Air ticket sales trend deep learning analysis system

Abstract

The invention relates to the technical field of air ticket sales analysis, and provides an air ticket sales trend deep learning analysis system. The method comprises the following specific implementation steps: collecting and cleaning historical airline ticket sales data, operation data and passenger data of an airline company through an airline ticket related data acquisition module; the air ticket related data processing module analyzes key air ticket attributes and operation attributes of different passenger cabin levels, calculates passenger space probability indexes, and forms historical sales data sets of different passenger cabin levels; the air ticket sales trend analysis module constructs a sales volume prediction model and optimizes the sales volume prediction model so as to predict future sales volumes of different passenger cabin levels; and the air ticket sales strategy regulation and control module determines an optimal sales strategy by using a reinforcement learning algorithm in combination with an economic state space, a sales strategy action space and a sales strategy value function. The method can effectively predict the future sales trend, optimizes the sales strategy, and improves the overall operation benefits of airlines.

Description

Technical Field

[0001] The present invention relates to the technical field of air ticket sales analysis, and particularly to a deep learning analysis system for air ticket sales trends. Background Art

[0002] With the rapid development of the global aviation market, air ticket sales have gradually shifted from traditional manual ticket booking to an online sales model centered on the Internet. The rapid development of Internet technology and the popularization of mobile payment enable airlines to quickly and conveniently achieve air ticket sales through their own platforms or third-party platforms, improving the interaction efficiency between airlines and consumers. In order to enable airlines to better understand passenger needs, optimize route layouts and ticket pricing strategies, air ticket sales trend analysis technology has emerged. By analyzing factors such as historical sales data, market demand fluctuations, and consumer purchase behaviors, this technology can accurately predict future sales trends, thereby optimizing pricing strategies, adjusting flight schedules, and enhancing passenger satisfaction.

[0003] Currently, existing sales trend analysis technologies focus on analyzing the impact of holidays, seasonal fluctuations, and passengers' historical travel behaviors on air ticket sales trends, ignoring the impact of specific flight schedules, airlines' service quality, as well as passengers' cabin type selections and flight purposes on air ticket sales situations, resulting in a decrease in the accuracy of air ticket sales trend prediction, which may lead to subsequent flight regulation and air ticket pricing errors, further causing customer loss and economic losses.

[0004] Therefore, a deep learning analysis system for air ticket sales trends is proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide a deep learning analysis system for air ticket sales trends, which collects and cleans historical air ticket sales data, operation data, and passenger data of airlines through an air ticket-related data collection module; then, the air ticket-related data processing module analyzes the key ticket attributes and operation attributes of different cabin classes, calculates the passenger cabin probability index, and forms a historical sales dataset for different cabin classes; then, the air ticket sales trend analysis module constructs and optimizes a sales volume prediction model to predict the future sales volume of different cabin classes; finally, the air ticket sales strategy regulation module uses a reinforcement learning algorithm, combined with an economic state space, a sales strategy action space, and a sales strategy value function, to determine the optimal sales strategy. The present invention can effectively predict future sales trends, optimize sales strategies, and improve the overall operational efficiency of airlines.

[0006] A deep learning analysis system for air ticket sales trends, comprising:

[0007] The air ticket related data collection module collects the air ticket sales data, operation data and passenger data of airlines within a historical time period, and performs data annotation and data cleaning to obtain historical air ticket sales data, historical operation data and historical passenger data;

[0008] The air ticket related data processing module obtains the air ticket attribute correlation degree according to the air ticket attributes of the historical air ticket sales data and the historical sales data of different cabin classes, and combines the constructed sales volume-air ticket attribute regression model to determine the key air ticket attributes of different cabin classes; by constructing a sales volume-operation attribute regression model and operation attribute screening, determine the key operation attributes of different cabin classes; calculate the passenger cabin probability index of different passengers, divide the historical data of passengers in different cabin classes, and combine the data of the corresponding key air ticket attributes and key operation attributes to obtain the historical sales data sets of different cabin classes;

[0009] The air ticket sales trend analysis module constructs a sales volume prediction model, performs model training and optimization on the historical sales data sets of different cabin classes, obtains the trained sales volume prediction model, and predicts the future sales volume of different cabin classes;

[0010] The air ticket sales strategy regulation module sets an economic state space, a sales strategy action space and a sales strategy value function, and adjusts the sales strategy in combination with the economic state space; calculates the adjusted air ticket profit index, and updates the sales strategy value function through a reinforcement learning algorithm to determine the optimal sales strategy.

[0011] Preferably, the air ticket sales data includes the departure time, arrival time, flight duration, date, air ticket fare and sales volume of each flight; the operation data includes the aircraft type, failure rate and cabin service quality; the passenger data includes the cabin class, ticket purchase frequency and ticket purchase purpose; the process of data annotation is to annotate the corresponding holidays according to the date of each flight, and replace the date attribute in the air ticket sales data with the holidays.

[0012] Preferably, the specific implementation process of the air ticket related data processing module includes:

[0013] Divide the sales volume in the historical air ticket sales data according to the cabin class to obtain the historical sales volume of economy class, business class and first class;

[0014] For the historical ticket sales data, calculate the ticket attribute correlation degree between different ticket attributes through the Pearson correlation coefficient; construct the sales volume - ticket attribute regression model, and determine the influence degree of different ticket attributes on the historical sales volume of different cabin classes through historical data fitting, so as to obtain the economy class ticket attribute influence degree, the business class ticket attribute influence degree, and the first class ticket attribute influence degree; in combination with the ticket attribute correlation degree, screen the ticket attributes corresponding to different cabin classes respectively to determine the economy class key ticket attributes, the business class key ticket attributes, and the first class key ticket attributes;

[0015] For the historical operation data, construct the sales volume - operation attribute regression model, and determine the economy class key operation attributes, the business class key operation attributes, and the first class key operation attributes through historical data fitting and operation attribute screening operations;

[0016] For the historical passenger data, calculate the passenger cabin probability index according to the cabin class, ticket purchase frequency, and ticket purchase purpose. The specific calculation formula is:

[0017]

[0018] where PC i represents the passenger cabin probability index of passenger i; represents the value of the cabin class c when the number of occurrences of the cabin class sequence C i is the highest, f() is the frequency function; exp() represents the exponential function with the natural constant as the base; β represents the average ticket purchase frequency of the ticket purchase frequency sequence F i ; represents the exponential weight of the ticket purchase purpose sequence P i ; p represents the specific encoded value of the ticket purchase purpose in the ticket purchase purpose sequence P 1 ; α i represents the influence factor of the cabin class sequence C 2 of passenger i, α i represents the influence factor of the ticket purchase frequency sequence F 3 of passenger i, α i represents the influence factor of the ticket purchase purpose sequence P 2 of passenger i, 0 < α 3 < α 1 < 1 and α 1 + α 2 + α 3 = 1;

[0019] According to the passenger cabin probability index, divide the historical passenger data into economy class passenger historical data, business class passenger historical data, and first class passenger historical data;

[0020] Integrate the data of passenger historical data, key ticket attributes, and key operation attributes corresponding to different cabin classes respectively to obtain the economy class historical sales dataset, business class historical sales dataset, and first class historical sales dataset.

[0021] Preferably, the sales volume prediction model includes a multi-dimensional feature extraction layer, a feature analysis and fusion layer, and a sales trend prediction layer;

[0022] The multi-dimensional feature extraction layer is used to extract ticket sales features, flight operation features, aircraft operation features, and passenger features from the historical sales datasets of different cabin classes;

[0023] The feature analysis and fusion layer is used to analyze the influence of the flight operation features, the aircraft operation features, and the passenger features on the ticket sales features of different cabin classes, and perform feature weighted fusion;

[0024] The sales trend prediction layer is used to predict the future sales volumes of different cabin classes.

[0025] Preferably, the specific implementation process of feature analysis and feature weighted fusion through the feature analysis and fusion layer includes:

[0026] Input the flight operation features and the passenger features into the cross-attention mechanism to capture the potential influence of different flights on passengers' ticket-buying behaviors and obtain the passenger-flight preference weights; input the aircraft operation features and the passenger features into the cross-attention mechanism to analyze the preferences of different passengers for aircraft operation conditions and obtain the passenger-aircraft preference weights; combine the passenger-flight preference weights and the passenger-aircraft preference weights, and perform weighted fusion on the passenger features, the aircraft operation features, and the flight operation features to obtain passenger preference features; through the LSTM network, mine the influence of the passenger preference features on the ticket sales features to obtain multi-dimensional fusion features of ticket sales.

[0027] Preferably, when adjusting the sales strategies in the sales strategy action space, re-predict the future sales volumes of different cabin classes through the ticket sales trend analysis module, and update the economic state space;

[0028] Among them, the economic state space includes the future sales volume of the economy class, the future sales volume of the business class, the future sales volume of the first class, the oil price, and the flight distance; the sales strategy action space includes ticket price adjustment, flight frequency adjustment, and aircraft type adjustment.

[0029] Preferably, the specific implementation process of calculating the adjusted ticket profit index, updating the sales strategy value function, and determining the optimal sales strategy includes:

[0030] After adjusting the sales strategy, calculate the ticket profit index; update the sales strategy value function through a reinforcement learning algorithm to obtain the optimal sales strategy; the specific formula for updating the sales strategy value function is:

[0031]

[0032] R u = es u · ep u + bs u · bp u + fs u · fp u -(es yc · ep ys + bs yc · bp ys + fs yc · fp ys ) + l u · Δop;

[0033] Where Q(S u , A u ) represents the value when selecting the sales strategy action A u under the current economic state S u , and Q() is the sales strategy value function; γ represents the learning rate and 0 < γ < 1; R u represents the immediate reward when selecting the sales strategy action A u under the current economic state S u , that is, the ticket profit index; δ represents the discount factor and 0 < δ < 1; represents the maximum value of the value under the next economic state S u+1 , represents the sales strategy action with the maximum value under the next economic state S u+1 ; es u , bs u and fs u respectively represent the future sales volume of economy class, business class and first class when selecting the sales strategy action A u ; ep u , bp u and fp u respectively represent the ticket prices of economy class, business class and first class when selecting the sales strategy action A u ; es yc , bs yc and fs yc respectively represent the future sales volume of economy class, business class and first class without adjusting the sales strategy; ep ys , bpys and fp ys are the economy class ticket price, business class ticket price, and first class ticket price respectively when no sales strategy adjustment is made; l u represents the current economic state S u the flight distance under; Δop represents the current economic state S u the difference in oil price under and the oil price in the previous economic state.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] 1. The ticket-related data processing module proposed by the present invention, in combination with the sales volume - ticket attribute regression model, sales volume - operation attribute regression model, and passenger cabin probability index, improves the accuracy of ticket sales trend prediction. By conducting detailed historical data analysis on different cabin classes (economy class, business class, and first class), key ticket attributes and operation attributes are identified, and accurate classification and prediction are carried out based on the ticket purchase frequency, purpose of passengers, and influence factors of cabin classes. This module can accurately reflect the sales trends of each cabin class under different conditions, enhancing the prediction ability of ticket sales data. Through the integration of data sets, more reliable data support can be provided for subsequent flight regulation and ticket pricing.

[0036] 2. The sales volume prediction model proposed by the present invention can effectively improve the accuracy of ticket sales trend prediction through multi-dimensional feature analysis and feature weighted fusion. By using the cross-attention mechanism to analyze and fuse different flight operation features, aircraft operation features, and passenger features, the potential laws of passenger behavior and preferences can be deeply explored, and the mutual relationship between different flights and passengers can be accurately captured; combined with the LSTM network to further analyze the deep impact of passenger preferences on ticket sales, multi-dimensional information fusion and trend prediction are achieved, improving the accuracy of sales volume prediction.

[0037] 3. The present invention optimizes the sales strategy by combining the economic state space, sales strategy action space, and sales strategy value function, and combines the reinforcement learning algorithm to effectively adjust subsequent flights and ticket pricing. Based on the accurate prediction of future sales volume by the ticket sales trend analysis module, the economic state space can be updated in real time to ensure that the sales volume of different cabin classes is consistent with market demand, avoiding errors caused by market fluctuations. By optimizing the sales strategy and maximizing the ticket profit index, not only the operation efficiency and revenue of flights are improved, but also the risk of customer loss is effectively reduced, and economic losses are decreased. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a system structure diagram of a deep learning analysis system for ticket sales trends provided by an embodiment of the present invention application;

[0039] Figure 2 It is a flowchart of the steps of a deep learning analysis system for air ticket sales trends provided by an embodiment of the present invention application;

[0040] Figure 3 It is a schematic structural diagram of a sales volume prediction model provided by an embodiment of the present invention application. Detailed implementation manners

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Airline A introduced a deep learning analysis system for air ticket sales trends to effectively manage and adjust its air ticket sales strategies and analyze the air ticket sales trends of different cabin classes. To illustrate the effectiveness of the deep learning analysis system for air ticket sales trends proposed by the present invention, it will be specifically described in conjunction with the accompanying drawings of this embodiment and the following two embodiments.

[0043] Embodiment 1

[0044] An embodiment of the present application discloses a deep learning analysis system for air ticket sales trends, which is used to analyze the air ticket sales trends of Airline A so as to optimize the sales strategy. Refer to Figure 1 and Figure 2 , which are respectively the system structure diagram and the flowchart of the steps of a deep learning analysis system for air ticket sales trends; the specific implementation steps of the system include: S1. Collect air ticket sales data, operation data, and passenger data within a historical time, and perform data annotation and data cleaning; S2. Construct a sales volume-air ticket attribute regression model and a sales volume-operation attribute regression model, and determine the key air ticket attributes and key operation attributes of different cabin classes; divide the historical passenger data through the passenger cabin probability index to obtain the historical sales data sets of different cabin classes; S3. Construct a sales volume prediction model and predict the future sales volumes of different cabin classes; S4. Set the economic state space, the sales strategy action space, and the sales strategy value function, and adjust the sales strategy in combination with the economic state space; S5. Calculate the adjusted air ticket profit index, optimize the sales strategy through a reinforcement learning algorithm, and determine the optimal sales strategy.

[0045] Further, the air ticket-related data collection module collects the air ticket sales data, operation data, and passenger data of Airline A within a historical time period, and performs data annotation and data cleaning to obtain historical air ticket sales data, historical operation data, and historical passenger data, corresponding to the above S1 step; the air ticket sales data includes the departure time, arrival time, flight duration, date, air ticket price, and sales volume of each flight; the operation data includes the aircraft type, failure rate, and cabin service quality; the passenger data includes the cabin class, ticket purchase frequency, and ticket purchase purpose; the process of data annotation is to annotate the corresponding holidays according to the date of each flight, and replace the date attribute in the air ticket sales data with the holidays.

[0046] Tables 1, 2, and 3 list some examples of economy class air ticket sales data, operation data, and passenger data.

[0047] Table 1. Some examples of economy class air ticket sales data

[0048]

[0049] Table 2. Some examples of operation data

[0050] Flight Number Aircraft Type Failure Rate (%) Cabin Service Quality AF2288 Airbus A320 0.05 8.5 AF2264 Boeing 737-800 0.02 9 AF2248 Airbus A321 0.1 7.5 AF2276 Boeing 737 0.03 8

[0051] Table 3. Some examples of passenger data

[0052] Flight Number Cabin Class Ticket Purchase Frequency (times / year) Purpose of Ticket Purchase AF2288 Economy Class 10 Business Trip AF2264 Business Class 5 Travel AF2248 Business Class 8 Business Trip AF2276 Economy Class 6 Visiting Relatives

[0053] Through the collection and processing of the historical air ticket sales data, operation data, and passenger data of Airline A, a solid data foundation can be provided for subsequent demand forecasting and optimization decisions. The process of data cleaning and annotation not only eliminates the noise in the original data, but also introduces a new dimension to the air ticket sales data through the annotation of holidays. Combining the air ticket sales volume, operation data, and passenger behavior data can reveal the potential laws of flights, such as high-demand periods, aircraft type suitability, and the impact of service quality on passenger choices, thus providing data support for subsequent air ticket sales analysis.

[0054] Further, corresponding to the above S2 step, the specific implementation process of the air ticket-related data processing module includes:

[0055] Divide the sales volume in the historical air ticket sales data according to the cabin class to obtain the historical sales volume of the economy class, the historical sales volume of the business class, and the historical sales volume of the first class;

[0056] For historical ticket sales data, calculate the ticket attribute correlation degree between different ticket attributes through the Pearson correlation coefficient; construct a sales volume - ticket attribute regression model, and determine the influence degree of different ticket attributes on the historical sales volume of different cabin classes through fitting with historical data, obtaining the economy class ticket attribute influence degree, business class ticket attribute influence degree, and first class ticket attribute influence degree; the specific formula of the sales volume - ticket attribute regression model is:

[0057] AS j =κ j,1 ·T dep +κ j,2 ·T land +κ j,3 ·D flig +κ j,4 ·H+κ j,5 ·AT j +ε j ;

[0058] Among them, AS j represents the ticket sales volume of cabin class j, and cabin class j is economy class or business class or first class; T dep represents the departure time of the flight; κ j,1 represents the influence factor of the departure time on the ticket sales volume of cabin class j, that is, the ticket attribute influence degree of the departure time for cabin class j; T land represents the arrival time of the flight; κ j,2 represents the influence factor of the arrival time on the ticket sales volume of cabin class j; D flig represents the flight duration of the flight; κ j,3 represents the influence factor of the flight duration on the ticket sales volume of cabin class j; H represents the holiday corresponding to the departure date of the flight; κ j,4 represents the influence factor of the holiday on the ticket sales volume of cabin class j; AT j represents the ticket price of cabin class j in the flight; κ j,5 represents the influence factor of the ticket price on the ticket sales volume of cabin class j; ε j represents the residual term of the ticket attribute influence degree of cabin class j;

[0059] Combined with the ticket attribute correlation degree, screen the ticket attributes corresponding to different cabin classes respectively to determine the key ticket attributes for economy class, business class, and first class; if the ticket attribute correlation degree of two ticket attributes is greater than the predetermined correlation degree threshold, then compare the ticket attribute influence degrees corresponding to the two ticket attributes, and delete the ticket attribute with the smaller ticket attribute influence degree;

[0060] For historical operation data, construct a sales volume - operation attribute regression model. Through historical data fitting and operation attribute screening operations, determine the key operation attributes for economy class, business class, and first class; the specific formula of the sales volume - operation attribute regression model is:

[0061] AS j =ε j,1 ·LX airc +ε j,2 ·FR airc +ε j,3 ·CS+η j ;

[0062] Among them, LX airc represents the aircraft type of the flight - corresponding aircraft; ε j,1 represents the influence factor of the aircraft type on the ticket sales volume of cabin class j; FR airc represents the failure rate of the flight - corresponding aircraft; ε j,2 represents the influence factor of the failure rate on the ticket sales volume of cabin class j; CS represents the cabin service quality of the flight - corresponding aircraft; ε j,3 represents the influence factor of the cabin service quality on the ticket sales volume of cabin class j; η j represents the residual term of the influence degree of operation attributes on the ticket sales volume of cabin class j;

[0063] The operation attribute screening operation is to analyze the influence factor of operation attributes on the ticket sales volume of cabin class j; if the corresponding influence factor is less than the predetermined influence degree threshold, delete this operation attribute;

[0064] For historical passenger data, calculate the passenger cabin probability index according to the cabin class, ticket - buying frequency, and ticket - buying purpose. The specific calculation formula is:

[0065]

[0066] Among them, PC i represents the passenger cabin probability index of passenger i; represents the value of cabin class c when the number of occurrences of cabin class sequence C i is the highest, f() is the frequency function; exp() represents the exponential function with the natural constant as the base; β represents the exponential weight of the average ticket - buying frequency of ticket - buying frequency sequence F i and is set according to expert experience; p represents the specific coding value of the ticket - buying purpose in ticket - buying purpose sequence P ; α i represents the influence factor of passenger i on cabin class sequence C 1 , α i represents the influence factor of passenger i on ticket - buying frequency sequence F 2 ​i influence factor, α 3 represents the ticket - buying purpose sequence P of passenger i i influence factor, 0 < α 2 < α 3 < α 1 < 1 and α 1 + α 2 + α 3 = 1;

[0067] According to the passenger cabin probability index, historical passenger data is divided into economy - class passenger historical data, business - class passenger historical data, and first - class passenger historical data; specifically, if the passenger cabin probability index belongs to the economy - class coding range, the historical passenger data is divided into economy - class passenger historical data; if the passenger cabin probability index belongs to the business - class coding range, the historical passenger data is divided into business - class passenger historical data; if the passenger cabin probability index belongs to the first - class coding range, the historical passenger data is divided into first - class passenger historical data;

[0068] Integrate the data of historical passenger data, key ticket attributes, and key operation attributes corresponding to different cabin classes respectively, to obtain an economy - class historical sales data set, a business - class historical sales data set, and a first - class historical sales data set.

[0069] By making a detailed division of historical ticket sales data and combining the regression models of ticket attributes and operation attributes, the impact of various factors on sales volume can be accurately analyzed. Using the passenger cabin probability index, passenger groups can be segmented, and the needs of passengers in different cabin classes can be deeply understood. This process not only improves the data reliability of subsequent sales forecasts but also enhances the airline's response ability to market dynamics.

[0070] Furthermore, corresponding to Figure 2 step S3, the ticket sales trend analysis module constructs a sales volume prediction model, trains and optimizes the model on the historical sales data sets of different cabin classes, obtains a trained sales volume prediction model, and predicts the future sales volume of different cabin classes;

[0071] Among them, the sales volume prediction model includes a multi - dimensional feature extraction layer, a feature analysis and fusion layer, and a sales trend prediction layer. Refer to Figure 3 , which is a structural schematic diagram of the sales volume prediction model;

[0072] The multi - dimensional feature extraction layer is used to extract ticket sales features, flight operation features, aircraft operation features, and passenger features from the historical sales data sets of different cabin classes;

[0073] The feature analysis and fusion layer is used to analyze the influence of flight operation features, aircraft operation features, and passenger features on the ticket sales features of different cabin classes, and perform feature weighted fusion;

[0074] A sales trend prediction layer for predicting future sales volumes of different cabin classes.

[0075] Specifically, the multi-dimensional feature extraction layer uses a convolutional neural network to extract different features from the historical sales datasets of different cabin classes; the sales trend prediction layer captures the temporal information of the output features of the feature analysis and fusion layer through an RNN network; according to the fully connected layer and the prediction head, the future sales volumes of different cabin classes are predicted; among them, the prediction result of the future sales volume can be the sales volume of a certain day in the future, or the sales volume of each day in a certain period in the future.

[0076] To clearly illustrate the accuracy of the sales volume prediction model proposed by the present invention, Table 4 exemplarily gives the predicted results of the economy class sales volume on different dates.

[0077] Table 4. Partial examples of the predicted results of the economy class sales volume on different dates

[0078] Date Predicted Sales Volume (tickets) Actual Sales Volume (tickets) Prediction Error Rate (%) 2024 / 1 / 15 150 145 3.45 2024 / 2 / 15 180 175 2.86 2024 / 3 / 15 130 120 8.33 2024 / 4 / 15 160 155 3.23 2024 / 5 / 15 146 153 4.58

[0079] By constructing a sales volume prediction model and training and optimizing it on the historical sales datasets of different cabin classes, the future sales volumes of different cabin classes can be accurately predicted. The multi-dimensional feature extraction layer effectively extracts ticket sales features, flight operation features, aircraft operation features, and passenger features, thus comprehensively considering the impacts of various factors on sales volume. The feature analysis and fusion layer performs weighted fusion on different features to ensure that each feature obtains reasonable weights and impacts in the prediction process, making the model have higher accuracy and adaptability. The sales trend prediction layer can perform accurate sales trend prediction according to the optimized model, providing a reliable decision-making basis for subsequent sales strategy regulation, thereby improving sales efficiency and reducing the empty seat rate.

[0080] Furthermore, the specific implementation process of feature analysis and feature weighted fusion through the feature analysis and fusion layer includes:

[0081] Input the flight operation features and passenger features into the cross-attention mechanism to capture the potential impacts of different flights on passengers' ticket-buying behaviors and obtain the passenger-flight preference weights; input the aircraft operation features and passenger features into the cross-attention mechanism to analyze the preferences of different passengers for aircraft operation conditions and obtain the passenger-aircraft preference weights; combine the passenger-flight preference weights and the passenger-aircraft preference weights, and perform weighted fusion on the passenger features, aircraft operation features, and flight operation features to obtain the passenger preference features; through the LSTM network, mine the impacts of the passenger preference features on the ticket sales features to obtain the multi-dimensional fusion features of ticket sales.

[0082] Through the combination of the cross-attention mechanism and the LSTM network, the embodiments of the present application can capture the personalized needs and preferences of passengers more comprehensively, which helps to deeply analyze the potential impact of different flight and aircraft operation conditions on passengers' purchasing behaviors. By weighted fusion of passenger characteristics, flight operation characteristics, and aircraft operation characteristics, the multi-dimensional preferences of passengers can be better described, improving the prediction accuracy.

[0083] Further, the air ticket sales strategy regulation module sets an economic state space, a sales strategy action space, and a sales strategy value function, and adjusts the sales strategy in combination with the economic state space, corresponding to Figure 2 step S4; when adjusting the sales strategy in the sales strategy action space, the future sales volumes of different cabin classes are re-predicted through the air ticket sales trend analysis module to update the economic state space;

[0084] Among them, the economic state space includes the future sales volume of economy class, the future sales volume of business class, the future sales volume of first class, oil price, and flight distance; the sales strategy action space includes air ticket fare adjustment, flight frequency adjustment, and aircraft type adjustment.

[0085] Specifically, first, the agent observes the current economic state, that is, factors such as the future sales volumes of different cabins and oil price. Then, based on the current state, the agent selects an appropriate action from the sales strategy action space, such as adjusting the air ticket fare or flight frequency, etc.; after selecting this action, the agent predicts the future sales volumes of each cabin through the air ticket sales trend analysis module, thereby updating the economic state space. Then, according to the sales strategy value function, the agent calculates the immediate reward obtained by taking this action, that is, the profit value calculated according to the current economic state and the taken sales strategy action. This reward value is used to guide the agent's behavior selection, thereby optimizing the sales strategy. In this process, the agent uses the reinforcement learning algorithm to gradually adjust its strategy by combining the method of trying different sales strategies and selecting the currently known optimal strategy, so as to maximize the cumulative air ticket profit index.

[0086] By setting the economic state space, the sales strategy action space, and the sales strategy value function, the dynamic optimization and fine-tuning of the air ticket sales strategy are realized. Through the analysis of the economic state space, the module can grasp the sales trends of different cabins in real time and adjust the sales strategy in a timely manner according to market changes, thereby enhancing the profitability and resource utilization efficiency of the flight. In addition, the inclusion of factors such as oil price and flight distance makes the sales strategy more in line with the actual operation conditions, thereby effectively reducing costs and increasing revenues.

[0087] Further, calculate the adjusted air ticket profit index, update the sales strategy value function through the reinforcement learning algorithm, and determine the optimal sales strategy; corresponding to Figure 2In step S5, the specific implementation process includes:

[0088] After adjusting the sales strategy, the computer ticket profit index; through the reinforcement learning algorithm, update the sales strategy value function to obtain the optimal sales strategy; the specific formula for updating the sales strategy value function is:

[0089]

[0090] R u = es u · ep u + bs u · bp u + fs u · fp u -(es yc · ep ys + bs yc · bp ys + fs yc · fp ys ) + l u · Δop;

[0091] Where Q(S u , A u ) represents the value when selecting the sales strategy action A u in the current economic state S u , and Q() is the sales strategy value function; γ represents the learning rate and 0 < γ < 1; R u represents the immediate reward when selecting the sales strategy action A u in the current economic state S u , that is, the computer ticket profit index; δ represents the discount factor and 0 < δ < 1; represents the maximum value of the value in the next economic state S u+1 , represents the sales strategy action with the maximum value in the next economic state S u+1 ; es u , bs u and fs u respectively represent the future sales volume of economy class, business class, and first class when selecting the sales strategy action A u ; ep u , bp u and fp u respectively represent the ticket price of economy class, business class, and first class when selecting the sales strategy action A u ; es yc , bs yc and fs ycrespectively represent the future sales volume of economy class, business class, and first class without sales strategy adjustment; ep ys 、bp ys and fp ys respectively represent the ticket prices of economy class, business class, and first class without sales strategy adjustment; l u represents the flight distance under the current economic state S u ; Δop represents the difference between the oil price under the current economic state S u and the oil price under the previous economic state.

[0092] By optimizing the sales strategy through the reinforcement learning algorithm, the maximization of the ticket profit index can be achieved, thereby enhancing the profitability. According to multiple factors such as the current economic state, market demand, and oil price, the sales strategy is dynamically adjusted to ensure the maximization of the sales profit of tickets in different economic environments. By combining information such as immediate rewards, discount factors, and future sales volume, the sales strategies of economy class, business class, and first class can be precisely adjusted to achieve precise pricing and cabin allocation, improving the overall operational efficiency.

[0093] Example 2

[0094] In this embodiment of the application, a deep learning analysis system for ticket sales trends is applied to the flight control system of A Airlines to analyze the historical ticket sales data to effectively control flight schedules and ticket prices.

[0095] Specifically, the ticket-related data collection module collects the ticket sales data, operation data, and passenger data of A Airlines over a historical period, and performs data annotation and data cleaning to obtain historical ticket sales data, historical operation data, and historical passenger data.

[0096] Furthermore, the ticket-related data processing module obtains the ticket attribute correlation based on the ticket attributes of the historical ticket sales data and the historical sales data of different cabin classes, and combines the constructed sales volume - ticket attribute regression model to determine the key ticket attributes of different cabin classes; by constructing a sales volume - operation attribute regression model and operation attribute screening, the key operation attributes of different cabin classes are determined; the passenger cabin probability index of different passengers is calculated, the historical passenger data of different cabin classes is divided, and combined with the data of the corresponding key ticket attributes and key operation attributes, the historical sales data sets of different cabin classes are obtained.

[0097] Furthermore, the ticket sales trend analysis module constructs a sales volume prediction model, trains and optimizes the model on the historical sales data sets of different cabin classes to obtain a trained sales volume prediction model, and predicts the future sales volume of different cabin classes.

[0098] Furthermore, the air ticket sales strategy regulation module sets an economic state space, a sales strategy action space, and a sales strategy value function, adjusts the sales strategy in combination with the economic state space, calculates the adjusted air ticket profit index, updates the sales strategy value function through a reinforcement learning algorithm, and determines the optimal sales strategy.

[0099] Specifically, when adjusting the sales strategy in the sales strategy action space, the future sales volume of different cabin classes is re-predicted through the air ticket sales trend analysis module, and the economic state space is updated.

[0100] Among them, the economic state space includes the future sales volume of economy class, business class, first class, oil price, and flight distance; the sales strategy action space includes air ticket fare adjustment, flight frequency adjustment, and aircraft type adjustment.

[0101] Furthermore, the specific implementation process of calculating the adjusted air ticket profit index, updating the sales strategy value function, and determining the optimal sales strategy includes:

[0102] After adjusting the sales strategy, calculate the air ticket profit index; update the sales strategy value function through a reinforcement learning algorithm to obtain the optimal sales strategy; the specific formula for updating the sales strategy value function is:

[0103]

[0104] R u =es u ·ep u +bs u ·bp u +fs u ·fp u -(es yc ·ep ys +bs yc ·bp ys +fs yc ·fp ys )+l u ·Δop;

[0105] Among them, Q(S u ,A u ) represents the value when selecting the sales strategy action A u in the current economic state S u , Q() is the sales strategy value function; γ represents the learning rate and 0 < γ < 1; R u represents the immediate reward for selecting the sales strategy action A u in the current economic state S u , that is, the air ticket profit index; δ represents the discount factor and 0 < δ < 1; represents the next economic state Su+1 The maximum value of the next value Indicates the next economic state S u+1 The sales strategy action when the next value is the largest; es u , bs u and fs u Respectively represent the future sales volume of economy class, business class and first class when selecting the sales strategy action A u ; ep u , bp u and fp u Respectively represent the future sales volume of economy class, business class and first class when selecting the sales strategy action A u ; es yc , bs yc and fs yc Respectively represent the future sales volume of economy class, business class and first class when no sales strategy adjustment is made; ep ys , bp ys and fp ys Respectively represent the ticket prices of economy class, business class and first class when no sales strategy adjustment is made; l u Indicates the flight distance under the current economic state S u ; Δop represents the difference between the oil price under the current economic state S u and the oil price in the previous economic state.

[0106] Through in-depth analysis of historical ticket sales data, operation data and passenger data, the embodiments of the present application can accurately identify the key ticket attributes and operation attributes of different cabin classes, and effectively predict the sales trend by using deep learning and regression models. Among them, combining deep learning algorithms and multi-dimensional feature extraction can accurately predict the future sales volume of different cabin classes, and extract key ticket attributes, operation attributes and passenger behavior patterns based on historical data to provide accurate sales predictions for airlines. At the same time, using reinforcement learning algorithms to dynamically adjust sales strategies to maximize the ticket profit index, and optimize resource allocation and improve operation efficiency by adjusting ticket prices, flight frequencies and aircraft types.

[0107] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A deep learning analysis system for air ticket sales trends, characterized in that: include: The air ticket related data collection module collects the airline's historical air ticket sales data, operation data and passenger data, and performs data labeling and data cleaning to obtain historical air ticket sales data, historical operation data and historical passenger data; The air ticket related data processing module obtains the air ticket attribute correlation according to the air ticket attributes of the historical air ticket sales data and the historical sales data of different cabin classes, and determines the key air ticket attributes of different cabin classes by combining the sales volume-air ticket attribute regression model; determines the key operational attributes of different cabin classes by building a sales volume-operation attribute regression model and operational attribute screening; calculates the passenger class probability index of different passengers, divides the historical data of passengers of different cabin classes, and combines the data of corresponding key air ticket attributes and key operational attributes to obtain the historical sales data set of different cabin classes; The ticket sales trend analysis module builds a sales volume prediction model, performs model training and optimization on the historical sales data set of different cabin classes, obtains the trained sales volume prediction model, and predicts the future sales volume of different cabin classes; The air ticket sales strategy control module sets the economic state space, the sales strategy action space and the sales strategy value function, and adjusts the sales strategy in combination with the economic state space; calculates the adjusted air ticket profit index, updates the sales strategy value function through the reinforcement learning algorithm, and determines the optimal sales strategy.

2. A deep learning analysis system for air ticket sales trends according to claim 1, characterized in that: The ticket sales data includes the take-off time, landing time, flight duration, date, ticket fare and sales volume of each flight; the operation data includes aircraft type, failure rate and cabin service quality; the passenger data includes cabin class, ticket purchase frequency and ticket purchase purpose; the data marking process is to mark the corresponding holidays according to the date of each flight, and replace the date attributes in the ticket sales data with holidays.

3. The deep learning analysis system for air ticket sales trend according to claim 1, characterized in that: The specific implementation process of the ticket related data processing module includes: Dividing the sales volume in the historical ticket sales data according to cabin class to obtain the historical sales volume of economy class, the historical sales volume of business class and the historical sales volume of first class; For the historical ticket sales data, the ticket attribute correlation between different ticket attributes is calculated by using the Pearson correlation coefficient; the sales volume-ticket attribute regression model is constructed, and the influence of different ticket attributes on the historical sales volume of different cabin classes is determined by fitting historical data, and the influence of economy class ticket attributes, business class ticket attributes and first class ticket attributes are obtained; based on the ticket attribute correlation, the ticket attributes corresponding to different cabin classes are screened respectively to determine the key ticket attributes of economy class, business class and first class; For the historical operation data, construct the sales volume-operation attribute regression model, and determine the key operation attributes of economy class, business class and first class by historical data fitting and operation attribute screening operations; For the historical passenger data, the passenger class probability index is calculated according to the cabin class, ticket purchase frequency and ticket purchase purpose. The specific calculation formula is: Among them, PC i represents the passenger class probability index of passenger i; Indicates cabin class sequence C i The value of cabin class c when it appears the most times, f() is the frequency function; exp() represents the exponential function with a natural constant as the base; β represents the ticket purchase frequency sequence F i Average frequency of ticket purchase The exponential weight of the ticket; p represents the sequence of ticket purchase purposes P i The specific code value of the ticket purchase purpose; α1 represents the cabin class sequence C of passenger i i The influencing factor of α2 is the ticket purchase frequency sequence F of passenger i. i The influence factor of α3 represents the ticket purchase purpose sequence P of passenger i. i The impact factor is 0<α2<α3<α1<1 and α1+α2+α3=1; According to the passenger class probability index, the historical passenger data is divided into economy class passenger historical data, business class passenger historical data and first class passenger historical data; The historical passenger data, key ticket attributes and key operation attribute data corresponding to different cabin classes are integrated to obtain the economy class historical sales data set, business class historical sales data set and first class historical sales data set.

4. The deep learning analysis system for ticket sales trends according to claim 1, characterized in that: The sales volume prediction model includes a multi-dimensional feature extraction layer, a feature analysis fusion layer and a sales trend prediction layer; The multidimensional feature extraction layer is used to extract ticket sales features, flight operation features, aircraft operation features and passenger features from the historical sales data set of different cabin classes; The feature analysis and fusion layer is used to analyze the impact of the flight operation characteristics, the aircraft operation characteristics and the passenger characteristics on the ticket sales characteristics of different cabin classes, and perform feature weighted fusion; The sales trend prediction layer is used to predict the future sales volume of different cabin classes.

5. A deep learning analysis system for air ticket sales trends according to claim 4, characterized in that: The specific implementation process of performing feature analysis and feature weighted fusion through the feature analysis fusion layer includes: The flight operation characteristics and the passenger characteristics are input into the cross-attention mechanism to capture the potential impact of different flights on the passenger's ticket purchasing behavior and obtain the passenger-flight preference weight; the aircraft operation characteristics and the passenger characteristics are input into the cross-attention mechanism to analyze the preferences of different passengers for aircraft operation conditions and obtain the passenger-aircraft preference weight; the passenger characteristics, the aircraft operation characteristics and the flight operation characteristics are weightedly fused by combining the passenger-flight preference weight and the passenger-aircraft preference weight to obtain the passenger preference characteristics; through the LSTM network, the influence of the passenger preference characteristics on the ticket sales characteristics is explored to obtain the multi-dimensional fusion characteristics of ticket sales.

6. The deep learning analysis system for ticket sales trends according to claim 1, characterized in that: When adjusting the sales strategy in the sales strategy action space, re-predicting the future sales volume of different cabin classes through the ticket sales trend analysis module, and updating the economic status space; The economic status space includes future sales volume of economy class, future sales volume of business class, future sales volume of first class, oil price and flight distance; the sales strategy action space includes air ticket fare adjustment, flight schedule adjustment and aircraft type adjustment.

7. The deep learning analysis system for ticket sales trends according to claim 1, characterized in that: The specific implementation process of calculating the adjusted ticket profit index, updating the sales strategy value function, and determining the optimal sales strategy includes: After adjusting the sales strategy, the ticket profit index is calculated; the sales strategy value function is updated through the reinforcement learning algorithm to obtain the optimal sales strategy; the specific formula for updating the sales strategy value function is: R u =es u ·ep u +bs u ·bp u +fs u ·fp u -(es yc ·ep ys +bs yc ·bp ys +fs yc ·fp ys )+l u ·Δop; Among them, Q(S u ,A u ) represents the current economic status S u Next, select Sales Strategy Action A u The value at that time, Q() is the sales strategy value function; γ represents the learning rate and 0<γ<1; R u Represents the current economic status S u Next, select Sales Strategy Action A u The instant reward is the ticket profit index; δ represents the discount factor and 0<δ<1; Indicates the next economic state S u+1 The maximum value of the next value, Indicates the next economic state S u+1 The sales strategy action when the value is the largest; u ,bs u and fs u Respectively represent the selection of sales strategy action A u The future sales volume of economy class, business class and first class at that time; u , bp u and fp u Respectively represent the selection of sales strategy action A u Economy class airfare, business class airfare and first class airfare at that time; yc ,bs yc and fs yc They represent the future sales volume of economy class, business class and first class when no sales strategy adjustment is made; ep ys , bp ys and fp ys Economy class airfares, business class airfares and first class airfares before sales strategy adjustment; u Represents the current economic status S u The flight distance under the condition; Δop represents the current economic status S u The difference between the oil price under the current economic state and the oil price under the previous economic state.

Images