Dynamic self-adaptive point burying system

Abstract

The invention provides a dynamic self-adaptive point burying system, and the system comprises a point burying data collection module which is used for collecting multi-dimensional passenger data and multi-dimensional scene data in a cruise ship in real time; the data anonymous encryption module is used for carrying out desensitization processing on the multi-dimensional passenger data according to a preset authority rule and a privacy authority set by a passenger; the edge calculation processing module is used for storing multi-dimensional passenger data and multi-dimensional scene data generated by each time of navigation, constructing a ship three-dimensional coordinate mapping system based on a multi-modal acquisition matrix, connecting the multi-dimensional passenger data and the multi-dimensional scene data of each passenger in series according to a time axis, generating an ordered behavior chain, and sending the ordered behavior chain to the ship. Mapping the multi-dimensional scene data into a scene thermodynamic diagram in real time in a dimension-by-dimension manner; and the data centralized analysis module is used for respectively constructing a navigation stage identification model, a passenger behavior analysis model and a passenger demand prediction model. The resource scheduling strategy is adjusted in real time according to different navigation stages of the cruise ship and passenger behavior changes, and the operation efficiency and passenger experience are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of smart cruise ship, and particularly relates to a dynamic adaptive burying point system. BACKGROUND

[0002] In the process of digitalization of cruise ship passenger transportation industry, user behavior analysis is the core basis for optimizing service experience and improving operation efficiency, and burying point technology as the key link between user behavior and data assets can convert scattered user operations into analyzable data assets by implanting collection rules in data collection points, but traditional burying point schemes have a series of serious technical bottlenecks.

[0003] Firstly, the traditional scheme has a rigid configuration problem, which causes a long and complex demand submission, development test and version release cycle, resulting in a serious lag in business response and difficulty in adapting to rapidly changing market demands. Secondly, due to the lack of a unified data identification and management system, user data is scattered in ticketing, consumption, questionnaire and other isolated systems, forming a strong data island effect, which makes it difficult for operators to build a complete user journey map throughout the pre-trip, in-trip and post-trip, and there is a serious blind area in the analysis perspective. Thirdly, the privacy compliance risk is increasingly prominent, and most existing schemes use a "full collection, post-de-sensitization" extensive mode, which often collects sensitive information such as user biological features and travel trajectory without explicit authorization. Finally, the analysis timeliness is insufficient, and there is usually a delay of tens of hours from data collection to data analysis, completely losing the ability to intervene and make decisions in real time in sudden scenarios such as passenger flow diversion, emergency warning and fault diagnosis. SUMMARY

[0004] The purpose of the present application is to provide a dynamic adaptive burying point system, which through the cooperative operation of each module makes the system have dynamic adaptive capability, so that the resource scheduling strategy can be adjusted in real time according to different navigation stages and passenger behavior changes, and finally realizes the dual goals of improving cruise ship operation efficiency and improving passenger experience.

[0005] To achieve the above purpose, the present application provides a dynamic adaptive burying point system, which comprises: A burying point data collection module is used to form a multi-modal collection matrix by pre-setting a plurality of data collection points in a cruise ship, and to collect multi-dimensional passenger data and multi-dimensional scene data in real time. A data anonymization encryption module is used to desensitize the multi-dimensional passenger data according to the privacy permissions set by the passenger according to the pre-set permission rules. The edge computing processing module is configured to store multi-dimensional passenger data and multi-dimensional scene data generated in each voyage, construct a ship three-dimensional coordinate mapping system based on a multi-modal acquisition matrix, concatenate multi-dimensional passenger data and multi-dimensional scene data of each passenger in a time axis, generate an ordered behavior chain, and map multi-dimensional scene data into a scene heat map in real time. The data centralized analysis module is configured to construct a voyage stage identification model, a passenger behavior analysis model and a passenger demand prediction model, input multi-dimensional scene data into the voyage stage identification model to obtain a voyage stage of the cruise ship, input the voyage stage, the ordered behavior chain and the scene heat map into the passenger behavior analysis model to obtain a scene preference of passengers in each voyage stage, and input the voyage stage and the scene preference into the passenger demand prediction model to obtain a resource scheduling suggestion.

[0006] Further, the buried point data acquisition module comprises a wireless beacon and a mobile terminal, and the wireless beacon is configured to acquire multi-dimensional scene data and multi-dimensional passenger data in real time according to the movement of the mobile terminal.

[0007] Further, the multi-dimensional passenger data and the multi-dimensional scene data generated in each voyage are stored, and specifically include: Based on a continuous static detection rule, when the acquired multi-dimensional passenger data and multi-dimensional scene data are less than or equal to a preset threshold value from quantized data of a previous reserved acquisition point, only the first appearing static point is stored; Based on a state switching detection rule, the state switching point, the key maneuvering point and the stay point in the stored multi-dimensional passenger data and multi-dimensional scene data are forcibly reserved; The reserved multi-dimensional passenger data and multi-dimensional scene data are concatenated into a series of trajectory points according to a time axis, the trajectory points are regarded as a broken line, a deviation threshold value is set for each straight line in the broken line, and points greater than the deviation threshold value are deleted; The remaining multi-dimensional passenger data and multi-dimensional scene data are compressed to generate a compressed trajectory package, and then stored offline.

[0008] Further, the ship three-dimensional coordinate mapping system is constructed based on the multi-modal acquisition matrix, and specifically includes: The cruise ship as a whole is periodically scanned to generate and output a ship three-dimensional point cloud model based on a ship main coordinate system; Based on the ship main coordinate system, a corresponding coordinate of each data acquisition point in the multi-modal acquisition matrix is calibrated, and a corresponding three-dimensional coordinate of each passenger is generated; The acquired multi-dimensional passenger data and multi-dimensional scene data are respectively assigned to a passenger anonymous ID, a scene space ID and a time slice ID, and matched with a three-dimensional space region of the ship three-dimensional point cloud model.

[0009] Further, the ordered behavior chain is generated, and specifically includes: Continuously acquire multi-dimensional passenger data and multi-dimensional scene data of the same passenger in the whole journey, and assign the passenger anonymous ID, scene space ID and time slice ID to each data to form a unified space-time index; Based on the unified space-time index, the multi-dimensional passenger data and multi-dimensional scene data are mapped into atomic events with event types in real time according to a preset rule; The atomic events are sorted according to a preset rule with the passenger anonymous ID as the primary key and the time slice ID as the sorting key, and an ordered behavior chain of the passenger is generated.

[0010] Further, the multi-dimensional scene data is mapped into a scene heat map in each dimension, specifically including: The data of each dimension of the multi-dimensional scene data is formed into an independent data stream according to the dimension category; Taking the unified space grid divided by the three-dimensional point cloud model of the ship as the reference, the scene data in each dimension data stream is aggregated in a set time window to the corresponding grid in real time to generate the current dimension value and its weight of each grid; After Gaussian weight convolution smoothing processing of the aggregated grid value, the smoothed grid value is mapped to a predefined color scale, and the transparency is calculated in real time according to the weight to generate the scene heat map layer corresponding to the dimension.

[0011] Further, the multi-dimensional scene data is input into a navigation stage recognition model to obtain the navigation stage of the cruise ship, specifically including: From the real-time collected multi-dimensional scene data, feature dimensions related to the navigation stage of the ship are extracted, which at least include the position, attitude, environmental condition and mechanical state of the ship; In a set time window, the statistics and frequency domain features of the feature dimensions are calculated to generate a fixed feature vector of fixed dimensions; The fixed feature vector is input into a pre-trained navigation stage recognition model to output the category label and confidence of the current navigation stage in real time; When the confidence is higher than a preset threshold, the current navigation stage is directly output, and when the confidence is between the preset threshold interval, the final navigation stage is determined through a time window voting mechanism.

[0012] Further, the navigation stage, ordered behavior chain and scene heat map are input into a passenger behavior analysis model to obtain the scene preference of each passenger in each navigation stage, specifically including: According to the real-time output navigation stage, the ordered behavior chain of each passenger and the scene heat map of the ship are divided into data segments corresponding to each navigation stage on the time axis; The behavior sequence features of the ordered behavior chain and the spatial heat features of the scene heat map in each data segment are extracted, and the context features are spliced to form a multi-modal feature vector of the corresponding stage; The multi-modal feature vector is input into a passenger behavior analysis model to output a scene preference probability distribution of the passenger in each voyage phase.

[0013] Further, the voyage phase and scene preference are input into a passenger demand prediction model to obtain resource scheduling suggestions, specifically including: The real-time voyage phase, scene preference probability distribution, historical demand curve, and additional variables are spliced in real time to form multi-modal prediction features; The multi-modal prediction features are input into a passenger demand prediction model to output demand prediction values and corresponding prediction interval values of each scene in a future set time window; The resource surplus of each scene is obtained and compared with the demand prediction values to obtain a resource gap vector; Real-time solving is performed under the constraints of current ship schedulable crew, inventory transfer time, and safe load to generate resource scheduling suggestions including personnel scheduling, inventory replenishment, and passenger flow guiding actions.

[0014] Compared with the prior art, the beneficial effects of the present application are: The dynamic adaptive burying point system provided by the present application provides comprehensive and three-dimensional basic data support for the system by real-time collection of multi-dimensional passenger data and multi-dimensional scene data through a multi-modal collection matrix; the data anonymization encryption module protects passenger data through privacy permission rules and desensitization processing, effectively avoids passenger privacy leakage risks, enhances passenger trust in cruise service, and improves user experience; the edge computing processing module realizes local data storage, three-dimensional coordinate mapping, ordered behavior chain generation, and real-time heat map mapping through edge computing, improving the spatio-temporal correlation and real-time of data processing; the data centralized analysis module realizes voyage phase identification, behavior preference analysis, and demand prediction through three models, driving the intelligentization and precision of cruise operation. The present application realizes the dual goals of improving cruise operation efficiency and passenger experience by the collaborative operation of the modules, so as to adjust the resource scheduling strategy in real time according to different voyage phases and passenger behavior changes, and ultimately realize the dual goals of improving cruise operation efficiency and passenger experience. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without creating any inventive labor.

[0016] Figure 1 A dynamic adaptive burying point system structure schematic diagram provided by the present application embodiment; Figure 2 A flowchart of a process for storing multi-dimensional passenger data and multi-dimensional scene data generated by each voyage is provided for an embodiment of the present application. Figure 3 A flowchart of a process for constructing a ship three-dimensional coordinate mapping system based on a multi-modal acquisition matrix is provided for an embodiment of the present application. Figure 4 A flowchart of a process for generating an ordered behavior chain is provided for an embodiment of the present application. Figure 5 A flowchart of a process for mapping multi-dimensional scene data into a scene heat map dimension by dimension is provided for an embodiment of the present application. Figure 6 A flowchart of a process for inputting multi-dimensional scene data into a voyage phase recognition model to obtain a cruise ship voyage phase is provided for an embodiment of the present application. Figure 7 A flowchart of a process for inputting a voyage phase, an ordered behavior chain, and a scene heat map into a passenger behavior analysis model to obtain a scene preference of a passenger in each voyage phase is provided for an embodiment of the present application. Figure 8 A flowchart of a process for inputting a voyage phase and a scene preference into a passenger demand prediction model to obtain a resource scheduling suggestion is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0017] The present application will be further described below in conjunction with the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present application, but not to limit the present application. In addition, it should be noted that, for the convenience of description, only the parts related to the present application are shown in the drawings, but not all the structures.

[0018] With reference to Figure 1 The present embodiment provides a dynamic and adaptive burying point system, which comprises: A burying point data acquisition module is configured to form a multi-modal acquisition matrix by presetting a plurality of data acquisition points in a cruise ship, and to acquire multi-dimensional passenger data and multi-dimensional scene data in real time.

[0019] A data anonymization and encryption module is configured to perform desensitization processing on the multi-dimensional passenger data according to a privacy permission set by a passenger and a preset permission rule.

[0020] An edge computing processing module is configured to store multi-dimensional passenger data and multi-dimensional scene data generated by each voyage, to construct a ship three-dimensional coordinate mapping system based on a multi-modal acquisition matrix, to generate an ordered behavior chain by concatenating multi-dimensional passenger data and multi-dimensional scene data of each passenger along a time axis, and to map multi-dimensional scene data into a scene heat map dimension by dimension in real time.

[0021] The data centralized analysis module is used for respectively constructing a navigation stage recognition model, a passenger behavior analysis model and a passenger demand prediction model, inputting multi-dimensional scene data into the navigation stage recognition model to obtain a navigation stage of the cruise ship, inputting the navigation stage, an ordered behavior chain and a scene heat map into the passenger behavior analysis model to obtain a scene preference of passengers in each navigation stage, and inputting the navigation stage and the scene preference into the passenger demand prediction model to obtain resource scheduling suggestions.

[0022] In the embodiment, the buried point data acquisition module serves as a data entrance, captures multi-dimensional passenger data (such as behavior trajectory, interactive operation, etc.) and multi-dimensional scene data (such as regional passenger flow, facility state, etc.) in a multi-modal acquisition matrix of the cruise ship in real time, and synchronously transmits the data to the data anonymous encryption module. The data anonymous encryption module performs instant desensitization and encryption processing on passenger data according to preset permission rules and privacy permissions set by passengers, thereby guaranteeing data compliance and providing safe basic data for subsequent analysis. The processed encrypted data is transmitted to the edge computing processing module.

[0023] The preset permission rules are specifically based on a compliance strategy automatic switching mechanism (GDPR / CCPA, etc.) of a real-time geographic fence of the ship, and the original trajectory and identity information are blurred and desensitized by applying K-anonymity technology at the network edge layer, so as to realize dynamic privacy risk prevention and control and compliant data flow in a cross-country navigation scene. That is, an intelligent mechanism is designed to determine whether the ship enters a specific national territorial sea or jurisdictional area (such as the European Union and California, USA) according to the real-time geographic position (based on GPS / Beidou), and when the ship enters the corresponding area, the preset data acquisition compliance strategy (such as the "consent management" switch of GDPR and the "no sale of personal information" flag setting of CCPA) is automatically activated to adjust the data acquisition range, field or trigger a secondary confirmation process of the user. At the same time, dynamic anonymization technology (such as the K-anonymity model) is applied to ensure that a specific individual cannot be accurately located in the published user trajectory or behavior data (for example, the area stay information is blurred to the aggregated number of people in a certain functional area in a certain period of time, rather than a single user path), and the direct identifier (name, ID number) is strongly encrypted or desensitized.

[0024] The privacy permissions set by the passengers include layered pre-configuration before boarding (distinguishing between basic necessary and optional extended permissions), real-time management of the mobile terminal after boarding (including scenario-based switches, use range and life cycle settings), single temporary authorization of the terminal in the public area, and instant notification and data traceability function of permission change, thereby constructing a multi-channel and adjustable privacy control system. The passengers can clearly understand the data use and flexibly define the privacy boundary, and the system can obtain necessary data under the premise of compliance, thereby achieving a balance between privacy protection and service experience.

[0025] After receiving encrypted data, the edge computing processing module stores all data generated during each voyage. Simultaneously, it constructs a three-dimensional coordinate mapping system for the ship based on a multimodal acquisition matrix, linking the multidimensional data of each passenger into an ordered behavioral chain along the timeline. At the same time, it transforms scene data into a dynamic scene heatmap in real time. This integrated spatiotemporal correlated data (ordered behavioral chains, heatmaps) is synchronized in real time to the data central analysis module, providing it with refined input bearing spatiotemporal labels.

[0026] The centralized data analysis module first processes the scene data using a navigation phase identification model to determine the cruise ship's current navigation phase (e.g., departure, sea voyage). Then, it inputs the navigation phase, ordered behavior chain, and scene heatmap into a passenger behavior analysis model to accurately extract passenger scenario preferences for different phases. Finally, combining the navigation phase and scenario preferences, a passenger demand prediction model generates resource scheduling suggestions. These suggestions, in turn, guide the dynamic adjustment of the data collection module (e.g., optimizing collection point density in high-preference areas), forming an adaptive collaborative loop that enables the system to respond to business needs in real time and continuously optimize.

[0027] In a preferred embodiment, the embedded data acquisition module includes a wireless beacon and a mobile terminal. The wireless beacon collects multi-dimensional scene data and multi-dimensional passenger data in real time based on the movement of the mobile terminal.

[0028] In this embodiment, low-power wireless beacons (such as Bluetooth or UWB beacons) are deployed in various areas of the cruise ship (such as cabins, restaurants, decks, entertainment facilities, etc.). Each beacon has a preset unique spatial coordinate identifier (e.g., east entrance of the main restaurant on deck L4). When a passenger carrying a mobile terminal (phone, smart bracelet, etc.) logged into the cruise ship's app enters the beacon's signal coverage area, the terminal automatically establishes a low-power connection with the beacon. The beacon calculates the terminal's location in real time by receiving the RSSI (signal strength index) value sent by the terminal, and generates the passenger's three-dimensional spatial trajectory (location information in multi-dimensional passenger data) by combining the beacon's own coordinates. At the same time, the beacon periodically scans the number of surrounding terminals and the frequency of signal changes to count the real-time passenger flow density and dwell time in specific areas (human flow characteristics in multi-dimensional scene data).

[0029] Passengers' smart devices (especially smartphones or dedicated wristbands equipped with the ship's app) are not only used for scanning beacons for positioning, but their built-in sensors (such as accelerometers and gyroscopes) can also capture basic motion states (walking, standing still). Mobile terminals simultaneously collect data on passengers' active actions (such as in-app service reservations, payments, and event registrations), and transmit this data back to the system via beacons. This forms a collaborative data collection model of "beacon-sensed spatial behavior and terminal-recorded active behavior," enabling both seamless collection of passenger dynamic trajectories and regional preferences, and precise capture of their explicit needs, together forming a complete multi-dimensional data collection matrix.

[0030] Extend the capability of the buried point to the environmental perception device, deploy the passenger flow counter (such as based on visual or thermal imaging technology) at the entrance of the restaurant, install RFID or sensor to detect the use state at the entertainment facilities (such as massage chair, VR device), add contact sensor to the hatch to judge the opening and closing state (which can be used to identify the clustering behavior), deploy visual analysis camera at important passageways (for non-identifiable crowd analysis), etc. Establish real-time or quasi-real-time data channel with the core system of the ship, interface with ticketing system (passenger basic information, cabin information, travel package), point of sale (POS) system (dining, shopping, service consumption record), on-board reservation system (activities, shore excursion reservation), navigation system (ship position, navigation state), access control system (boarding and disembarking records), etc. to build a panoramic view of the user journey.

[0031] As a preferred embodiment, store the multi-dimensional passenger data and multi-dimensional scene data generated in each voyage, including: Based on the continuous stationary detection rule, when the collected multi-dimensional passenger data and multi-dimensional scene data are less than or equal to the preset threshold value compared with the quantitative data of the previous reserved collection point, only store the first appearing stationary point.

[0032] Based on the state switching detection rule, force to reserve the state switching point, key maneuvering point and stay point in the multi-dimensional passenger data and multi-dimensional scene data.

[0033] According to the time axis, connect the reserved multi-dimensional passenger data and multi-dimensional scene data into a series of trajectory points. Treat the trajectory points as a polyline, set a deviation threshold value for each straight line in the polyline, and delete the points greater than the deviation threshold value.

[0034] Compress the remaining multi-dimensional passenger data and multi-dimensional scene data, generate a compressed trajectory package, and then store it offline.

[0035] In this embodiment, to solve the problems of instability, high delay and high cost of offshore satellite network, an edge computing node / gateway is deployed locally on the cruise ship. The node has strong data caching capability (usually supporting 7 days or complete voyage offline storage), and locally pre-processes the raw behavior data when network conditions allow, applies intelligent compression algorithms, removes redundant and low-value location data points (such as repeated reporting in continuous stationary state), and only keeps key mobile path nodes, state switching points and stay point information. When the satellite network conditions permit (or through strategic timing / quantity synchronization), the edge node transmits the pre-processed, compressed and anonymized data packets to the cloud data center through an optimized protocol (efficient compression, breakpoint continuation). The cloud builds a powerful centralized analysis platform, which collects data from the entire fleet and multiple routes. It realizes long-term behavior trend analysis across voyages and ships (such as seasonal preference changes), more complex model training and optimization (such as personalized recommendation engines based on machine learning), and global perspective operational insights and strategic decision support (such as route optimization, facility renovation evaluation, and overall pricing strategy).

[0036] As a preferred embodiment, a ship three-dimensional coordinate mapping system is constructed based on a multi-modal acquisition matrix, specifically including: Periodically scan the entire cruise ship to generate and output a ship three-dimensional point cloud model based on the ship's main coordinate system.

[0037] Based on the ship's main coordinate system, calibrate the coordinates of each data acquisition point in the multi-modal acquisition matrix to generate corresponding three-dimensional coordinates for each passenger.

[0038] Assign the collected multi-dimensional passenger data and multi-dimensional scene data to passenger anonymous ID, scene space ID and time slice ID respectively, and match them with the three-dimensional space region of the ship three-dimensional point cloud model.

[0039] In this embodiment, a precise three-dimensional spatial topology model is established for the entire cruise ship. Each physical location (dining room, bar, theater seating area, cabin floor corridor, deck viewing area) is assigned a unique logical coordinate code (for example: coordinate system: L5-Bar-Aft-Port-Section2). All collected location data and event data (consumption, activity participation) are mapped to this logical coordinate system.

[0040] As a preferred embodiment, the generation of an ordered behavior chain includes: Continuously acquire multi-dimensional passenger data and multi-dimensional scene data of the same passenger throughout the entire voyage, and assign each data with passenger anonymous ID, scene space ID and time slice ID to form a unified space-time index.

[0041] Based on the unified space-time index, the multi-dimensional passenger data and the multi-dimensional scene data are mapped into atomic events with event types in real time according to preset rules.

[0042] The atomic events are sorted according to preset rules with the passenger anonymous ID as the primary key and the time slice ID as the sorting key, to generate the ordered behavior chain of the passenger.

[0043] In this embodiment, within the scope of the ship, the beacon trajectory, consumption events, cabin door opening and closing events, etc. are preliminarily associated based on the timestamp and ID, and data format standardization and basic cleaning are performed. By assigning the passenger anonymous ID, scene space ID and time slice ID to the multi-dimensional passenger data and the multi-dimensional scene data, a unified space-time index is constructed, the ternary relationship of "person-time-space" is accurately associated, and the problem of scattered and isolated original data is solved; on this basis, the data are mapped into atomic events according to the rules, which can convert scattered data into readable behavior units, eliminate information ambiguity and realize structured processing; further, the ordered behavior chain is generated with the anonymous ID as the primary key and the time slice ID as the sorting key, which can completely present the timing and continuity of passenger behavior. It can not only mine passenger preferences through the behavior chain to support personalized services, but also analyze group behavior rules to optimize scene resource allocation and improve operational efficiency, and identify abnormal behavior chains to strengthen security risk early warning, while balancing data value mining and privacy protection through anonymous ID, and finally convert fragmented data into the core ability to drive the fine management, experience upgrading and risk prevention of cruise ships.

[0044] As a preferred embodiment, the multi-dimensional scene data is mapped into a scene heat map dimension by dimension, specifically including: The data of each dimension of the multi-dimensional scene data is formed into an independent data stream according to the dimension category.

[0045] Taking the unified space grid divided by the three-dimensional point cloud model of the ship as the reference, the scene data in each dimension data stream is aggregated to the corresponding grid in the set time window in real time to generate the current dimension value and its weight of each grid.

[0046] After the aggregated grid value is subjected to Gaussian weight convolution smoothing processing, the smoothed grid value is mapped to a predefined color scale, and the transparency is calculated in real time according to the weight to generate the scene heat map layer corresponding to this dimension.

[0047] In this embodiment, first, by splitting the multi-dimensional scene data into independent data streams according to dimensions (such as people flow density, environmental temperature, facility utilization rate, etc.), the mutual interference of different dimensional data is avoided, and the operator can analyze the spatial distribution characteristics of a single scene element (such as separately viewing “pool area water temperature distribution” or “restaurant seat utilization rate”) and accurately locate the problem or advantage area of each dimension. Secondly, the data is aggregated based on the unified spatial grid of the three-dimensional point cloud model of the ship, ensuring that the scene data of different dimensions are aligned in the same spatial coordinate system, and then cross-dimension comparison and analysis is realized, for example, intuitively associating “people flow heat map of a certain area” with “Wi-Fi signal strength heat map of the area” to quickly determine whether the network congestion is caused by dense people flow.

[0048] Thirdly, real-time aggregation within a time window supports dynamic tracking of scene changes (such as real-time updating of the heat distribution of people flow on the deck in a certain period), which facilitates the timely discovery of sudden situations (such as sudden congestion in a certain passage); and Gaussian weight convolution smoothing processing can filter data noise (such as isolated abnormal values or equipment errors), making the heat map more consistent with the real scene trend and avoiding misjudgment caused by local fluctuations. Finally, by mapping the data intensity with color scale and adjusting the transparency with weight, the abstract scene data is converted into intuitive visual heat map layers (such as highlighting high people flow areas in red and marking low temperature areas in blue), reducing the interpretation threshold of the operator. In combination with the heat map, key information such as “spatial distribution of high-frequency use facilities”, “location of environmental abnormal areas”, and “aggregation point of people flow peak” can be quickly identified, providing precise data support for resource scheduling (such as increasing the number of service personnel in hot areas), facility optimization (such as adjusting air conditioners in low temperature areas), and safety control (such as dredging congested passages), thereby improving the efficiency and pertinence of scene management.

[0049] As a preferred embodiment, the multi-dimensional scene data is input into a navigation phase recognition model to obtain the navigation phase of the cruise ship, specifically including: From the real-time collected multi-dimensional scene data, feature dimensions related to the navigation phase of the ship are extracted, including at least the position, attitude, environmental conditions, and mechanical state of the ship.

[0050] Within a set time window, statistical quantities and frequency domain features of the feature dimensions are calculated to generate fixed-dimensional fixed feature vectors.

[0051] The fixed feature vectors are input into a pre-trained navigation phase recognition model to output the category label and confidence of the current navigation phase in real time.

[0052] When the confidence is higher than a preset threshold, the current navigation phase is directly output; when the confidence is within a preset threshold interval, the final navigation phase is determined through a time window voting mechanism.

[0053] In this embodiment, by focusing on core features such as ship position, attitude, environmental conditions and mechanical state, the key information strongly related to the sailing stage is directly locked, irrelevant data interference is avoided, high-quality input is provided for the model, and identification errors are reduced from the source. The statistical quantity and frequency domain feature of the feature is calculated within a set time window, and the instantaneous data is converted into a description of the trend of the scene within a period of time. For example, the speed during the port approach stage will continue to decrease, and the distance from the shore will approach 0. The trend feature is more easily identified by the model after aggregation through the time window, avoiding the influence of single-point data anomalies on judgment, and enhancing the robustness of identification.

[0054] The confidence of the model output can intuitively reflect the reliability of the identification result. When the confidence is high, it is directly output to ensure efficient decision-making in normal scenarios. When the confidence is in the middle interval, the time window voting mechanism is used to offset instantaneous errors to avoid misjudgment caused by a single low-confidence result. For example, a short-term storm may cause the model's confidence in offshore cruising to decrease, but most of the time within the window is still determined to be cruising, and the correct stage is finally confirmed through voting, significantly reducing the risk of misjudgment. The entire process from feature extraction to stage identification is real-time processing, which can dynamically track the switching of the sailing stage and provide immediate feedback for cruise ship operations. For example, when the port approach stage is identified, the processes of preparing for boarding and disembarking, luggage handling scheduling, etc. are triggered; when the offshore sailing stage is identified, the security level or entertainment facility opening strategy is adjusted accordingly, realizing fine and dynamic management based on the sailing stage, and improving the operation efficiency and safety guarantee capability.

[0055] Specifically, the key stages of cruise ship sailing include: Embarkation, starting from the first scanning of tickets / certificates at the dock, to the window of ship departure; Sailing At Sea, the sailing stage from ship departure to arrival at the next destination port; Port Visit / Docked, the period during which the ship is docked at the port; Onshore, the period when passengers pass through the gate and go ashore for onshore activities; Pre-Disembarkation, the last day of the voyage or the return journey; Disembarkation, when passengers disembark and end the voyage; this stage is automatically triggered based on ship navigation system signals (departure / arrival), gate access records, reservation system status (service end), etc.

[0056] As a preferred embodiment, the sailing stage, the ordered behavior chain and the scene heat map are input into the passenger behavior analysis model to obtain the scene preference of each passenger in each sailing stage, which specifically includes: According to the real-time output of the sailing stage, the ordered behavior chain of each passenger and the scene heat map of the ship are divided into data segments corresponding to each sailing stage on the time axis.

[0057] The behavior sequence features of the ordered behavior chain and the spatial heat features of the scene heat map in each data segment are extracted, and context features are spliced to form a multi-modal feature vector corresponding to the stage.

[0058] The multi-modal feature vector is input into a passenger behavior analysis model to output a scene preference probability distribution of passengers in each navigation stage.

[0059] In this embodiment, by dividing the data on the time axis according to the navigation stage, the passenger behavior can be deeply bound to the specific navigation scene. The passenger demand in different stages is different, the divided data segment can focus on the real behavior of passengers in a certain stage, avoid the preference misjudgment caused by the mixing of cross-stage data, and make the preference analysis more in line with the characteristics of the stage scene. The ordered behavior chain provides individual-level behavior sequence features, and the scene heat map provides group-level spatial heat features. The multi-modal feature vector formed by splicing the two can capture the uniqueness of individual behavior and reflect the common trend of group behavior.

[0060] Multi-dimensional perspective avoids one-sidedness of a single data dimension, making preference description more three-dimensional and more accurate. The scene preference probability distribution of passengers in each navigation stage output by the model provides a quantifiable basis for operational decision-making. The cruise ship can dynamically optimize resource allocation according to different stages: increase the number of service personnel on the viewing platform and supplies during the port stage; adjust the opening time of the restaurant and the type of food during the cruise stage; and focus on ensuring the operation efficiency of entertainment facilities during the open sea stage. Precise matching of passengers' core needs in different scenes not only improves passenger experience satisfaction, but also reduces waste caused by resource mismatch, ultimately achieving cost reduction and efficiency increase of fine operation.

[0061] As a preferred embodiment, the navigation stage and scene preference are input into a passenger demand prediction model to obtain resource scheduling suggestions, specifically including: Real-time navigation stage, scene preference probability distribution, historical demand curve, and additional variables are spliced in real time to form a multi-modal prediction feature.

[0062] The multi-modal prediction feature is input into a passenger demand prediction model to output demand prediction values and corresponding prediction interval values for each scene in a future set time window.

[0063] The resource surplus of each scene is obtained and compared with the demand prediction value to obtain a resource gap vector.

[0064] With the goal of minimizing the shortage penalty and excessive configuration cost, real-time solving is performed under the constraints of current ship schedulable crew, inventory transfer time, and safety load to generate resource scheduling suggestions including personnel scheduling, inventory replenishment, and passenger flow guidance actions.

[0065] In this embodiment, by comparing the current resource reserves of each scene with the demand prediction value, the generated resource gap vector directly locates the specific scene of resource supply and demand imbalance and the gap size, avoiding the problem of fuzzy allocation by experience in traditional scheduling. For example, if it is predicted that the demand for the viewing platform will surge during the port approach stage, the gap vector can directly point to the resource gap of the sunshades and rest seats of the viewing platform. The operator does not need to blindly allocate all ship resources, but can accurately supplement the shortage of resources in the target scene, thereby improving the scheduling efficiency. The model takes minimizing the shortage penalty and over-provisioning cost as the goal, and simultaneously takes into account the actual constraints of the ship to solve in real time, so that the optimal balance between cost and service can be achieved under the premise of limited resources. For example, if the demand for the theater increases during the open sea navigation stage, but there is not enough adjustable crew, the model will prioritize the core services rather than blindly increase the non-essential positions, thereby controlling the cost and ensuring the service experience of the key scenes.

[0066] The entire process from feature splicing, demand prediction to scheduling suggestion is real-time processing, which can dynamically respond to the switching of the navigation stage and changes in demand. For example, when the navigation stage changes from "leaving the port" to "open sea cruising", and the scene preference changes from "boarding passage" to "dining room and shop", the model can quickly update the demand prediction and generate a real-time suggestion for scheduling personnel from the boarding area to the dining room. If a sudden rainstorm causes the demand for the outdoor swimming pool to drop sharply, the model can immediately adjust the prediction and reduce the resource allocation of the swimming pool area, and transfer the manpower to the indoor entertainment area. Dynamic adaptation ensures that resource scheduling is always consistent with passenger demand, avoiding waste or shortage caused by resource solidification, and ultimately achieving fine and flexible management of the entire voyage service, thereby reducing operating costs and significantly improving passenger experience and satisfaction.

[0067] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A dynamically adaptive data embedding system, characterized in that, The system includes: The embedded data acquisition module is used to pre-set several data acquisition points in the cruise ship to form a multimodal acquisition matrix, and to collect multi-dimensional passenger data and multi-dimensional scene data in real time. The data anonymization and encryption module is used to desensitize multi-dimensional passenger data according to preset permission rules and passenger privacy settings; The edge computing processing module is used to store multi-dimensional passenger data and multi-dimensional scene data generated during each voyage. It constructs a ship three-dimensional coordinate mapping system based on a multi-modal acquisition matrix, and links the multi-dimensional passenger data and multi-dimensional scene data of each passenger along the time axis to generate an ordered behavior chain. It also maps the multi-dimensional scene data into a scene heat map in real time. The data central analysis module is used to construct a navigation phase identification model, a passenger behavior analysis model, and a passenger demand prediction model, respectively. Multi-dimensional scene data is input into the navigation phase identification model to obtain the navigation phase of the cruise ship. The navigation phase, ordered behavior chain, and scene heat map are input into the passenger behavior analysis model to obtain the scene preferences of passengers in each navigation phase. The navigation phase and scene preferences are input into the passenger demand prediction model to obtain resource scheduling suggestions.

2. The dynamically adaptive embedding system according to claim 1, characterized in that, The embedded data acquisition module includes a wireless beacon and a mobile terminal. The wireless beacon collects multi-dimensional scene data and multi-dimensional passenger data in real time based on the movement of the mobile terminal.

3. The dynamically adaptive embedding system according to claim 1, characterized in that, It stores multi-dimensional passenger data and multi-dimensional scene data generated from each voyage, specifically including: Based on the continuous stationary detection rule, when the collected multidimensional passenger data and multidimensional scene data are less than or equal to the preset threshold compared with the quantized data of the previous retained collection point, only the first stationary point is stored. Based on state transition detection rules, state transition points, key maneuver points, and dwell points in multi-dimensional passenger data and multi-dimensional scene data are forcibly retained and stored. The retained multi-dimensional passenger data and multi-dimensional scene data are linked together into a series of trajectory points according to the time axis. The trajectory points are regarded as a broken line. A deviation threshold is set for each straight line segment in the broken line. If the deviation threshold is greater than the deviation threshold, the point is deleted. The remaining multidimensional passenger data and multidimensional scene data are compressed to generate compressed trajectory packages, which are then stored offline.

4. The dynamically adaptive embedding system according to claim 1, characterized in that, A ship three-dimensional coordinate mapping system is constructed based on a multimodal acquisition matrix, specifically including: The entire cruise ship is periodically scanned to generate and output a three-dimensional point cloud model of the ship based on the ship's main coordinate system; Using the ship's main coordinate system as a reference, the coordinates of each data acquisition point in the multimodal acquisition matrix are calibrated to generate corresponding three-dimensional coordinates for each passenger. The collected multidimensional passenger data and multidimensional scene data are assigned passenger anonymous ID, scene space ID, and time slice ID, respectively, and then matched with the three-dimensional spatial region of the ship's three-dimensional point cloud model.

5. The dynamically adaptive embedding system according to claim 1, characterized in that, The generation of ordered behavior chains specifically includes: Continuously acquire multi-dimensional passenger data and multi-dimensional scene data of the same passenger throughout the entire flight, and assign each data point an anonymous passenger ID, scene space ID, and time slice ID to form a unified spatiotemporal index; Based on a unified spatiotemporal index, multidimensional passenger data and multidimensional scene data are mapped in real time into atomic events with event types according to preset rules; Using the passenger's anonymous ID as the primary key and the time slice ID as the sorting key, atomic events are sorted according to preset rules to generate an ordered chain of passenger behavior.

6. The dynamically adaptive embedding system according to claim 1, characterized in that, Mapping multidimensional scene data dimension by dimension into scene heatmaps, specifically including: The data for each dimension of the multi-dimensional scene data is divided into independent data streams according to the dimension category; Based on the unified spatial grid defined by the ship's 3D point cloud model, the scene data in each dimension data stream is aggregated to the corresponding grid in real time within a set time window, generating the current dimension value and its weight for each grid. After smoothing the aggregated raster values ​​using Gaussian weighted convolution, the smoothed raster values ​​are mapped to predefined color levels, and the transparency is calculated in real time according to the weights to generate the scene heatmap layer corresponding to that dimension.

7. The dynamically adaptive embedding system according to claim 1, characterized in that, Multi-dimensional scene data is input into the navigation phase recognition model to obtain the navigation phase of the cruise ship, specifically including: From real-time collected multi-dimensional scene data, feature dimensions related to the ship's navigation stage are extracted. These feature dimensions include at least the ship's position, attitude, environmental conditions, and mechanical state. Within a set time window, calculate the statistics and frequency domain features of the feature dimensions to generate a fixed feature vector of fixed dimensions. Input a fixed feature vector into a pre-trained navigation phase recognition model, and output the category label and confidence level of the current navigation phase in real time; When the confidence level is higher than the preset threshold, the current navigation stage is output directly. When the confidence level is within the preset threshold range, the final navigation stage is determined through a time window voting mechanism.

8. The dynamically adaptive embedding system according to claim 1, characterized in that, By inputting the flight phase, ordered behavior chain, and scenario heatmap into the passenger behavior analysis model, the scenario preferences of passengers in each flight phase are obtained, specifically including: Based on the real-time output of the navigation phase, the orderly behavior chain of each passenger and the scene heat map of the ship are divided into data segments corresponding to each navigation phase on the time axis. Extract the behavioral sequence features of the ordered behavioral chain and the spatial thermal features of the scene heatmap within each data segment, and then concatenate the context features to form a multimodal feature vector for the corresponding stage. The multimodal feature vectors are input into the passenger behavior analysis model, which outputs the probability distribution of passenger scenario preferences for each flight phase.

9. The dynamically adaptive embedding system according to claim 1, characterized in that, By inputting the flight phase and scenario preferences into the passenger demand prediction model, resource scheduling suggestions are obtained, specifically including: The real-time navigation phase, scenario preference probability distribution, historical demand curve, and additional variables are spliced ​​together in real time to form multimodal prediction features; Input the multimodal prediction features into the passenger demand prediction model, and output the demand prediction values ​​and corresponding prediction interval values ​​for each scenario within a future set time window; Obtain the current resource reserves for each scenario and compare them with the predicted demand values ​​to obtain a resource gap vector; With the goal of minimizing shortage penalties and over-configuration costs, the system performs real-time solutions under constraints of available crew, inventory transfer time, and safety load, generating resource scheduling recommendations that include personnel scheduling, inventory replenishment, and passenger flow guidance.

Images