Flight sales income prediction method and system

By dividing the flight sales period into multiple sales time points, combining model constraints and refund rates, predicting flight revenues at each time point, it solves the complexity and risk of flight revenue forecasting caused by dynamic changes in the air ticket market, and improves the prediction accuracy .

CN118917528APending Publication Date: 2024-11-08FEIYOU TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410949794.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-16
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Dynamic changes in the air ticket sales market have led to complex flight revenue forecasts and high difficulty and risks. They are affected by a variety of factors, such as seasonal fluctuations, market promotions, economic conditions and competitive environment, resulting in low prediction accuracy.

Method used

A method for predicting flight sales revenue is proposed. By obtaining flight sales period and flight number, dividing sales time points, obtaining model constraints and minimum fare prices, using a pre-trained sales quantity prediction model to calculate sales quantity and revenue, and considering the refund rate. Forecast revenue at each time point to improve prediction accuracy.

Benefits of technology

By predicting flight revenue at each time point and predicting the final sales revenue based on the revenue and refund rate of each sales point, the accuracy of flight revenue forecasts is improved and the forecast risk is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917528A_ABST
    Figure CN118917528A_ABST
Patent Text Reader

Abstract

The invention discloses a flight sales income prediction method and system, and the method comprises the steps: obtaining a flight sales time period T of a to-be-predicted sales income and a corresponding flight number, and dividing the flight sales time period T into a plurality of sales time points, namely T = {t1, t2,..., ti,..., tn}; obtaining model constraints corresponding to each sales time point ti of the flight number from the lookup table according to the flight number and the plurality of sales time points, wherein the model constraints include but are not limited to purchase intention intensity a and standard deviation sigma of a Gaussian function; and inputting the model constraint and the plurality of sales time points into a pre-trained sales quantity prediction model according to a time point sequence to obtain a sales quantity S (ti) corresponding to each sales time point ti. According to the flight sales revenue prediction method and system, the revenue of the flight is predicted according to the sales time points one by one, the total sales revenue is predicted based on the revenue of each sales point and the refunding rate r (ti) of each sales point, and the accuracy of flight revenue prediction is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of flight revenue forecasting, and in particular to a flight sales revenue forecasting method and system. Background Art

[0002] Airline ticket sales is a dynamic market, and ticket prices and sales are affected by many factors, such as seasonal fluctuations, market promotions, economic conditions, and competitive environment. For airlines, the revenue of each flight is an integral part of the airline's revenue, and the accuracy of the forecast will affect the effect of aircraft scheduling and future flight schedule application plans. Usually, when forecasting flight revenue, due to multi-dimensional factors such as different periods, different shared flights, and different times, the forecasting work is more difficult and there are also greater risks. Summary of the invention

[0003] In order to solve the technical problems existing in the background technology, the present invention proposes a flight sales revenue forecasting method and system.

[0004] A flight sales revenue forecasting method proposed by the present invention comprises the following steps:

[0005] S1. Obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n};

[0006] S2. Obtain each sales time point t of the flight number from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include but are not limited to the purchase intention intensity a and the standard deviation σ of the Gaussian function;

[0007] S3, input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in time order, and obtain the sales quantity prediction model for each sales time point t i The corresponding sales quantity S(t i );

[0008] S4. Query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i );

[0009] S5, based on each sales time point t i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(ti );

[0010] S6. Get each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.

[0011] Preferably, the lookup table is pre-configured with each sales time point t i A one-to-one mapping relationship between the model constraints.

[0012] Preferably, it also includes:

[0013] S7. Update the model constraints in the lookup table according to a preset strategy within a preset time period; the preset strategy specifically includes:

[0014]

[0015] Among them, η1 is the learning rate corresponding to the purchase intention intensity a; is the gradient of the loss function with respect to parameter a;

[0016]

[0017] Among them, η2 is the learning rate corresponding to δ; is the gradient of the loss function with respect to the parameter δ.

[0018] Preferably, the training process of the sales quantity prediction model is as follows:

[0019] Collect historical sales data of each route and form a training sample set with the historical sales data. The historical sales data includes but is not limited to the total number of sales of each flight N, the lowest fare at each sales time point, flight date, time, destination, aircraft model, and promotional information;

[0020] The total number of sales people N and the purchase desire index D(t i ) is the input, sales quantity S(t i ) is the output, based on the training sample set and loss function Perform weak supervision training on the preset sales quantity prediction model to obtain a trained sales quantity prediction model.

[0021] Preferably, the loss function Specifically:

[0022] The actual sales revenue Ractual (t i ) and adjusted forecast sales revenue R adjusted (t i ) is used as the loss function.

[0023] Loss Function The calculation process is as follows:

[0024]

[0025] Among them, R actual (t i ) is the actual sales revenue; R adjusted (t i ) is the adjusted forecast sales revenue.

[0026] Preferably, each sales time point t i The corresponding sales forecast revenue R(t i ) is calculated as follows:

[0027] R(t i )=P(t i )×S(t i );

[0028] Among them, R(t i ) for each sales time point t i The corresponding sales forecast revenue; P(t i ) for each sales time point t i The corresponding minimum fare; S(t i ) for each sales time point t i The corresponding sales quantity.

[0029] Preferably, the above method is based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R of the flight sales period T, including:

[0030] Based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate each sales time point t after considering the refund factor i The corresponding sales forecast revenue R′(t i ), R′(t i ) is calculated as follows:

[0031] R′(t i )=R(t i )×(1-r(t i ));

[0032] Set the flight sales period T = {t1, t2, …, t i ,…,t n} for each sales time point t i The corresponding sales forecast revenue R′(t i )The sum of the two is taken as the sales forecast revenue R for flight sales period T.

[0033] The present invention provides a flight sales revenue forecasting system, comprising:

[0034] The data collection module is used to obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n};

[0035] The data query module is used to obtain the flight number and multiple sales time points from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include but are not limited to the purchase intention intensity a and the standard deviation σ of the Gaussian function;

[0036] The first processing module is used to input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in the order of time points, and obtain each sales time point t i The corresponding sales quantity S(t i );

[0037] The second processing module is used to query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i );

[0038] The third processing module is used to process the i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i );

[0039] Data generation module, used to obtain each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(ti ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.

[0040] In the present invention, the proposed flight sales revenue prediction method and system obtain the flight sales period T and the corresponding flight number of the flight sales revenue to be predicted, and divide the flight sales period T into multiple sales time points; according to the flight number and the multiple sales time points, obtain the flight number from the lookup table for each sales time point t i Corresponding model constraints; input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in time point order, and obtain the sales quantity prediction model for each sales time point t i The corresponding sales quantity S(t i ) ; query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i ); Based on each sales time point t i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i );Get each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) calculates the sales forecast revenue R of the flight sales period T. By predicting the revenue of the flight at each sales point in time, and based on the revenue of each sales point and the refund rate r(t i ) predicts the final sales revenue and improves the accuracy of flight revenue forecasts. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the workflow structure of a flight sales revenue forecasting method proposed by the present invention;

[0042] Figure 2 A schematic diagram of the system architecture of a flight sales revenue forecasting system proposed by the present invention. DETAILED DESCRIPTION

[0043] Reference Figure 1 and Figure 2 The present invention proposes a flight sales revenue forecasting method, comprising the following steps:

[0044] S1. Obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n}.

[0045] S2. Obtain each sales time point t of the flight number from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include but are not limited to the purchase intention intensity a and the standard deviation σ of the Gaussian function.

[0046] In this embodiment, the lookup table is pre-configured with each sales time point t i A one-to-one mapping relationship between the model constraints.

[0047] In this embodiment, based on the assumption that ticket prices can reflect passengers' willingness to buy, that is, the higher the market ticket price, the higher the passenger's willingness to buy, historical ticket price data P(t) is analyzed to determine its relationship with the desire to buy. The time point Tpeak with the highest willingness to buy is determined. A Gaussian function is used to simulate the purchase desire index D(t), and the formula is:

[0048]

[0049] Where a is the peak value of purchasing desire at Tpeak, and σ is the standard deviation of the Gaussian distribution.

[0050] S3, input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in time order, and obtain the sales quantity prediction model for each sales time point t i The corresponding sales quantity S(t i ).

[0051] In this embodiment, since we usually only know the final sales total number after the flight ends, reverse calculation allows us to simulate the possible sales situation at each time point in the sales cycle. Based on the final sales total number N and the purchase desire index D(t), the sales quantity S(t) at each time point ti is calculated. i ).

[0052] The formula for the reverse calculation algorithm is:

[0053]

[0054] Among them, S(t i ) represents the sales quantity at time point ti; N is the known final sales total number, that is, the total number of tickets actually sold after the flight takes off; D(t i ) is the time point t iThe purchase desire index reflects the passengers’ willingness to buy tickets at that time point; Represents the sum of the purchase desire index at all time points in the entire sales cycle as a normalization factor.

[0055] In this embodiment, the training process of the sales quantity prediction model is as follows:

[0056] Collect historical sales data of each route and form a training sample set with the historical sales data. The historical sales data includes but is not limited to the total number of sales of each flight N, the lowest fare at each sales time point, flight date, time, destination, aircraft model, and promotional information;

[0057] The total number of sales people N and the purchase desire index D(t i ) is the input, sales quantity S(t i ) is the output, based on the training sample set and loss function Perform weak supervision training on the preset sales quantity prediction model to obtain a trained sales quantity prediction model.

[0058] In this embodiment, the sources of historical sales data include: the airline's sales interface, which provides information such as the lowest fare and flight details at each time point. The airline's refund interface, which provides relevant information such as the refund quantity and refund time. The online travel agency (OTA) sales interface, which obtains sales data through OTA channels. The OTA refund interface, which obtains refund data through OTA channels.

[0059] In this embodiment, the collected historical sales data will be cleaned, including removing outliers, filling missing values, and unifying the format to ensure the quality of the data. At the same time, standardized time point processing will be performed to ensure that the data is compared and analyzed on the same time scale, which is helpful for subsequent data analysis and model building. Special analysis is performed on the refund data to explore the regularity of refund behavior and factors that may affect the refund rate, such as seasonality, weather conditions, flight delays, etc. The sales data and refund data are integrated to provide comprehensive data support for the subsequent purchase desire index modeling and sales forecasting.

[0060] In this embodiment, the loss function Specifically:

[0061] The actual sales revenue R actual (t i ) and adjusted forecast sales revenue R adjusted (t i ) is used as the loss function.

[0062] Loss Function The calculation process is as follows:

[0063]

[0064] Among them, R actual (t i ) is the actual sales revenue; R adjusted (t i ) is the adjusted forecast sales revenue.

[0065] S4. Query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i ).

[0066] S5, based on each sales time point t i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i ).

[0067] In this embodiment, each sales time point t i The corresponding sales forecast revenue R(t i ) is calculated as follows:

[0068] R(t i )=P(t i )×S(t i );

[0069] Among them, R(t i ) for each sales time point t i The corresponding sales forecast revenue; P(t i ) for each sales time point t i The corresponding minimum fare; S(t i ) for each sales time point t i The corresponding sales quantity.

[0070] S6. Get each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.

[0071] In this embodiment, ticket refunds directly affect the actual revenue of the airline, because after the sold tickets are refunded, the airline will lose the originally expected revenue. The refund rate refers to the ratio of the number of tickets refunded to the total number of tickets sold at a specific time point or time period. If the refund rate is not taken into account, the prediction model may overestimate the actual sales revenue because the model only predicts revenue based on the sales volume. By introducing the refund rate, the predicted sales revenue can be adjusted to more accurately reflect the expected revenue of the airline. The refund rate can be obtained by analyzing historical refund data, which can be collected from the airline's ticketing system or OTA platform. Factors that need to be considered include the impact of seasonal changes, special events, flight cancellations or delays, etc. on the refund rate. The refund rate may change over time and market conditions, so the refund rate data needs to be updated regularly to maintain the accuracy of the prediction model.

[0072] In this embodiment, based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R of the flight sales period T, including:

[0073] Based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate each sales time point t after considering the refund factor i The corresponding sales forecast revenue R′(t i ), R′(t i ) is calculated as follows:

[0074] R′(t i )=R(t i )×(1-r(t i ));

[0075] Set the flight sales period T = {t1, t2, …, t i ,…,t n} for each sales time point t i The corresponding sales forecast revenue R′(t i )The sum of the two is taken as the sales forecast revenue R for flight sales period T.

[0076] In this embodiment, it also includes:

[0077] S7. Update the model constraints in the lookup table according to a preset strategy within a preset time period; the preset strategy specifically includes:

[0078]

[0079] Among them, η1 is the learning rate corresponding to the purchase intention intensity a; is the gradient of the loss function with respect to parameter a;

[0080]

[0081] Among them, η2 is the learning rate corresponding to δ; is the gradient of the loss function with respect to the parameter δ.

[0082] In this embodiment, the model constraint is used to update the data so that the loss function Minimum.

[0083] Reference Figure 1 and Figure 2 The present invention proposes a flight sales revenue forecasting system, comprising:

[0084] The data collection module is used to obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n};

[0085] The data query module is used to obtain the flight number and multiple sales time points from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include but are not limited to the purchase intention intensity a and the standard deviation σ of the Gaussian function;

[0086] The first processing module is used to input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in the order of time points, and obtain each sales time point t i The corresponding sales quantity S(t i );

[0087] The second processing module is used to query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i );

[0088] The third processing module is used to process the i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i );

[0089] Data generation module, used to obtain each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.

[0090] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A flight sales revenue forecasting method, characterized in that: The following steps are involved: S1. Obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n }; S2. Obtain each sales time point t of the flight number from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include, but are not limited to, the purchase intention intensity a, the standard deviation of the σ Gaussian function; S3, input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in time order, and obtain the sales quantity prediction model for each sales time point t i The corresponding sales quantity S(t i ); S4. Query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i ); S5, based on each sales time point t i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i ); S6. Get each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.

2. The flight sales revenue forecasting method according to claim 1, characterized in that: The lookup table is pre-configured with each sales time point t i A one-to-one mapping relationship between the model constraints.

3. The flight sales revenue forecasting method according to claim 1, characterized in that: Also includes: S7, updating the model constraints in the lookup table according to a preset strategy within a preset time period; The preset strategy specifically includes: Among them, η1 is the learning rate corresponding to the purchase intention intensity a; is the gradient of the loss function with respect to parameter a; Among them, η2 is the learning rate corresponding to δ; is the gradient of the loss function with respect to the parameter δ.

4. The flight sales revenue forecasting method according to claim 1, characterized in that: The training process of the sales quantity prediction model is as follows: Collect historical sales data of each route and form a training sample set with the historical sales data. The historical sales data includes but is not limited to the total number of sales of each flight N, the lowest fare at each sales time point, flight date, time, destination, aircraft model, and promotional information; The total number of sales people N and the purchase desire index D(t i ) is the input, sales quantity S(t i ) is the output, based on the training sample set and loss function Perform weak supervision training on the preset sales quantity prediction network to obtain a trained sales quantity prediction model.

5. The flight sales revenue forecasting method according to claim 4, characterized in that: The loss function Specifically: The actual sales revenue R actual (t i ) and adjusted forecast sales revenue R adjusted (t i ) is used as the loss function. Loss Function The calculation process is as follows: Among them, R actual (t i ) is the actual sales revenue; R adjusted (t i ) is the adjusted forecast sales revenue.

6. The flight sales revenue forecasting method according to claim 1, characterized in that: Each sales time point t i The corresponding sales forecast revenue R(t i ) is calculated as follows: R(t i )=P(t i )×S(t i ); Among them, R(t i ) for each sales time point t i The corresponding sales forecast revenue; P(t i ) for each sales time point t i The corresponding minimum fare; S(t i ) for each sales time point t i The corresponding sales quantity.

7. The flight sales revenue forecasting method according to claim 1, characterized in that: Based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R of the flight sales period T, including: Based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate each sales time point t after considering the refund factor i The corresponding sales forecast revenue R′(t i ), R′(t i ) is calculated as follows: R′(t i )=R(t i )×(1-r(t i )); Set the flight sales period T = {t1, t2, …, t i ,…,t n } for each sales time point t i The corresponding sales forecast revenue R′(t i )The sum of the two is taken as the sales forecast revenue R for flight sales period T.

8. A flight sales revenue forecasting system, characterized in that: include: The data collection module is used to obtain the flight sales period T and the corresponding flight number for the sales revenue to be predicted, and divide the flight sales period T into multiple sales time points, that is, T = {t1, t2, ..., t i ,…,t n }; The data query module is used to obtain the flight number and multiple sales time points from the lookup table according to the flight number and multiple sales time points. i The corresponding model constraints include but are not limited to the purchase intention intensity a and the standard deviation σ of the Gaussian function; The first processing module is used to input the model constraints and multiple sales time points into the pre-trained sales quantity prediction model in the order of time points, and obtain each sales time point t i The corresponding sales quantity S(t i ); The second processing module is used to query each sales time point t from the flight history database according to the flight number i The corresponding minimum fare P(t i ); The third processing module is used to process the i The corresponding minimum fare P(t i ) and sales quantity S(t i ) Calculate each sales time point t i The corresponding sales forecast revenue R(t i ); Data generation module, used to obtain each sales time point t i The corresponding refund rate r(t i ), and based on each sales time point t i The corresponding refund rate r(t i ), each sales time point t i The corresponding sales forecast revenue R(t i ) Calculate the sales forecast revenue R for flight sales period T.