Air internet cabin resource dynamic arrangement system based on flow-experience coupling

Abstract

This invention relates to a dynamic cabin resource orchestration system for aviation internet based on traffic-experience coupling, belonging to the field of aviation internet resource scheduling technology. The system includes: a data acquisition and association module, which collects real-time traffic, session status, and passenger identity, seat, and policy group information of device IPs, dynamically binds them, and forms real-time operational data; a multi-dimensional experience evaluation module, which receives this data and calculates a network experience index for IP sessions using a preset evaluation model, comprehensively evaluating connection quality, bandwidth sufficiency, and service compatibility, and classifying experience levels accordingly; a dynamic policy decision-making and generation module, which dynamically generates resource orchestration instructions from a pre-set policy library based on the experience level, combined with global load and flight phase; and a policy execution and feedback module, responsible for issuing and executing instructions and collecting new data to feed back to the front end, forming a closed loop. This system achieves intelligent, dynamic, and on-demand allocation of cabin network resources, improving the passenger internet experience.

Description

Technical Field

[0001] This invention belongs to the field of aviation internet resource scheduling technology, specifically relating to an aviation internet cabin resource dynamic scheduling system based on traffic-experience coupling. Background Technology

[0002] With the digital upgrade of air transport, in-flight internet has become a core competitive advantage for airlines. Passengers' diverse internet needs in the cabin, such as web browsing, video streaming, real-time communication, and large file downloads, are growing rapidly, and their demands for differentiated and stable network experiences are continuously increasing. Currently, major airlines have launched tiered service packages such as Business Class Premium and Economy Class Standard, but existing cabin network resource scheduling technologies still have many limitations and are difficult to adapt to the complex needs of cabin scenarios.

[0003] Existing technologies mostly employ static bandwidth allocation or single-dimensional QoS scheduling strategies, lacking dynamic coupling and adaptation to flight operation phases, global network load, and user experience. For example, network transmission conditions differ significantly between takeoff / climb and level flight, but traditional solutions do not specifically limit high-bandwidth-consuming services. During level flight, under medium load, premium sessions struggle to obtain dedicated bandwidth guarantees, while standard sessions are prone to experience degradation due to bandwidth contention under high load. Furthermore, existing systems can only simply identify service types and cannot effectively handle behaviors such as P2P downloads that consume large amounts of satellite link bandwidth. They also struggle to handle abnormal scenarios where single-IP traffic exceeds thresholds and there are no compliant services to match, often resulting in problems such as "sufficient bandwidth but poor experience" and "some users excessively consuming resources, affecting fairness."

[0004] In terms of security and compliance, existing solutions are slow to respond to malicious attacks such as port scanning and lack an immediate mandatory intervention mechanism. Real-name authentication and resource scheduling are disconnected, failing to form a closed loop of "compliance verification - resource allocation - anomaly handling." Furthermore, the feedback mechanism of traditional scheduling systems only compares data before and after intervention, failing to exclude confounding factors such as changes in flight phases and interference from other sessions, resulting in a lack of precise basis for strategy optimization. For example, economy class passengers are in restricted classes during low-load periods but cannot receive bandwidth increases, and malicious port scanning cannot be blocked in a timely manner, all reflecting the shortcomings of existing technologies in dynamic adaptation, fairness assurance, and security protection.

[0005] To address these issues, there is an urgent need to build a dynamic orchestration system that integrates traffic characteristics and user experience. This system should combine multiple factors such as flight operation phases, overall load, and service types, and use refined decision-making rules to achieve intelligent resource allocation. This system should balance improved experience, bandwidth utilization, and security compliance, thereby meeting the differentiated operational needs of airlines and the diverse internet access needs of passengers. Summary of the Invention

[0006] To address the aforementioned problems in the existing technology, this invention provides a dynamic scheduling system for in-flight internet cabin resources based on traffic-experience coupling. The objective of this invention can be achieved through the following technical solutions: A dynamic cabin resource scheduling system based on traffic-experience coupling for aviation internet includes: The data acquisition and association module collects traffic data and session status of the device IP addresses on the flight in real time, and obtains the passenger's real-name authentication information, seat information and policy group information. It dynamically binds and associates the device IP address with the passenger's identity and flight itinerary information to form structured real-time operation data. The multi-dimensional experience evaluation module receives the real-time running data and dynamically calculates the network experience index for IP sessions based on a preset experience evaluation model. The experience evaluation model comprehensively evaluates the connection quality dimension, bandwidth sufficiency dimension, and service adaptation dimension, and classifies the session into the corresponding experience level according to the evaluation results. The dynamic policy decision and generation module, based on the session experience level output by the multi-dimensional experience evaluation module, combined with the global network load and flight operation phase, dynamically makes decisions from the preset policy library and generates resource orchestration instructions, including switching policy groups for specific sessions, adjusting bandwidth parameters, or performing forced intervention. The policy execution and feedback module sends the resource orchestration instructions to the airborne network equipment for execution, and collects new session data and status information generated after policy execution. The new session data and status information are then fed back to the data acquisition and association module as input, and continuous evaluation and decision-making are performed based on the updated data.

[0007] As a preferred embodiment of the present invention, the dynamic binding and association specifically includes: When a passenger device first accesses the onboard network and completes real-name authentication, a globally unique session identifier is generated based on the timestamp, flight identifier, and device MAC address. The session identifier, device IP address, passenger ID number, and real-time seat number obtained from the cabin service system are recorded in a dynamic association record with a version identifier and mapped to the logical topology of the flight. When a device with the same ID number is detected to reconnect or a passenger is detected to change seats, the IP and seat fields in the dynamically associated record are atomically updated by cross-comparing the onboard radio controller data with the cabin service log, and the version change is recorded.

[0008] Specifically, the real-time operational data includes: The uplink traffic bytes, downlink traffic bytes, current session duration, signal strength index, and packet loss rate of each device IP address are collected at a preset sampling frequency through the airborne network equipment interface. The following information is retrieved and linked in real time from the airline's passenger service system: the valid ID number used by the passenger when boarding, the status and time of real-name authentication completed through the ID, and the physical seat number assigned to the passenger. Based on real-time data from the flight management system, the collected IP-level data and passenger identity data are bound to the currently active flight number, aircraft registration tail number, current flight segment, and flight stage to generate structured data records with spatiotemporal stamps.

[0009] Specifically, the experience evaluation model includes: a core decision tree model that uses online incremental learning to learn experience mapping from real-time data streams; an outer rule engine that embeds operational rules and security constraints; and outputs a quantified network experience index and feature attribution.

[0010] Specifically, the experience evaluation model also incorporates business identification and intent inference for collaborative correction during calculation: Maintain an extensible business type feature library, including static traffic fingerprints and dynamic behavior sequences; For each IP session, streaming clustering is used to match its traffic characteristics and behavior sequences to identify the dominant business type and infer intent in real time; When calculating the network experience index, based on the identified service type and intent, corresponding benchmark parameters and multi-dimensional tolerance thresholds are dynamically selected to weight and correct the calculation results of bandwidth adequacy and connection stability factors.

[0011] Specifically, the comprehensive evaluation includes connection quality, bandwidth adequacy, and service compatibility dimensions. The evaluation of connection quality is based on a weighted calculation of the number of abnormal disconnections and the average signal strength. The evaluation of bandwidth adequacy is based on a comparison of the historical traffic baseline of the IP session with the real-time traffic occupancy rate. The evaluation of service compatibility is based on the matching degree identification by analyzing the traffic feature packets of the IP and the preset service type feature library.

[0012] Specifically, the preset policy library in the dynamic policy decision and generation module contains policy groups with different names. Each policy group defines different bandwidth limits, priority queues, allowed ports or protocol sets, and security rules.

[0013] Specifically, the decision-making logic of the dynamic policy decision-making and generation module adopts a multi-objective optimization strategy search, which includes: optimizing global experience fairness, bandwidth utilization, and high-priority service guarantee rate; when the session experience level is limited and the global load is lower than the preset value, searching and generating instructions to increase bandwidth in the Pareto solution set that meets the constraints; when P2P downloads are identified as affecting fairness, generating instructions to bind and prohibit P2P policy groups; and when port scanning attacks are detected, triggering security routines to generate instructions to force offline and add to a time-limited temporary blacklist.

[0014] Specifically, the dynamic strategy decision-making and generation module also receives input from the flight cabin service system when making decisions. The input includes at least passenger cabin class information or special service requests marked by flight attendants, and uses this as an additional weighting factor for strategy decision-making.

[0015] Specifically, the policy execution and feedback module issues instructions in the following ways: by sending configuration commands based on SNMP or a specific API interface to the airborne network gateway or wireless access point on the target flight through a secure communication protocol, and modifying the policy group affiliation or quality of service parameters of the specified IP address in real time.

[0016] Specifically, after sending the instruction, the policy execution and feedback module starts a monitoring timer to verify the parameter changes of the target IP session within a preset time according to the instruction, and records the delay time for the policy to take effect.

[0017] Specifically, the strategy execution and feedback module implements a structured feedback loop, including: After the instruction is issued, monitor and collect new data from multiple dimensions following the intervention of the target session; By constructing a pre- and post-intervention comparison sequence and using the difference method to control for confounding factors, the net effect of the strategy instruction is estimated, and a feedback record containing effect assessment is generated. The feedback records and new data are packaged and pushed back to the stream processing buffer of the data acquisition and correlation module via the data bus; After receiving the data, the data acquisition and association module integrates it into the real-time data stream and triggers online adjustment of relevant parameters in the experience evaluation and decision-making module.

[0018] The beneficial effects of this invention are as follows: Significantly improve the accuracy and satisfaction of user experience. Differentiated strategies are developed for different flight phases, load conditions, and experience levels. During periods of moderate load in cruising conditions, 10Mbps dedicated bandwidth is allocated to premium sessions to ensure a smooth experience for high-value users; bandwidth is increased for restricted users during low-load scenarios. Simultaneously, during periods of high load, priority is given to core services such as video streaming to avoid widespread buffering, ensuring a suitable experience for passengers with different needs.

[0019] Optimize bandwidth utilization and allocation fairness. By dynamically adjusting bandwidth parameters and service restrictions, resource waste and excessive usage are avoided. During takeoff and climb, bandwidth for all sessions is limited and P2P and large file downloads are prohibited to adapt to the transmission constraints of this phase. For continuous high-bandwidth P2P downloads, a binding restriction policy is enforced, disconnecting invalid occupied links and ensuring that satellite link bandwidth is allocated to effective services. The system significantly improves the overall bandwidth utilization of the cabin network and effectively solves the fairness issues caused by single-user resource contention.

[0020] Strengthen security protection and compliance management capabilities. Establish a rapid-response security intervention mechanism to immediately enforce shutdown and add malicious behaviors such as port scanning to a temporary blacklist. Bind abnormal interception groups to traffic exceeding thresholds that does not match compliant business and trigger flight attendant confirmation, balancing security control with flexible handling, fully meeting the Civil Aviation Administration's compliance audit requirements.

[0021] Enhance the intelligence and response efficiency of resource scheduling. Decision commands are updated via the MQTT protocol to update gateway configurations, atomically execute policy switching and bandwidth adjustments, and adapt to dynamic changes in the cabin environment. A feedback loop mechanism verifies execution effectiveness in real time, and combined with business type identification and behavior analysis, accurately triggers adaptation strategies. This reduces manual intervention costs while ensuring the timeliness and accuracy of scheduling decisions, achieving a dual improvement in service quality and operational efficiency. Attached Figure Description

[0022] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0023] Figure 1 This is a flowchart illustrating a dynamic scheduling system for cabin resources in aviation internet based on traffic-experience coupling, according to the present invention. Figure 2 This is a diagram of the experience evaluation model architecture of the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0025] Please see Figure 1-2 A dynamic cabin resource scheduling system for aviation internet based on traffic-experience coupling includes: The data acquisition and association module collects traffic data and session status of the device IP addresses on the flight in real time, and obtains the passenger's real-name authentication information, seat information and policy group information. It dynamically binds and associates the device IP address with the passenger's identity and flight itinerary information to form structured real-time operation data. The multi-dimensional experience evaluation module receives the real-time running data and dynamically calculates the network experience index for IP sessions based on a preset experience evaluation model. The experience evaluation model comprehensively evaluates the connection quality dimension, bandwidth sufficiency dimension, and service adaptation dimension, and classifies the session into the corresponding experience level according to the evaluation results. The dynamic policy decision and generation module, based on the session experience level output by the multi-dimensional experience evaluation module, combined with the global network load and flight operation phase, dynamically makes decisions from the preset policy library and generates resource orchestration instructions, including switching policy groups for specific sessions, adjusting bandwidth parameters, or performing forced intervention. The policy execution and feedback module sends the resource orchestration instructions to the airborne network equipment for execution, and collects new session data and status information generated after policy execution. The new session data and status information are then fed back to the data acquisition and association module as input, and continuous evaluation and decision-making are performed based on the updated data.

[0026] This embodiment applies to an airline's wide-body aircraft (A350-900, tail number B-305X) on an intercontinental route (flight number MU5101, Shanghai Pudong PVG-Los Angeles LAX). This flight is equipped with an independently controllable in-flight internet platform, containing 180 cabin Wi-Fi access points (APs, Wi-Fi codes G2TSIO13WR to G2TSIO150WR), using LEO satellite links for transmission, with a maximum available bandwidth of 100Mbps, and adaptable to differentiated services for economy, business, and first class cabins.

[0027] (I) Data Acquisition and Correlation Module Data Acquisition Interface and Protocol: Traffic data is collected through the NetFlowv9 protocol interface of the airborne network gateway (Cisco ISR4000 series). The AP mirror port captures packet characteristics using the Lippcap library, and the sampling frequency is dynamically adjusted (500ms / sample under high load, 1 second / sample under medium and low load). A RESTful API with SSL / TLS 1.3 encryption is used to interface with the Civil Aviation Administration's real-name database, airline passenger service systems (PSS), and flight management systems (FMS). Three retries are allowed after a timeout, with a 2-second interval between each attempt.

[0028] Data Collection and Binding Logic: Collect traffic data including IP address, uplink and downlink traffic bytes, peak traffic, signal strength (dBm), packet loss rate, and transmission latency; associate with passenger real-name authentication information (anonymized ID number, authentication channel, authentication time), seat number, cabin class, and product subscription information; bind flight number, aircraft tail number, and flight phase (code 0 = takeoff and climb, 1 = level flight, 2 = descent and landing), generating JSON-formatted structured data with millisecond-level spatiotemporal stamps. Upon initial access and completion of real-name authentication, a UUIDv4 session identifier is generated and bound to the IP address, anonymized ID number, seat number, and Wi-Fi code, stored in a Redis cache (validity period = flight duration + 24 hours), and atomically updated upon device reconnection or seat change.

[0029] (II) Multidimensional Experience Assessment Module Evaluation Model Architecture: A hybrid architecture of "rule engine + lightweight random forest model" is adopted. The rule engine is based on the Drools framework, filtering invalid data (such as signal strength > -30dBm or < -90dBm) and marking candidate anomalies; the lightweight random forest model is optimized by TensorFlowLite (10 decision trees, tree depth ≤ 5 layers), with inference time ≤ 10ms. It takes three types of features as input: connection quality, bandwidth, and service adaptability, and outputs a network experience index of 0-10.

[0030] Three-dimensional assessment and grading: Connection quality dimension (weight 40%) = (1 - number of abnormal disconnections / total number of connections) × 0.6 + (-signal strength + 90) / 60 × 0.4; Bandwidth sufficiency dimension (weight 35%) = real-time bandwidth utilization / bandwidth ratio required by business needs; Business adaptation dimension (weight 25%) = matching degree between traffic characteristics and business type characteristic library. Experience levels are divided according to thresholds: 8-10 points are Premium, 6-8 points are Standard, 4-6 points are Restricted, and <4 points are Abnormal.

[0031] (III) Dynamic Strategy Decision-Making and Generation Module Pre-configured strategy library: The strategy library contains 4 core strategy groups. Core decision rule execution logic: During the cruising period (flight phase 1) + medium load (bandwidth utilization 60%-85%) + premium sessions: Business Class Premium Group will be assigned, with a bandwidth limit of 10Mbps; During the plateau period, under high load (bandwidth utilization ≥ 85%), and with standard-level sessions, bandwidth is reduced to 3Mbps, and service priority scheduling is enabled (video streaming > real-time communication > web browsing). Takeoff and climb phase (flight phase 0): All sessions are bound to a restricted experience group, with bandwidth ≤2Mbps, and P2P (ports 6881-6889) and large file downloads (data packets >4096Byte) are prohibited. If a single IP experiences peak traffic of ≥20Mbps within 5 minutes and has no matching compliant services: bind it to an abnormal interception group and trigger flight attendant confirmation through the cabin service system.

[0032] (iv) Strategy Execution and Feedback Module Command Issuance and Execution: Commands are issued via the AES-256 encrypted MQTT protocol (QoS=2). The gateway modifies the QoS configuration via the SNMPv3 protocol, and the AP adjusts resource allocation via its private API. Bandwidth adjustment steps are 1 Mbps / second, and policy switching adopts a "connect first, disconnect later" mechanism. A 10-second monitoring timer is started after the command is issued to dual-verify the configuration effectiveness and changes in traffic metrics.

[0033] Feedback closed-loop mechanism: Traffic data, latency, packet loss rate, and execution status codes after strategy execution are collected at a dynamic acquisition frequency. Differential methods are used to control confounding factors and estimate the net effect of the strategy. Feedback data is packaged into structured information packets and pushed to the data acquisition module via the data bus to update the real-time running data pool, triggering the next round of evaluation and decision-making. The closed-loop cycle is ≤2 seconds.

[0034] III. Specific Decision Implementation Examples Example 1: Limited load + low load → Increase bandwidth Passenger B (seat 35B, economy class, 1-hour experience package) browsed the web after connecting to Wi-Fi. The system detected a signal strength of -55dBm and a packet loss rate of 1.2%, with an initial experience evaluation of 4.8 points (limited level). At this time, the flight was in the pre-descent transition phase, with a global bandwidth utilization of 52% (low load, preset threshold 60%). The system triggered a decision rule, generating a command to "increase bandwidth to 3Mbps and switch to the standard economy class group". After the command was sent via the MQTT protocol, the gateway configuration was updated within 1 second. Feedback data showed that the webpage loading latency decreased from 800ms to 450ms, and the experience index improved to 6.2 points (standard level).

[0035] Example 2: Continuous high-bandwidth P2P download → Bind a P2P blocking policy group After passenger C (seat 28C, economy class, no subscription) connected, the system, through feature matching, found that their IP (10.189.56.23) continuously occupied ports 6881-6889, with data packets exceeding 4096 bytes and a peak traffic of 15Mbps in 5 minutes, indicating high-bandwidth P2P downloading. The system generated a command to bind them to the economy class standard group (including P2P blocking rules), limiting their bandwidth to 1Mbps. After the command was executed, the P2P connection was disconnected, retaining only basic service permissions to avoid consuming satellite link bandwidth.

[0036] Example 3: Same IP port scan → Forced offline + temporary blacklist The onboard safety monitoring module detected IP 10.189.78.45 scanning 20 core ports (21, 22, 8080, etc.) of the gateway within one minute, consistent with port scanning characteristics. The system generated a mandatory intervention command, immediately disconnecting the IP connection, adding its MAC address (00:1B:44:11:3A:B7) to a 2-hour temporary blacklist, and simultaneously transmitting the intervention log to the ground security platform to ensure cabin network security.

[0037] As a preferred embodiment of the present invention, the dynamic binding and association specifically includes: When a passenger device first accesses the onboard network and completes real-name authentication, a globally unique session identifier is generated based on the timestamp, flight identifier, and device MAC address. The session identifier, device IP address, passenger ID number, and real-time seat number obtained from the cabin service system are recorded in a dynamic association record with a version identifier and mapped to the logical topology of the flight. When a device with the same ID number is detected to reconnect or a passenger is detected to change seats, the IP and seat fields in the dynamically associated record are atomically updated by cross-comparing the onboard radio controller data with the cabin service log, and the version change is recorded.

[0038] Specifically, the real-time operational data includes: The uplink traffic bytes, downlink traffic bytes, current session duration, signal strength index, and packet loss rate of each device IP address are collected at a preset sampling frequency through the airborne network equipment interface. The following information is retrieved and linked in real time from the airline's passenger service system: the valid ID number used by the passenger when boarding, the status and time of real-name authentication completed through the ID, and the physical seat number assigned to the passenger. Based on real-time data from the flight management system, the collected IP-level data and passenger identity data are bound to the currently active flight number, aircraft registration tail number, current flight segment, and flight stage to generate structured data records with spatiotemporal stamps.

[0039] Specifically, the experience evaluation model includes: a core decision tree model that uses online incremental learning to learn experience mapping from real-time data streams; an outer rule engine that embeds operational rules and security constraints; and outputs a quantified network experience index and feature attribution.

[0040] Specifically, the experience evaluation model also incorporates business identification and intent inference for collaborative correction during calculation: Maintain an extensible business type feature library, including static traffic fingerprints and dynamic behavior sequences; For each IP session, streaming clustering is used to match its traffic characteristics and behavior sequence to identify the dominant business type and inferred intent in real time (intention is inferred based on session duration, traffic fluctuation patterns, and access target address, such as 10 consecutive minutes of high bandwidth transmission and access to the office cloud platform, inferred to be an urgent office intent). When calculating the network experience index, based on the identified service type and intent, corresponding benchmark parameters and multi-dimensional tolerance thresholds are dynamically selected to weight and correct the calculation results of bandwidth adequacy and connection stability factors.

[0041] This embodiment is based on the MU5101 (PVG-LAX) intercontinental route operated by an A350-900 wide-body aircraft (tail number B-305X) of a certain airline. The flight is equipped with 180 cabin Wi-Fi APs (wificodes: G2TSIO13WR to G2TSIO150WR), using LEO satellite links (maximum bandwidth 100Mbps). The onboard server is configured with an Intel Atomx7-E3950 CPU and 8GB of memory, adapting to differentiated services for economy class, business class, and first class, and strictly adhering to the Civil Aviation Administration's real-name compliance and security audit requirements. The experience evaluation model serves as the core module, deeply integrated with the data collection and decision-making modules, supporting dynamic resource orchestration throughout the entire process.

[0042] (I) Core Architecture: Online Incremental Learning Decision Tree + Outer Rule Engine Online incremental learning decision tree model Basic algorithm selection: The optimized C4.5 decision tree algorithm is adopted, which is lightweighted through the TensorFlowLite framework to adapt to the limited computing power of the onboard. The core parameters are configured as follows: number of decision trees = 8, tree depth ≤ 4 layers, feature dimension = 6 dimensions (connection stability, bandwidth sufficiency, service matching degree, signal strength, packet loss rate, session duration), and single session inference time ≤ 8ms.

[0043] Incremental learning mechanism: The model is initialized by loading baseline data of "traffic characteristics-experience feedback" from 1,000+ flights over the past 3 months (including passenger ratings, number of stutters, and other labeled information); during runtime, it receives real-time data streams every 5 seconds and dynamically updates the model weights through a "sample sliding window" (window size = 100 session data), without the need for manual intervention on the ground, adapting to the dynamic fluctuations of cabin traffic (such as peak traffic during level flight and sudden drop in traffic during descent).

[0044] Feature attribution output: While calculating the network experience index, output the contribution percentage of each feature (e.g., "insufficient bandwidth contributes 60%, signal fluctuation contributes 30%, service adaptation deviation contributes 10%), providing a precise optimization basis for the decision-making module.

[0045] outer rule engine Framework and rule configuration: Built on the Drools 7.59 framework, embedding two types of core rules: Operational rules include: experience level thresholds (Premium 8-10 points, Standard 6-8 points, Restricted 4-6 points, Abnormal <4 points), and data validity filtering (when signal strength is >-30dBm or <-90dBm, replace with the average of the last 5 samples; packet loss rate >50% is marked as invalid data and triggers resampling). Security constraint rules: Unauthenticated sessions are directly marked as "abnormal level", port scanning behavior is associated with the "feature attribution - security risk" label, and connection stability score is automatically reduced if a single IP reconnects ≥5 times within 1 minute.

[0046] Rule execution priority: security constraint rules (P0 level) > data validity rules (P1 level) > operational rules (P2 level), ensuring compliance and security take precedence.

[0047] (ii) Collaborative Correction Mechanism for Business Identification and Intent Inference Extensible Business Type Feature Library Feature library structure: It adopts Redis cluster storage, supports remote iterative updates, and contains two-dimensional features of "static traffic fingerprint + dynamic behavior sequence" for 6 core business categories; Streaming clustering recognition and intent inference Identification algorithm: The MiniBatchK-Means streaming clustering algorithm (batch size = 50 data points, number of clusters = 6) is used to analyze the traffic characteristics of IP sessions in real time and perform similarity calculation with the feature library (cosine similarity threshold ≥ 88%).

[0048] Intent inference logic: Combining "business type + behavioral context" for dual determination, for example: Large file download service + access to office cloud platform (target IP matches the airline's office network segment) + session duration ≤ 15 minutes → infer "urgent office intent"; Large file download service + access to video resource platform + session duration ≥ 30 minutes → infer "leisure download intention"; Real-time communication service + stable signal strength + session time during the day → infer "business meeting intent".

[0049] Weighted correction calculation The baseline parameters and tolerance thresholds are dynamically invoked, and a preset parameter table is matched based on "business type + intent"; Corrected formula: Final network experience index = initial model score × (1 + weighted correction coefficient) + feature attribution adjustment value (±0.1~±0.5 points).

[0050] (III) Model Collaborative Computation Process Data input: Receive structured data (including IP traffic data, real-name authentication status, flight stage, and seat information) from the data acquisition module; Rule engine preprocessing: Performs data validity filtering and security constraint verification, and marks abnormal candidate sessions; Business identification and intent inference: By matching the feature library through streaming clustering, the dominant business type, intent and confidence level are output (≥90% direct confirmation, 70%-90% correction based on historical business preferences). Incremental learning model calculation: Input the preprocessed feature set, output the initial experience score and feature attribution; Weighted adjustment: Invoke the corresponding baseline parameters and tolerance thresholds to calculate the final experience index and level; Output results: Synchronize experience level, feature attribution, and business intent to the dynamic strategy decision-making and generation module.

[0051] III. Example of Decision Logic Implementation (Combined with Model Workflow) Example 1: Limited load + low load → Increase bandwidth Data input: Passenger B (seat 35B, economy class, 1-hour experience package) connects to Wi-Fi, real-name authentication is successful, system collects data: IP [10.189.45.78] (10.189.45.78), signal strength -55dBm, packet loss rate 1.2%, session duration 2 minutes, business traffic characteristics match "web browsing" (similarity over 90%), flight stage is the pre-descent transition stage (code 2), global bandwidth utilization is about 50% (low load, preset threshold 60%).

[0052] Model calculation process: Rule engine: If the data is valid and there are no security constraints, it is marked as "web browsing + general intent"; Streaming clustering: Confirms the business type as web browsing and the intent as general browsing; Incremental learning model: Input feature set, calculate initial score, feature attribution "insufficient bandwidth contributes 70%, signal fluctuation contributes 30%"; Weighted correction: Using the web browsing baseline parameters (≥2Mbps), with a correction factor of 0, the final score is 4.8 (restricted level).

[0053] Decision and Execution: The system identifies “restricted level + low load” and generates the instruction to “increase bandwidth to 3Mbps and switch to the economy class standard group”. The gateway QoS configuration is updated within 1 second via the MQTT protocol. Feedback data shows that the webpage loading latency has decreased from 800ms to 450ms. The model recalculates the score to 6.2 points (standard level), and the feature attribution is updated to “sufficient bandwidth contributes 80% and signal fluctuation contributes 2 layers”.

[0054] Example 2: Continuous high-bandwidth P2P download → Bind a P2P blocking policy group Data input: Passenger C (seat 28C, economy class, no subscription) accesses Wi-Fi, IP [10.189.56.23](10.189.56.23). The system collected the following data: concurrent connections on ports 6883-6885, data packet size 2048-4096 bytes, peak traffic of 15Mbps in 5 minutes, session duration of 12 minutes, flight stage is level flight (code 1), global bandwidth utilization is over 70% (medium load).

[0055] Model calculation process: Rule engine: Triggered without security constraints, marked as "P2P download candidate"; Streaming clustering: The similarity to the matching feature library is over 90%, confirming the business type as P2P download and the intent as leisure (no access records from office network segments), with a confidence level of over 90%. Incremental learning model: Input feature set, calculate initial score of 3.2, feature attribution "P2P bandwidth usage contributes 90%, other factors contribute 10%"; Weighted correction: The P2P download correction factor is reduced by 0.4, resulting in a final score of 2.8 (abnormal level).

[0056] Decision and Execution: The system triggers the "continuous high-bandwidth P2P download" decision logic, generating the instruction "bind economy class standard group (including P2P prohibition rule) bandwidth limit 1Mbps"; after execution, the P2P connection is disconnected, only basic business permissions are retained, the model is re-evaluated and scored 5.1 points (restricted level), and the global bandwidth utilization rate drops to below 70%.

[0057] Specifically, the comprehensive evaluation includes connection quality, bandwidth adequacy, and service compatibility dimensions. The evaluation of connection quality is based on a weighted calculation of the number of abnormal disconnections and the average signal strength. The evaluation of bandwidth adequacy is based on a comparison of the historical traffic baseline of the IP session with the real-time traffic occupancy rate. The evaluation of service compatibility is based on the matching degree identification by analyzing the traffic feature packets of the IP and the preset service type feature library.

[0058] Specifically, the preset policy library in the dynamic policy decision and generation module contains policy groups with different names. Each policy group defines different bandwidth limits, priority queues, allowed ports or protocol sets, and security rules.

[0059] Specifically, the decision-making logic of the dynamic policy decision-making and generation module adopts a multi-objective optimization strategy search, which includes: optimizing global experience fairness, bandwidth utilization, and high-priority service guarantee rate; when the session experience level is limited and the global load is lower than the preset value, searching and generating instructions to increase bandwidth in the Pareto solution set that meets the constraints; when P2P downloads are identified as affecting fairness, generating instructions to bind and prohibit P2P policy groups; and when port scanning attacks are detected, triggering security routines to generate instructions to force offline and add to a time-limited temporary blacklist.

[0060] This embodiment utilizes an airline's A350-900 wide-body aircraft (tail number B-305X) on an intercontinental route (flight number MU5101, Shanghai Pudong PVG-Los Angeles LAX). The flight is equipped with 180 cabin Wi-Fi access points (APs, Wi-Fi codes: G2TSIO13WR to G2TSIO150WR), using LEO satellite links with a maximum available bandwidth of 100Mbps. The onboard server is configured with an Intel Atomx7-E3950 CPU and 8GB of RAM, adapting to the differentiated service needs of economy, business, and first class cabins. The system must simultaneously meet multiple optimization requirements: overall user experience fairness, bandwidth utilization ≥80%, and high-priority service (real-time communication / business office) guarantee rate ≥95%, while strictly adhering to the Civil Aviation Administration's real-name registration compliance, security audit, and avionics system electromagnetic compatibility constraints.

[0061] The dynamic strategy decision-making and generation module, as the core decision-making unit of the system, receives the session experience level (premium 8-10 points, standard 6-8 points, restricted 4-6 points, abnormal <4 points), global network load data (high ≥85% / medium 60%-85% / low <60%) and flight operation phase (0=takeoff and climb period, 1=level flight period, 2=descent and landing period) output by the multi-dimensional experience evaluation module, and generates accurate resource orchestration instructions through multi-objective optimization strategy search.

[0062] (I) Quantitative Definition of Multi-Objective Optimization Objectives The module uses a combination of quantitative indicators and weight allocation to clearly define the optimization objectives, with the specific settings as follows: Global experience fairness: Measured using the Gini coefficient, with a target value ≤ 0.2 (the closer the Gini coefficient is to 0, the better the fairness), and a weight of 35%. Bandwidth utilization: Target value ≥80% and ≤95% (to avoid excessive usage leading to link congestion), weight 35%; High-priority service reliability: The transmission success rate of high-priority services (business class real-time communication, paid business office packages) is ≥95%, with a weight of 30%.

[0063] (II) Pareto solution set construction and search logic The module constructs Pareto solution sets based on NSGA-II (Non-dominated sorting genetic algorithm II). The core process is as follows: Constraint settings include: bandwidth limit constraint (maximum bandwidth per session ≤ 10Mbps), service permission constraint (high bandwidth services are prohibited for unauthenticated sessions), and flight phase constraint (P2P / large file downloads are prohibited during takeoff / descent). Initial solution generation: Based on the pre-set policy library (4 types of core policy groups), 100 initial resource allocation schemes (including policy group switching and bandwidth adjustment parameters) are generated. Non-dominated sorting: The initial solutions are sorted according to the three optimization objectives, and solutions that are not dominated by other solutions (i.e., a solution has at least one objective that is better than other solutions and no objective that is worse than other solutions) are selected to form the Pareto solution set; Optimal solution selection: From the Pareto solution set, select the solution that "satisfies all target thresholds + has the highest comprehensive score" (comprehensive score = fairness score × 0.35 + bandwidth utilization score × 0.35 + high priority guarantee rate score × 0.3) to generate the final resource orchestration instruction.

[0064] (III) Pre-configured strategy library core configuration The strategy library provides basic strategy support for multi-objective optimization, including four core strategy groups. (III. Examples of Multi-Objective Optimization Decision Execution) Example 1: Restricted session + low load → Pareto unset search generates bandwidth-boosting instructions 1. Input data Session information: Passenger B (seat 35B, economy class, 1-hour basic experience package), IP 10.189.45.78, experience level restricted (score 4.8), current activity is web browsing; Global Status: Flights are in cruising phase (Phase 1), global bandwidth utilization is 52% (low load, preset threshold <60%), current experience index Gini coefficient is 0.18 (good fairness), and high-priority service guarantee rate is close to 100%. Constraints: Maximum bandwidth per session ≤ 4Mbps (economy class limit), no security risk flags.

[0065] 2. Multi-objective optimization decision-making process Initial solution generation: 100 solutions are generated based on the policy library, including "Maintaining Limited Experience Group (2Mbps)", "Switching to Economy Class Standard Group (3Mbps)", "Switching to Economy Class Standard Group (4Mbps)", etc. Pareto solution set filtering: exclude the "4Mbps plan" (which may lead to insufficient bandwidth reservation when the load increases later), and filter out the "3Mbps plan" and "maintain 2Mbps plan" to enter the solution set; Optimal solution selection: Comparing two options: ① 3Mbps option: bandwidth utilization increased to 58% (still meeting the target threshold of ≥80%), fairness Gini coefficient 0.17 (optimized), high priority guarantee rate 98% (no impact), overall score 0.92; ② Maintain 2Mbps option: bandwidth utilization 52%, Gini coefficient 0.18, overall score 0.88; The final option selected was "switching to the economy class standard group and increasing bandwidth to 3Mbps".

[0066] 3. Command Execution and Feedback Command generated: The command is issued via the MQTT protocol (QoS=2) encrypted with AES-256. Command ID: CMD20250815103001. The core content is "IP10.189.45.78 switch to the Economy Class Standard Group, bandwidth limit 3Mbps".

[0067] Feedback results: The gateway QoS configuration was updated within 1 second, the webpage loading latency was reduced from 800ms to 450ms, and the experience index was improved to 6.2 points (standard level); the global bandwidth utilization rate increased to 58%, the Gini coefficient was 0.17, and the high-priority service guarantee rate remained at 98%, all three optimization targets were met.

[0068] Example 2: P2P downloads affect fairness → Generate a command to bind a policy group that prohibits P2P. 1. Input data Session information: Passenger C (seat 28C, economy class, no paid subscription), IP 10.189.56.23, continuously performing P2P downloads (ports 6883-6885, peak traffic of 15Mbps in 5 minutes), causing the experience level of 10 surrounding sessions to drop (3 standard level to restricted level). Global Status: During the flat period, the global bandwidth utilization rate is 75% (medium load), the Gini coefficient of the experience index rises to 0.32 (exceeding the target value of 0.2, indicating an imbalance in fairness), and the high-priority service guarantee rate drops to 92% (below the target of 95%).

[0069] 2. Multi-objective optimization decision-making process The module detected that "P2P downloads caused the dual objectives of fairness and high-priority guarantee rate to fail to meet the targets," and initiated targeted optimizations: Constraints: Prioritize fairness and high-priority business; prohibit P2P protocols. Pareto unpacking and filtering: The core solution is to "bind the session to the economy class standard group (including the P2P ban rule) + bandwidth limit of 1Mbps". This solution can quickly terminate P2P occupation and release bandwidth resources. Optimal solution verification: After the solution is implemented, the estimated results are: the Gini coefficient drops to 0.21 (close to the target value), the bandwidth utilization rate is 72% (meets the standard), the high priority guarantee rate rises to 96% (meets the standard), and the overall score is 0.91, which is the optimal solution.

[0070] 3. Command Execution and Feedback Command generated: Bind to Economy Class Standard Group, disable P2P protocol, bandwidth limit 1Mbps. Upon execution, P2P connections are immediately disconnected, and the experience level of the surrounding 10 sessions is restored to Standard level; global Gini coefficient is 0.20, high priority guarantee rate is close to 100%, and multiple objectives are met.

[0071] Specifically, the dynamic strategy decision-making and generation module also receives input from the flight cabin service system when making decisions. The input includes at least passenger cabin class information or special service requests marked by flight attendants, and uses this as an additional weighting factor for strategy decision-making.

[0072] Specifically, the policy execution and feedback module issues instructions in the following ways: by sending configuration commands based on SNMP or a specific API interface to the airborne network gateway or wireless access point on the target flight through a secure communication protocol, and modifying the policy group affiliation or quality of service parameters of the specified IP address in real time.

[0073] Specifically, after sending the instruction, the policy execution and feedback module starts a monitoring timer to verify the parameter changes of the target IP session within a preset time according to the instruction, and records the delay time for the policy to take effect.

[0074] Specifically, the strategy execution and feedback module implements a structured feedback loop, including: After the instruction is issued, monitor and collect new data from multiple dimensions following the intervention of the target session; By constructing a pre- and post-intervention comparison sequence and using the difference method to control for confounding factors, the net effect of the strategy instruction is estimated, and a feedback record containing effect assessment is generated. The feedback records and new data are packaged and pushed back to the stream processing buffer of the data acquisition and correlation module via the data bus; After receiving the data, the data acquisition and association module integrates it into the real-time data stream and triggers online adjustment of relevant parameters in the experience evaluation and decision-making module.

[0075] This embodiment utilizes an A350-900 wide-body aircraft (tail number B-305X) on the MU5101 (Shanghai Pudong PVG-Los Angeles LAX) intercontinental route. The flight is equipped with 180 cabin Wi-Fi access points (APs, Wi-Fi codes: G2TSIO13WR to G2TSIO150WR), using LEO satellite links (maximum available bandwidth 100Mbps). The onboard server is configured with an Intel Atomx7-E3950 CPU and 8GB of memory. The strategy execution and feedback module, as the core of the system's closed loop, receives resource orchestration instructions generated by the dynamic strategy decision-making module. Through a structured closed loop of "execution-monitoring-evaluation-feedback-optimization," it ensures the accuracy and adaptability of cabin network resource scheduling, strictly complying with the Civil Aviation Administration's avionics system computing power constraints and compliance audit requirements.

[0076] Core technology implementation details (structured feedback closed loop) The module adopts a "layered architecture + standardized protocol" to achieve a structured feedback loop. The core is divided into four progressive stages, and the technical details of each stage are as follows: (I) Monitoring and collecting new data from multiple dimensions after the instruction is issued 1. Data collection triggering mechanism: The "dual timer" mechanism is activated the moment the command is issued. The first is an instant verification timer within 10 seconds (used to confirm whether the command is effective), and the second is an effect tracking timer that lasts for 3 minutes (used to collect stable data after intervention). 2. Data Collection Content and Dimensions: Covering three types of data: "session performance + network global + device status", specifically including: real-time bandwidth, latency, packet loss rate, signal strength, and service transmission success rate of the target session; bandwidth utilization, Gini coefficient, and high-priority service guarantee rate of the global network; CPU utilization, memory usage, and instruction execution status codes (0 = success, 1 = failure, 2 = partial success) of the airborne gateway / AP. 3. Data Acquisition Technology and Frequency: Gateway / AP device status is acquired via SNMPv3 protocol, session traffic characteristics are captured using the Lippcap library, and the sampling frequency is dynamically adjusted according to the load (500ms / time for high load, 1 second / time for medium and low load); all data is timestamped in milliseconds and stored in JSON format to ensure traceability.

[0077] (II) Construction of pre- and post-intervention comparative sequences and estimation of net effect 1. Contrast sequence construction: Using the instruction issuance time as the dividing point, construct a "baseline sequence for the first 3 minutes + intervention sequence for the last 3 minutes". Each sequence contains 20 key data sampling points (extracted at a frequency of 1 second / time) to ensure the representativeness of the data samples; 2. Confounding Factor Identification and Control: The "Dual Difference (DID)" method is used to control confounding factors. The core confounding factors identified include: flight phase transitions (such as transitioning from cruising to descent), bursts of other high-bandwidth sessions, and satellite link latency fluctuations. By introducing a "concurrent uninterrupted reference group" (selecting 3-5 uninterrupted sessions of the same cabin class and service type as a control), the interference of confounding factors on the effect evaluation is eliminated. 3. Net effect estimation logic: Net effect = (mean of the post-intervention series - mean of the pre-intervention series) - (mean of the post-intervention series in the reference group - mean of the pre-intervention series in the reference group), and calculate the effect confidence interval (confidence level ≥ 95% is considered an effective effect); finally, generate the effect assessment results containing "absolute effect value, relative improvement rate, and confidence interval".

[0078] (III) Packaging and Pushing Feedback Records 1. Feedback record structure: Adopting a standardized data format, it includes four parts: "basic instruction information + comparison data before and after intervention + net effect assessment + equipment status"; 2. Push Mechanism: Push is made through the Kafka data bus (Topic: cabin_network_feedback), using AES-256 encryption for transmission, with a message QoS level of 2 (ensuring that the message is consumed only once); after push timeout, it will automatically retry 3 times with a retry interval of 2 seconds, and if it fails, an alarm log will be recorded and synchronized to the ground operation and maintenance platform.

[0079] (iv) Data integration and online adjustment of relevant parameters 1. Data integration: After receiving feedback data, the data acquisition and association module integrates it with the original real-time data stream according to "session ID + timestamp", updates the real-time running data pool, and stores it in the onboard local database (retaining 30 days of data for traceability). 2. Online parameter adjustment: Based on the net effect assessment results, the parameters of the experience assessment and decision-making module are adaptively adjusted. The preset adjustment rules are as follows: If the net effect is positive (e.g., the experience index increases by ≥1 point) and the confidence level is ≥95%, optimize the correction coefficient for the corresponding business type in the experience evaluation model (e.g., the correction coefficient for web browsing business +0.05) and the bandwidth adjustment step size for similar scenarios in the decision module (e.g., from 1Mbps / time to 1.2Mbps / time). If the net effect is negative (e.g., latency increases by ≥200ms) or has no effect: restore the parameters before adjustment, and mark the policy scenario. Add a verification step when similar decisions are triggered in the future. If there is valid feedback regarding security commands (such as forced shutdown): update the trigger threshold of security rules (e.g., reduce the threshold for the number of ports detected by port scanning from 20 to 18 to improve response sensitivity).

[0080] Example of Structured Feedback Closed Loop Implementation Example 1: Feedback loop for bandwidth enhancement commands 1. Issuance of instructions: The decision module generates an instruction "IP10.189.45.78 (passenger B, economy class, web browsing) switch to the economy class standard group, bandwidth increase from 2Mbps to 3Mbps", instruction ID: CMD20250820144901.

[0081] 2. Multi-dimensional new data collection: Start the instant verification timer and effect tracking timer to collect data: baseline data 3 minutes before intervention (bandwidth 2Mbps, latency 800ms, packet loss rate 1.2%); data 3 minutes after intervention (bandwidth 3Mbps, latency 450ms, packet loss rate 0.8%); global network bandwidth utilization increased from 52% to 58%, and the Gini coefficient of experience index decreased from 0.18 to 0.17; gateway CPU utilization was 45%, and the execution status code was 0 (success).

[0082] 3. Contrast sequence construction and net effect estimation: Three web browsing sessions in the same cabin were selected as a reference group, with the difference between the data before and after being (bandwidth +0.1Mbps, latency -20ms). The net effect was calculated using the double difference method: the net increase in bandwidth was 0.9Mbps (1-0.1), and the net decrease in latency was 330ms (350-20). With a confidence level close to 100%, it was determined to be a valid positive effect.

[0083] 4. Feedback push and parameter adjustment: A feedback record (FB20250820145001) is generated and pushed to the data acquisition module via the Kafka bus. After data integration, parameter adjustments are triggered: the correction coefficient for "Economy Class Web Browsing Service" in the experience evaluation model is adjusted from 0 to +0.05, and the bandwidth increase step size for the "Restricted Class + Low Load" scenario in the decision module is optimized from 1Mbps to 1.1Mbps.

[0084] Example 2: Disable the feedback loop of P2P download commands 1. Issuance of instructions: The decision module generates the instruction "IP10.189.56.23 (passenger C, economy class, P2P download) binds to the economy class standard group, prohibits P2P protocol, and limits bandwidth to 1Mbps", instruction ID: CMD20250820151001.

[0085] 2. Multi-dimensional new data collection: Data collected: Before intervention (P2P download status: bandwidth 15Mbps, surrounding session latency 1200ms); After intervention (P2P connection disconnected, bandwidth 1Mbps, latency of 10 surrounding sessions restored to within 500ms); Global bandwidth utilization dropped from 75% to less than 70%, and high-priority service guarantee rate rose from over 90% to nearly 100%; AP execution status code 0 (success).

[0086] 3. Contrast sequence construction and net effect estimation: The reference group selected data from the same load period without P2P interference and calculated the net effect: the global bandwidth utilization rate was significantly reduced, the high-priority service guarantee rate was effectively improved, and the net latency of surrounding sessions was reduced by 700ms. The confidence level was close to 100%, and the effect was significant.

[0087] 4. Feedback push and parameter adjustment: After the feedback record is pushed, the decision module parameters are adjusted: the feature matching threshold for "P2P download identification" is lowered from 88% to 85% to identify and intervene in advance; at the same time, the P2P protocol interception rules for the standard economy class group are optimized, and the interception range of ports 6890-6899 is added.

[0088] Example 3: Feedback loop of forced offline command IP10.189.78.45 was forcibly taken offline due to port scanning. The feedback loop collected a net effect of "no similar attack behavior after intervention, and the network security risk level was reduced from high risk to low risk". Subsequently, security rule optimization was triggered, which shortened the detection time threshold of port scanning from 1 minute to 40 seconds, improving the security protection response speed.

[0089] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic scheduling system for cabin resources in aviation internet based on traffic-experience coupling, characterized in that, include: The data acquisition and association module collects traffic data and session status of the device IP addresses on the flight in real time, and obtains the passenger's real-name authentication information, seat information and policy group information. It dynamically binds and associates the device IP address with the passenger's identity and flight itinerary information to form structured real-time operation data. The multi-dimensional experience evaluation module receives the real-time running data and dynamically calculates the network experience index for IP sessions based on a preset experience evaluation model. The experience evaluation model comprehensively evaluates the connection quality dimension, bandwidth sufficiency dimension, and service adaptation dimension, and classifies the session into the corresponding experience level according to the evaluation results. The dynamic policy decision and generation module, based on the session experience level output by the multi-dimensional experience evaluation module, combined with the global network load and flight operation phase, dynamically makes decisions from the preset policy library and generates resource orchestration instructions, including switching policy groups for specific sessions, adjusting bandwidth parameters, or performing forced intervention. The policy execution and feedback module sends the resource orchestration instructions to the airborne network equipment for execution, and collects new session data and status information generated after policy execution. The new session data and status information are then fed back to the data acquisition and association module as input, and continuous evaluation and decision-making are performed based on the updated data.

2. The system according to claim 1, characterized in that, The dynamic binding and association specifically include: When a passenger device first accesses the onboard network and completes real-name authentication, a globally unique session identifier is generated based on the timestamp, flight identifier, and device MAC address. The session identifier, device IP address, passenger ID number, and real-time seat number obtained from the cabin service system are recorded in a dynamic association record with a version identifier and mapped to the logical topology of the flight. When a device with the same ID number is detected to reconnect or a passenger is detected to change seats, the IP and seat fields in the dynamically associated record are atomically updated by cross-comparing the onboard radio controller data with the cabin service log, and the version change is recorded.

3. The system according to claim 1, characterized in that, The real-time operational data specifically includes: The uplink traffic bytes, downlink traffic bytes, current session duration, signal strength index, and packet loss rate of each device IP address are collected at a preset sampling frequency through the airborne network equipment interface. The following information is retrieved and linked in real time from the airline's passenger service system: the valid ID number used by the passenger when boarding, the status and time of real-name authentication completed through the ID, and the physical seat number assigned to the passenger. Based on real-time data from the flight management system, the collected IP-level data and passenger identity data are bound to the currently active flight number, aircraft registration tail number, current flight segment, and flight stage to generate structured data records with spatiotemporal stamps.

4. The system according to claim 1, characterized in that, The experience evaluation model specifically includes: a core decision tree model that uses online incremental learning to learn experience mapping from real-time data streams; an outer rule engine that embeds operational rules and security constraints; and outputs a quantified network experience index and feature attribution.

5. The system according to claim 4, characterized in that, The experience evaluation model also incorporates business identification and intent inference for collaborative correction during calculation: Maintain an extensible business type feature library, including static traffic fingerprints and dynamic behavior sequences; For each IP session, streaming clustering is used to match its traffic characteristics and behavior sequences to identify the dominant business type and infer intent in real time; When calculating the network experience index, based on the identified service type and intent, corresponding benchmark parameters and multi-dimensional tolerance thresholds are dynamically selected to weight and correct the calculation results of bandwidth adequacy and connection stability factors.

6. The system according to claim 1, characterized in that, The evaluation is conducted across three dimensions: connection quality, bandwidth adequacy, and service compatibility. Specifically, the evaluation of connection quality is based on a weighted calculation of the number of abnormal session disconnections and the average signal strength. The evaluation of bandwidth adequacy is based on a comparison of the historical traffic baseline of the IP session with the real-time traffic occupancy rate. The evaluation of service compatibility is conducted by analyzing the matching degree of IP traffic feature packets with a preset service type feature library.

7. The system according to claim 1, characterized in that, The preset policy library in the dynamic policy decision and generation module contains policy groups with different names. Each policy group defines different bandwidth limits, priority queues, allowed ports or protocol sets, and security rules.

8. The system according to claim 1, characterized in that, The decision-making logic of the dynamic policy decision-making and generation module adopts a multi-objective optimization strategy search, specifically including: optimizing global experience fairness, bandwidth utilization, and high-priority service guarantee rate; when the session experience level is limited and the global load is lower than the preset value, searching and generating instructions to increase bandwidth in the Pareto solution set that meets the constraints; when P2P downloads are identified as affecting fairness, generating instructions to bind and prohibit P2P policy groups; when port scanning attacks are detected, triggering security routines to generate instructions to force offline and add to a time-limited temporary blacklist.

9. The system according to claim 1, characterized in that, When making decisions, the dynamic strategy decision-making and generation module also receives input from the flight cabin service system. The input includes at least passenger cabin class information or special service requests marked by flight attendants, and uses this as an additional weighting factor for strategy decision-making.

10. The system according to claim 1, characterized in that, The policy execution and feedback module issues instructions in the following ways: by sending configuration commands based on SNMP or a specific API interface to the airborne network gateway or wireless access point on the target flight through a secure communication protocol, and modifying the policy group affiliation or quality of service parameters of the specified IP address in real time.

11. The system according to claim 10, characterized in that, After sending the instruction, the policy execution and feedback module starts a monitoring timer to verify the parameter changes of the target IP session within a preset time and records the delay time for the policy to take effect.

12. The system according to claim 1, characterized in that, The strategy execution and feedback module implements a structured feedback loop, specifically including: After the instruction is issued, monitor and collect new data from multiple dimensions following the intervention of the target session; By constructing a pre- and post-intervention comparison sequence and using the difference method to control for confounding factors, the net effect of the strategy instruction is estimated, and a feedback record containing effect assessment is generated. The feedback records and new data are packaged and pushed back to the stream processing buffer of the data acquisition and correlation module via the data bus; After receiving the data, the data acquisition and association module integrates it into the real-time data stream and triggers online adjustment of relevant parameters in the experience evaluation and decision-making module.

Images