Method and system for dynamically generating air travel price

Abstract

The invention, which relates to the technical field of air transportation and income management, discloses a dynamic generation method and system for an air travel price, and the system comprises a user behavior fine-grained tracking module, a dynamic pricing decision engine module, a cross-channel cooperative control module, and a compliance auditing and feedback module. Through a real-time data stream fusion technology, multivariate signals such as competition dynamic signals, user behavior signals and external environment signals are integrated, and second-level strategy response is realized in combination with the online training capability of a reinforcement learning model; through dynamic state space modeling, variables such as demand popularity, user sensitivity and external risk are coded into six-dimensional vectors, and the limitation of a fixed formula is broken through in combination with the nonlinear mapping capability of a deep Q network; through three measures of dynamic modeling, elastic constraint and cross-chain cooperation, the problems of response lag, high compliance risk and split user experience of a traditional pricing technology are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of air transportation and revenue management, and in particular to a method and system for dynamically generating air travel fares. Background Art

[0002] Demand in the air travel market is influenced by numerous factors, including season, holidays, and destination popularity. Travel demand is high during peak tourist seasons and holidays, while demand is relatively low during off-seasons or for less popular destinations. With the continuous advancement of information technology, data processing capabilities have greatly improved. Airlines are able to collect extensive user data, such as passengers' age, gender, location, and historical spending history, as well as market data such as competitor ticket availability and flight inventory. Leveraging this data, combined with advanced data processing and analysis techniques, more accurate pricing models can be constructed, enabling the dynamic generation of air travel fares.

[0003] The aviation industry is highly competitive, and airlines need to adopt flexible pricing strategies to attract passengers and increase market share. Dynamic pricing allows airlines to set more competitive prices based on their operating costs, market demand, and competitive dynamics in different market conditions, thereby gaining an advantage in the market.

[0004] However, there are still problems that need to be solved: the existing systems mostly use segmented pricing and lack fine-grained analysis of individual user behavior; cross-channel price consistency is relatively poor, and the price difference between the official website and third-party platforms has caused user complaints. Summary of the Invention

[0005] In order to solve the above technical problems, a method and system for dynamically generating air travel fares are provided. This technical solution solves the problems of insufficient fusion of multi-source data and simplification of post-ripening characteristic modeling.

[0006] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0007] A system for dynamically generating air travel fares, comprising:

[0008] Fine-grained user behavior tracking module: This module uses tracking technology to capture user interactions on the official website, mobile applications, and third-party OTA platforms in real time, including search frequency, page jump paths, historical order cancellation rates, and payment method preferences. Based on graph neural networks, it models user behavior data, generates dynamic tags, and correlates user cross-device behavior.

[0009] Dynamic Pricing Decision Engine Module: A reinforcement learning model unit whose input layer integrates real-time market data and user behavior feature vectors to output base fares and confidence intervals through a deep Q-network. A flexible constraint unit with a built-in Civil Aviation Administration pricing rule base and airline cost models limits fare fluctuations.

[0010] Cross-channel collaborative control module: Deployed on alliance chain nodes on the official website and third-party platforms, it synchronizes freight rate adjustment instructions in real time and generates an unalterable pricing log. Using a real-time streaming computing engine, it compares cross-channel prices. When the price difference exceeds a preset threshold, it triggers an automatic calibration mechanism and issues price difference compensation to users.

[0011] Compliance audit and feedback module: Antitrust rule detector, based on the knowledge graph, identifies whether the freight rate strategy constitutes price monopoly; extracts keywords from complaint content through natural language processing, generates optimization suggestions and feeds them back to the decision engine.

[0012] Preferably, the cross-channel collaborative control module specifically includes:

[0013] Blockchain smart contract unit: Consortium chain network, a permissioned chain jointly maintained by airlines, third-party platforms and regulators, adopts the PBFT consensus algorithm; smart contract design, pricing synchronization contract, when the official website or third-party platform initiates a fare adjustment, triggers multi-node verification, and after passing, it is written into the blockchain ledger and broadcast to the entire network; log evidence contract, records the operator IP address, timestamp, original price and new price of each price adjustment, and generates a hash fingerprint for audit call.

[0014] Preferably, the cross-channel collaborative control module specifically includes:

[0015] Price difference monitoring unit: A distributed stream processing cluster built on Apache Flink supports processing millions of price events per second. The unit also sets a price difference calculation model and dynamic thresholds, adjusting them based on route popularity and time periods. Furthermore, it performs multi-dimensional comparisons, verifying the consistency of additional services in addition to price values.

[0016] Automatic calibration and compensation unit: If the price difference continues to exceed the threshold, the smart contract verification is initiated and a calibration instruction is sent to the high-price node; the execution strategy includes active price adjustment, which prioritizes adjusting the price of the third-party platform to be consistent with the official website; passive compensation, if the third-party platform refuses to adjust the price, the official website automatically generates a price difference coupon; the compensation strategy is based on user classification, and the compensation voucher is instantly pushed to the user account via the HTTPS encrypted interface, supporting automatic verification.

[0017] Preferably, the dynamic pricing decision engine module specifically includes:

[0018] Reinforcement Learning Model Unit:

[0019] The input layer data fusion sub-unit includes real-time market data such as competitive prices, remaining seats, and external environmental indicators; user behavior feature vectors include short-term behavior and long-term profiles; and uses Min-Max normalization and one-hot encoding.

[0020] The six-dimensional vector of the deep Q network subunit in the state space includes demand heat, average bidding price, remaining seat rate, user sensitivity, time window, and external risk level.

[0021] Preferably, the dynamic pricing decision engine module specifically includes:

[0022] Flexible constraint unit: The Civil Aviation Administration of China's pricing rule base sets hard boundaries for minimum and maximum fares; a dynamic rule engine supports XML-configured rules and monitors policy changes in real time; sets an airline cost model, with cost components including fuel costs, crew manpower, airport landing fees, maintenance and depreciation, and others; cost forecasting uses the ARIMA model to predict short-term cost fluctuations; a constraint execution mechanism provides real-time interception, triggering the rule engine to force corrections when the DQN output price exceeds the constraint boundary; and early warning feedback pushes price adjustment exception reports to operators.

[0023] Preferably, the user behavior fine-grained tracking module specifically includes:

[0024] Omnichannel tracking coverage unit: On the official website and mobile terminals, front-end SDK integration and lightweight tracking code based on JavaScript and React Native capture user micro-behaviors such as clicks, scrolling, and interruptions in form filling. Events are categorized into core events, including search, browsing, favorites, ordering, payment, cancellation, feedback, and sharing. Third-party OTA platforms and API data bridging are used to obtain user behavior logs from third-party platforms through OAuth2.0 authorization.

[0025] Real-time data collection unit: In the collection layer, edge computing nodes are deployed in the CDN to process data nearby and filter out invalid events. In the transport layer, high-concurrency data stream transmission is achieved through the Apache Kafka cluster. In the storage layer, the time series database stores real-time behavior logs, and the graph database stores user relationship networks.

[0026] Behavioral graph construction unit: User nodes are divided into ID, sensitivity level, and device fingerprint; behavioral event nodes are divided into event type, timestamp, and geographic location; external entity nodes are divided into flight number, competing airline, and weather event; dynamic tags are generated, and price sensitivity is calculated based on historical order price elasticity regression analysis; travel demand is predicted through LSTM to obtain the probability of emergency travel; cross-platform price comparison frequency and the final order channel are correlated and analyzed to obtain channel loyalty.

[0027] Preferably, the user behavior fine-grained tracking module specifically includes:

[0028] Cross-device behavioral correlation unit:

[0029] User identity mapping uses a strong authentication system. Account-logged users are directly associated through their UserID. For logged-in users, the system matches device fingerprints and behavioral patterns. Federated learning integration works with mobile phone manufacturers to associate the same user's mobile phone, tablet, and PC behaviors through device-level federated learning.

[0030] Cross-device behavior analysis and typical path mining include flight searches on PC, price comparisons on the app, and mini-program ordering; browsing on OTA platforms, and paying on official websites; device preference strategies: iOS users have a higher average payment premium rate than Android users; and tablet users are more receptive to recommendations for additional services.

[0031] Preferably, the compliance audit and feedback module specifically includes:

[0032] Knowledge graph construction unit: Data source integration, including structured extraction of key constraints based on the regulatory database; annotation of violation patterns and penalty basis based on the historical case database; real-time access to Civil Aviation Administration of China announcements, airline financial reports, and industry research reports based on industry dynamics to identify potential risk signals; graph relationships, including entities such as airlines, routes, pricing strategies, regulatory agencies, and legal provisions;

[0033] Monopolistic behavior detection unit: Real-time monitoring, inputting the fare strategy, competitor price fluctuation data, and market share changes output by the dynamic pricing engine; detection logic, horizontal collaborative analysis, through Granger causality test, to determine whether there is a statistically significant correlation between the price changes of multiple airlines on the same route; vertical abuse detection, identifying whether airlines use their market dominance to implement predatory pricing.

[0034] Preferably, the compliance audit and feedback module specifically includes:

[0035] Complaint Analysis and Feedback Unit: Complaint data processing, multi-source data access, including structured data, customer service ticket classification, and user ratings; unstructured data, complaint text, and social media public opinion; NLP technology stack, pre-trained models, and domain adaptation models based on RoBERTa; keyword extraction, combining TF-IDF and LDA topic models to identify high-frequency issues; optimization suggestion generation, root cause analysis including cross-channel price differences caused by data synchronization delays on third-party platforms, the excessive weighting of short-term profits in reinforcement learning models, which leads to large fluctuations in dynamic pricing, and lack of transparency in the prices of additional services, which leads to hidden costs.

[0036] Furthermore, a method for dynamically generating air travel fares is provided, for implementing a system for dynamically generating air travel fares according to claims 1-9, comprising:

[0037] Use tracking technology to collect real-time user interaction behavior data on official websites, mobile terminals, and third-party OTA platforms, including search frequency, page jump paths, payment interruption rates, and cross-device behavior tracks;

[0038] Build a user behavior graph, use graph neural networks to analyze behavioral correlations, and generate dynamic tags, including price sensitivity level, emergency travel probability, and channel loyalty;

[0039] Integrate real-time market data and user behavior feature vectors, input into the reinforcement learning model to generate basic fares; based on the Civil Aviation Administration of China's pricing rule base and airline cost models, set the fare elasticity boundary and dynamically modify the DQN output results;

[0040] Through the alliance chain smart contract, the official website and third-party platform freight rate instructions are synchronized, and a real-time streaming computing engine is deployed to monitor cross-channel price differences. If the price difference exceeds the dynamic threshold, automatic calibration is triggered and users are compensated.

[0041] Identify the monopoly risks of freight rate strategies based on knowledge graphs and generate compliance correction suggestions; extract user complaint keywords through natural language processing and feed them back to the decision-making model to optimize the reward function weight.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] This invention proposes to achieve second-level policy response by integrating multiple signals such as competitor dynamics, user behavior, and the external environment through real-time data stream fusion technology, combined with the online training capabilities of the reinforcement learning model; through dynamic state space modeling, variables such as demand heat, user sensitivity, and external risks are encoded into six-dimensional vectors, combined with the nonlinear mapping capabilities of the deep Q network to break through the limitations of fixed formulas; through a flexible constraint engine, the Civil Aviation Administration's rigid rules, airline cost models, and dynamic policy changes are integrated into the reinforcement learning training process. When the model output price approaches the regulatory upper limit, the constraint unit intervenes in advance to make corrections, and feedback is fed back to the reward function to optimize the long-term strategy and avoid the risk of subsequent penalties.

[0044] Through consortium chain smart contracts, prices are synchronized across the entire network. Combined with a streaming computing engine, it monitors inter-channel discrepancies in real time. When OTA platforms experience price delays due to cache delays, the system automatically triggers a calibration command to ensure that the user's price comparison result is consistent with the final payment, significantly reducing disputes. A multi-objective reward function factors user retention rate and complaint keyword analysis into decision-making weights. For example, for users who frequently compare prices but don't place an order, the system automatically generates appropriate discount coupons, stimulating conversions while avoiding excessive profit concessions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is an internal framework diagram of a system for dynamically generating air travel fares;

[0046] Figure 2 This is the internal framework diagram of the cross-channel collaborative control module;

[0047] Figure 3 The present invention is a flowchart of a method for dynamically generating air travel fares. DETAILED DESCRIPTION

[0048] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0049] Reference Figure 1 As shown, a system for dynamically generating air travel fares includes:

[0050] Fine-grained user behavior tracking module: This module uses tracking technology to capture user interactions on the official website, mobile applications, and third-party OTA platforms in real time, including search frequency, page jump paths, historical order cancellation rates, and payment method preferences. Based on graph neural networks, it models user behavior data, generates dynamic tags, and correlates user cross-device behavior.

[0051] Dynamic Pricing Decision Engine Module: A reinforcement learning model unit whose input layer integrates real-time market data and user behavior feature vectors to output base fares and confidence intervals through a deep Q-network. A flexible constraint unit with a built-in Civil Aviation Administration pricing rule base and airline cost models limits fare fluctuations.

[0052] Cross-channel collaborative control module: Deployed on alliance chain nodes on the official website and third-party platforms, it synchronizes freight rate adjustment instructions in real time and generates an unalterable pricing log. Using a real-time streaming computing engine, it compares cross-channel prices. When the price difference exceeds a preset threshold, it triggers an automatic calibration mechanism and issues price difference compensation to users.

[0053] Compliance audit and feedback module: Antitrust rule detector, based on the knowledge graph, identifies whether the freight rate strategy constitutes price monopoly; extracts keywords from complaint content through natural language processing, generates optimization suggestions and feeds them back to the decision engine.

[0054] It should be noted that the input layer: user behavior data (micro-decision paths), market data (competitor dynamics), and external environment data (weather, policies) form multi-dimensional inputs;

[0055] Processing layer: Graph neural networks model behavioral correlations, reinforcement learning integrates business goals and compliance constraints, and blockchain and stream computing achieve cross-channel collaboration;

[0056] Output layer: Dynamic freight rate strategy → User terminal reach → Behavior feedback → Compliance audit → Model iteration, forming a closed loop of "perception-decision-execution-optimization".

[0057] The collaboration between blockchain and reinforcement learning allows pricing decision results to be stored on the chain in real time, serving as a trusted source of DQN training data; the complementarity between GNN and knowledge graphs allows the user behavior graph to be linked to the antitrust knowledge graph to predict potential violations.

[0058] Reference Figure 2 As shown, the cross-channel collaborative control module specifically includes:

[0059] Blockchain smart contract unit: A consortium chain network, a permissioned chain jointly maintained by airlines, third-party platforms, and regulators, uses the PBFT consensus algorithm. Smart contract design includes a pricing synchronization contract. When the official website or third-party platform initiates a fare adjustment, multi-node verification is triggered. Once verified, the verification is written to the blockchain ledger and broadcast to the entire network. A log evidence contract records the operator's IP address, timestamp, original price, and new price for each price adjustment, generating a hash fingerprint for audit access.

[0060] Price difference monitoring unit: A distributed stream processing cluster built on Apache Flink supports processing millions of price events per second. The unit also sets a price difference calculation model and dynamic thresholds, adjusting them based on route popularity and time periods. Furthermore, it performs multi-dimensional comparisons, verifying the consistency of additional services in addition to price values.

[0061] Automatic calibration and compensation unit: If the price difference continues to exceed the threshold, the smart contract verification is initiated and a calibration instruction is sent to the high-price node; the execution strategy includes active price adjustment, which prioritizes adjusting the price of the third-party platform to be consistent with the official website; passive compensation, if the third-party platform refuses to adjust the price, the official website automatically generates a price difference coupon; the compensation strategy is based on user classification, and the compensation voucher is instantly pushed to the user account via the HTTPS encrypted interface, supporting automatic verification.

[0062] It should be noted that the consortium chain network design consists of three nodes: airlines (master nodes), third-party platforms (such as Ctrip and Fliggy, verification nodes), and the Civil Aviation Administration of China (regulatory node). The PBFT consensus algorithm is used to achieve second-level verification.

[0063] To ensure data security, inter-node communication is encrypted using the national secret SM9 algorithm, and price adjustment instructions require signature confirmation from at least 3 / 4 nodes (including regulatory nodes).

[0064] Smart contract logic, pricing synchronization contract, the triggering conditions are freight rate adjustment range ≥ 0.5% or cumulative price adjustment ≥ 2 times within the time window, and it is irreversible after being written to the blockchain (block generation time ≤ 1.5 seconds); log evidence contract, hash fingerprint is generated based on the SHA-3 algorithm, and supports real-time auditing by the Civil Aviation Administration.

[0065] Stream computing engine:

[0066] Event processing, based on Apache Flink's CEP (Complex Event Processing) module, identifies cross-channel price linkage patterns (e.g., platform A lowers its price, and platform B follows suit within 5 seconds);

[0067] Dynamic window, divided into monitoring granularity according to route popularity, 1-second window for popular routes and 30-second window for unpopular routes;

[0068] Sliding window mean, used to determine the persistence of the price difference:

[0069] Where μΔP is the mean price difference within the sliding window, which is used to measure the average change in the price difference. Its unit is consistent with that of the price difference. ΔP t is the price difference change at the t-th time point; N is the size of the sliding window in seconds, where N=60 represents a 60-second window; t is the index of the time point, which is used to mark each time point in the window.

[0070] Calibration strategy and execution logic, with priority rules, prioritizes the official website. If the price difference between the official website and third-party platforms exceeds the limit, the third-party price will be calibrated first (due to the 12% higher user loyalty premium on the official website). Time is sensitive, within 48 hours before flight departure, to prevent the loss of expiring users.

[0071] Compensation grading mechanism:

[0072]

[0073] Automated process, instant compensation, pushed to user accounts via HTTPS two-way authentication interface (end-to-end encryption, latency ≤ 200ms); blockchain verification and evidence storage, compensation voucher usage records are stored on the chain to prevent duplicate redemption (hash collision rate < 10^-6).

[0074] The dynamic pricing decision engine module specifically includes:

[0075] Reinforcement Learning Model Unit:

[0076] The input layer data fusion sub-unit includes real-time market data such as competitive prices, remaining seats, and external environmental indicators; user behavior feature vectors include short-term behavior and long-term profiles; and uses Min-Max normalization and one-hot encoding.

[0077] The six-dimensional vector of the deep Q network subunit state space includes demand heat, average bidding price, remaining seat rate, user sensitivity, time window, and external risk level;

[0078] Flexible constraint unit: The Civil Aviation Administration of China's pricing rule base sets hard boundaries for minimum and maximum fares; a dynamic rule engine supports XML-configured rules and monitors policy changes in real time; sets an airline cost model, with cost components including fuel costs, crew manpower, airport landing fees, maintenance and depreciation, and others; cost forecasting uses the ARIMA model to predict short-term cost fluctuations; a constraint execution mechanism provides real-time interception, triggering the rule engine to force corrections when the DQN output price exceeds the constraint boundary; and early warning feedback pushes price adjustment exception reports to operators.

[0079] It should be noted that the input layer data fusion subunit:

[0080] Real-time market data source and processing, competitive pricing, crawling data from platforms such as Ctrip and Fliggy through APIs, and cleaning outliers; remaining seats, connecting to the real-time inventory interface of the flight management system, accurate to the cabin level such as economy class and business class; external environmental indicators, typhoon warning level (China Meteorological Administration API), fuel price volatility (ICE Brent crude oil futures real-time data).

[0081] Construction of user behavior feature vectors, short-term behavior (last 24 hours): search frequency (statistics by route), page dwell time (heat map analysis), and price comparison tool usage depth (e.g., number of filter conditions);

[0082] Long-term profile (annual dimension): Total spending stratification (VIP level), order cancellation rate (differentiating between voluntary and involuntary cancellations), cabin preference (users with more than 90% economy class seats are marked as price sensitive);

[0083] Data encoding strategies include Min-Max normalization, which is suitable for numerical data, such as fuel prices normalized to the range [0, 1]. One-hot encoding is suitable for handling discrete variables, such as weather conditions classified into four categories: "sunny / rainy / snowy / typhoon."

[0084] State space definition dimensions

[0085]

[0086]

[0087] Action space design, with price adjustment range: -5% to +8% (0.5% step size), with the optimal range verified based on historical data; dynamic step size adjustment, with the step size expanded to 1% in peak season (accelerated response) and reduced to 0.25% in off-season;

[0088] Civil Aviation Administration of China pricing rule base, hard boundary rules:

[0089] Minimum freight rate: flight cost × 1.15 + fixed taxes (e.g. ￥50 airport construction fee);

[0090] Maximum freight rate: the smaller of the Civil Aviation Administration of China's guidance price × 0.9 and the average competitive price × 1.1;

[0091] Special scenario rules: The maximum price for red-eye flights (00:00-06:00) is reduced by 20%.

[0092] Dynamic rule engine and XML-configured templates support rapid updates, policy change monitoring, and real-time analysis of Civil Aviation Administration of China official website announcements (NLP keyword extraction).

[0093] Airline cost model, cost structure and dynamic factors:

[0094]

[0095] Cost forecast, ARIMA model, autoregressive integrated moving average formula:

[0096] C t =φ1C t-1 +φ2C t-2 +θ1∈ t-1 +θ2∈ t-2 +∈ t

[0097] Where C t is the cost value at time point t, the forecast target, in monetary units; φ1 is the autoregressive coefficient 1, which measures the linear impact of the cost in period t-1 on the cost in period t; φ2 is the autoregressive coefficient 2, which measures the linear impact of the cost in period t-2 on the cost in period t; C t-1 is the cost value at time point t-1, which is used to construct the lag term of the cost in period t; C t-2 is the cost value at time point t-2, which is used to construct the lag term of the cost in period t; θ1 is the moving average coefficient 1, which measures the linear influence of the error term in period t-1 on the cost in period t; θ2 is the moving average coefficient 2, which measures the linear influence of the error term in period t-2 on the cost in period t; ∈t is the white noise error term at time point t, which represents the random fluctuation that cannot be explained by the model, with a mean of 0 and a constant variance; ∈ t-1 is the white noise error term at time point t-1; ∈ t-2 is the white noise error term at time point t-2; parameters: p = 3 (autoregressive order), d = 1 (number of differences), q = 2 (moving average order).

[0098] Constraint execution mechanism, real-time interception logic, DQN output price → rule engine verification → if out of bounds, replace with boundary value; correction record blockchain evidence (including original price, correction reason, operation timestamp);

[0099] The early warning feedback process includes push channels such as corporate WeChat / email / SMS three-level alerts (graded according to the severity of the violation); the report content includes the illegal price value, related flights, and the recommended adjustment range.

[0100] The user behavior fine-grained tracking module specifically includes:

[0101] Omnichannel tracking coverage unit: On the official website and mobile terminals, front-end SDK integration and lightweight tracking code based on JavaScript and React Native capture user micro-behaviors such as clicks, scrolling, and interruptions in form filling. Events are categorized into core events, including search, browsing, favorites, ordering, payment, cancellation, feedback, and sharing. Third-party OTA platforms and API data bridging are used to obtain user behavior logs from third-party platforms through OAuth2.0 authorization.

[0102] Real-time data collection unit: In the collection layer, edge computing nodes are deployed in the CDN to process data nearby and filter out invalid events. In the transport layer, high-concurrency data stream transmission is achieved through the Apache Kafka cluster. In the storage layer, the time series database stores real-time behavior logs, and the graph database stores user relationship networks.

[0103] Behavioral graph construction unit: User nodes are divided into ID, sensitivity level, and device fingerprint; behavioral event nodes are divided into event type, timestamp, and geographic location; external entity nodes are divided into flight number, competing airline, and weather event; dynamic labels are generated, and price sensitivity is calculated based on historical order price elasticity regression analysis; travel demand is predicted using LSTM to obtain the probability of emergency travel; cross-platform price comparison frequency and the final order channel are analyzed to obtain channel loyalty;

[0104] Cross-device behavioral correlation unit:

[0105] User identity mapping uses a strong authentication system. Account-logged users are directly associated through their UserID. For logged-in users, the system matches device fingerprints and behavioral patterns. Federated learning integration works with mobile phone manufacturers to associate the same user's mobile phone, tablet, and PC behaviors through device-level federated learning.

[0106] Cross-device behavior analysis and typical path mining include flight searches on PC, price comparisons on the app, and mini-program ordering; browsing on OTA platforms, and paying on official websites; device preference strategies: iOS users have a higher average payment premium rate than Android users; and tablet users are more receptive to recommendations for additional services.

[0107] It should be noted that the tracking of official website and mobile terminals is refined, and the SDK is lightweight: the JavaScript (Web) and React Native (App) tracking code size is compressed to less than 30KB, and the page loading delay is ≤100ms;

[0108] Micro-behavior capture, click heat map tracking (accurate to pixel-level coordinates); scroll depth analysis (recording the page stay area and duration); form filling interruption detection, such as if the user stays on the payment page for more than 2 minutes without submitting, it is marked as "high churn risk".

[0109] Event classification logic, core event attribute extension, such as associating "search events" with departure, destination, and number of filter conditions; data standardization protocol (following OpenTelemetry specifications and compatible with third-party analysis tools).

[0110] Data bridging with third-party OTA platforms and optimized OAuth2.0 authorization processes: After user authorization, behavior logs, such as Ctrip browsing history and Fliggy price comparison paths, are synchronized hourly, with data latency less than 5 minutes.

[0111] Privacy-enhancing technology, sensitive field encryption, ID number encryption using AES-256, key sharding storage; behavioral log desensitization, such as retaining only the first two segments of the IP address, and geographic location blurring to the city level.

[0112] Three-tier architecture design, collection layer (edge ​​computing), CDN node deployment strategy: 10 edge nodes deployed in first-tier cities, 5 in second- and third-tier cities; invalid event filtering, crawler traffic identification (based on behavioral fingerprints, such as abnormal mouse movement trajectory); high-frequency request interception, the same user >50 clicks within 10 seconds is considered abnormal;

[0113] The transport layer (Kafka cluster) has a peak throughput of 800,000 queries per second (QPS). The partitioning strategy is based on user region hash distribution. Data compression uses the Snappy algorithm, reducing bandwidth usage by 65%.

[0114] The storage layer (dual database collaboration) includes a time series database that stores raw behavior logs and supports time range queries, such as the distribution of users who canceled their orders in the past hour. The graph database builds a user-behavior-entity relationship network and supports three-hop association queries, such as "user A → search → flight X → competing airline Y."

[0115] Behavior event node deepening:

[0116] Timestamp granularity, accurate to milliseconds, supports micro-behavior sequence analysis, such as the time taken for the "search → price comparison → save → abandon" path; geographic location association, combining IP positioning and LBS data, identifies users' resident cities and travel patterns;

[0117] Dynamic updates from external entities, including data from competing airlines, crawling prices from the entire network every hour, and marking promotional labels (such as "limited-time flash sales"); impact of weather events, access to typhoon path forecast data, and warning of the probability of route cancellations.

[0118] Dynamic label generation mechanism:

[0119] Price sensitivity calculation, feature engineering, historical order price variance, discount coupon usage rate, and cross-platform price comparison frequency; model iteration and weekly incremental training to ensure prediction accuracy;

[0120] Emergency travel probability prediction, LSTM parameters: time window = 7 days, hidden layer units = 128, predicting demand for the next 3 days (AUC = 0.93);

[0121] Channel loyalty analysis, funnel model, number of price comparison platforms → final ordering channel, loyalty = 1 / (number of price comparison platforms + 1); strategy optimization, low-loyalty users trigger exclusive channel subsidies.

[0122] Breakthrough in identity mapping technology enhances the verification system:

[0123] Login user: UserID + device fingerprint two-factor verification;

[0124] Not logged in user:

[0125] Behavioral pattern matching: comparing 23 features such as screen resolution, time zone, and language settings;

[0126] Device fingerprint clustering: The DBSCAN algorithm identifies clusters of devices belonging to the same user.

[0127] Federated learning integration, in collaboration with Huawei and Xiaomi, enables device-level model training to correlate mobile phone and tablet behaviors without transmitting raw data.

[0128] Reference Figure 3 As shown, a method for dynamically generating air travel fares includes:

[0129] Use tracking technology to collect real-time user interaction behavior data on official websites, mobile terminals, and third-party OTA platforms, including search frequency, page jump paths, payment interruption rates, and cross-device behavior tracks;

[0130] Build a user behavior graph, use graph neural networks to analyze behavioral correlations, and generate dynamic tags, including price sensitivity level, emergency travel probability, and channel loyalty;

[0131] Integrate real-time market data and user behavior feature vectors, input into the reinforcement learning model to generate basic fares; based on the Civil Aviation Administration of China's pricing rule base and airline cost models, set the fare elasticity boundary and dynamically modify the DQN output results;

[0132] Through the alliance chain smart contract, the official website and third-party platform freight rate instructions are synchronized, and a real-time streaming computing engine is deployed to monitor cross-channel price differences. If the price difference exceeds the dynamic threshold, automatic calibration is triggered and users are compensated.

[0133] Identify the monopoly risks of freight rate strategies based on knowledge graphs and generate compliance correction suggestions; extract user complaint keywords through natural language processing and feed them back to the decision-making model to optimize the reward function weight.

[0134] It should be noted that the full link data drive

[0135] Input layer: User behavior data: tracking points cover official websites, apps, and third-party platforms, collecting micro-behaviors such as search frequency and payment interruption rate; market and cost data: competitive prices, number of remaining seats, and fuel costs.

[0136] Processing layer:

[0137] User profile generation: Graph neural networks mine cross-device behavioral correlations, such as PC search → App ordering; pricing decision optimization, deep Q network integrates multi-source data and outputs price confidence intervals; output and feedback: freight rate strategy → cross-channel synchronization → user behavior feedback → compliance audit → model iteration, forming a closed loop.

[0138] In summary, the advantages of this invention lie in: It focuses on the integration of real-time market signals, integrating competitive behavior, user intent, and external environmental variables to construct a dynamic game scenario. Through the autonomous evolutionary capabilities of the reinforcement learning model, it achieves continuous optimization of pricing strategies, capturing short-term profit windows while mitigating long-term market imbalance risks.

[0139] Through the deep coupling of rules and intelligence, the Civil Aviation Administration's rigid rules are directly internalized as model constraints, intercepting illegal price adjustments in real time. Cost forecasts are dynamically linked with pricing baselines to ensure bottom-line profits. The policy semantic parsing engine automatically loads new regulations to avoid manual delays. With risk prediction and self-healing mechanisms, the knowledge graph scans for monopolistic behavior patterns in real time to prevent risks in advance. Keyword analysis of user complaints reversely optimizes decision-making weights, achieving a closed "strategy-feedback" loop.

[0140] Cross-channel consistency is guaranteed, and blockchain smart contracts achieve price synchronization across the entire network in seconds, eliminating price differences caused by cache delays in traditional solutions; the streaming computing engine calibrates abnormal deviations in real time, automatically compensates user losses, and rebuilds the trust chain; long-term value is deeply cultivated, and the multi-objective reward function balances revenue and user retention, accurately identifies highly sensitive groups and triggers targeted discounts; cross-device behavior correlation technology restores the user's complete decision-making path, supporting the precise matching of "demand forecasting - service recommendation - dynamic pricing".

[0141] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A system for dynamically generating air travel fares, characterized in that: include: Fine-grained user behavior tracking module: This module uses tracking technology to capture user interactions on official websites, mobile applications, and third-party OTA platforms in real time, including search frequency, page jump paths, historical order cancellation rates, and payment method preferences. Based on graph neural networks, user behavior data is modeled, dynamic tags are generated, and user cross-device behavior is associated; Dynamic Pricing Decision Engine Module: A reinforcement learning model unit whose input layer integrates real-time market data and user behavior feature vectors to output base fares and confidence intervals through a deep Q-network. A flexible constraint unit with a built-in Civil Aviation Administration pricing rule base and airline cost models limits fare fluctuations. Cross-channel collaborative control module: Deployed on alliance chain nodes on the official website and third-party platforms, it synchronizes freight rate adjustment instructions in real time and generates an unalterable pricing log. Using a real-time streaming computing engine, it compares cross-channel prices. When the price difference exceeds a preset threshold, it triggers an automatic calibration mechanism and issues price difference compensation to users. Compliance audit and feedback module: Antitrust rule detector, based on knowledge graph, identifies whether freight rate strategies constitute price monopoly; Through natural language processing, keywords in complaint content are extracted, optimization suggestions are generated and fed back to the decision-making engine.

2. The system for dynamically generating air travel fares according to claim 1, characterized in that: The cross-channel collaborative control module specifically includes: Blockchain smart contract unit: Consortium chain network, a permissioned chain jointly maintained by airlines, third-party platforms and regulators, adopts the PBFT consensus algorithm; smart contract design, pricing synchronization contract, when the official website or third-party platform initiates a fare adjustment, triggers multi-node verification, and after passing, it is written into the blockchain ledger and broadcast to the entire network; log evidence contract, records the operator IP address, timestamp, original price and new price of each price adjustment, and generates a hash fingerprint for audit call.

3. The system for dynamically generating air travel fares according to claim 2, characterized in that: The cross-channel collaborative control module specifically includes: Price difference monitoring unit: A distributed stream processing cluster built on Apache Flink supports processing millions of price events per second. The unit also sets a price difference calculation model and dynamic thresholds, adjusting them based on route popularity and time periods. Furthermore, it performs multi-dimensional comparisons, verifying the consistency of additional services in addition to price values. Automatic calibration and compensation unit: If the price difference continues to exceed the threshold, the smart contract verification is initiated and a calibration instruction is sent to the high-price node; the execution strategy includes active price adjustment, which prioritizes adjusting the price of the third-party platform to be consistent with the official website; passive compensation, if the third-party platform refuses to adjust the price, the official website automatically generates a price difference coupon; the compensation strategy is based on user classification, and the compensation voucher is instantly pushed to the user account via the HTTPS encrypted interface, supporting automatic verification.

4. The system for dynamically generating air travel fares according to claim 3, characterized in that: The dynamic pricing decision engine module specifically includes: Reinforcement Learning Model Unit: The input layer data fusion sub-unit includes real-time market data such as competitive prices, remaining seats, and external environmental indicators; user behavior feature vectors include short-term behavior and long-term profiles; and uses Min-Max normalization and one-hot encoding. The six-dimensional vector of the deep Q network subunit in the state space includes demand heat, average bidding price, remaining seat rate, user sensitivity, time window, and external risk level.

5. The system for dynamically generating air travel fares according to claim 4, characterized in that: The dynamic pricing decision engine module specifically includes: Flexible constraint unit: The Civil Aviation Administration of China's pricing rule base sets hard boundaries for minimum and maximum fares; a dynamic rule engine supports XML-configured rules and monitors policy changes in real time; sets an airline cost model, with cost components including fuel costs, crew manpower, airport landing fees, maintenance and depreciation, and others; cost forecasting uses the ARIMA model to predict short-term cost fluctuations; a constraint execution mechanism provides real-time interception, triggering the rule engine to force corrections when the DQN output price exceeds the constraint boundary; and early warning feedback pushes price adjustment exception reports to operators.

6. The system for dynamically generating air travel fares according to claim 5, characterized in that: The user behavior fine-grained tracking module specifically includes: Omnichannel tracking coverage unit: On the official website and mobile terminals, front-end SDK integration and lightweight tracking code based on JavaScript and React Native capture user micro-behaviors such as clicks, scrolling, and interruptions in form filling. Events are categorized into core events, including search, browsing, favorites, ordering, payment, cancellation, feedback, and sharing. Third-party OTA platforms and API data bridging are used to obtain user behavior logs from third-party platforms through OAuth2.0 authorization. Real-time data collection unit: In the collection layer, edge computing nodes are deployed in the CDN to process data nearby and filter out invalid events. In the transport layer, high-concurrency data stream transmission is achieved through the Apache Kafka cluster. In the storage layer, the time series database stores real-time behavior logs, and the graph database stores user relationship networks. Behavioral graph construction unit: User nodes are divided into ID, sensitivity level, and device fingerprint; behavioral event nodes are divided into event type, timestamp, and geographic location; external entity nodes are divided into flight number, competing airline, and weather event; dynamic tags are generated, and price sensitivity is calculated based on historical order price elasticity regression analysis; travel demand is predicted through LSTM to obtain the probability of emergency travel; cross-platform price comparison frequency and the final order channel are correlated and analyzed to obtain channel loyalty.

7. The system for dynamically generating air travel fares according to claim 6, characterized in that: The user behavior fine-grained tracking module specifically includes: Cross-device behavioral correlation unit: User identity mapping uses a strong authentication system. Account-logged users are directly associated through their UserID. For logged-in users, the system matches device fingerprints and behavioral patterns. Federated learning integration works with mobile phone manufacturers to associate the same user's mobile phone, tablet, and PC behaviors through device-level federated learning. Cross-device behavior analysis and typical path mining include flight searches on PC, price comparisons on the app, and mini-program ordering; browsing on OTA platforms, and paying on official websites; device preference strategies: iOS users have a higher average payment premium rate than Android users; and tablet users are more receptive to recommendations for additional services.

8. The system for dynamically generating air travel fares according to claim 7, characterized in that: The compliance audit and feedback module specifically includes: Knowledge graph construction unit: Data source integration, including structured extraction of key constraints based on the regulatory database; annotation of violation patterns and penalty basis based on the historical case database; real-time access to Civil Aviation Administration of China announcements, airline financial reports, and industry research reports based on industry dynamics to identify potential risk signals; graph relationships, including entities such as airlines, routes, pricing strategies, regulatory agencies, and legal provisions; Monopolistic behavior detection unit: Real-time monitoring, inputting the fare strategy, competitor price fluctuation data, and market share changes output by the dynamic pricing engine; detection logic, horizontal collaborative analysis, through Granger causality test, to determine whether there is a statistically significant correlation between the price changes of multiple airlines on the same route; vertical abuse detection, identifying whether airlines use their market dominance to implement predatory pricing.

9. The system for dynamically generating air travel fares according to claim 8, characterized in that: The compliance audit and feedback module specifically includes: Complaint Analysis and Feedback Unit: Complaint data processing, multi-source data access, including structured data, customer service ticket classification, and user ratings; unstructured data, complaint text, and social media public opinion; NLP technology stack, pre-trained models, and domain adaptation models based on RoBERTa; keyword extraction, combining TF-IDF and LDA topic models to identify high-frequency issues; optimization suggestion generation, root cause analysis including cross-channel price differences caused by data synchronization delays on third-party platforms, the excessive weighting of short-term profits in reinforcement learning models, which leads to large fluctuations in dynamic pricing, and lack of transparency in the prices of additional services, which leads to hidden costs.

10. A method for dynamically generating air travel fares, according to a system for dynamically generating air travel fares according to claims 1-9, characterized in that: include: Use tracking technology to collect real-time user interaction behavior data on official websites, mobile terminals, and third-party OTA platforms, including search frequency, page jump paths, payment interruption rates, and cross-device behavior tracks; Build a user behavior graph, use graph neural networks to analyze behavioral correlations, and generate dynamic tags, including price sensitivity level, emergency travel probability, and channel loyalty; Integrate real-time market data and user behavior feature vectors, input into the reinforcement learning model to generate basic fares; based on the Civil Aviation Administration of China's pricing rule base and airline cost models, set the fare elasticity boundary and dynamically modify the DQN output results; Through the alliance chain smart contract, the official website and third-party platform freight rate instructions are synchronized, and a real-time streaming computing engine is deployed to monitor cross-channel price differences. If the price difference exceeds the dynamic threshold, automatic calibration is triggered and users are compensated. Identify the monopoly risk of freight rate strategies based on knowledge graphs and generate compliance correction suggestions; User complaint keywords are extracted through natural language processing and fed back into the decision model to optimize the reward function weight.

Images