A flight model adjustment optimization method

Through integer planning models and machine learning to predict flight data, flight models are automatically scheduled, which solves the problem of low manual adjustment efficiency, and achieves efficient optimization of flight model scheduling and adapts to the needs of large-scale route networks.

CN114462672BActive Publication Date: 2024-11-08CHINA SOUTHERN AIRLINES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111616158.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-27
Publication Date
2024-11-08
Estimated Expiration
2041-12-27

AI Technical Summary

Technical Problem

In the prior art, flight model adjustments rely on manual methods, which are inefficient and have poor results, and cannot meet the needs of large-scale and automated flight plan optimization.

Method used

The integer planning model is used combined with machine learning methods to automatically schedule the optimal flight model by predicting the number of seats, income and cost of flights, and meet the constraints and objective functions to realize the flight model scheduling plan.

Benefits of technology

It improves the efficiency and effectiveness of flight model adjustment, adapts to the needs of large-scale route networks, maximizes capacity resource allocation, reduces resource consumption, and enhances flight revenue.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114462672B_ABST
    Figure CN114462672B_ABST
Patent Text Reader

Abstract

The present invention discloses a flight model adjustment optimization method, comprising: step S1, reading the data of the target airline from the database of the target airline on the flight plan date; step S2, predicting: the predicted number of passengers and predicted revenue of the target flight f, and the predicted cost when the aircraft of model k is used to fly; and calculating the predicted revenue according to the prediction result; step S3, solving and obtaining the optimal flight model scheduling plan that meets the constraint conditions and the objective function; step S4, according to the optimal flight model scheduling plan, scheduling the aircraft of the corresponding model k for the target flight f to fly. The present invention can automatically generate the optimal flight model scheduling plan that meets the predetermined constraint conditions and the objective function, so as to schedule the aircraft of the corresponding model k for the target flight f in the target flight set F to fly according to the plan, and has the advantages of high efficiency and good effect, and can adapt to the current increasingly large-scale route network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of civil aviation passenger transport, and in particular to a flight model adjustment and optimization method. Background Art

[0002] Flight planning is the foundation and core of all production activities of airlines. It carries forward the strategic planning of airlines and connects to the actual revenue of airlines. It is a very complex and labor-intensive task, and the quality of its preparation directly affects the profits of airlines. The essence of the aircraft model optimization problem in flight planning is to adjust the aircraft models of flights according to the different seat numbers, operating costs and potential revenues of different aircraft models, with the goal of maximizing the marginal contribution of the route network, and ultimately achieve the maximization of the marginal contribution of the entire network or other goals.

[0003] When the route network reaches a certain scale, manual adjustment and optimization will lag far behind actual needs in terms of efficiency and effect. It is necessary to introduce the concepts and technologies of operations research and overall planning to achieve large-scale, automated global optimization of flight plans. Therefore, relying on manual experience to complete aircraft model optimization has low work efficiency and cannot achieve optimal resource allocation. Introducing advanced machine learning and operations research methods to solve aircraft model optimization problems is conducive to improving work efficiency, optimizing capacity allocation, and achieving the goal of reducing costs and increasing efficiency.

[0004] Flight planning and its optimization are key areas of operations research application. Currently, the world's leading companies in this field include SABRE in the United States, JEPPESEN under Boeing, and Lufthansa in Germany. However, the systems developed by foreign countries are quite different from the actual business of domestic airlines, and the application is not good. The domestic research level in this field is relatively backward, and it is urgent to make efforts to solve such problems.

[0005] As the size of airline fleets continues to expand, the complexity of route network management is gradually increasing, and the demand for medium- and short-term flight optimization services continues to grow. The existing route network management system can only solve daily flight scheduling management problems. The system functions are limited by conditions such as data, structure, user-friendliness, and rule support, and the business cannot complete more practical applications. At present, flight planning business personnel use manual experience to find interchangeable model flights, and the range of choices is small. Relying on manual experience to complete model optimization work is inefficient and cannot achieve optimal resource allocation. In order to establish an advanced route network planning process and improve airline flight revenue and work efficiency, it is urgent to implement a set of flight plan model adjustment functions that meet the business characteristics of airlines. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a flight model adjustment optimization method to solve the problem of low efficiency and poor effect of manually adjusting the flight operating model in the prior art.

[0007] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] A flight model adjustment optimization method, characterized by comprising:

[0009] Step S1, record the Tth day before the end time of the model adjustment target date as the flight plan date, T≥1, and the duration of the model adjustment target date is less than the value of T; on the flight plan date, read the following data of the target airline from the database of the target airline through the corresponding data interface: a target flight set F whose departure time is on the model adjustment target date, a model set K of all operable aircraft that can fly on the model adjustment target date, and historical flight data related to the target flight set F; wherein the model adjustment target date can be one day or multiple days.

[0010] Step S2, based on the historical flight data, predicting: the predicted number of passengers and the predicted revenue of each target flight f in the target flight set F, and the predicted cost of each target flight f when operated by an aircraft of model k, k∈K;

[0011] And, based on the prediction results, the predicted revenue of each target flight f when using an available aircraft of model k is calculated. That is, the difference between the predicted revenue and the predicted cost;

[0012] Step S3: Establish an integer programming model and use the target flight set F, aircraft model set K and predicted revenue The solution is performed to obtain an optimal flight model scheduling plan that satisfies the constraints and objective function of the integer programming model; wherein, the solution process of the optimal flight model scheduling plan can transmit data to a commercial solver through a network for solution, and then transmit the solution result back through the network; the data can be written in a standard format readable by commercial solvers such as .mps, .lp, etc., or the integer programming model can be established using the interface provided by the solver.

[0013] Step S4: on the target date for adjusting the aircraft type, according to the optimal flight aircraft type scheduling plan, dispatching an available aircraft of the corresponding aircraft type k for the target flight f in the target flight set F to fly.

[0014] Preferably: in step S3, the constraint conditions include:

[0015]

[0016]

[0017] and

[0018]

[0019]

[0020]

[0021] The objective function includes:

[0022]

[0023]

[0024]

[0025] Where, F represents the target flight set whose departure time of the target airline is on the target date of the aircraft model adjustment, f represents the target flight in the target flight set F, and the symbol It means any;

[0026] K represents the set of all aircraft types that the target airline can operate on the target aircraft type adjustment date, and k represents the aircraft type in the aircraft type set K;

[0027] A represents the set of take-off and landing airports of all the target flights f, a represents an element in the take-off and landing airport set A, that is, the take-off and landing airport, and the take-off and landing airport a is a take-off airport or a landing airport;

[0028] N k represents the take-off and landing time-space node set of all aircraft of model k in the model set K, and n represents the take-off and landing time-space node set N k The elements in , namely, the take-off and landing space-time nodes, the take-off and landing space-time node n is a take-off space-time node including the take-off time and the take-off airport or a landing space-time node including the landing time and the landing airport; The take-off time and space node of the first flight of the operable aircraft of model k on the target date of adjustment of the model; The take-off and landing time-space node set representing all the aircraft of model k in the aircraft set K that can fly at 3 a.m. on the target date for adjusting the aircraft model;

[0029] O(k, n) represents the set of flights that take off from the take-off and landing space-time node n using an aircraft of model k, and the take-off and landing space-time node n is the take-off space-time node; I(k, n) represents the set of flights that land at the take-off and landing space-time node n using an aircraft of model k, and the take-off and landing space-time node n is the landing space-time node;

[0030] F c represents the set of cancelable target flights f in the target flight set F; M represents the set F c The set of flights with the same flight number in the set M, m represents a subset of the set M, and the departure times of the target flights f in the subset m have the same number of days of the week, for example, the departure times are all Mondays or all Fridays;

[0031] represents the set of target flights f in flight at the take-off and landing time and space node n in the target flight set F;

[0032] represents the aircraft model flight decision variable, where When the value is 1, it means that the target flight f is operated by an aircraft of model k. When the value is 0, it means that the aircraft of model k is not used to fly the target flight f; all the model flight decision variables of the target flight f A value combination method of forms a flight model scheduling plan for the target flight set F;

[0033] Indicates the number of aircraft of model k that can fly before the take-off and landing time and space node n; Indicates the number of aircraft of model k that can fly after the take-off and landing time and space node n;

[0034] Relative to the initial quantity The absolute value of the number of new aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented; Relative to the initial quantity The absolute value of the reduction in the number of aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented;

[0035] The initial quantity The number of aircraft of type k that can fly at the take-off and landing airport a before executing the corresponding flight type scheduling plan, that is, the initial time of the target date of the type adjustment;

[0036] N represents the preset upper limit of the number of model adjustments;

[0037] N kIndicates the number of aircraft of model k in model set K that can fly, which is set according to the actual situation of model set K;

[0038] N m represents the minimum number of flights m in the preset set M;

[0039] represents the predicted revenue of the target flight f when the target flight f is operated by an aircraft of model k that can fly;

[0040] k f Indicates the aircraft type used by the target flight f in the original plan before the corresponding flight aircraft type scheduling plan is executed;

[0041] The process of solving the optimal flight model scheduling plan is:

[0042] The first step is to exhaustively enumerate all the aircraft model and flight decision variables of the target flight f under the premise of satisfying the constraints described in formulas 1 to 6. , so as to find a value combination that meets the formula 7; if there is only one value combination in the search result, then the value combination is the optimal flight model scheduling plan, and the solution process ends; if there are multiple value combinations in the search result, the found value combination is recorded as the first target flight model scheduling plan, and the second step is executed;

[0043] in,

[0044] Formula 1 indicates: flow balance, that is, for any aircraft type and any take-off and landing time and space node n, the number of aircraft before the take-off and landing time and space node n should be equal to the number of aircraft after the take-off and landing time and space node n;

[0045] Formula 2 means: Adjust the number of aircraft types, that is, the flight decision variable for the aircraft type that is different from the original aircraft type of the flight The sum should be less than the preset adjustment model number limit;

[0046] Formula 3 represents: flight coverage constraint, that is, for any non-cancellable flight, all the aircraft model flight decision variables related to the flight The sum should be equal to 1;

[0047] Formula 4 indicates: the aircraft model quantity limit, that is, for any checkpoint at 3 a.m. and any aircraft model, the sum of the number of aircraft on the ground plus the sum of aircraft in flight should be less than or equal to the number of aircraft of that model;

[0048] Formula 5 means: cancel the flight constraint, that is, for the flight set m and the non-virtual aircraft model flight decision variable The sum should be greater than the preset number;

[0049] Formula 6 indicates: the aircraft connection quantity constraint, that is, for each aircraft type and each airport (the number of aircraft + the number of aircraft reduced - the number of aircraft increased) should be equal to the input initial number of aircraft;

[0050] Formula 7 indicates that: the number of flights covered is the largest, which means that the decision variables of non-virtual aircraft models and flights are Maximize the sum.

[0051] Step 2: Find a flight model scheduling plan that meets the formula 8 from each of the first target flight model scheduling plans; if there is only one flight model scheduling plan in the search result, then the flight model scheduling plan is the optimal flight model scheduling plan, and the solution process ends; if there are multiple flight model scheduling plans in the search result, then the found flight model scheduling plan is recorded as the second target flight model scheduling plan, and the third step is executed;

[0052] Wherein, the formula 8 indicates that: the deviation of the number of connecting aircraft types is minimized, that is, the sum of the number of aircraft increased or decreased by all aircraft types at all airports relative to the initial state is minimized;

[0053] The third step is to find a flight model scheduling plan that meets the formula nine from each of the third target flight model scheduling plans. The search result is the optimal flight model scheduling plan. If there are multiple search results, one of them can be selected as the optimal flight model scheduling plan.

[0054] Wherein, the formula nine indicates: maximum profit, that is, the sum of the aircraft model-flight profit coefficient multiplied by the aircraft model-flight decision variable is maximized.

[0055] As a preferred embodiment of the present invention: in the step S2, the predicted number of passengers for any target flight f is predicted using the passenger seat prediction method of step S2-1;

[0056] Step S2-1, the method for predicting the number of passengers seated comprises:

[0057] Step S2-1-1: Training steps of the trained attendance rate linear regression model:

[0058] Step S2-1-1A, establish the attendance rate linear regression model:

[0059] y load =b load +b T_load x T_load ;

[0060] In the formula, y loadrepresents the predicted occupancy rate of the flight on the departure day, x T_load represents the scheduled occupancy rate of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, T≥1, and the scheduled occupancy rate refers to the ratio of the number of scheduled passengers on the flight to the number of seats on the aircraft; b load and b T_load Represent the intercept and coefficient to be regressed respectively;

[0061] Step S2-1-1B: Filter the historical flight data within 60 days before the scheduled flight date according to the following four screening conditions in turn, and use the historical flight data with attendance rate filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the attendance rate linear regression model: load and coefficient b T_load Performing a first-order linear fitting to obtain the trained attendance rate linear regression model;

[0062] The historical flight attendance data includes: the scheduled attendance of the historical flight on its scheduled date as the scheduled attendance x T_load , the actual load factor of the historical flight on its departure day is taken as the predicted load factor y load ;

[0063] Furthermore, when the trained load factor linear regression model is used to predict the predicted number of passengers for the target flight f, the scheduled load factor of the target flight f on the flight schedule date is substituted into the scheduled load factor x T_load , the calculated predicted attendance rate y load The product of the number of seats on the corresponding aircraft is the predicted number of passengers on the target flight f;

[0064] The sample quantity requirement is: the number of historical flight occupancy rate data filtered out by the screening conditions is above the preset minimum training sample quantity threshold; the problem addressed by the sample quantity requirement is: due to the large difference in the number of samples of different flights, different sample screening methods are used for the prediction of a flight number-date-segment, that is, the following four screening conditions with decreasing screening detail; more detailed sample screening means that the model reflects the characteristics of the prediction object to a higher degree, but it is also easy for the model to have poor robustness due to too few samples. The problem can be solved by using the sample quantity requirement for judgment.

[0065] The four screening conditions are:

[0066] First screening condition: the historical flights have the same flight number and the same flight segment, and the deviation between the actual load factor and the scheduled load factor of the historical flights is within plus or minus 10%;

[0067] The second screening condition: the flight segments of the historical flights are the same, and the reserved seat occupancy levels of the historical flights are the same as those of the target flight f. The reserved seat occupancy levels include five levels with reserved seat occupancy rates of 0-20%, 21-40%, 41-60%, 61-80%, and 81-100% respectively;

[0068] The third screening condition: the departure airports of the historical flights are the same, and the reserved seat occupancy levels of the historical flights are the same as those of the target flight f;

[0069] The fourth screening condition: the reserved seat occupancy levels of the historical flights are the same as those of the target flight f;

[0070] Step S2-1-2, the training steps of the trained seat number DOW linear regression model:

[0071] Step S2-1-2A, establish the seat number DOW linear regression model:

[0072]

[0073] In the formula, ynum represents the predicted seat number of the flight on the departure date, x T _ num represents the reserved seat number of the flight on the reservation date. The reservation date is the Tth day before the departure date, where T≥1; represents the actual seat number of the flight with the same flight number as the flight corresponding to the predicted seat number ynum and taking off h days before the departure date of the corresponding flight; b num represents the intercept to be regressed, b T _ num and both represent the coefficients to be regressed. And if h<T, then is directly taken as 0, that is, the corresponding is removed from the seat number DOW linear regression model;

[0074] Step S2-1-2B, from the historical flight data within 60 days before the flight schedule date, screen according to the four screening conditions in turn, and use the seat number historical flight data screened by the screening condition that first meets the sample quantity requirement as the training data to perform first-order linear fitting on the intercept b num of the seat number DOW linear regression model, the coefficient b T _ num and the coefficient to obtain the trained seat number DOW linear regression model;

[0075] The historical flight data of the number of passengers includes: the number of passengers scheduled on the historical flight on its scheduled date as the number of passengers scheduled x T_num The actual occupancy rate of the historical flight on its departure day is taken as the predicted number of passengers y num The actual number of passengers on the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight is taken as the actual number of passengers on the flight.

[0076] Furthermore, when the trained passenger number DOW linear regression model is used to predict the predicted passenger number of the target flight f, the scheduled passenger number of the target flight f on the flight schedule date is substituted into the scheduled passenger number x T_num The actual number of passengers on the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute the actual number of passengers The calculated predicted number of occupants y num That is, the predicted number of passengers for the target flight f;

[0077] Step S2-1-3: Establish the mean model of the number of people seated:

[0078]

[0079] In the formula, y num2 It indicates the predicted number of passengers on the flight under test on its departure date. It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual number of passengers on flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0080] Step S2-1-4: Establish the DOW mean model of the number of passengers:

[0081]

[0082] In the formula, y num3 represents the predicted number of passengers on the flight under test on its departure date, F i represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. For example, if the tested flight takes off on Tuesday, when i=7 and i=14, the corresponding F i That is, the flights that took off on Tuesday one week and two weeks before the flight being tested. Denotes the flight set F i The number of flights in, f′ represents the flight set F i Flights in represents the actual number of passengers on flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0083] Step S2-1-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained passenger number XGBOOST model, the trained passenger number XGBOOST model takes the scheduled passenger number of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted passenger number of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f;

[0084] Step S2-1-6, using historical flight data, calculate the root mean square error (RMSE) of the five passenger seat prediction models, namely, the trained passenger seat linear regression model, the trained passenger seat DOW linear regression model, the passenger seat mean model, the passenger seat DOW mean model, and the trained passenger seat XGBOOST model, and record the model with the smallest root mean square error (RMSE), i.e., the smallest prediction error, as the optimal passenger seat prediction model;

[0085] The calculation formula of the root mean square error RMSE is as follows:

[0086]

[0087] - represents the error of model m for flight f segment od

[0088] y f,od,i - represents the actual value of flight flight segment od the i days before

[0089] - represents the predicted value of flight f segment od j days in advance i days before the flight, where j=T.

[0090] Step S2-1-7: Use the optimal passenger seat prediction model to predict the predicted passenger seat number of the target flight f.

[0091] Furthermore, in step S2-1-7, the predicted number of passengers may be adjusted according to the aircraft type of the target flight f in the following manner to make the prediction result more accurate:

[0092] For direct flights, the predicted value of a flight-segment combination on a certain aircraft type is the minimum value of the flight-segment predicted value and the number of seats on the aircraft type. For example, if the predicted number of passengers on a flight is 200, and the number of seats on the predicted aircraft type is only 180, the predicted number of passengers will be adjusted to 180.

[0093] The calculation method for connecting flights is:

[0094]

[0095]

[0096]

[0097] - indicates the predicted value of the long segment od of flight f

[0098] - represents the adjusted forecast value of the long leg od of flight f

[0099] - indicates the predicted value of the short segment of flight f from the starting point o to the transfer point m

[0100] - represents the adjusted forecast value of the short segment of flight f from the starting point o to the transfer point m

[0101] - indicates the predicted value of the short leg of flight f from transfer point m to the destination d

[0102] - represents the adjusted forecast value of the short leg of flight f from transfer point m to the destination d

[0103] S a - Indicates the number of seats on aircraft type a

[0104] For example: a flight departs from Airport A and arrives at Airport C, with a stopover at Airport B. The number of seats for a certain aircraft model is 180. The predicted number of passengers for the AC segment is 100, the predicted number of passengers for the AB segment is 80, and the predicted number of passengers for the BC segment is 100. The total predicted number of passengers on the BC segment is 200, which exceeds the number of seats of the aircraft model by 180. According to the passenger composition ratio on the BC segment, the number of passengers on the BC segment is adjusted to 180*(100 / (100+100))=90, the number of passengers on the AC segment is adjusted to 180*(100 / (100+100))=90, and the number of passengers on the AB segment remains unchanged at 80.

[0105] Preferably: in the step S2, the revenue prediction method of step S2-2 is used to predict the predicted revenue of any target flight f;

[0106] Step S2-2, the income forecasting method comprises:

[0107] Step S2-2-1: Training steps of the trained income linear regression model:

[0108] Step S2-2-1A: Establish an income linear regression model:

[0109] y revenue =b revenue +b T_revenue x T_revenue ;

[0110] In the formula, y revenue represents the predicted revenue of the flight on the departure day, x T_revenue represents the scheduled revenue of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, and T≥1; b revenue and b T_revenue Represent the intercept and coefficient to be regressed respectively;

[0111] Step S2-2-1B: Filter the historical flight data within 60 days before the flight schedule date according to the following four screening conditions in turn, and use the income historical flight data filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the income linear regression model. revenue and coefficient b T_revenue Performing a first-order linear fitting to obtain the trained income linear regression model;

[0112] The historical flight revenue data includes: the scheduled revenue of the historical flight on its scheduled date as the scheduled revenue x T_load , the actual income of the historical flight on its departure date is taken as the predicted income y load ;

[0113] Furthermore, when the trained revenue linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue x T_revenue , the calculated predicted income y revenue That is, the predicted revenue of the target flight f;

[0114] The requirements for the number of samples are as follows: the number of historical flight data on revenue screened by the screening conditions is above the preset minimum training sample quantity threshold; the problem addressed by this requirement for the number of samples is that due to significant differences in the number of samples for different flights, for the prediction of a flight number - date - flight segment, different sample screening methods are used, namely the following four screening conditions with gradually decreasing screening granularity; more meticulous sample screening means a higher degree of reflecting the characteristics of the prediction object by the model, but it is also prone to poor model robustness due to过少 samples. This problem can be solved by using the requirement for the number of samples for judgment.

[0115] The four screening conditions are as follows in sequence:

[0116] The first screening condition: the flight numbers of the historical flights are the same, the flight segments are the same, and furthermore, the deviation between the actual occupancy rate and the scheduled occupancy rate of the historical flights is within plus or minus 10%;

[0117] The second screening condition: the flight segments of the historical flights are the same, the scheduled occupancy classes of the historical flights are the same as those of the target flight f, and the scheduled occupancy classes include five levels with scheduled occupancy rates of 0 - 20%, 21 - 40%, 41 - 60%, 61 - 80%, and 81 - 100% respectively;

[0118] The third screening condition: the departure airports of the historical flights are the same, and the scheduled occupancy classes of the historical flights are the same as those of the target flight f;

[0119] The fourth screening condition: the scheduled occupancy classes of the historical flights are the same as those of the target flight f;

[0120] Step S2 - 2 - 2, the training steps of the trained revenue DOW linear regression model:

[0121] Step S2 - 2 - 2A, establishing the revenue DOW linear regression model:

[0122]

[0123] In the formula, y revenue2 represents the predicted revenue of the flight on the departure date, x T_revenue2 represents the scheduled revenue of the flight on the scheduled date, and the scheduled date is the T - th day before the departure date, where T ≥ 1; represents the actual revenue of the flight with the same flight number as the flight corresponding to the predicted revenue y revenue2 and taking off h days before the departure date of the corresponding flight; b revenue2 represents the intercept to be regressed, b T_revenue2 and both represent the coefficients to be regressed, and if h < T, then is directly taken as 0, that is, the corresponding Removed from the income DOW linear regression model;

[0124] Step S2-2-2B: Filter the historical flight data within 60 days before the flight schedule date according to the four screening conditions in turn, and use the income historical flight data filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the income DOW linear regression model. revenue2 , coefficient b T_revenue2 and coefficient Performing a first-order linear fitting to obtain the trained income DOW linear regression model;

[0125] The historical flight revenue data includes: the scheduled revenue of the historical flight on its scheduled date as the scheduled revenue x T_revenue2 , the actual load factor of the historical flight on its departure date is used as the predicted revenue y revenue2 The actual revenue of the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight is taken as the actual revenue

[0126] Furthermore, when the trained revenue DOW linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue x T_revenue2 The actual revenue of the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute the actual revenue The calculated predicted income y revenue2 That is, the predicted revenue of the target flight f;

[0127] Step S2-2-3: Establish the mean income model:

[0128]

[0129] In the formula, y revenue3 represents the predicted revenue of the tested flight on its departure day, It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0130] Step S2-2-4: Establish the income DOW mean model:

[0131]

[0132] In the formula, y revenue4 represents the predicted revenue of the tested flight on its departure day, F i represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. For example, if the tested flight takes off on Tuesday, when i=7 and i=14, the corresponding F i That is, the flights that took off on Tuesday one week and two weeks before the flight being tested. Denotes the flight set F i The number of flights in, f′ represents the flight set F i Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0133] Step S2-2-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained income XGBOOST model, the trained income XGBOOST model takes the scheduled income of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted income of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f;

[0134] Step S2-2-6, using historical flight data, calculate the root mean square error (RMSE) of the five revenue prediction models, namely, the trained revenue linear regression model, the trained revenue DOW linear regression model, the revenue mean model, the revenue DOW mean model, and the trained revenue XGBOOST model, and record the model with the smallest root mean square error (RMSE), that is, the smallest prediction error, as the optimal revenue prediction model;

[0135] Step S2-2-7, using the optimal revenue prediction model to predict the predicted revenue of the target flight f;

[0136] Step S2-2-8, using the historical flight data, calculate the historical average ticket price of flights with the same flight number as the target flight f, and divide the calculation result of step S2-2-7 by the predicted number of passengers of the target flight f to obtain the predicted ticket price; if the deviation between the predicted ticket price and the historical average ticket price is within the preset revenue deviation threshold, then the calculation result of step S2-2-7 is directly used as the predicted revenue of the target flight f; otherwise, the product of the predicted number of passengers of the target flight f and the historical average ticket price is used as the predicted revenue of the target flight f.

[0137] As a preferred embodiment of the present invention: in the step S2, the cost prediction method of step S2-3 is used to predict the predicted cost of any target flight f when the aircraft model is aircraft type k;

[0138] Step S2-3, the cost prediction method comprises:

[0139] Step S2-3-1, computer-based mean cost prediction result y cost1 and the model mean volatility A1;

[0140]

[0141]

[0142] In the formula, F 1,30 represents the set of flights that have the same route and are operated by the same aircraft type as the target flight f within 30 days before the departure of the target flight f, |F 1,30 | represents the flight set F 1,30 The number of flights in, f′ represents the flight set F 1,30 Flights in represents the actual cost of flight f′;

[0143] Furthermore, it is determined whether the average volatility A1 of the model is approximately the preset volatility threshold; if the determination result is no, the average cost prediction result y of the model is used. cost1 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-2;

[0144] Step S2-3-2: Calculate the seat grade mean cost prediction result y cost2 and seat-level mean volatility A2;

[0145]

[0146]

[0147] In the formula, It indicates the set of flights with the same route and the same seat quantity level as the target flight f within 30 days before the departure of the target flight f, and the difference between any two adjacent seat quantity levels is 10 seats, for example: the first-level seat quantity level is 100 seats, and the second-level seat quantity level is 110 seats; Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual cost of flight f′;

[0148] Furthermore, it is determined whether the average volatility A2 of the seat grade is approximately equal to a preset volatility threshold; if the determination result is no, the seat grade average cost prediction result y is used. cost2 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-3;

[0149] Step S2-3-3, using the trained cost linear regression model to calculate the predicted cost of the target flight f when the aircraft model is aircraft type k, including: Step S2-3-3A, establishing a cost linear regression model:

[0150] y cost =b cost +b distance x distance ;

[0151] In the formula, y cost represents the predicted flight cost of the flight, x distance Indicates the flight distance of the flight; b cost and b distance Represent the intercept and coefficient to be regressed respectively;

[0152] Step S2-3-3B: Filter the historical flight data within 60 days before the flight plan date according to the following three screening conditions in turn, and use the cost historical flight data selected by the first screening condition that meets the cost prediction sample quantity requirement as training data, and calculate the intercept b of the cost linear regression model. cost and coefficient b distance Performing a first-order linear fitting to obtain the trained cost linear regression model;

[0153] The cost historical flight data includes: the flight distance of the historical flight as the flight distance x distance , the actual flight cost of the historical flight is used as the predicted flight cost y cost ;

[0154] The requirement for the number of cost prediction samples is: the number of cost historical flight data selected by the screening condition is above the preset minimum training sample number threshold;

[0155] The three screening conditions are:

[0156] Screening condition 1: The aircraft model of the historical flight is the same as that of the target flight f, and the stopover classification of the historical flight is the same as that of the target flight f, and the international flight classification of the historical flight is the same as that of the target flight f;

[0157] Screening condition 2: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f, and the historical flight has the same international flight classification as the target flight f;

[0158] Screening condition three: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f;

[0159] Step S2-3-3C: Substitute the flight distance of the target flight f into the flight distance x of the trained cost linear regression model. distance , the calculated predicted flight cost y cost That is, it is the predicted cost when the target flight f is operated by an aircraft of model k.

[0160] Compared with the prior art, the present invention has the following beneficial effects:

[0161] First, the present invention can automatically generate an optimal flight model scheduling plan that meets predetermined constraints and objective functions through steps S1 to S4, so as to schedule the corresponding model k of the flyable aircraft for the target flight f in the target flight set F to fly according to the plan. It has the advantages of high efficiency and good effect, and can adapt to the current increasingly large-scale route network.

[0162] Second, the present invention forms constraints through Formula 1 to Formula 6, and forms an objective function through Formula 7 to Formula 9 to establish an integer programming model, and obtains the optimal flight model scheduling plan through the process of the first step to the third step, which is adapted to the actual status of the current development of the domestic aviation industry, and can further improve the effect of flight model scheduling to maximize the capacity demand of the target flight set F, and reduce the resource consumption caused by the flight model scheduling for the connecting model scheduling, thereby maximizing the flight revenue.

[0163] Third, the present invention adopts the passenger seat prediction method described in step S2-1 to predict the predicted passenger seat number of target flight f. By providing an innovative training data screening method for each model, the prediction accuracy of the predicted passenger seat number can be improved, so that the target flight f in the target flight set F can be better scheduled according to the optimal flight model scheduling plan, and it is more in line with the actual situation of the target date of the model adjustment.

[0164] Fourth, the present invention adopts the revenue prediction method described in step S2-2 to predict the predicted revenue of the target flight f. By providing an innovative training data screening method for each model, the prediction accuracy of the predicted revenue can be improved, so that the target flight f in the target flight set F can be better scheduled according to the optimal flight model scheduling plan, and it is more in line with the actual situation of the target date of the model adjustment.

[0165] Fifth, the present invention adopts the cost prediction method described in step S2-3 to predict the predicted cost of the target flight f, which can improve the prediction accuracy of the predicted cost, so that the target flight f in the target flight set F can be better scheduled according to the optimal flight model scheduling plan, and it is more in line with the actual situation of the target date of the model adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0166] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0167] Figure 1 The flowchart is a flowchart of the flight model adjustment optimization method of the present invention. DETAILED DESCRIPTION

[0168] The present invention is described in detail below in conjunction with the embodiments and the accompanying drawings to help those skilled in the art better understand the inventive concept of the present invention. However, the protection scope of the claims of the present invention is not limited to the following embodiments. For those skilled in the art, all other embodiments obtained without creative work without departing from the inventive concept of the present invention belong to the protection scope of the present invention.

[0169] Embodiment 1

[0170] like Figure 1 As shown, the present invention discloses a flight model adjustment optimization method, comprising:

[0171] Step S1, record the Tth day before the end time of the model adjustment target date as the flight plan date, T≥1, and the duration of the model adjustment target date is less than the value of T; on the flight plan date, read the following data of the target airline from the database of the target airline through the corresponding data interface: a target flight set F whose departure time is on the model adjustment target date, a model set K of all operable aircraft that can fly on the model adjustment target date, and historical flight data related to the target flight set F; wherein the model adjustment target date can be one day or multiple days.

[0172] Step S2, based on the historical flight data, predicting: the predicted number of passengers and the predicted revenue of each target flight f in the target flight set F, and the predicted cost of each target flight f when operated by an aircraft of model k, k∈K;

[0173] And, based on the prediction results, the predicted revenue of each target flight f when using an available aircraft of model k is calculated. That is, the difference between the predicted revenue and the predicted cost;

[0174] Step S3: Establish an integer programming model and use the target flight set F, aircraft model set K and predicted revenue The solution is performed to obtain an optimal flight model scheduling plan that satisfies the constraints and objective function of the integer programming model; wherein, the solution process of the optimal flight model scheduling plan can transmit data to a commercial solver through a network for solution, and then transmit the solution result back through the network; the data can be written in a standard format readable by commercial solvers such as .mps, .lp, etc., or the integer programming model can be established using the interface provided by the solver.

[0175] Step S4: on the target date for adjusting the aircraft type, according to the optimal flight aircraft type scheduling plan, dispatching an available aircraft of the corresponding aircraft type k for the target flight f in the target flight set F to fly.

[0176] The above is a basic implementation of the first embodiment of the present invention, and further optimization, improvement and limitation can be made on the basis of the basic implementation:

[0177] Preferably: in step S3, the constraint conditions include:

[0178]

[0179]

[0180] and

[0181]

[0182]

[0183]

[0184] The objective function includes:

[0185]

[0186]

[0187]

[0188] Where, F represents the target flight set whose departure time of the target airline is on the target date of the aircraft model adjustment, f represents the target flight in the target flight set F, and the symbol It means any;

[0189] K represents the set of all aircraft types that the target airline can operate on the target aircraft type adjustment date, and k represents the aircraft type in the aircraft type set K;

[0190] A represents the set of take-off and landing airports of all the target flights f, a represents an element in the take-off and landing airport set A, that is, the take-off and landing airport, and the take-off and landing airport a is a take-off airport or a landing airport;

[0191] N k represents the take-off and landing time-space node set of all aircraft of model k in the model set K, and n represents the take-off and landing time-space node set N k The elements in , namely, the take-off and landing space-time nodes, the take-off and landing space-time node n is a take-off space-time node including the take-off time and the take-off airport or a landing space-time node including the landing time and the landing airport; The take-off time and space node of the first flight of the operable aircraft of model k on the target date of adjustment of the model; The take-off and landing time-space node set representing all the aircraft of model k in the aircraft set K that can fly at 3 a.m. on the target date for adjusting the aircraft model;

[0192] O(k, n) represents the set of flights that take off from the take-off and landing space-time node n using an aircraft of model k, and the take-off and landing space-time node n is the take-off space-time node; I(k, n) represents the set of flights that land at the take-off and landing space-time node n using an aircraft of model k, and the take-off and landing space-time node n is the landing space-time node;

[0193] F c represents the set of cancelable target flights f in the target flight set F; M represents the set F c The set of flights with the same flight number in the set M, m represents a subset of the set M, and the departure times of the target flights f in the subset m have the same number of days of the week, for example, the departure times are all Mondays or all Fridays;

[0194] represents the set of target flights f that are flying at the take-off and landing time and space node n in the target flight set F;

[0195] represents the aircraft model flight decision variable, where When the value is 1, it means that the target flight f is operated by an aircraft of model k. When the value is 0, it means that the aircraft of model k is not used to fly the target flight f; all the model flight decision variables of the target flight f A value combination method of forms a flight model scheduling plan for the target flight set F;

[0196] Indicates the number of aircraft of model k that can fly before the take-off and landing time and space node n; Indicates the number of aircraft of model k that can fly after the take-off and landing time and space node n;

[0197] Relative to the initial quantity The absolute value of the number of new aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented; Relative to the initial quantity The absolute value of the reduction in the number of aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented;

[0198] The initial quantity The number of aircraft of type k that can fly at the take-off and landing airport a before executing the corresponding flight type scheduling plan, that is, the initial time of the target date of the type adjustment;

[0199] N represents the preset upper limit of the number of model adjustments;

[0200] N k Indicates the number of aircraft of model k in model set K that can fly, which is set according to the actual situation of model set K;

[0201] N m represents the minimum number of flights m in the preset set M;

[0202] represents the predicted revenue of the target flight f when it is operated by an aircraft of model k;

[0203] k f Indicates the aircraft type used by the target flight f in the original plan before the corresponding flight aircraft type scheduling plan is executed;

[0204] The process of solving the optimal flight model scheduling plan is:

[0205] The first step is to exhaustively enumerate all the aircraft model and flight decision variables of the target flight f under the premise of satisfying the constraints described in formulas 1 to 6. , so as to find a value combination that meets the formula 7; if there is only one value combination in the search result, then the value combination is the optimal flight model scheduling plan, and the solution process ends; if there are multiple value combinations in the search result, the found value combination is recorded as the first target flight model scheduling plan, and the second step is executed;

[0206] in,

[0207] Formula 1 indicates: flow balance, that is, for any aircraft type and any take-off and landing time and space node n, the number of aircraft before the take-off and landing time and space node n should be equal to the number of aircraft after the take-off and landing time and space node n;

[0208] Formula 2 means: Adjust the number of aircraft types, that is, the flight decision variable for the aircraft type that is different from the original aircraft type of the flight The sum should be less than the preset adjustment model number limit;

[0209] Formula 3 represents: flight coverage constraint, that is, for any non-cancellable flight, all the flight decision variables related to the flight The sum should be equal to 1;

[0210] Formula 4 indicates: the aircraft model quantity limit, that is, for any checkpoint at 3 a.m. and any aircraft model, the sum of the number of aircraft on the ground plus the sum of aircraft in flight should be less than or equal to the number of aircraft of that model;

[0211] Formula 5 means: cancel the flight constraint, that is, for the flight set m and the non-virtual aircraft model flight decision variable The sum should be greater than the preset number;

[0212] Formula 6 indicates: the aircraft connection quantity constraint, that is, for each aircraft type and each airport (the number of aircraft + the number of aircraft reduced - the number of aircraft increased) should be equal to the input initial number of aircraft;

[0213] Formula 7 indicates that: the number of flights covered is the largest, which means that the decision variables of non-virtual aircraft models and flights are Maximize the sum.

[0214] Step 2: Find a flight model scheduling plan that meets the formula 8 from each of the first target flight model scheduling plans; if there is only one flight model scheduling plan in the search result, then the flight model scheduling plan is the optimal flight model scheduling plan, and the solution process ends; if there are multiple flight model scheduling plans in the search result, then the found flight model scheduling plan is recorded as the second target flight model scheduling plan, and the third step is executed;

[0215] Wherein, the formula 8 indicates that: the deviation of the number of connecting aircraft types is minimized, that is, the sum of the number of aircraft increased or decreased by all aircraft types at all airports relative to the initial state is minimized;

[0216] The third step is to find a flight model scheduling plan that meets the formula nine from each of the third target flight model scheduling plans. The search result is the optimal flight model scheduling plan. If there are multiple search results, one of them can be selected as the optimal flight model scheduling plan.

[0217] Wherein, the formula nine indicates: maximum profit, that is, the sum of the aircraft model-flight profit coefficient multiplied by the aircraft model-flight decision variable is maximized.

[0218] Embodiment 2

[0219] On the basis of the above-mentioned embodiment 1, this embodiment 2 also adopts the following preferred implementation manner:

[0220] In the step S2, the predicted number of passengers for any target flight f is predicted using the passenger seat prediction method of step S2-1;

[0221] Step S2-1, the method for predicting the number of passengers seated comprises:

[0222] Step S2-1-1: Training steps of the trained attendance rate linear regression model:

[0223] Step S2-1-1A, establish the attendance rate linear regression model:

[0224] y load =b load +b T_load x T_load ;

[0225] In the formula, y load represents the predicted occupancy rate of the flight on the departure day, x T_load represents the scheduled occupancy rate of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, T≥1, and the scheduled occupancy rate refers to the ratio of the number of scheduled passengers on the flight to the number of seats on the aircraft; b load and b T_load Represent the intercept and coefficient to be regressed respectively;

[0226] Step S2-1-1B: Filter the historical flight data within 60 days before the scheduled flight date according to the following four screening conditions in turn, and use the historical flight data with attendance rate filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the attendance rate linear regression model: load and coefficient b T_loadPerforming a first-order linear fitting to obtain the trained attendance rate linear regression model;

[0227] The historical flight attendance data includes: the scheduled attendance of the historical flight on its scheduled date as the scheduled attendance x T_load , the actual load factor of the historical flight on its departure day is taken as the predicted load factor y load ;

[0228] Furthermore, when the trained load factor linear regression model is used to predict the predicted number of passengers for the target flight f, the scheduled load factor of the target flight f on the flight schedule date is substituted into the scheduled load factor x T_load , the calculated predicted attendance rate y load The product of the number of seats on the corresponding aircraft is the predicted number of passengers on the target flight f;

[0229] The sample quantity requirement is: the number of historical flight occupancy rate data filtered out by the screening conditions is above the preset minimum training sample quantity threshold; the problem addressed by the sample quantity requirement is: due to the large difference in the number of samples of different flights, different sample screening methods are used for the prediction of a flight number-date-segment, that is, the following four screening conditions with decreasing screening detail; more detailed sample screening means that the model reflects the characteristics of the prediction object to a higher degree, but it is also easy for the model to have poor robustness due to too few samples. The problem can be solved by using the sample quantity requirement for judgment.

[0230] The four screening conditions are:

[0231] First screening condition: the historical flights have the same flight number and the same flight segment, and the deviation between the actual load factor and the scheduled load factor of the historical flights is within plus or minus 10%;

[0232] The second screening condition: the flight segments of the historical flights are the same, the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f, and the scheduled seating level includes five levels of scheduled seating rates of 0-20%, 21-40%, 41-60%, 61-80%, and 81-100% respectively;

[0233] The third screening condition: the departure airport of the historical flights is the same, and the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f;

[0234] Fourth screening condition: the scheduled seating level of the historical flight is the same as the scheduled seating level of the target flight f;

[0235] Step S2-1-2: Training steps of the trained seat occupancy DOW linear regression model:

[0236] Step S2-1-2A: Establish a linear regression model for the number of passengers on the upper deck DOW:

[0237]

[0238] In the formula, y num represents the predicted number of passengers on the flight on the departure date, and x T_num represents the reserved number of passengers on the flight on the reservation date. The reservation date is the Tth day before the departure date, where T ≥ 1; represents the actual number of passengers on the flight with the same flight number as the flight corresponding to the predicted number of passengers y num and taking off h days before the departure date of the corresponding flight; b num represents the intercept to be regressed, and b T_num and both represent the coefficients to be regressed. And if h < T, then is directly taken as 0, that is, the corresponding is removed from the linear regression model of the number of passengers on the upper deck DOW;

[0239] Step S2-1-2B: From the historical flight data within 60 days before the flight schedule date, filter according to the four filtering conditions in sequence, and use the historical flight data of the number of passengers filtered by the filtering condition that first meets the sample quantity requirement as the training data, and perform first-order linear fitting on the intercept b num , coefficient b T_num and coefficient of the linear regression model of the number of passengers on the upper deck DOW to obtain the trained linear regression model of the number of passengers on the upper deck DOW;

[0240] Among them, the historical flight data of the number of passengers includes: the reserved number of passengers on the historical flight on its reservation date as the reserved number of passengers x T_num , the actual occupancy rate of the historical flight on its departure date as the predicted number of passengers y num , and the actual number of passengers on the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight as the actual number of passengers

[0241] Moreover, when predicting the predicted number of passengers of the target flight f using the trained linear regression model of the number of passengers on the upper deck DOW, the reserved number of passengers of the target flight f on the flight schedule date is used to substitute into the reserved number of passengers x T_num , and the actual number of passengers on the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute into the actual number of passengers The predicted number of passengers y calculated numThat is, the predicted number of passengers for the target flight f;

[0242] Step S2-1-3: Establish the mean model of the number of people seated:

[0243]

[0244] In the formula, y num2 It indicates the predicted number of passengers on the flight under test on its departure date. It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual number of passengers on flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0245] Step S2-1-4: Establish the DOW mean model of the number of passengers:

[0246]

[0247] In the formula, y num3 represents the predicted number of passengers on the flight under test on its departure date, F i represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. For example, if the tested flight takes off on Tuesday, when i=7 and i=14, the corresponding F i That is, the flights that took off on Tuesday one week and two weeks before the flight being tested. Denotes the flight set F i The number of flights in, f′ represents the flight set F i Flights in represents the actual number of passengers on flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0248] Step S2-1-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained passenger number XGBOOST model, the trained passenger number XGBOOST model takes the scheduled passenger number of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted passenger number of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f;

[0249] Step S2-1-6, using historical flight data, calculate the root mean square error (RMSE) of the five passenger seat prediction models, namely, the trained passenger seat linear regression model, the trained passenger seat DOW linear regression model, the passenger seat mean model, the passenger seat DOW mean model, and the trained passenger seat XGBOOST model, and record the model with the smallest root mean square error (RMSE), i.e., the smallest prediction error, as the optimal passenger seat prediction model;

[0250] The calculation formula of the root mean square error RMSE is as follows:

[0251]

[0252] - represents the error of model m for flight f segment od

[0253] y f,od,i - represents the actual value of flight flight segment od the i days before

[0254] - represents the predicted value of flight f segment od j days in advance i days before the flight, where j=T.

[0255] Step S2-1-7: Use the optimal passenger seat prediction model to predict the predicted passenger seat number of the target flight f.

[0256] Furthermore, in step S2-1-7, the predicted number of passengers may be adjusted according to the aircraft type of the target flight f in the following manner to make the prediction result more accurate:

[0257] For direct flights, the predicted value of a flight-segment combination on a certain aircraft type is the minimum value of the flight-segment predicted value and the number of seats on the aircraft type. For example, if the predicted number of passengers on a flight is 200, and the number of seats on the predicted aircraft type is only 180, the predicted number of passengers will be adjusted to 180.

[0258] The calculation method for connecting flights is:

[0259]

[0260]

[0261]

[0262] - indicates the predicted value of the long segment od of flight f

[0263] - represents the adjusted forecast value of the long leg od of flight f

[0264] - indicates the predicted value of the short segment of flight f from the starting point o to the transfer point m

[0265] - represents the adjusted forecast value of the short segment of flight f from the starting point o to the transfer point m

[0266] - indicates the predicted value of the short leg of flight f from transfer point m to the destination d

[0267] - represents the adjusted forecast value of the short leg of flight f from transfer point m to the destination d

[0268] S a - Indicates the number of seats on aircraft type a

[0269] For example: a flight departs from Airport A and arrives at Airport C, with a stopover at Airport B. The number of seats for a certain aircraft model is 180. The predicted number of passengers for the AC segment is 100, the predicted number of passengers for the AB segment is 80, and the predicted number of passengers for the BC segment is 100. The total predicted number of passengers on the BC segment is 200, which exceeds the number of seats of the aircraft model by 180. According to the passenger composition ratio on the BC segment, the number of passengers on the BC segment is adjusted to 180*(100 / (100+100))=90, the number of passengers on the AC segment is adjusted to 180*(100 / (100+100))=90, and the number of passengers on the AB segment remains unchanged at 80.

[0270] The above is a basic implementation of the second embodiment, and further optimization, improvement and limitation can be made on the basis of the basic implementation:

[0271] Preferably: in the step S2, the revenue prediction method of step S2-2 is used to predict the predicted revenue of any target flight f;

[0272] Step S2-2, the income forecasting method comprises:

[0273] Step S2-2-1: Training steps of the trained income linear regression model:

[0274] Step S2-2-1A: Establish an income linear regression model:

[0275] y revenue =b revenue +b T_revenue x T_revenue ;

[0276] In the formula, y revenue represents the predicted revenue of the flight on the departure day, x T_revenue represents the scheduled revenue of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, and T≥1; b revenueand b T_revenue Represent the intercept and coefficient to be regressed respectively;

[0277] Step S2-2-1B: Filter the historical flight data within 60 days before the flight schedule date according to the following four screening conditions in turn, and use the income historical flight data filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the income linear regression model. revenue and coefficient b T_revenue Performing a first-order linear fitting to obtain the trained income linear regression model;

[0278] The historical flight revenue data includes: the scheduled revenue of the historical flight on its scheduled date as the scheduled revenue x T_load , the actual income of the historical flight on its departure date is taken as the predicted income y load ;

[0279] Furthermore, when the trained revenue linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue x T_revenue , the calculated predicted income y revenue That is, the predicted revenue of the target flight f;

[0280] The sample quantity requirement is: the number of revenue historical flight data screened out by the screening conditions is above the preset minimum training sample quantity threshold; the problem addressed by the sample quantity requirement is: since the number of samples of different flights varies greatly, different sample screening methods are used for the prediction of a flight number-date-segment, that is, the following four screening conditions with decreasing screening levels; more detailed sample screening means that the model reflects the characteristics of the prediction object to a higher degree, but it is also easy for the model to have poor robustness due to too few samples. This problem can be solved by using the sample quantity requirement for judgment.

[0281] The four screening conditions are:

[0282] First screening condition: the historical flights have the same flight number and the same flight segment, and the deviation between the actual load factor and the scheduled load factor of the historical flights is within plus or minus 10%;

[0283] The second screening condition: the flight segments of the historical flights are the same, the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f, and the scheduled seating level includes five levels of scheduled seating rates of 0-20%, 21-40%, 41-60%, 61-80%, and 81-100% respectively;

[0284] The third screening condition: The departure airport of the historical flight is the same, and the reserved seat class of the historical flight is the same as that of the target flight f;

[0285] The fourth screening condition: The reserved seat class of the historical flight is the same as that of the target flight f;

[0286] Step S2-2-2, the training step of the trained revenue DOW linear regression model:

[0287] Step S2-2-2A, establish the revenue DOW linear regression model:

[0288]

[0289] In the formula, y revenue2 represents the predicted revenue of the flight on the departure date, x T_revenue2 represents the reserved revenue of the flight on the reservation date, and the Tth day before the departure date is the reservation date, T≥1; represents the actual revenue of the flight with the same flight number as the flight corresponding to the predicted revenue y revenue2 and taking off h days before the departure date of the corresponding flight; b revenue2 represents the intercept to be regressed, b T_revenue2 and both represent the coefficients to be regressed, and if h<T, then is directly taken as 0, that is, the corresponding is removed from the revenue DOW linear regression model;

[0290] Step S2-2-2B, from the historical flight data within 60 days before the flight schedule date, screen in turn according to the four screening conditions, and use the revenue historical flight data screened by the screening condition that first meets the sample quantity requirement as the training data, and perform first-order linear fitting on the intercept b revenue2 , coefficient b T_revenue2 and coefficient to obtain the trained revenue DOW linear regression model;

[0291] Among them, the revenue historical flight data includes: the reserved revenue of the historical flight on its reservation date as the reserved revenue x T_revenue2 , the actual occupancy rate of the historical flight on its departure date as the predicted revenue y revenue2 , and the actual revenue of the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight as the actual revenue

[0292] Furthermore, when the trained revenue DOW linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue x T_revenue2 The actual revenue of the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute the actual revenue The calculated predicted income y revenue2 That is, the predicted revenue of the target flight f;

[0293] Step S2-2-3: Establish the mean income model:

[0294]

[0295] In the formula, y revenue3 represents the predicted revenue of the tested flight on its departure day, It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0296] Step S2-2-4: Establish the income DOW mean model:

[0297]

[0298] In the formula, y revenue4 represents the predicted revenue of the tested flight on its departure day, F i represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. For example, if the tested flight takes off on Tuesday, when i=7 and i=14, the corresponding F i That is, the flights that took off on Tuesday one week and two weeks before the flight being tested. Denotes the flight set F i The number of flights in, f′ represents the flight set F i Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f;

[0299] Step S2-2-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained income XGBOOST model, the trained income XGBOOST model takes the scheduled income of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted income of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f;

[0300] Step S2-2-6, using historical flight data, calculate the root mean square error (RMSE) of the five revenue prediction models, namely, the trained revenue linear regression model, the trained revenue DOW linear regression model, the revenue mean model, the revenue DOW mean model, and the trained revenue XGBOOST model, and record the model with the smallest root mean square error (RMSE), that is, the smallest prediction error, as the optimal revenue prediction model;

[0301] Step S2-2-7, using the optimal revenue prediction model to predict the predicted revenue of the target flight f;

[0302] Step S2-2-8, using the historical flight data, calculate the historical average ticket price of flights with the same flight number as the target flight f, and divide the calculation result of step S2-2-7 by the predicted number of passengers of the target flight f to obtain the predicted ticket price; if the deviation between the predicted ticket price and the historical average ticket price is within the preset revenue deviation threshold, then the calculation result of step S2-2-7 is directly used as the predicted revenue of the target flight f; otherwise, the product of the predicted number of passengers of the target flight f and the historical average ticket price is used as the predicted revenue of the target flight f.

[0303] Embodiment 3

[0304] Based on the above-mentioned embodiment 1 or embodiment 2, this embodiment 3 also adopts the following preferred implementation manner:

[0305] In the step S2, the cost prediction method of step S2-3 is used to predict the predicted cost of any target flight f when the aircraft model is aircraft type k;

[0306] Step S2-3, the cost prediction method comprises:

[0307] Step S2-3-1, computer-based mean cost prediction result y cost1 and the model mean volatility A1;

[0308]

[0309]

[0310] In the formula, F 1,30 represents the set of flights that have the same route and are operated by the same aircraft type as the target flight f within 30 days before the departure of the target flight f, |F 1,30 | represents the flight set F 1,30 The number of flights in, f′ represents the flight set F 1,30 Flights in represents the actual cost of flight f′;

[0311] Furthermore, it is determined whether the average volatility A1 of the model is approximately the preset volatility threshold; if the determination result is no, the average cost prediction result y of the model is used. cost1 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-2;

[0312] Step S2-3-2: Calculate the seat grade mean cost prediction result y cost2 and seat-level mean volatility A2;

[0313]

[0314]

[0315] In the formula, It indicates the set of flights with the same route and the same seat quantity level as the target flight f within 30 days before the departure of the target flight f, and the difference between any two adjacent seat quantity levels is 10 seats, for example: the first-level seat quantity level is 100 seats, and the second-level seat quantity level is 110 seats; Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual cost of flight f′;

[0316] Furthermore, it is determined whether the average volatility A2 of the seat grade is approximately equal to a preset volatility threshold; if the determination result is no, the seat grade average cost prediction result y is used. cost2 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-3;

[0317] Step S2-3-3, using the trained cost linear regression model to calculate the predicted cost of the target flight f when the aircraft model is aircraft type k, including: Step S2-3-3A, establishing a cost linear regression model:

[0318] y cost=b cost +b distance x distance ;

[0319] In the formula, y cost represents the predicted flight cost of the flight, x distance Indicates the flight distance of the flight; b cost and b distance Represent the intercept and coefficient to be regressed respectively;

[0320] Step S2-3-3B: Filter the historical flight data within 60 days before the flight plan date according to the following three screening conditions in turn, and use the cost historical flight data selected by the first screening condition that meets the cost prediction sample quantity requirement as training data, and calculate the intercept b of the cost linear regression model. cost and coefficient b distance Performing a first-order linear fitting to obtain the trained cost linear regression model;

[0321] The cost historical flight data includes: the flight distance of the historical flight as the flight distance x distance , the actual flight cost of the historical flight is used as the predicted flight cost y cost ;

[0322] The requirement for the number of cost prediction samples is: the number of cost historical flight data selected by the screening condition is above the preset minimum training sample number threshold;

[0323] The three screening conditions are:

[0324] Screening condition 1: The aircraft model of the historical flight is the same as that of the target flight f, and the stopover classification of the historical flight is the same as that of the target flight f, and the international flight classification of the historical flight is the same as that of the target flight f;

[0325] Screening condition 2: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f, and the historical flight has the same international flight classification as the target flight f;

[0326] Screening condition three: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f;

[0327] Step S2-3-3C: Substitute the flight distance of the target flight f into the flight distance x of the trained cost linear regression model. distance , the calculated predicted flight cost y cost That is, it is the predicted cost when the target flight f is operated by an aircraft of model k.

[0328] The present invention is not limited to the above-mentioned specific implementation modes. According to the above-mentioned contents, in accordance with the common technical knowledge and customary means in the field, without departing from the above-mentioned basic technical ideas of the present invention, the present invention can also make other various forms of equivalent modifications, replacements or changes, all of which fall within the protection scope of the present invention.

Claims

1. A flight model adjustment optimization method, characterized in that: include: Step S1, record the Tth day before the end time of the model adjustment target date as the flight plan date, T≥1, and the duration of the model adjustment target date is less than the value of T; on the flight plan date, read the following data of the target airline: a target flight set F whose departure time is on the model adjustment target date, a model set K of all operable aircraft that can fly on the model adjustment target date, and historical flight data related to the target flight set F; Step S2, based on the historical flight data, predicting: the predicted number of passengers and the predicted revenue of each target flight f in the target flight set F, and the predicted cost of each target flight f when operated by an aircraft of model k, k∈K; And, based on the prediction results, the predicted revenue of each target flight f when using an available aircraft of model k is calculated. In the step S2, the predicted number of passengers for any target flight f is predicted using the passenger seat prediction method of step S2-1; Step S2-1, the method for predicting the number of passengers seated comprises: Step S2-1-1: Training steps of the trained attendance rate linear regression model: Step S2-1-1A, establish the attendance rate linear regression model: y load =b load +b T_load x T_load ; In the formula, y load represents the predicted occupancy rate of the flight on the departure day, x T_load represents the scheduled occupancy rate of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, and T≥1; b load and b T_load Represent the intercept and coefficient to be regressed respectively; Step S2-1-1B: Filter the historical flight data within 60 days before the scheduled flight date according to the following four screening conditions in turn, and use the historical flight data with attendance rate filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the attendance rate linear regression model: load and coefficient b T_load Performing a first-order linear fitting to obtain the trained attendance rate linear regression model; The historical flight attendance data includes: the scheduled attendance of the historical flight on its scheduled date as the scheduled attendance x T_load , the actual load factor of the historical flight on its departure day is taken as the predicted load factor y load ; Furthermore, when the trained load factor linear regression model is used to predict the predicted number of passengers for the target flight f, the scheduled load factor of the target flight f on the flight schedule date is substituted into the scheduled load factor x T_load , the calculated predicted attendance rate y load The product of the number of seats on the corresponding aircraft is the predicted number of passengers on the target flight f; The sample quantity requirement is that the number of historical flight occupancy rate data selected by the screening condition is above the preset minimum training sample quantity threshold; The four screening conditions are: First screening condition: the historical flights have the same flight number and the same flight segment, and the deviation between the actual load factor and the scheduled load factor of the historical flights is within plus or minus 10%; The second screening condition: the flight segments of the historical flights are the same, the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f, and the scheduled seating level includes five levels of scheduled seating rates of 0-20%, 21-40%, 41-60%, 61-80%, and 81-100% respectively; The third screening condition: the departure airport of the historical flights is the same, and the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f; Fourth screening condition: the scheduled seating level of the historical flight is the same as the scheduled seating level of the target flight f; Step S2-1-2: Training steps of the trained seat occupancy DOW linear regression model: Step S2-1-2A: Establish the linear regression model of the number of passengers DOW: Wherein, y num represents the predicted number of passengers on the flight on the departure date, and x T_num represents the reserved number of passengers on the flight on the reserved date, where the reserved date is the T-th day before the departure date, and T≥1; represents the actual number of passengers on a flight with the same flight number as the flight corresponding to the predicted number of passengers y num and taking off h days before the departure date of the corresponding flight; b num represents the intercept to be regressed, and b T_num and both represent the coefficients to be regressed, and if h<T, then is directly taken as 0; Step S2-1-2B, from the historical flight data within 60 days before the scheduled date of the flight, filter according to the four screening conditions in turn, and use the historical flight data of the number of passengers screened out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the DOW linear regression model of the number of passengers num , coefficient b T_num and coefficient Perform a first-order linear fitting to obtain the trained seat occupancy DOW linear regression model; The historical flight data of the number of passengers includes: the number of passengers scheduled on the historical flight on its scheduled date as the number of passengers scheduled x T_num The actual occupancy rate of the historical flight on its departure day is taken as the predicted number of passengers y num The actual number of passengers on the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight is taken as the actual number of passengers on the flight. Furthermore, when the trained passenger number DOW linear regression model is used to predict the predicted passenger number of the target flight f, the scheduled passenger number of the target flight f on the flight schedule date is substituted into the scheduled passenger number x T_num The actual number of passengers on the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute the actual number of passengers The calculated predicted number of occupants y num That is, the predicted number of passengers for the target flight f; Step S2-1-3: Establish the mean model of the number of people seated: In the formula, y num2 It indicates the predicted number of passengers on the flight under test on its departure date. It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a collection of flights The number of flights in f ′ represents the flight set Flights in Indicate flight f '; wherein the measured flight is a historical flight in the historical flight data or the target flight f; Step S2-1-4: Establish the DOW mean model of the number of passengers: In the formula, y num3 represents the predicted number of passengers on the flight under test on its departure date, F i It represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. Denotes the flight set F i The number of flights in f ′ represents the flight set F i Flights in Indicate flight f '; wherein the measured flight is a historical flight in the historical flight data or the target flight f; Step S2-1-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained passenger number XGBOOST model, the trained passenger number XGBOOST model takes the scheduled passenger number of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted passenger number of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f; Step S2-1-6, using historical flight data, calculate the root mean square error of the five passenger seat prediction models, namely, the trained passenger seat linear regression model, the trained passenger seat DOW linear regression model, the passenger seat mean model, the passenger seat DOW mean model, and the trained passenger seat XGBOOST model, and record the model with the smallest root mean square error as the optimal passenger seat prediction model; Step S2-1-7, using the optimal passenger seat prediction model to predict the predicted passenger seat number of the target flight f; In the step S2, the revenue prediction method of step S2-2 is used to predict the predicted revenue of any target flight f; Step S2-2, the income forecasting method comprises: Step S2-2-1: Training steps of the trained income linear regression model: Step S2-2-1A: Establish an income linear regression model: In the formula, y revenue represents the forecasted revenue of the flight on the departure day, represents the scheduled revenue of the flight on the scheduled date, where the Tth day before the departure date is the scheduled date, and T≥1; b revenue and Represent the intercept and coefficient to be regressed respectively; Step S2-2-1B: Filter the historical flight data within 60 days before the flight schedule date according to the following four screening conditions in turn, and use the income historical flight data filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the income linear regression model. revenue and coefficient Performing a first-order linear fitting to obtain the trained income linear regression model; The historical flight revenue data includes: the scheduled revenue of the historical flight on its scheduled date as the scheduled revenue x T_load , the actual income of the historical flight on its departure date is taken as the predicted income y load ; Furthermore, when the trained revenue linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue The calculated predicted income y revenue That is, the predicted revenue of the target flight f; The sample quantity requirement is: the number of revenue historical flight data screened by the screening condition is above the preset minimum training sample quantity threshold; The four screening conditions are: First screening condition: the historical flights have the same flight number and the same flight segment, and the deviation between the actual load factor and the scheduled load factor of the historical flights is within plus or minus 10%; The second screening condition: the flight segments of the historical flights are the same, the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f, and the scheduled seating level includes five levels of scheduled seating rates of 0-20%, 21-40%, 41-60%, 61-80%, and 81-100% respectively; The third screening condition: the departure airport of the historical flights is the same, and the scheduled seating level of the historical flights is the same as the scheduled seating level of the target flight f; Fourth screening condition: the scheduled seating level of the historical flight is the same as the scheduled seating level of the target flight f; Step S2-2-2: Training steps of the trained DOW linear regression model: Step S2-2-2A: Establish the income DOW linear regression model: where y revenue2 represents the predicted revenue of the flight on the departure date, x T_revenue2 represents the booked revenue of the flight on the booking date, and the booking date is the T-th day before the departure date, where T≥1; represents the actual revenue of a flight with the same flight number as the flight corresponding to the predicted revenue y revenue2 and taking off h days before the departure date of the corresponding flight; b revenue2 represents the intercept to be regressed, b T_revenue2 and both represent the coefficients to be regressed, and if h<T, then it is directly taken as 0; Step S2-2-2B: Filter the historical flight data within 60 days before the flight schedule date according to the four screening conditions in turn, and use the income historical flight data filtered out by the first screening condition that meets the sample quantity requirement as training data, and calculate the intercept b of the income DOW linear regression model. revenue2 , coefficient b T_revenue2 and coefficient Performing a first-order linear fitting to obtain the trained income DOW linear regression model; The historical flight revenue data includes: the scheduled revenue of the historical flight on its scheduled date as the scheduled revenue x T_revenue2 , the actual load factor of the historical flight on its departure date is used as the predicted revenue y revenue2 The actual revenue of the flight with the same flight number as the historical flight and taking off h days before the departure date of the historical flight is taken as the actual revenue Furthermore, when the trained revenue DOW linear regression model is used to predict the predicted revenue of the target flight f, the scheduled revenue of the target flight f on the flight schedule date is substituted into the scheduled revenue x T_revenue2 The actual revenue of the flight with the same flight number as the target flight f and taking off h days before the departure date of the target flight f is used to substitute the actual revenue The calculated predicted income y revenue2 That is, the predicted revenue of the target flight f; Step S2-2-3: Establish the mean income model: In the formula, y revenue3 represents the predicted revenue of the tested flight on its departure day, It represents the set of flights with the same flight number and the same flight segment as the flight under test within 30 days before the departure date. Represents a flight collection The number of flights in, f′ represents the flight set Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f; Step S2-2-4: Establish the income DOW mean model: In the formula, y revenue4 represents the predicted revenue of the tested flight on its departure day, F i It represents the set of flights with the same flight number, same flight segment, and same departure day and week as the tested flight within 30 days before the departure date. Denotes the flight set F i The number of flights in, f′ represents the flight set F i Flights in represents the actual revenue of flight f′; wherein the measured flight is a historical flight in the historical flight data or the target flight f; Step S2-2-5, using the historical flight data within 90 days before the scheduled date of the flight to train the XGBOOST model to obtain a trained income XGBOOST model, the trained income XGBOOST model takes the scheduled income of the tested flight T days before its departure date, information on whether the tested flight is a direct flight, and whether the tested flight has a long stop as input, and takes the predicted income of the tested flight as output; wherein the tested flight is a historical flight in the historical flight data or the target flight f; Step S2-2-6, using historical flight data, calculate the root mean square deviation of the five revenue prediction models, namely, the trained revenue linear regression model, the trained revenue DOW linear regression model, the revenue mean model, the revenue DOW mean model, and the trained revenue XGBOOST model, and record the model with the smallest root mean square deviation as the optimal revenue prediction model; Step S2-2-7, using the optimal revenue prediction model to predict the predicted revenue of the target flight f; Step S2-2-8, using the historical flight data, calculate the historical average ticket price of flights with the same flight number as the target flight f, and divide the calculation result of step S2-2-7 by the predicted number of passengers of the target flight f to obtain the predicted ticket price; if the deviation between the predicted ticket price and the historical average ticket price is within a preset revenue deviation threshold, directly use the calculation result of step S2-2-7 as the predicted revenue of the target flight f; otherwise, use the product of the predicted number of passengers of the target flight f and the historical average ticket price as the predicted revenue of the target flight f; In the step S2, the cost prediction method of step S2-3 is used to predict the predicted cost of any target flight f when the aircraft model is aircraft type k; Step S2-3, the cost prediction method comprises: Step S2-3-1, computer-based mean cost prediction result y cost1 and the model mean volatility A1; In the formula, F 1,30 represents the set of flights that have the same route and are operated by the same aircraft type as the target flight f within 30 days before the departure of the target flight f, |F 1,30 | represents the flight set F 1,30 The number of flights in, f′ represents the flight set F 1,30 Flights in represents the actual cost of flight f′; Furthermore, it is determined whether the average volatility A1 of the model is approximately the preset volatility threshold; if the determination result is no, the average cost prediction result y of the model is used. cost1 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-2; Step S2-3-2: Calculate the seat grade mean cost prediction result y cost2 and seat-level mean volatility A2; In the formula, It represents the set of flights with the same route and the same seat quantity level as the target flight f within 30 days before the departure of the target flight f, and the difference between any two adjacent seat quantity levels is 10 seats. Represents a collection of flights The number of flights in, f′ represents the flight set Flights in represents the actual cost of flight f′; Furthermore, it is determined whether the average volatility A2 of the seat grade is approximately equal to a preset volatility threshold; if the determination result is no, the seat grade average cost prediction result y is used. cost2 as the predicted cost of the target flight f when the aircraft model is aircraft type k; if the judgment result is yes, go to step S2-3-3; Step S2-3-3, using the trained cost linear regression model to calculate the predicted cost of the target flight f when the aircraft model is aircraft type k, including: Step S2-3-3A, establishing a cost linear regression model: y cost =b cost +b distance x distance ; In the formula, y cost represents the predicted flight cost of the flight, x distance Indicates the flight distance of the flight; b cost and b distance Represent the intercept and coefficient to be regressed respectively; Step S2-3-3B: Filter the historical flight data within 60 days before the flight plan date according to the following three screening conditions in turn, and use the cost historical flight data selected by the first screening condition that meets the cost prediction sample quantity requirement as training data, and calculate the intercept b of the cost linear regression model. cost and coefficient b distance Performing a first-order linear fitting to obtain the trained cost linear regression model; The cost historical flight data includes: the flight distance of the historical flight as the flight distance x distance , the actual flight cost of the historical flight is used as the predicted flight cost y cost ; The requirement for the number of cost prediction samples is: the number of cost historical flight data selected by the screening condition is above the preset minimum training sample number threshold; The three screening conditions are: Screening condition 1: The aircraft model of the historical flight is the same as that of the target flight f, and the stopover classification of the historical flight is the same as that of the target flight f, and the international flight classification of the historical flight is the same as that of the target flight f; Screening condition 2: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f, and the historical flight has the same international flight classification as the target flight f; Screening condition three: the historical flight has the same seat quantity level as the target flight f, and the historical flight has the same stopover classification as the target flight f; Step S2-3-3C: Substitute the flight distance of the target flight f into the flight distance x of the trained cost linear regression model. distance , the calculated predicted flight cost y cost That is, the predicted cost when the target flight f is operated by an aircraft of model k; Step S3: Establish an integer programming model and use the target flight set F, aircraft model set K and predicted revenue Solving the problem to obtain an optimal flight model scheduling plan that satisfies the constraints and objective function of the integer programming model; In step S3, the constraint conditions include: The objective function includes: Wherein, A represents the set of take-off and landing airports of all the target flights f, a represents the take-off and landing airports in the take-off and landing airport set A, and the take-off and landing airport a is a departure airport or a landing airport; N k represents the take-off and landing time-space node set of all aircraft of model k in the model set K, and n represents the take-off and landing time-space node set N k The take-off and landing space-time node in the take-off and landing space-time node n is a take-off space-time node including the take-off time and the take-off airport or a landing space-time node including the landing time and the landing airport; The take-off time and space node of the first flight of the operable aircraft of model k on the target date of adjustment of the model; The take-off and landing time-space node set representing all the aircraft of model k in the aircraft set K that can fly at 3 a.m. on the target date for adjusting the aircraft model; O(k,n) represents the set of flights that take off from the take-off and landing space-time node n using an aircraft of type k that can fly; I(k,n) represents the set of flights that land at the take-off and landing space-time node n using an aircraft of type k that can fly; F c represents the set of cancelable target flights f in the target flight set F; M represents the set F c The set of flights with the same flight number in , m represents a subset of the set M, and the departure times of the target flights f in the subset m have the same weekday; represents the set of target flights f that are flying at the take-off and landing time and space node n in the target flight set F; represents the aircraft model flight decision variable, where When the value is 1, it means that the target flight f is operated by an aircraft of model k. When the value is 0, it means that the aircraft of model k is not used to fly the target flight f; all the model flight decision variables of the target flight f A value combination method of forms a flight model scheduling plan for the target flight set F; Indicates the number of aircraft of model k that can fly before the take-off and landing time and space node n; Indicates the number of aircraft of model k that can fly after the take-off and landing time and space node n; Relative to the initial quantity The absolute value of the number of new aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented; Relative to the initial quantity The absolute value of the reduction in the number of aircraft of type k that can fly at take-off and landing airport a after the corresponding flight type scheduling plan is implemented; The initial quantity The number of aircraft of type k that can fly at the take-off and landing airport a before the corresponding flight type scheduling plan is executed; N represents the preset upper limit of the number of model adjustments; N k represents the number of aircraft of model k in the model set K that can fly; N m represents the minimum number of flights m in the preset set M; k f Indicates the aircraft type used by the target flight f in the original plan before the corresponding flight aircraft type scheduling plan is executed; The process of solving the optimal flight model scheduling plan is: The first step is to exhaustively enumerate all the aircraft model and flight decision variables of the target flight f under the premise of satisfying the constraints described in formulas 1 to 6. , so as to find a value combination that meets the formula 7; if there is only one value combination in the search result, then the value combination is the optimal flight model scheduling plan, and the solution process ends; if there are multiple value combinations in the search result, the found value combination is recorded as the first target flight model scheduling plan, and the second step is executed; Step 2: Find a flight model scheduling plan that meets the formula 8 from each of the first target flight model scheduling plans; if there is only one flight model scheduling plan in the search result, then the flight model scheduling plan is the optimal flight model scheduling plan, and the solution process ends; if there are multiple flight model scheduling plans in the search result, then the found flight model scheduling plan is recorded as the second target flight model scheduling plan, and the third step is executed; Step 3: Find a flight model scheduling plan that meets the formula 9 from each of the second target flight model scheduling plans, and the search result is the optimal flight model scheduling plan; Step S4: on the target date for adjusting the aircraft type, according to the optimal flight aircraft type scheduling plan, dispatching an available aircraft of the corresponding aircraft type k for the target flight f in the target flight set F to fly.