A high-precision positioning system and method for a subway station based on a 5G public and private network

Abstract

The application provides a subway station high-precision positioning system and method based on 5G public and private networks, and belongs to the field of wireless communication and indoor positioning technology.The method comprises the following steps: performing multi-dimensional signal coverage analysis on the area in the subway station to generate a public network and private network signal strength distribution atlas; deploying a positioning terminal supporting 5G public network and private network dual-mode access according to the atlas, constructing a hybrid positioning network cooperated by the public network and the private network, and synchronously collecting double-network signal fingerprint feature data; and through 5G public network and private network dual-mode access and signal fingerprint dynamic fusion, the positioning precision and coverage range in the subway station are significantly improved, low-cost high-precision positioning can be realized in the public network coverage area, and data security and global consistency can be ensured in the private network deployment scene.

Description

Technical Field

[0001] This invention proposes a high-precision positioning system and method for subway stations based on 5G public and private networks, belonging to the field of wireless communication and indoor positioning technology. Background Technology

[0002] Against the backdrop of accelerated urbanization and rapid expansion of subway networks, accurate positioning of people and goods within subway stations has become a core requirement for smart subway operations. However, existing 5G indoor positioning technology still faces significant limitations: on the one hand, insufficient deployment density of public network base stations leads to fluctuations in positioning accuracy; while private network construction can improve performance, it faces high costs and global calibration challenges; the lack of a public-private network collaboration mechanism further restricts the universality of positioning services. On the other hand, the complex architectural structure within subway stations causes severe multipath effects and non-line-of-sight propagation; traditional UTDOA / RTT algorithms are insufficiently adaptable to this environment, and positioning results are prone to jumps and drifts, making it difficult to meet the needs of high-precision scenarios. Even more serious is the fact that when positioning terminals enter signal blind spots such as equipment rooms or deep tunnels, the current technology's complete reliance on signal transmission leads to service interruptions, lacking effective auxiliary positioning methods. Although some solutions attempt to optimize performance by reusing 5G networks or deploying edge computing, none have resolved the core contradictions of public-private network resource integration, multipath interference suppression, and service continuity assurance. Therefore, there is an urgent need for a new positioning method that deeply integrates the signal characteristics of public and private networks, has the ability to resist multipath interference, and can maintain service continuity, so as to provide subway stations with seamless, high-precision location awareness support. Summary of the Invention

[0003] This invention provides a high-precision positioning system and method for subway stations based on 5G public and private networks, to solve the problems mentioned in the background section above: Example 1

[0004] This invention proposes a high-precision positioning method for subway stations based on 5G public and private networks, the method comprising: S1. Conduct multi-dimensional signal coverage analysis of the area within the subway station to generate a signal strength distribution map of the public network and private network; deploy positioning terminals that support dual-mode access of 5G public network and private network based on the map, construct a hybrid positioning network that coordinates public and private networks, and simultaneously collect dual-network signal fingerprint feature data; S2. Upload the dual-network signal fingerprint feature data to the edge computing (MEC) platform, and construct a three-dimensional spatial signal propagation model of the subway station through digital twin technology; based on the model, perform multipath pre-compensation processing on the original signal to generate enhanced signal data that is resistant to multipath interference; S3. Using enhanced signal data, combined with a dynamically weighted UTDOA / RTT hybrid positioning algorithm, an initial position is calculated to generate preliminary positioning results. At the same time, external context information, including video surveillance and access control systems, is fused through a graph neural network to perform spatial correlation verification on the preliminary positioning results, generating optimized high-precision positioning data. S4. When the positioning terminal enters a signal blind zone, the blind zone perception mechanism is activated. The MEC platform continuously aggregates effective positioning data and external context information from the past week. The time-series reasoning capability of the graph neural network is used to continuously generate intelligent inference location data during signal loss to ensure the spatiotemporal continuity of the positioning service. S5. Upload high-precision positioning data and intelligent inferred location data to the positioning management platform in a unified manner. Through a dynamic weight allocation mechanism, the dual-source location information is integrated to generate a seamless positioning result across the entire domain. Based on this result, the real-time location of personnel / materials, trajectory playback, and abnormal behavior warnings are output to provide full-scenario location awareness services for subway station operations.

[0005] This invention proposes a high-precision positioning system for subway stations based on a 5G public / private network. The system includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above.

[0006] The beneficial effects of this invention are as follows: By combining 5G public and private network dual-mode access with dynamic signal fingerprint fusion, the positioning accuracy and coverage within subway stations are significantly improved. This achieves low-cost, high-precision positioning in public network coverage areas while ensuring data security and global consistency in private network deployment scenarios. The multipath pre-compensation algorithm, combined with digital twins, effectively reduces multipath interference caused by complex building structures, enhancing the stability of positioning results and reducing jumps and drift phenomena. When the positioning terminal enters a signal blind zone, the intelligent inference mechanism driven by graph neural networks continuously generates location data, avoiding monitoring failures caused by service interruptions and ensuring the continuity of personnel and material trajectory tracking. This method not only reuses existing 5G network resources to reduce construction costs but also improves environmental adaptability through multi-source data fusion, effectively avoiding core problems in traditional solutions such as public and private network fragmentation, poor algorithm robustness, and service vulnerability. It provides a more reliable and efficient all-domain location awareness capability for smart subway operations. Attached Figure Description

[0007] Figure 1 This is a diagram illustrating the steps of the method described in this invention. Detailed Implementation

[0008] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0009] One embodiment of the present invention, such as Figure 1 As shown, a high-precision positioning method for subway stations based on a 5G public / private network is described, the method comprising: S1. Conduct multi-dimensional signal coverage analysis of the area within the subway station to generate a signal strength distribution map of the public network and private network; deploy positioning terminals that support dual-mode access of 5G public network and private network based on the map, construct a hybrid positioning network that coordinates public and private networks, and simultaneously collect dual-network signal fingerprint feature data; S2. Upload the dual-network signal fingerprint feature data to the edge computing (MEC) platform, and construct a three-dimensional spatial signal propagation model of the subway station through digital twin technology; based on the model, perform multipath pre-compensation processing on the original signal to generate enhanced signal data that is resistant to multipath interference; S3. Using enhanced signal data, combined with a dynamically weighted UTDOA / RTT hybrid positioning algorithm, an initial position is calculated to generate preliminary positioning results. At the same time, external context information, including video surveillance and access control systems, is fused through a graph neural network to perform spatial correlation verification on the preliminary positioning results, generating optimized high-precision positioning data. S4. When the positioning terminal enters a signal blind zone, the blind zone perception mechanism is activated. The MEC platform continuously aggregates effective positioning data and external context information from the past week. The time-series reasoning capability of the graph neural network is used to continuously generate intelligent inference location data during signal loss to ensure the spatiotemporal continuity of the positioning service. S5. Upload high-precision positioning data and intelligent inferred location data to the positioning management platform in a unified manner. Through a dynamic weight allocation mechanism, the dual-source location information is integrated to generate a seamless positioning result across the entire domain. Based on this result, the real-time location of personnel / materials, trajectory playback, and abnormal behavior warnings are output to provide full-scenario location awareness services for subway station operations.

[0010] The working principle of the above technical solution is as follows: First, signal scanning is performed on the entire area of ​​the subway station, including the platform, concourse, passageways, and equipment rooms, in layers, zones, and heights. Simultaneously, full-area indicators such as transmission attenuation, obstruction loss, and signal strength of the 5G public and private networks are collected to complete multi-dimensional signal coverage analysis, forming a signal strength distribution map of the public and private networks that intuitively reflects signal strength. Based on the map, weak signal areas, overlapping areas, and optimal areas are marked. 5G positioning terminals supporting dual-mode access are deployed at corresponding locations to establish a collaborative communication link between the public and private networks, opening up data interaction channels between the two networks and forming a hybrid public-private network positioning network. The positioning terminals move at a uniform speed throughout the station to sample, continuously receiving downlink positioning reference signals from the public and private networks, extracting information such as signal arrival time, arrival angle, and phase changes, and simultaneously collecting and storing dual-network signal fingerprint feature data to provide raw input for subsequent positioning calculations. The positioning terminal first establishes a dedicated secure transmission link with the edge computing (MEC) platform, completing port initialization, identity authentication, and link detection. Then, it performs time-series processing, block encapsulation, and integrity processing on the signal fingerprint feature data, uploading it packet by packet to the local MEC node via a stable link. The platform then performs reception verification and data retention. The MEC platform loads all scene parameters, including the subway station's building structure, facility layout, and wall obstructions. Using digital twin technology, it creates a three-dimensional replica of the physical space, digitally mapping signal propagation constraints to the virtual space. It combines model components and completes parameter matching to construct a three-dimensional signal propagation model of the subway station. This model is used to traverse and analyze all signal transmission directions and reflection scenarios within the station, accurately identifying all multipath propagation paths. The original signal is input into the model to complete multipath feature matching, separating multipath interference components from the original signal. Subsequently, phase adjustment and amplitude calibration are performed on the interference components to eliminate signal distortion caused by multipath propagation. Finally, the corrected components are fused with the original signal to complete multipath pre-compensation. Finally, noise filtering and abnormal fluctuation removal are performed on the compensated signal to output clean and stable standardized data, generating enhanced signal data resistant to multipath interference, significantly improving the positioning robustness in complex indoor environments. Two core features, uplink time difference and round-trip time, are extracted from the enhanced signal data as the basic input for positioning calculation. Dynamic indicators of the current transmission environment are collected to generate environment adaptation weight factors, allocating the weight of the two types of features in the calculation. UTDOA and RTT are used in parallel computation: UTDOA is used to perform propagation time difference numerical calculation, and RTT is used to perform propagation distance numerical calculation. The two sets of parallel results are then fused, and local coordinate conversion and point fitting within the station are performed to generate preliminary positioning results. Simultaneously, external devices such as in-station video surveillance and access control systems are accessed to collect real-time status information and form an external context information set. A target spatial association topology is constructed using a graph neural network. The preliminary positioning results are compared spatially with information such as video footage, access control switch status, and device location to identify and correct position offsets and abnormal deviations, generating optimized high-precision positioning data.The system monitors the signal reception quality, data transmission rate, and reporting stability of the positioning terminal in real time. When these indicators fall below a critical level, the terminal is determined to have entered a signal blind zone, and the blind zone perception mechanism is immediately and automatically activated, triggering the data retrieval process of the MEC platform. The MEC platform collects valid historical positioning data and external context information for the corresponding area of ​​the blind zone within the past week, completing data cleaning, deduplication, format unification, and feature extraction. The processed historical time-series data is input into a graph neural network, converted into a learnable feature set, and iteratively trained to capture the target's movement behavior patterns and regional spatial constraint features. The behavioral features and spatial constraints are then fused to construct a basic model for location estimation. During the continuous period of complete signal loss, the model continuously outputs the corresponding spatial coordinates, uninterruptedly generating intelligent inferred location data, maintaining the spatiotemporal continuity of the positioning service without interruption, jumps, or omissions. The system synchronously uploads high-precision positioning data and intelligent inferred location data to the positioning management platform through a unified channel. Within the platform, dynamic weights are configured based on signal quality, data confidence, and scene stability to perform weighted fusion calculations on the two types of location information, eliminating single-point errors and time-series gaps, and generating seamless full-domain positioning results. Based on the seamless positioning results across the entire domain, the system marks the coordinates of personnel and materials in real time on 2D / 3D visualized maps and refreshes the location display; it reads historical coordinate sequences and reassembles them chronologically to generate queryable and playable trajectory playback content; it compares the target coordinates with the preset area boundaries in real time, detects abnormal states such as boundary crossing, lingering, and deviation from the route, and triggers early warnings. Finally, it outputs three core capabilities: real-time location, trajectory playback, and abnormal behavior early warning, providing full-scenario, all-weather, and highly reliable location awareness support for the daily operation, emergency response, operation supervision, and material scheduling of subway stations.

[0011] The effects of the above technical solution are as follows: Using a 5G public-private network collaborative positioning method can improve the integrity of signal coverage within subway stations, ensuring stable operation of positioning services in complex indoor scenarios. Hybrid public-private network networking can improve the access efficiency of positioning terminals and reduce service interruptions caused by single network failures. Utilizing digital twins and multipath pre-compensation processing can enhance signal anti-interference capabilities, improve the accuracy of positioning results, and reduce positional shifts caused by multipath reflections. Dynamic weighted hybrid positioning algorithms can improve the response speed of location calculation and reduce computational errors in complex environments. Fusion of external contextual information through graph neural networks can improve the reliability of positioning verification and prevent erroneous location data from affecting system judgment. Intelligent blind zone inference can maintain the spatiotemporal continuity of positioning services, preventing positioning failures caused by signal obstruction or loss. Dual-source data fusion can improve the smoothness of global positioning and reduce data jumps and gaps. The entire solution can reduce deployment and construction costs within stations, strengthen operational safety management capabilities, and improve emergency response and personnel and material management efficiency.

[0012] In one embodiment of the present invention, S1 includes: S11. Scan the physical space of the entire subway station in layers and zones, collect 5G public network signal transmission parameters at different locations and heights, collect 5G private network signal transmission parameters at different locations and heights, and complete the full-area survey and measurement of the signal propagation environment within the station. S12. Integrate the measured signal attenuation data and obstruction loss data, conduct multi-dimensional signal coverage analysis, and form a public network and private network signal strength distribution map covering the entire station; S13. Divide the signal adaptation area according to the signal strength distribution map, and deploy positioning terminals that support dual-mode access of 5G public network and private network at the corresponding locations to complete the full-area deployment of the perception layer hardware. S14. Establish a collaborative communication link between the public network and the private network, open up the data interaction channel between the two networks, and build a hybrid positioning network that integrates the public and private networks. S15. Drive the positioning terminal to conduct continuous sampling throughout the entire area of ​​the station, record the characteristic parameters of the dual-network signals in real time, and synchronously collect and store complete dual-network signal fingerprint characteristic data.

[0013] The working principle and effects of the above technical solution are as follows: By scanning the physical space within the subway station in a layered and zoned manner, signal transmission parameters can be comprehensively collected, completing a comprehensive assessment of the signal environment, improving the completeness of signal measurement, and avoiding the loss of signal data in local areas. Integrating multi-dimensional signal data to form a distribution map can improve the accuracy of signal coverage analysis and reduce deployment errors caused by signal assessment biases. Deploying dual-mode positioning terminals according to the map can improve the rationality of hardware layout and reduce resource waste. Establishing a collaborative communication link between public and private networks can enhance network transmission stability and avoid the impact of a single network interruption on data collection. Driving the terminal to continuously sample the entire area can improve the comprehensiveness of signal fingerprint feature collection, reduce feature data omissions, provide sufficient data support for subsequent positioning calculations, and ensure that signal collection conforms to the actual propagation environment, guaranteeing the smooth progress of subsequent positioning processes.

[0014] In one embodiment of the present invention, S2 includes: S21. Establish a secure transmission link from the positioning terminal to the edge computing MEC platform, and upload the collected dual-network signal fingerprint feature data to the local node of the platform. S22. Load the building structure and facility layout parameters of the subway station, use digital twin technology to restore the physical space and signal propagation constraints within the station, and construct a three-dimensional spatial signal propagation model of the subway station. S23. Identify multipath propagation paths of signals using a three-dimensional spatial signal propagation model, and separate multipath interference components from the original signal; S24. Perform phase and amplitude correction on the multipath interference components to complete the multipath pre-compensation processing of the original signal; filter out redundant noise and abnormal fluctuations in the compensated signal to generate enhanced signal data resistant to multipath interference.

[0015] The working principle and effects of the above technical solution are as follows: By establishing a secure transmission link from the terminal to the edge computing platform, the complete upload of dual-network signal fingerprint feature data can be ensured, improving the security and stability of data transmission and preventing data loss or leakage during transmission. Loading in-station parameters and constructing a three-dimensional spatial signal propagation model can improve the realism of signal environment simulation and restore the propagation patterns in complex scenarios. Identifying multipath propagation paths and separating interference components through the model can improve the targeting of signal processing and reduce the impact of multipath effects on the original data. Performing phase and amplitude correction on multipath interference, and then filtering noise and abnormal fluctuations, can enhance the purity and reliability of the signal and prevent interference data from affecting subsequent positioning accuracy. The entire processing flow can not only accelerate signal preprocessing speed and reduce the computational pressure on edge nodes, but also provide a high-quality data foundation for subsequent location calculation, making the overall positioning effect more stable.

[0016] In one embodiment of the present invention, step S21 includes: S211. Enable the dedicated transmission interface of the edge computing MEC platform and complete the interface parameter configuration and transmission channel initialization; S212. Establish a point-to-point transmission connection between the positioning terminal and the edge computing MEC platform to form a stable and secure data transmission link; S213. Perform timing processing and data encapsulation on the dual-network signal fingerprint feature data collected by the positioning terminal to improve transmission efficiency and integrity. S214. Upload the encapsulated dual-network signal fingerprint feature data packet by packet to the local node of the edge computing MEC platform through the established secure transmission link; S215. Complete data reception and integrity verification on the edge computing MEC platform, and retain usable dual-network signal fingerprint feature data.

[0017] The working principle and effects of the above technical solution are as follows: Enabling a dedicated transmission interface and completing initialization improves the adaptability of the data channel and avoids transmission interruptions caused by interface incompatibility. Establishing a point-to-point transmission connection enhances the stability and security of data transmission and reduces data risks from external network intrusion. Time-series processing and encapsulation of signal fingerprint feature data improves transmission efficiency and avoids parsing anomalies caused by data clutter. Uploading encapsulated data packet by packet reduces the load of a single transmission and prevents the loss of large data packets. Platform-side reception and integrity verification promptly detects missing or incorrect data, preventing invalid data from entering subsequent processing stages. This entire process ensures the complete and reliable uploading and retention of signal fingerprint data, simplifies the transmission process, improves the overall efficiency of data preprocessing, and provides solid data support for subsequent model building and signal optimization.

[0018] In one embodiment of the present invention, S214 includes: Read the real-time transmission parameters of the secure transmission link, match the transmission specifications of the encapsulated dual-network signal fingerprint feature data, and generate a data transmission scheme that adapts to the current link status. According to the data transmission scheme, the encapsulated dual-network signal fingerprint feature data is time-series grouped to form a data unit with continuous transmission capability. Send the completed data units to the local node of the edge computing MEC platform through the established secure transmission link, and synchronously record the transmission timing and transmission location of the data units; Track the transmission status of data units in the transmission link, promptly replenish data units lost due to transmission interruption, and maintain the smooth operation of the transmission process; All data units are pushed to the local node of the edge computing MEC platform to complete the upload of this round of encapsulated data.

[0019] The working principle and effects of the above technical solution are as follows: Matching transmission specifications with transmission parameters and generating a transmission plan improves the adaptability of data transmission and avoids transmission stuttering caused by incompatible transmission specifications. Sequential grouping of encapsulated data enhances the continuity of data transmission and reduces data congestion and packet loss probability. Sending data units and synchronously recording their timing and location improves the traceability of the transmission process and facilitates subsequent troubleshooting. Real-time tracking of transmission status and replenishment of lost units maintains the stability of the transmission process and prevents overall upload failure due to partial data loss. Completely pushing all data units and completing the upload ensures the comprehensiveness of data upload and prevents the loss of key feature data. The entire process improves the success rate and smoothness of dual-network signal fingerprint data upload, reduces manual intervention, and makes the data upload process more efficient and reliable, providing a complete data source for subsequent signal processing and model calculations.

[0020] In one embodiment of the present invention, step S22 includes: Import relevant parameters of the subway station building structure, complete the loading and format unification of the parameters in the edge computing MEC platform, and generate a standardized set of building space parameters. Import relevant parameters of subway station facility layout, complete the loading and format unification of parameters in the edge computing MEC platform, and generate a standardized facility layout parameter set; By integrating standardized building space parameter sets and standardized facility layout parameter sets, a digital basic framework for the physical space of subway stations is constructed. Digital twin technology is used to create a three-dimensional spatial replica of the digital infrastructure framework, generating a virtual space within the station that is consistent with the physical scene. By embedding signal propagation constraints into the virtual space within the station, the internal parameters of the model are adjusted and the scene is matched to generate a three-dimensional spatial signal propagation model for the subway station.

[0021] The working principle and effects of the above technical solution are as follows: Importing and unifying the two types of parameter formats improves data compatibility within the platform and avoids parameter confusion affecting subsequent framework construction. The fusion and generation of a digital physical space framework enhances the completeness of spatial reconstruction and reduces simulation deviations caused by missing scene information. Using digital twin technology to replicate 3D space improves the fit between the virtual scene and the physical reality, making the signal propagation environment closer to reality. Embedding signal propagation constraints in the virtual space and completing debugging and matching improves the accuracy of model calculations and avoids environmental parameter distortion affecting signal processing results. The entire construction process not only allows the 3D spatial signal propagation model to fully reflect the actual environment within the station but also simplifies the complexity of model building, improves the efficiency of signal simulation and preprocessing, and provides accurate model support for subsequent multipath interference suppression.

[0022] In one embodiment of the present invention, S3 includes: S31. Extract the core features of uplink arrival time difference and round-trip time from the enhanced signal data to provide basic computational input for positioning solution; S32. Set dynamic weighting rules to allocate calculation weights, use a hybrid positioning algorithm combining UTDOA and RTT to perform calculations, complete the initial position calculation and generate preliminary positioning results; S33. Connect to external systems such as in-station video surveillance and access control systems, and collect real-time status information to form a set of external context information; S34. Construct a target spatial association topology through a graph neural network, and verify the consistency between the preliminary positioning results and external context information; correct the position offset and abnormal deviation found during the verification process, and generate optimized high-precision positioning data.

[0023] The working principle and effects of the above technical solution are as follows: Extracting core features from enhanced signals improves the effectiveness of positioning calculation input and reduces resource consumption caused by invalid data participating in the calculation. Employing a dynamic weighted hybrid algorithm for position calculation improves the rationality of the initial positioning results and reduces computational bias caused by a single algorithm. Connecting to external systems and aggregating real-time status information enriches the reference dimensions for positioning verification, preventing positioning data from deviating from the actual situation on site. Utilizing graph neural networks to complete spatial correlation and consistency verification can promptly correct position offsets and abnormal deviations, improving the accuracy of the final positioning data. This entire processing flow not only makes the position calculation more closely aligned with changes in the on-site environment but also enhances the reliability of the positioning results, reduces interference from abnormal position data on operation and management, and provides a stable and reliable location basis for on-site safety supervision.

[0024] In one embodiment of the present invention, S32 includes: Collect dynamic change indicators of enhanced signal data under the current transmission environment, and generate environment adaptation weight factors; Incorporate environmental adaptation weighting factors into the calculation process and allocate the calculation weight of uplink arrival time difference and round-trip time in the solution process; The allocated computational weights are loaded to drive the UTDOA algorithm to perform numerical calculations of the signal propagation time difference; the RTT algorithm is simultaneously driven to perform numerical calculations of the signal propagation distance, resulting in two sets of parallel solution results. By integrating the results of two parallel calculations, the numerical conversion of local coordinates within the station and the point fitting are completed, generating preliminary positioning results.

[0025] The working principle and effects of the above technical solution are as follows: Collecting dynamic environmental indicators and generating adaptive weighting factors improves the algorithm's adaptability to on-site transmission conditions, preventing fixed parameters from failing to match environmental changes. Allocating the calculation weights of the two types of features according to their respective weights enhances the rationality of the solution process and reduces positioning errors caused by a single dominant feature. Parallel computation of the two algorithms improves the comprehensiveness of the location calculation and reduces the possibility of a single algorithm failing in complex scenarios. Fusion of the two sets of results and completion of coordinate transformation and point fitting improves the smoothness and accuracy of the initial positioning results, preventing point jumps from affecting subsequent verification. The entire computational process not only makes the positioning calculation more closely fit the complex indoor environment of the subway but also improves the response speed of the location calculation, providing solid intermediate results for subsequent high-precision positioning optimization.

[0026] In one embodiment of the present invention, step S4 includes: S41. Monitor the signal reception quality and data transmission status of the positioning terminal in real time to determine whether the terminal has entered a signal blind zone; S42. When it is determined that the blind zone has been entered, the blind zone perception mechanism is automatically activated, triggering the data retrieval and aggregation process of the MEC platform. S43. Collect effective positioning data and external context information of the corresponding blind area within the past week through the MEC platform to complete the cleaning and organization of historical data; S44. Utilize graph neural networks to perform deep learning on historical time-series data to extract target movement patterns and regional spatial constraint features; continuously output position estimation results during the continuous period of signal loss, and continuously generate intelligent inference position data.

[0027] The working principle and effects of the above technical solution are as follows: Real-time monitoring of terminal signals and transmission status improves the sensitivity of blind spot identification, promptly detects signal anomalies, and prevents system unresponsiveness after signal loss. Automatic activation of the blind spot perception mechanism improves response speed and reduces delays caused by manual intervention. Collecting and cleaning historical data and contextual information enhances the reference value of inferred data and avoids interference from invalid historical information in calculations. Through graph neural network deep learning and continuous output of inference results, the spatiotemporal continuity of positioning services is maintained, preventing positioning interruptions caused by signal blind spots. This entire mechanism covers all positioning needs within the station, eliminating management gaps caused by blind spots, and ensures continuous output of location data, providing uninterrupted location support for operational supervision and emergency response.

[0028] In one embodiment of the present invention, S44 includes: The cleaned and organized historical time-series data is loaded, and the data is input into a graph neural network to complete the feature vector transformation, generating a learnable time-series data feature set; The graph neural network is driven by a time-series data feature set to carry out iterative training, capture the movement behavior features of the target in the station space, and generate a target behavior feature set; By combining the spatial structure information within the station, the correlation attributes between the target's movement path and spatial distribution are mined to generate a regional spatial constraint feature set; By integrating the target behavior feature set and the regional spatial constraint feature set, an in-station location estimation calculation model is constructed, and a basic location estimation model is generated. Based on the location estimation model, the system continuously outputs corresponding coordinate results during periods of signal loss, forming continuous and complete intelligent location estimation data.

[0029] The working principle and effects of the above technical solution are as follows: Converting historical time-series data into a learnable feature set improves the effectiveness of model training and avoids the impact of messy raw data on learning outcomes. Iterative training to capture movement behavior features improves the accuracy of location inference, making it closer to the actual movement of the target. Combining spatial structure mining with related attributes enhances the rationality of location estimation constraints and reduces unfounded coordinate offsets. Integrating the two types of features to construct the inference model improves the stability of the model output and reduces abnormal fluctuations in inference results. Continuously outputting coordinates by the model maintains the integrity and continuity of location data, avoiding location gaps during periods of signal loss. This entire processing method not only makes blind-zone location inference more consistent with the spatial rules of the site but also ensures uninterrupted location services throughout the entire process, providing reliable support for full-scene location awareness within the station.

[0030] In one embodiment of the present invention, step S5 includes: S51. Establish a unified upload channel for high-precision positioning data and intelligent inferred location data to the positioning management platform to complete the synchronous aggregation of dual-source data; S52. Configure dynamic weights based on signal quality and data reliability to complete the fusion calculation of dual-source location information and generate a seamless global positioning result. S53. Based on the seamless global positioning results, mark the target coordinates on the visualization map, refresh and display the real-time location of personnel and materials; read the historical location sequence for time-series reconstruction, and generate queryable and playable trajectory playback content; S54. Detect the spatial relationship between the target location and the preset area, identify abnormal behavior and trigger early warning, and provide full-scene location awareness services for subway station operation.

[0031] The working principle and effects of the above technical solution are as follows: Establishing a unified upload channel to achieve synchronous aggregation of dual-source data reduces processing time caused by the dispersed transmission of multiple types of data and minimizes data reception errors. Dynamic weights are adjusted based on signal quality and data reliability to perform fusion calculations, effectively mitigating the shortcomings of single data sources, reducing the frequency of location data gaps and point jumps, and obtaining seamless, comprehensive location content. Real-time annotation of target coordinates on a visual page and simultaneous organization of historical location sequences complete trajectory integration, improving the intuitiveness of personnel and material management within the station and facilitating quick retrieval of past movement records by management personnel. Continuous comparison of the spatial differences between the target's location and the designated area promptly detects unauthorized lingering or boundary crossings, providing early risk warnings. This entire processing model not only unifies the management standards of all station location data but also enriches the service dimensions of subway operation management, steadily strengthening the overall level of daily management and security control within the station.

[0032] One embodiment of the present invention discloses a high-precision positioning system for subway stations based on a 5G public / private network, the system comprising: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method described in any one of the above. Example 2

[0033] In one embodiment of the present invention, the reliability redistribution process is further performed on the dynamically weighted UTDOA / RTT hybrid localization algorithm.

[0034] In the complex indoor environment of subway stations, the UTDOA algorithm is susceptible to the effects of base station synchronization status, uplink time difference stability, and the consistency of timestamps from multiple base stations, while the RTT algorithm is susceptible to fluctuations in round-trip time measurement, non-line-of-sight propagation, and the extension of reflection paths. If the calculation weights of UTDOA and RTT are allocated solely based on signal quality, link strength, or general environmental adaptation weights, then when the public and private network link states are inconsistent, a certain type of signal experiences short-term anomalies, or the uplink time difference characteristics and round-trip time characteristics contradict each other, a higher weight may still be assigned to the positioning results with lower actual reliability. This can lead to the initial positioning results drifting, jumping, or exhibiting abnormal deviations from the actual travel path near the station hall, platform, transfer passage, or equipment room entrances and exits.

[0035] To address this, this embodiment incorporates a multi-dimensional confidence-driven UTDOA / RTT weight redistribution mechanism during the initial location calculation process. This ensures that both UTDOA and RTT positioning results are constrained by the consistency of public and private network links and the continuity of historical trajectories. Furthermore, UTDOA positioning results are further constrained by the stability of uplink time difference, and RTT positioning results are further constrained by the non-line-of-sight sensitivity of round-trip time. This results in a confidence assessment method that combines unified environmental constraints with differentiated algorithmic constraints.

[0036] Specifically, the method described in this embodiment includes steps S1-S8. Among them, S1-S5 are the original steps in Embodiment 1, and S6-S8 are further improved steps based on Embodiment 1. In actual execution, S6-S8 are embedded after the initial position calculation process in step S3 of Embodiment 1 and before the graph neural network spatial association verification process, and are used to generate a hybrid positioning result after confidence correction, and use the result as input data for subsequent spatial association verification.

[0037] S1. Conduct multi-dimensional signal coverage analysis of the area within the subway station to generate a signal strength distribution map of the public network and private network; deploy positioning terminals that support dual-mode access of 5G public network and private network based on the map, construct a hybrid positioning network that coordinates public and private networks, and simultaneously collect dual-network signal fingerprint feature data.

[0038] S2. Upload the dual-network signal fingerprint feature data to the edge computing MEC platform, and construct a three-dimensional spatial signal propagation model of the subway station through digital twin technology; based on the model, perform multipath pre-compensation processing on the original signal to generate enhanced signal data that resists multipath interference.

[0039] S3. Utilize enhanced signal data and combine it with a dynamically weighted UTDOA / RTT hybrid positioning algorithm to perform initial position calculation, generating UTDOA positioning results and RTT positioning results respectively; at the same time, connect to external systems such as video surveillance and access control systems to collect real-time status information to form an external context information set.

[0040] S4. When the positioning terminal enters a signal blind zone, the blind zone perception mechanism is activated. The effective positioning data and external context information within the past week are continuously aggregated through the edge computing MEC platform. The time series reasoning capability of the graph neural network is used to continuously generate intelligent inference location data during the signal loss period.

[0041] S5. Upload high-precision positioning data and intelligent inferred location data to the positioning management platform in a unified manner. Integrate dual-source location information through a dynamic weight allocation mechanism to generate a seamless positioning result across the entire domain. Based on this result, output the real-time location of personnel or materials, trajectory playback, and early warning of abnormal behavior.

[0042] S6. Within the edge computing MEC platform, based on the consistency deviation between public and private network links, the continuity deviation of historical trajectories, and the stability deviation of uplink arrival time difference, a UTDOA credibility coefficient is generated to determine whether the current UTDOA positioning result is suitable for increasing the computational weight in the hybrid positioning process.

[0043] Among them, the consistency deviation between public and private network links is used to characterize whether the signal reception quality, transmission delay, packet loss status, and positioning reference signal stability reflected by the public and private network links are consistent at the same sampling time for the same positioning terminal; the historical trajectory continuity deviation is used to characterize whether there are abrupt changes in the current UTDOA positioning result relative to the effective location sequence of the previous time period that do not conform to the topology of the station channel, access control status, or personnel movement patterns; and the uplink arrival time difference stability deviation is used to characterize whether there are jitter, abrupt changes, or synchronization residual abnormalities in the uplink arrival time difference of the multiple base stations on which the UTDOA algorithm depends within the sliding observation window. After uniformly normalizing the above three types of deviations, the edge computing MEC platform generates the UTDOA credibility coefficient according to the principle that the larger the deviation, the lower the credibility. Therefore, even if the local signal strength is high in a certain area, as long as the dual network link status is inconsistent, the trajectory continuity is abnormal, or the uplink arrival time difference is unstable, the system will not simply increase the calculation weight of the UTDOA positioning result.

[0044] S7. Within the edge computing MEC platform, an RTT reliability coefficient is generated based on the consistency deviation between public and private network links, the continuity deviation of historical trajectories, and the non-line-of-sight sensitive deviation of round-trip time. This coefficient is used to determine whether the current RTT positioning result is suitable for increasing its computational weight in the hybrid positioning process.

[0045] Among these, the consistency deviation between public and private network links and the continuity deviation of historical trajectories are kept consistent with S6, ensuring that UTDOA credibility judgment and RTT credibility judgment are under the same environmental and motion constraint benchmarks. The non-line-of-sight sensitive deviation of round-trip time is used to characterize whether the round-trip time measurement relied upon by the RTT algorithm is affected by wall reflections, train obstructions, station hall corners, equipment room shielding, or multi-path detours. The edge computing MEC platform generates the non-line-of-sight sensitive deviation of round-trip time based on the degree of abnormal extension of the round-trip time series, the degree of deviation between the round-trip distance and the station topology path, and the degree of difference in RTT ranging results under public and private network links. It then generates the RTT credibility coefficient according to the principle that the larger the deviation, the lower the credibility. Therefore, when the propagation path of RTT measurement is lengthened due to non-line-of-sight propagation, even if its short-term ranging data appears continuous, it will not be directly assigned an excessively high fusion weight.

[0046] S8. Based on the UTDOA confidence coefficient and RTT confidence coefficient, the UTDOA positioning result and RTT positioning result are fused with confidence normalization to generate a confidence-corrected hybrid positioning result, and this result is used as the input data for subsequent graph neural network spatial association verification.

[0047] Specifically, when the UTDOA confidence coefficient is higher than the RTT confidence coefficient, the proportion of UTDOA positioning results in the hybrid positioning results is increased; when the RTT confidence coefficient is higher than the UTDOA confidence coefficient, the proportion of RTT positioning results in the hybrid positioning results is increased; when both confidence coefficients are at a low level, neither algorithm result is directly used as the final preliminary positioning result. Instead, the sampling time is marked as a low-confidence sampling point and further verified by a graph neural network combining external context information, historical trajectory continuity, and station spatial correlation topology. This approach avoids unreasonable jumps in positioning results between the station hall, platform, passageway, or equipment room due to short-term anomalies in a single algorithm.

[0048] Furthermore, S6 specifically includes the following steps: S61. Establish a sliding observation window corresponding to the current sampling time in the edge computing MEC platform, and read the public network enhanced signal data, private network enhanced signal data, historical valid positioning data, and the positioning results output by the current UTDOA algorithm corresponding to the positioning terminal in the sliding observation window.

[0049] S62. Based on the differences in signal reception quality, transmission delay, data packet loss status, and positioning reference signal stability between public network enhanced signal data and private network enhanced signal data, a consistency deviation between the public network and private network links is generated. The larger the consistency deviation between the public network and private network links, the more inconsistent the current dual networks' responses to the same positioning terminal status, and the less suitable the current sampling environment is for simply increasing the weight of either algorithm's result.

[0050] S63. Based on historical valid positioning data, station passage topology, access control status, and the target's movement direction in the previous time period, determine whether the current UTDOA positioning result exceeds the reasonable movement range, and generate historical trajectory continuity deviation. The larger the historical trajectory continuity deviation, the more likely the current UTDOA positioning result is to be a short-term anomaly, a jump point, or a location that does not match the station's passage structure.

[0051] S64. Based on the jitter of the uplink arrival time difference sequence of multiple base stations within the sliding observation window, the timestamp synchronization residual, and the amplitude of time difference variation between adjacent sampling points, an uplink arrival time difference stability deviation is generated. The larger the uplink arrival time difference stability deviation, the more unstable the basic time difference characteristics on which the UTDOA algorithm relies for calculation are.

[0052] S65. The edge computing MEC platform inputs the consistency deviation between public and private network links, the continuity deviation of historical trajectories, and the stability deviation of uplink time difference of arrival into the UTDOA reliability evaluation process, and generates a UTDOA reliability coefficient according to the principle that the three types of deviations jointly weaken the reliability of UTDOA. Specifically, when all three types of deviations are small, it means that the current UTDOA positioning result simultaneously meets the requirements of dual-network link consistency, historical trajectory continuity, and stable uplink time difference of arrival, and the system increases the UTDOA reliability coefficient; when any deviation increases significantly, the system decreases the UTDOA reliability coefficient; when the uplink time difference of arrival stability deviation continues to increase, even if the public or private network signal strength is good, the system still limits the proportion of UTDOA positioning results in the hybrid positioning process.

[0053] With S6, UTDOA weights are no longer determined by a single signal strength or general environment adaptation factor, but are simultaneously constrained by dual-network consistency, target movement continuity, and the stability of UTDOA's own time difference. When there is a conflict between public and private network states, discontinuous target trajectories, or unstable uplink arrival time differences, the system can automatically reduce the reliability of UTDOA positioning results to prevent them from being erroneously amplified.

[0054] Furthermore, S7 specifically includes the following steps: S71. Within the same sliding observation window as S6, read public network enhanced signal data, private network enhanced signal data, historical valid positioning data, and the positioning results output by the current RTT algorithm, so that the RTT confidence judgment and the UTDOA confidence judgment maintain the same data time base.

[0055] S72. Continuing with the consistency deviation between public and private network links generated in S6, the RTT reliability evaluation is also subject to the consistency constraints of the dual-network links. When the public and private networks do not reflect the transmission status of the same positioning terminal consistently, the system reduces its reliance on a single RTT measurement result.

[0056] S73. Utilizing the historical trajectory continuity deviation generated in S6, the RTT reliability evaluation is also constrained by the historical movement patterns of the target. When the RTT positioning result shows a sudden change relative to the valid trajectory of the previous time period that does not conform to the positional relationships of station passages, platform boundaries, turnstile areas, or equipment room entrances / exits, the system reduces its reliability.

[0057] S74. Based on the abnormal extension of the round-trip time series within the sliding observation window, the degree of deviation between the round-trip distance and the in-station topology path, and the degree of difference in RTT ranging results for the same target under public network links and private network links, a non-line-of-sight sensitive bias for round-trip time is generated. The larger the non-line-of-sight sensitive bias for round-trip time, the more likely the RTT measurement is to be affected by non-line-of-sight propagation, reflection detours, or temporary obstruction.

[0058] The S75 edge computing MEC platform inputs the consistency deviation between public and private network links, the continuity deviation of historical trajectories, and the non-line-of-sight sensitive deviation of round-trip time into the RTT reliability evaluation process, and generates an RTT reliability coefficient according to the principle that the three types of deviations jointly weaken RTT reliability. Specifically, when the dual network links are consistent, the historical trajectory is continuous, and the round-trip time does not show abnormal extension, the system increases the RTT reliability coefficient; when the RTT ranging results of public and private networks are significantly inconsistent, the current RTT positioning result breaks through the spatial constraints within the station, or the round-trip time shows abnormal extension, the system decreases the RTT reliability coefficient. Since the RTT reliability evaluation and the UTDOA reliability evaluation both use the consistency deviation between public and private network links and the continuity deviation of historical trajectories, the reliability of the two algorithms can be compared on the same environmental basis; and since the RTT reliability evaluation further introduces the non-line-of-sight sensitive deviation of round-trip time, it can suppress the problem that RTT itself is susceptible to the influence of non-line-of-sight propagation.

[0059] With S7, the RTT weight is no longer determined solely by the availability of round-trip time measurements, but rather by a comprehensive assessment combining dual-network consistency, historical trajectory continuity, and non-line-of-sight propagation sensitivity. When the propagation path of an RTT measurement is prolonged due to wall reflections, train obstructions, equipment room shielding, or passageway corners, the system can automatically reduce the reliability of the RTT positioning result, preventing it from being incorrectly weighted in hybrid positioning.

[0060] Furthermore, S8 specifically includes the following steps: S81. Read the positioning results output by the UTDOA algorithm and the RTT algorithm in S3, and read the UTDOA confidence coefficient generated in S6 and the RTT confidence coefficient generated in S7.

[0061] S82. Normalize the UTDOA credibility coefficient and RTT credibility coefficient to obtain the actual fusion weight of the two algorithms at the current sampling time. This fusion weight is no longer equivalent to the general environment adaptation weight in Example 1, but is jointly determined by the consistency of public network and private network links, the continuity of historical trajectories, the stability of uplink arrival time difference, and the non-line-of-sight sensitivity of round-trip time.

[0062] S83. When the spatial difference between the UTDOA positioning result and the RTT positioning result is within the allowable range, a hybrid positioning result is generated according to the fusion weight after confidence normalization. When the spatial difference between the UTDOA positioning result and the RTT positioning result exceeds the allowable range, the trend information of the side with higher confidence is retained first, and the positioning result of the side with lower confidence is used as the reference value to be verified and input into the graph neural network spatial correlation verification process.

[0063] S84. Based on the UTDOA positioning result, RTT positioning result, UTDOA confidence coefficient, and RTT confidence coefficient, generate a confidence-corrected hybrid positioning result. Specifically, the system allocates the fusion ratio of the two types of positioning results according to the relative magnitude of the UTDOA confidence coefficient and the RTT confidence coefficient; the higher the UTDOA confidence coefficient, the greater the influence of the UTDOA positioning result on the hybrid positioning result; the higher the RTT confidence coefficient, the greater the influence of the RTT positioning result on the hybrid positioning result; when the confidence levels of the two are close, the system maintains a balanced fusion of the two types of positioning results; when both are below the preset confidence level, the system does not directly output a high-confidence preliminary positioning result, but instead marks the sampling point as a positioning point to be verified.

[0064] S85. Input the confidence-corrected hybrid positioning result into the graph neural network spatial association verification process. The graph neural network combines external context information, access control opening and closing status, video surveillance target location, station spatial association topology, and historical motion trajectory to further verify the confidence-corrected hybrid positioning result and generate optimized high-precision positioning data.

[0065] With S8, UTDOA and RTT are no longer simply fused according to a fixed ratio or a single signal quality ratio. Instead, they are dynamically redistributed based on the actual reliability of the two algorithms in the current sampling environment. When UTDOA is distorted due to base station synchronization residuals or uplink arrival time difference jitter, the system automatically reduces the proportion of UTDOA positioning results. When RTT is distorted due to non-line-of-sight propagation or extended round-trip path, the system automatically reduces the proportion of RTT positioning results. When both algorithms have low reliability risks, the system does not directly amplify either positioning result. Instead, it further verifies the result by combining graph neural networks with external context information, thereby reducing initial positioning result drift, jumps, and subsequent trajectory discontinuities.

[0066] The working principle of the above embodiment is as follows: The positioning terminal continuously collects public network and private network signal fingerprint feature data in the subway station, and obtains enhanced signal data after digital twin multipath pre-compensation by the edge computing MEC platform. After the enhanced signal data enters the UTDOA / RTT hybrid positioning process, the system does not directly allocate the calculation weight of UTDOA and RTT according to the conventional signal quality. Instead, it establishes a sliding observation window in the edge computing MEC platform to jointly analyze the public network link status, private network link status, historical valid trajectory, uplink arrival time difference sequence, and round-trip time sequence at the same sampling time. For the UTDOA algorithm, the system focuses on determining whether it is affected by the inconsistency of the two network links, sudden changes in the target trajectory, and instability of the uplink arrival time difference; for the RTT algorithm, the system focuses on determining whether it is affected by the inconsistency of the two network links, sudden changes in the target trajectory, and abnormally prolonged round-trip time caused by non-line-of-sight propagation. Since both algorithms incorporate public and private network link consistency deviations and historical trajectory continuity deviations, they ensure that both types of reliability assessments have unified environmental and motion constraints. Furthermore, because UTDOA introduces uplink time difference stability deviations and RTT introduces round-trip time non-line-of-sight sensitive deviations, they can differentiate and suppress the different failure mechanisms of the two algorithms. Finally, the system normalizes and fuses the two types of positioning results using the UTDOA reliability coefficient and the RTT reliability coefficient to obtain a reliability-corrected hybrid positioning result, which is then fed into the graph neural network spatial association verification process. This improves the stability and reliability of the initial positioning results in complex indoor environments.

[0067] The technical effects of the above-described embodiment two are as follows: By introducing a multi-dimensional confidence-driven UTDOA / RTT weight redistribution mechanism, the problem of misjudgment when simply calculating weight allocation based on signal quality or general environment adaptation weights can be solved. When the quality of public network and private network links is inconsistent, the system can reduce its dependence on positioning results supported by a single link; when the target trajectory exhibits abrupt changes that do not conform to the subway station's traffic structure, the system can reduce the fusion ratio of abnormal positioning results; when the UTDOA time difference sequence is unstable, the system can suppress the erroneous amplification of UTDOA results; when the RTT round-trip time is affected by non-line-of-sight propagation, the system can suppress the erroneous amplification of RTT results. As a result, the preliminary positioning results are smoother and more stable in complex areas such as station halls, platforms, transfer passages, and equipment room entrances and exits, reducing position drift, point jumps, and trajectory breaks caused by short-term anomalies, and reducing the correction pressure of subsequent graph neural network spatial association verification, thereby improving the reliability of the final high-precision positioning data.

[0068] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A high-precision positioning method for subway stations based on 5G public / private networks, characterized in that, The method includes: S1. Conduct multi-dimensional signal coverage analysis of the area within the subway station to generate a signal strength distribution map of the public network and private network; deploy positioning terminals that support dual-mode access of 5G public network and private network based on the map, construct a hybrid positioning network that coordinates public and private networks, and simultaneously collect dual-network signal fingerprint feature data; S2. Upload the dual-network signal fingerprint feature data to the edge computing platform, and construct a three-dimensional spatial signal propagation model of the subway station through digital twin technology; based on the model, perform multipath pre-compensation processing on the original signal to generate enhanced signal data that is resistant to multipath interference; S3. Using enhanced signal data, combined with a dynamically weighted UTDOA / RTT hybrid positioning algorithm, the initial position is calculated to generate preliminary positioning results; at the same time, external context information is fused through graph neural networks to perform spatial correlation verification on the preliminary positioning results, generating optimized high-precision positioning data. S4. When the positioning terminal enters a signal blind zone, the blind zone perception mechanism is activated, and the effective positioning data and external context information within the past week are continuously aggregated through the MEC platform; the time series reasoning capability of the graph neural network is used to continuously generate intelligent inference location data during the signal loss period. S5. Upload high-precision positioning data and intelligent inferred location data to the positioning management platform. Merge the dual-source location information through a dynamic weight allocation mechanism to generate a seamless positioning result across the entire domain. Based on this result, output the real-time location of personnel / materials, trajectory playback, and abnormal behavior warnings. The S3 includes: S31. Extract the core features of uplink arrival time difference and round-trip time from the enhanced signal data to provide basic computational input for positioning solution; S32. Set dynamic weighting rules to allocate calculation weights, use a hybrid positioning algorithm combining UTDOA and RTT to perform calculations, complete the initial position calculation and generate preliminary positioning results; S33. Connect to external systems and collect real-time status information to form an external context information set; S34. Construct a target spatial association topology through a graph neural network, and verify the consistency between the preliminary positioning results and external context information; correct the position offset and abnormal deviation found during the verification process, and generate optimized high-precision positioning data. S32 includes: Collect dynamic change indicators of enhanced signal data under the current transmission environment, and generate environment adaptation weight factors; Incorporate environmental adaptation weighting factors into the calculation process and allocate the calculation weight of uplink arrival time difference and round-trip time in the solution process; The allocated computational weights are loaded to drive the UTDOA algorithm to perform numerical calculations of the signal propagation time difference; the RTT algorithm is simultaneously driven to perform numerical calculations of the signal propagation distance, resulting in two sets of parallel solution results. By integrating the results of two parallel calculations, the numerical conversion of local coordinates within the station and the point fitting are completed, generating preliminary positioning results.

2. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 1, characterized in that, S1 includes: S11. Scan the physical space of the entire subway station in layers and zones, collect 5G public network signal transmission parameters at different locations and heights, collect 5G private network signal transmission parameters at different locations and heights, and complete the full-area survey and measurement of the signal propagation environment within the station. S12. Integrate the measured signal attenuation data and obstruction loss data, conduct multi-dimensional signal coverage analysis, and form a public network and private network signal strength distribution map covering the entire station; S13. Divide the signal adaptation area according to the signal strength distribution map, and deploy positioning terminals that support dual-mode access of 5G public network and private network at the corresponding locations to complete the full-area deployment of the perception layer hardware. S14. Establish a collaborative communication link between the public network and the private network, open up the data interaction channel between the two networks, and build a hybrid positioning network that integrates the public and private networks. S15. Drive the positioning terminal to conduct continuous sampling throughout the entire area of ​​the station, record the characteristic parameters of the dual-network signals in real time, and synchronously collect and store complete dual-network signal fingerprint characteristic data.

3. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 1, characterized in that, The S2 includes: S21. Establish a secure transmission link from the positioning terminal to the edge computing MEC platform, and upload the collected dual-network signal fingerprint feature data to the local node of the platform. S22. Load the building structure and facility layout parameters of the subway station, use digital twin technology to restore the physical space and signal propagation constraints within the station, and construct a three-dimensional spatial signal propagation model of the subway station. S23. Identify multipath propagation paths of signals using a three-dimensional spatial signal propagation model, and separate multipath interference components from the original signal; S24. Perform phase and amplitude correction on the multipath interference components to complete the multipath pre-compensation processing of the original signal; filter out redundant noise and abnormal fluctuations in the compensated signal to generate enhanced signal data resistant to multipath interference.

4. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 3, characterized in that, S21 includes: S211. Enable the dedicated transmission interface of the edge computing MEC platform and complete the interface parameter configuration and transmission channel initialization; S212. Establish a point-to-point transmission connection between the positioning terminal and the edge computing MEC platform to form a stable and secure data transmission link; S213. Perform timing processing and data encapsulation on the dual-network signal fingerprint feature data collected by the positioning terminal to improve transmission efficiency and integrity. S214. Upload the encapsulated dual-network signal fingerprint feature data packet by packet to the local node of the edge computing MEC platform through the established secure transmission link; S215. Complete data reception and integrity verification on the edge computing MEC platform, and retain usable dual-network signal fingerprint feature data.

5. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 4, characterized in that, S214 includes: Read the real-time transmission parameters of the secure transmission link, match the transmission specifications of the encapsulated dual-network signal fingerprint feature data, and generate a data transmission scheme that adapts to the current link status. According to the data transmission scheme, the encapsulated dual-network signal fingerprint feature data is time-series grouped to form a data unit with continuous transmission capability. Send the completed data units to the local node of the edge computing MEC platform through the established secure transmission link, and synchronously record the transmission timing and transmission location of the data units; Track the transmission status of data units in the transmission link, promptly replenish data units lost due to transmission interruption, and maintain the smooth operation of the transmission process; All data units are pushed to the local node of the edge computing MEC platform to complete the upload of this round of encapsulated data.

6. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 1, characterized in that, The S4 includes: S41. Monitor the signal reception quality and data transmission status of the positioning terminal in real time to determine whether the terminal has entered a signal blind zone; S42. When it is determined that the blind zone has been entered, the blind zone perception mechanism is automatically activated, triggering the data retrieval and aggregation process of the MEC platform. S43. Collect effective positioning data and external context information of the corresponding blind area within the past week through the MEC platform to complete the cleaning and organization of historical data; S44. Utilize graph neural networks to perform deep learning on historical time-series data to extract target movement patterns and regional spatial constraint features; continuously output position estimation results during the continuous period of signal loss, and continuously generate intelligent inference position data.

7. The high-precision positioning method for subway stations based on 5G public / private networks according to claim 1, characterized in that, The S5 includes: S51. Establish a unified upload channel for high-precision positioning data and intelligent inferred location data to the positioning management platform to complete the synchronous aggregation of dual-source data; S52. Configure dynamic weights based on signal quality and data reliability to complete the fusion calculation of dual-source location information and generate a seamless global positioning result. S53. Based on the seamless global positioning results, mark the target coordinates on the visualization map, refresh and display the real-time location of personnel and materials; read the historical location sequence for time-series reconstruction, and generate queryable and playable trajectory playback content; S54. Detect the spatial relationship between the target location and the preset area, identify abnormal behavior and trigger early warning, and provide full-scene location awareness services for subway station operation.

8. A high-precision positioning system for subway stations based on a 5G public / private network, characterized in that, The system includes: One or more processors; Memory, used to store one or more programs; Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 7.

Images