribution of the value of the transit time by using the transit time estimation model under the condition of the flight arrival information. On this basis, by calculating the expected value, we can get The possible values of the departure time. In addition, aiming at the characteristics of increasing flight data, the method used in the present invention can continuously learn the newly added data, and ensure the consistency of the learning result and the re-learning result of all data, so that the transit time estimation model can dynamically Adjust and regularly update transit time estimates to adapt to changing external conditions.

Description

A Dynamic Estimation Method of Flight Transit Time Based on Bayesian Network

technical field

The invention belongs to the technical field of civil aviation, in particular to a method for dynamically estimating flight transit time based on a Bayesian network.

Background technique

Flight delays, as the focus of air transport service disputes, have received more and more attention in recent years with the continuous growth of China's civil aviation traffic. Flight delays not only bring direct economic losses to airports and airlines, but also cause great inconvenience to passengers' normal travel, and seriously disrupt the normal order of airports. There are various reasons for flight delays, common reasons include weather, air traffic control, mechanical failure, aircraft deployment, flight planning, etc. Effective prediction of flight delays before they occur is a goal expected to be achieved in the development of the civil aviation industry. For airlines, each aircraft executes multiple flights within one day, and effectively predicts the delay of downstream flights in the case of delays in upstream flights, which is helpful for improving the service quality of airline companies and improving the airline's profitability. Competitiveness has important practical significance.

Flyontime.us is a free flight delay time analysis system open to the public in the United States. The system is free and open to the whole society, through which anyone can query and analyze the delay rate, airport waiting time, flight schedule and weather information of various flights in the United States. The data used by the system mainly comes from the U.S. Department of Transportation, and the time spent waiting in line at the security check is obtained through the submission of ordinary travelers to the system. The system mainly uses statistical methods to study the distribution of all flight delays at each airport in a certain period of time.

American company FlightCaster has developed a flight delay information service system, which uses an advanced algorithm to collect the data of each domestic flight in the past few years, and then matches it with the real-time situation to determine the delay of the flight, which can predict the next few hours in advance Flight status within. It also provides terminal applications for iPhone and Blackberry, providing flight delay forecasts.

Europe attaches great importance to flight information services. The European civil aviation management agency (EUROCONTROL) has implemented a scientific and technological plan for the research and development of the European AIS Database (EAD), which integrates and integrates the information of the national aviation information system transit time rule library. The world's largest centralized aviation information service system, EAD's online aviation information service has covered most countries in Europe.

Singapore Changi Airport is one of the busiest large-scale hub airports in Asia. The information release system of the airport can provide real-time and accurate flight dynamic information. , landing and other news) has been updated and released in real time.

In recent years, some domestic airports, such as Capital Airport and New Baiyun Airport, have provided passengers with dynamic flight query services through the establishment of websites and telephone call centers. Some also set up exclusive pages on Weibo and other social networking sites to explain and communicate to passengers through text and pictures when there is a flight delay.

Feiyou Technology (Civil Aviation Resource Network) has developed related products such as "Very Accurate" and "VariFlight" to provide passengers with free flight dynamic customization services, and also provide flight information services to some small airports.

However, the above-mentioned system has the following problems: in general, the information is incomplete, and the randomness of each influencing factor is relatively large. Therefore, there is currently a lack of a method that can use data mining to learn flight history data. Estimate the transit time under the influence, so as to effectively estimate the transit time when encountering the same situation.

Contents of the invention

In order to solve the above problems, the object of the present invention is to provide a method for dynamically estimating flight transit time based on a Bayesian network.

In order to achieve the above object, the flight time dynamic estimation method based on Bayesian network provided by the invention comprises the following steps carried out in order:

Step 1: Preprocess the historical flight data, and extract the data including the arrival delay time of the previous flight, the arrival time period of the previous flight, the planned transit time, the departure airport, the aircraft type and the actual transit time as influencing factors ;

Step 2: Assuming that the above factors have an influence on the actual transit time and the influence is independent of each other, then determine the topology of the Bayesian network; then use the data obtained in the previous step to obtain the Bayesian network topology by using the maximum likelihood estimation method. YES network parameters, thus obtaining the transit time estimation model;

Step 3: Reasoning the estimated transit time model obtained above to obtain estimated transit time values under different conditions;

Step 4: Predict the departure time of the incoming flight;

Step 5: After a period of time, use the method of step 1 to preprocess the new historical flight data to obtain new training samples; then use the previously obtained model as prior knowledge, combine with the new training samples, and use Bayesian The estimation method corrects the parameters of the Bayesian network; after the parameters are corrected, the new transit time estimation model is reasoned, and the transit time rule base is updated;

Step 6: Repeat step 5 regularly to dynamically update the transit time rule base.

In step 1, the flight history data includes flight number, aircraft number, flight planned departure time, planned landing time, actual departure time, actual landing time, departure airport, destination airport and flight seat number; previous flight arrival delay The time is the actual arrival time of the previous flight minus the planned arrival time of the previous flight, and the planned transit time is the planned departure time of the next flight in the flight schedule minus the planned arrival time of the previous flight.

The described method of adopting the maximum likelihood estimation method to obtain the Bayesian network parameters, thus obtaining the estimation model of the transit time is:

suppose is the vector composed of all parameters, n is the number of nodes, q _i is the number of value combinations of π(X _i ), r _i is the number of values of node X _i , θ _ijk =P(X _i =k/ π(X _i )=j) is the probability that when the value of the parent node of X _i is the jth value, the value of X _i is the kth value, D _i ,i=1,2,...,m is the sample data, then the vector The log-likelihood function for is:

$\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; log</span>}\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

When θ _ijk takes the following values, the logarithmic likelihood function achieves the maximum value:

Among them, m _ijk is the number of samples satisfying Xi = k, π(X _i ) = _j in the flight data _, and r _i is the value number of node Xi; if the historical flight data source The parameters are assumed to be uniformly distributed; the parameters of the Bayesian network are determined by formula (2), and the transit time estimation model is thus obtained.

Reasoning on the obtained transit time estimation model, the method to obtain the estimated value of transit time under different conditions is as follows:

3.1 Reasoning the transit time estimation model to obtain the probability distribution of transit time in different situations;

3.2 Find the expected value of the transit time:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Where E(t) is the expected value of transit time when other conditions are determined, T _i is the median value of the i-th transit time interval, P(i) is the probability of the transit time in the i-th interval, to obtain The expected value of the transit time E(t) is used as the estimated value of the transit time in this situation;

3.3 Insert the transit time estimates under different conditions into the transit time rule base to provide conditions for flight departure time prediction.

In step 4, the method of predicting the departure time of the incoming flight is as follows; if there is no arrival delay on the previous flight or the delay time is less than 10 minutes, then the estimated departure time of the next flight is the planned time; if the delay time If it is greater than 10 minutes, the estimated transit time under the corresponding conditions is obtained from the transit time rule library; if the actual arrival time plus the estimated transit time is less than the planned transit time of the next flight, the estimated transit time of the next flight The departure time is the planned departure time; if the actual arrival time of the previous flight plus the estimated transit time is greater than the planned departure time of the next flight, the estimated departure time of the next flight is the actual arrival time plus If the delay time is greater than 10 minutes and there is no corresponding transit time information in the transit time rule base, the estimated transit time is the planned transit time, and the estimated departure time of the next flight is the actual Arrival time plus planned transit time.

In step five, the method of correcting the parameters of the Bayesian network using the Bayesian estimation method is as follows:

suppose for the reason α _ijk is the number of samples satisfying Xi = k and π(X _i ) = _j in the prior knowledge, is a Dirichlet distribution From Bayesian formula:

$\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&Integral;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

in as a vector The prior probability distribution of , as a vector The posterior probability distribution of ; since Obey the Dirichlet distribution $\mathrm{D.}=[m_{\mathrm{ij}1}+{\mathrm{\&alpha;}}_{\mathrm{ij}1},m_{\mathrm{ij}2}+{\mathrm{\&alpha;}}_{\mathrm{ij}2},...,{m_{\mathrm{ijr}}}_i+{{\mathrm{\&alpha;}}_{\mathrm{ijr}}}_i],$ so:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Formula (5) is used to correct the parameters of the Bayesian network, where m _ijk is the number of samples satisfying Xi =k and π(X _i )= _j in the new samples.

The method for dynamically estimating flight transit time based on Bayesian networks provided by the present invention uses the method of data mining in the estimation of flight transit time, and first extracts several factors that have a significant impact on flight transit time, and uses Yesnet obtained the transit time estimation model, and then obtained the estimated transit time under different conditions. When estimating the departure time of a flight, it is only necessary to know the probability distribution of the value of the transit time by using the transit time estimation model under the condition of the flight arrival information. On this basis, by calculating the expected value, we can get The possible values of the departure time. In addition, aiming at the characteristics of increasing flight data, the method used in the present invention can continuously learn the newly added data, and ensure that the learning result is consistent with the re-learning result of all data, so that the transit time estimation model can dynamically Adjust and regularly update transit time estimates to adapt to changing external conditions.

Description of drawings

Fig. 1 is a flow chart of the method for dynamically estimating flight transit time based on the Bayesian network provided by the present invention.

Fig. 2 is a flow chart of flight departure time estimation in the method for dynamically estimating flight transit time based on Bayesian network provided by the present invention.

detailed description

The method for dynamically estimating flight transit time based on the Bayesian network provided by the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

As shown in Figure 1, the flight time dynamic estimation method based on Bayesian network provided by the invention comprises the following steps carried out in order:

Step 1: Preprocess the historical flight data, and extract the data including the arrival delay time of the previous flight, the arrival time period of the previous flight, the planned transit time, the departure airport, the aircraft type and the actual transit time as influencing factors ;The flight history data includes flight number, aircraft number, flight scheduled departure time, planned landing time, actual departure time, actual landing time, departure airport, destination airport and the number of seats on the flight; the arrival delay time of the previous flight is the actual arrival time of the previous flight Time minus the scheduled arrival time of the previous flight, and the planned transit time is the scheduled departure time of the next flight in the flight schedule minus the scheduled arrival time of the previous flight;

Step 2: Assuming that the above-mentioned factors have an influence on the actual transit time and the influence is independent of each other, the Bayesian network topology is determined; in the process of determining the Bayesian network topology, the nodes representing the aircraft type are based on There are four value states for the different number of seats on the flight, below 90 seats is model A, 90-160 seats is model B, 160-230 seats is model C, and more than 230 seats is model D; it represents the arrival time of the previous flight There are 24 states for the node value of a segment, and each state represents a time span of one hour. For example, H13 represents that the arrival time of the previous flight is between 13:00 and 13:59; the arrival delay time of the previous flight and the planned transit time 1. Each state of the actual transit time takes 10 minutes as an interval; each state of a node representing an airport represents an airport. After determining the topology of the Bayesian network, using the data obtained in the previous step, the parameters of the Bayesian network are obtained by using the maximum likelihood estimation method, and thus the transit time estimation model is obtained:

suppose is the vector composed of all parameters, n is the number of nodes, q _i is the number of value combinations of π(X _i ), r _i is the number of values of node X _i , θ _ijk =P(X _i =k/ π(X _i )=j) is the probability that when the value of the parent node of X _i is the jth value, the value of X _i is the kth value, D _i ,i=1,2,...,m is the sample data, then the parameter vector The log-likelihood function for is:

$\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; log</span>}\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Pi;</span>}}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; log</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; D.</span>}}_{\mathrm{<span\; class="notranslate">\; l</span>}}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

When θ _ijk takes the following values, the logarithmic likelihood function achieves the maximum value:

Among them, m _ijk is the number of samples satisfying Xi = k, π(X _i ) = _j in the flight data _, and r _i is the value number of node Xi; if the historical flight data source Then the parameters are assumed to be uniformly distributed. The parameters of the Bayesian network are determined by the formula (2), and the transit time estimation model is thus obtained;

Step 3: Infer the transit time estimation model obtained above to obtain the transit time estimates under different conditions. The specific steps are as follows:

3.1 Reasoning the transit time estimation model to obtain the probability distribution of transit time in different situations;

3.2 Find the expected value of the transit time:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Where E(t) is the expected value of transit time when other conditions are determined, T _i is the median value of the i-th transit time interval, P(i) is the probability of the transit time in the i-th interval, to obtain The expected value of the transit time E(t) is used as the estimated value of the transit time in this situation;

3.3 Insert the transit time estimates under different conditions into the transit time rule base to provide conditions for flight departure time prediction.

Step 4: Predict the departure time of the incoming flight; when predicting, if there is no arrival delay on the previous flight or the delay time is less than 10 minutes, then the estimated departure time of the next flight is the planned time; if the delay time is greater than 10 minutes Minutes, the estimated transit time under the corresponding situation is obtained from the transit time rule library; if the actual arrival time plus the estimated transit time is less than the planned transit time of the next flight, the estimated departure time of the next flight The time is the scheduled departure time; if the actual arrival time of the previous flight plus the estimated transit time is greater than the planned departure time of the next flight, the estimated departure time of the next flight is the actual arrival time plus the estimated Transit time; if the delay time is greater than 10 minutes, and there is no corresponding transit time information in the transit time rule base, the estimated transit time is the planned transit time, and the estimated departure time of the next flight is the actual arrival time Time plus planned transit time.

Step 5: After a period of time, use the method of step 1 to preprocess the new historical flight data to obtain new training samples; then use the previously obtained model as prior knowledge, combine with the new training samples, and use Bayesian The estimation method corrects the parameters of the Bayesian network;

suppose for the reason The composed sub-vector, α _ijk is the number of samples satisfying X _i =k and π(X _i )=j in the prior knowledge, is a Dirichlet distribution From Bayesian formula:

$\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&Integral;</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; d</span>}\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

in as a vector The prior probability distribution of , as a vector The posterior probability distribution of ; since Obey the Dirichlet distribution $\mathrm{D.}=[m_{\mathrm{ij}1}+{\mathrm{\&alpha;}}_{\mathrm{ij}1},m_{\mathrm{ij}2}+{\mathrm{\&alpha;}}_{\mathrm{ij}2},...,{m_{\mathrm{ijr}}}_i+{{\mathrm{\&alpha;}}_{\mathrm{ijr}}}_i],$ so:

$\mathrm{<span\; class="notranslate">\; E.</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&theta;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}}{\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; ijk</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

Formula (5) is used to correct the parameters of the Bayesian network, where m _ijk is the number of samples satisfying Xi =k and π(X _i )= _j in the new samples. After the parameters are corrected, the new transit time estimation model is reasoned and the transit time rule base is updated.

Step 6: Repeat step 5 regularly to dynamically update the transit time rule base.

Claims (6)

1. a kind of flight turnaround on airport method for dynamic estimation based on Bayesian network it is characterised in that: described method include by The following step that order is carried out:

Step one: pretreatment is carried out to history flight data, therefrom extract including before flight reach delay time at stop, front flight and arrive Reach time period, plan turnaround on airport, original base, type of airplane and actual turnaround on airport in interior data as influence factor；

Step 2: assume that above-mentioned each factor all has impact and affects to be separate on actual turnaround on airport, thereby determine that Go out Bayesian network topological structure；Then the data obtaining with previous step, obtains Bayesian network using maximum likelihood estimate Parameter, thus obtains turnaround on airport and estimates model；

Step 3: model makes inferences to be estimated to turnaround on airport obtained above, obtains front flight and reach delay time at stop, front flight The time of advent section, plan turnaround on airport, original base, this five variables of type of airplane take turnaround on airport during different value to estimate respectively Evaluation；

Step 4: the Departure airport of the flight that approaches is predicted；

Step 5: through after a period of time, the method using step one carries out pretreatment to new history flight data, obtains new Training sample；Then using the model obtaining before as priori, in conjunction with new training sample, using Bayes' assessment Revise Bayesian network parameter；After having revised parameter, model makes inferences to be estimated to new turnaround on airport, and update when missing the stop Between rule base；

Step 6: repeat at periodic or other desired step 5, dynamically update time rule storehouse of missing the stop.

2. the flight turnaround on airport method for dynamic estimation based on Bayesian network according to claim 1 it is characterised in that: In step one, described flight historical data includes flight number, aircraft number, the flight planning departure time, plan landing time, reality The border departure time, Actual Time Of Fall, original base, destination airport and airline seat number；The front flight arrival delay time at stop is front The flight actual time of arrival deducts front flight planning time of advent, and plan turnaround on airport is the meter of Next Flight in flight planning table Drawing the departure time deducts plan time of advent of a flight.

3. the flight turnaround on airport method for dynamic estimation based on Bayesian network according to claim 1 it is characterised in that: In step 2, described obtains Bayesian network parameter using maximum likelihood estimate, thus obtains turnaround on airport and estimates model Method is:

AssumeFor the vector of all parameters composition, n is node number, q_iFor π(x_i) valued combinations number, r_iFor node x_iValue number, θ_ijk=p (x_i=k/ π (x_i)=j) it is to work as x_iFather node value For j-th value, x_iValue is the probability of k-th value, d_i, i=1,2 ..., m is sample data, then vectorLogarithm seemingly So function is:

$l(\stackrel{\&rightarrow;}{\mathrm{\&theta;}}/d)=log\underset{l=1}{\overset{m}{\&pi;}}p(d_l/\stackrel{\&rightarrow;}{\mathrm{\&theta;}})=\underset{l=1}{\overset{m}{\&sigma;}}\log p(d_l/\stackrel{\&rightarrow;}{\mathrm{\&theta;}})---\left(1\right)$

Work as θ_ijkWhen taking following value, log-likelihood function acquirement maximum:

Wherein, m_ijkIt is in flight data, to meet x_i=k, π (x_iThe sample size of)=j；If in history flight data sourceThen setting parameter is to be uniformly distributed；Bayesian network parameter is determined by formula (2), thus obtains turnaround on airport and estimate mould Type.

4. the flight turnaround on airport method for dynamic estimation based on Bayesian network according to claim 1 it is characterised in that: In step 3, model makes inferences to be estimated to the turnaround on airport obtaining, when obtaining front flight arrival delay time at stop, the arrival of front flight Between section, plan turnaround on airport, original base, this five variables of type of airplane take turnaround on airport estimated value during different value respectively Method is:

3.1 pairs of turnaround on airport estimate that model makes inferences, and obtain front flight and reach delay time at stop, the front flight section time of advent, meter Draw turnaround on airport, original base, this five changes of type of airplane measure the probability distribution of turnaround on airport during a certain definite value；

3.2 pairs of turnaround on airport seek expected value:

$e\left(t\right)=\underset{i}{\&sigma;}p\left(i\right)*t_i---\left(3\right)$

Wherein e (t) is to reach delay time at stop, the front flight section time of advent, plan turnaround on airport, original base, fly in front flight This five changes of machine type measure turnaround on airport expected value during a certain definite value, t_iFor the intermediate value that i-th turnaround on airport is interval, p (i) is Probability in i-th interval for the turnaround on airport, in the hope of turnaround on airport expected value e (t) as front flight reach the delay time at stop, The estimation of turnaround on airport when the front flight section time of advent, plan turnaround on airport, original base, this five variables of type of airplane determine Value；

3.3 the turnaround on airport estimated value under different condition is inserted in turnaround on airport rule base, is flight for putting forth time prediction Offer condition.

5. the flight turnaround on airport method for dynamic estimation based on Bayesian network according to claim 1 it is characterised in that: In step 4, to the method that Departure airport of the flight that approaches is predicted it is；If a upper flight does not approach and being delayed or prolongs It is less than between mistaking 10 minutes, then the estimation departure time of Next Flight is planned time；If the delay time at stop is more than 10 minutes, The estimation turnaround on airport under respective conditions is drawn from turnaround on airport rule base；If actual enter ETA estimated time of arrival when missing the stop plus estimation Between less than Next Flight plan turnaround on airport, then the estimation Departure airport of Next Flight be plan the Departure airport；If upper one Flight actual enter ETA estimated time of arrival add that the turnaround on airport estimated is more than the plan Departure airport of Next Flight, then the estimating of Next Flight Meter the Departure airport be actual enter ETA estimated time of arrival add estimate turnaround on airport；If the delay time at stop is more than 10 minutes, and turnaround on airport There is no corresponding turnaround on airport information, then the turnaround on airport estimated is plan turnaround on airport, the estimation of Next Flight in rule base Departure airport be actual enter ETA estimated time of arrival add plan turnaround on airport.

6. the flight turnaround on airport method for dynamic estimation based on Bayesian network according to claim 1 it is characterised in that: In step 5, the method using Bayes' assessment correction Bayesian network parameter is:

AssumeIt is by θ_ij1,θ_ij2,...,The subvector being formed, α_ijkMeet x in priori_i=k and π (x_i)=j Sample size,For the distribution of Di Li CrayObtained by Bayesian formula:

$p(\stackrel{\&rightarrow;}{\mathrm{\&theta;}}/d)=\frac{p\left(\stackrel{\&rightarrow;}{\mathrm{\&theta;}}\right)p(d/\stackrel{\&rightarrow;}{\mathrm{\&theta;}})}{\&integral;p\left(d\stackrel{\&rightarrow;}{\mathrm{\&theta;}}\right)d\stackrel{\&rightarrow;}{\mathrm{\&theta;}}}=\frac{p\left(\stackrel{\&rightarrow;}{\mathrm{\&theta;}}\right)p(d/\stackrel{\&rightarrow;}{\mathrm{\&theta;}})}{p\left(d\right)}---\left(4\right)$

WhereinFor vectorPrior probability distribution,For vectorPosterior probability distribution；Due to Obey the distribution of Di Li CraySo:

$e({\mathrm{\&theta;}}_{ijk}/d)=\frac{m_{ijk}+{\mathrm{\&alpha;}}_{ijk}}{\underset{k=1}{\overset{r_i}{\&sigma;}}(m_{ijk}+{\mathrm{\&alpha;}}_{ijk})}---\left(5\right)$

Bayesian network parameter, wherein m are revised using formula (5)_ijkMeet x in new samples_i=k and π (x_iThe sample size of)=j.