Unified perception base-based intelligent park safe operation situation real-time monitoring system

Abstract

The application provides a smart park safe operation situation real-time monitoring system based on a unified perception base, belongs to the field of park monitoring management, and comprises an edge perception time sequence demand monitoring and dispatching optimization module, a park data transmission grading monitoring module, and a park safety risk early warning and timeliness evaluation module. The application carries out edge perception time sequence demand verification for quantifying the matching condition of the time sequence of a multi-source heterogeneous perception terminal and safety monitoring demand, so as to determine whether to take edge perception time sequence demand adjustment. After the edge perception time sequence demand verification ends, the park data transmission priority grading determination is carried out, so as to determine whether to start a dynamic grading transmission strategy. After the park data transmission priority grading determination ends, the park multi-modal feature extraction and fusion and the park safety risk early warning timeliness evaluation are sequentially carried out, so as to solve the technical problem of low safety situation perception and early warning timeliness of a smart park.

Description

Technical Field

[0001] This invention relates to the field of park monitoring and management technology, and in particular to a real-time monitoring system for the safe operation of smart parks based on a unified sensing platform. Background Technology

[0002] As smart parks become increasingly large-scale and complex, traditional decentralized and isolated monitoring models are no longer sufficient to achieve comprehensive perception and coordinated response to multi-dimensional security risks related to people, vehicles, objects, facilities, and the environment. Therefore, building a unified perception platform aims to break down data silos and achieve the fusion and analysis of multi-source heterogeneous data, thereby improving the real-time performance, accuracy, and intelligence of park security management. The existing implementation process for park security operation monitoring mainly involves: firstly, deploying high-definition AI (Artificial Intelligence) cameras, infrared sensors, and RFID (Radio Frequency Identification) sensors in key areas such as the park perimeter, buildings, computer rooms, and pipelines. Multimodal data is collected by various sensing devices, including identification (RFID) and positioning sensing terminals. The collected multimodal data is then aggregated and integrated by edge gateway devices (such as industrial-grade edge computing gateways and IoT edge gateways) on a unified sensing base, undergoing preliminary edge computing processing and standardization. Next, the processed multimodal data is transmitted with low latency to the park's security monitoring center. Based on data fusion technologies, such as weighted average fusion algorithms and deep learning feature fusion, spatiotemporal correlation and feature extraction are performed. Furthermore, based on artificial intelligence models, such as convolutional neural networks, time-series prediction models, and graph neural network situation assessment models, real-time identification, risk assessment, and situation analysis of security events are conducted. Finally, early warning information and handling suggestions are output to management personnel through a visual large screen or mobile sensing terminal.

[0003] For example, Chinese invention patent application CN117372223A discloses an IoT-based park management and monitoring system, which includes: a park energy monitoring system, a park security monitoring system, and a central control module; the central control module is electrically connected to the park energy monitoring system and the park security monitoring system respectively; the park energy monitoring system includes energy-related basic equipment, an energy monitoring module, and an energy adjustment module; the park security monitoring system includes an access monitoring system, a boundary monitoring system, an internal monitoring system, and an alarm system, and the access monitoring system, the boundary monitoring system, the internal monitoring system, and the alarm system are electrically connected to the central control module.

[0004] The above-mentioned technology has at least the following technical problems:

[0005] In the process of security and prevention, existing park monitoring integrates multiple security monitoring units and energy-related infrastructure equipment in various areas such as entrances and exits, boundaries, and within the park by building an architecture of park energy monitoring system, park security monitoring system and central control module. The central control module is used to realize the centralized aggregation and linkage management of data from various systems. Although the system achieves centralized management of energy and security monitoring through the central control module, its process essentially relies on aggregating all kinds of data collected by the park energy monitoring system and security monitoring system to the cloud or central server through electrical connection for unified processing and analysis. This centralized processing paradigm is effective when the data scale is limited. However, in the new context of the continuous deepening of the digitalization process of smart parks, its inherent bottlenecks are becoming increasingly prominent.

[0006] Specifically, during the monitoring of the park's safe operation, the volume of massive multimodal data (such as image frames recognized by AI cameras, ambient temperature monitored by infrared sensors, and vehicle positioning coordinates monitored by RFID devices) collected by high-definition video streams and high-frequency sensors within the park is growing explosively. However, the unified sensing base supporting this data collection suffers from inconsistencies in the collaborative acquisition timing of multi-source data at the edge and a lack of a synchronous scheduling mechanism. Existing technologies often employ independent data acquisition and parallel data upload by each sensing terminal, failing to dynamically schedule the acquisition frequency and upload timing of different sensing terminals according to the real-time needs of park safety monitoring. This results in inconsistencies in the collaborative acquisition timing of multi-source data. The existing technologies suffer from issues such as misaligned data collection cycles and insufficient data coverage. Furthermore, current technologies typically employ a full-data upload and centralized cloud analysis model rather than a dynamic, hierarchical transmission strategy. This can lead to network bandwidth being consumed by high-definition video streams and high-frequency sensor data, causing transmission congestion. Consequently, the data transmission latency from the edge to the cloud increases significantly. This further exacerbates the time lag in the identification and early warning of abnormal events during the security situation assessment process of artificial intelligence models based on lagging and time-disordered data. Ultimately, the real-time monitoring system for the park's security operation cannot respond to sudden security risks in a timely manner, resulting in low timeliness of smart park security situation perception and early warning. Summary of the Invention

[0007] In view of this, embodiments of the present invention provide a real-time monitoring system for the safe operation of smart parks based on a unified perception platform, which can improve the perception and timeliness of early warning of the safety situation in smart parks.

[0008] The technical solution of this invention is implemented as follows:

[0009] This invention provides a real-time monitoring system for the safe operation of a smart park based on a unified sensing platform, comprising: an edge sensing time-series demand monitoring and scheduling optimization module, a park data transmission hierarchical monitoring module, and a park security risk early warning and timeliness assessment module. The edge sensing time-series demand monitoring and scheduling optimization module initiates an edge sensing time-series demand verification to quantify the matching of the acquisition time sequence of multi-source heterogeneous sensing terminals with security monitoring requirements, and determines whether to adjust the edge sensing time-series demand based on the verification results. The park data transmission hierarchical monitoring module, after the edge sensing time-series demand verification is completed, performs a park data transmission priority hierarchical determination, and determines whether to activate a dynamic hierarchical transmission strategy based on the determination results. The park security risk early warning and timeliness assessment module, after the park data transmission priority hierarchical determination is completed, sequentially performs park multimodal feature extraction and fusion and park security risk early warning timeliness assessment.

[0010] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0011] 1. By verifying the timing requirements of edge sensing, and using the verification results to determine whether to implement collaborative timing calibration and data acquisition frequency scheduling, it helps ensure the timing consistency and acquisition accuracy of multi-source heterogeneous data in the park from the data acquisition source. This reduces data spatiotemporal misalignment caused by timing deviations of different sensing terminals and reduces the waste of network and computing resources caused by invalid data acquisition. After the edge sensing timing requirement verification is completed, the priority of data transmission in the park is classified and determined. The results determine whether to implement a dynamic hierarchical transmission strategy. This helps allocate network transmission resources, prioritizing the low-latency and high-reliability transmission of high-urgency core security data and alleviating link congestion. After the priority classification of data transmission in the park is completed, park data acquisition continues, and multimodal feature extraction and fusion are performed. After feature extraction and fusion, the timeliness of park security risk early warning is assessed. This helps improve the comprehensiveness and accuracy of park security risk identification, ensures the real-time and reliability of smart park security operation status monitoring, and solves the problem of low timeliness of smart park security status perception and early warning.

[0012] 2. By specifically selecting timing synchronization deviation indicators and scenario data demand matching indicators, this solution helps address the problems of existing technologies that often focus solely on timing synchronization or acquisition frequency without combining both as pre-assessment indicators for acquisition quality. This results in situations where timing is synchronized but data is redundant, or acquisition frequency is compatible but timing is disordered. This solution ensures data acquisition quality from two key dimensions: timing consistency and demand adaptability. It effectively identifies the core issues in the multi-source sensing data acquisition process within the park, laying a high-quality data foundation for subsequent data transmission and early warning analysis. The solution determines whether the timing synchronization deviation indicator exceeds a preset synchronization deviation threshold. If so, collaborative timing calibration is initiated. Conversely, it determines whether the scenario data demand matching indicator is less than a preset matching threshold. If so, acquisition frequency scheduling is initiated. Otherwise, priority classification for park data transmission is determined. This helps solve the problem of resource waste or inconsistent data quality caused by the lack of hierarchical processing logic in the acquisition process in existing technologies. This solution improves the efficiency and relevance of the solution, ensuring that data entering the transmission process meets the prerequisites of timing consistency and demand adaptability.

[0013] 3. When the temporal stability entropy of the transmission link from the edge of the unified sensing base to the cloud continues to exceed the preset secondary entropy threshold, it indicates that the network is congested. In full data upload mode, the data transmission latency from the edge to the cloud will increase significantly, failing to meet the real-time requirements of park security monitoring. Therefore, the second scheme for determining the priority of park data transmission needs to be implemented. By specifically acquiring the temporal stability entropy of the park data transmission link, it helps to overcome the limitations of existing technologies that rely solely on single static parameters such as transmission latency and bandwidth utilization to determine network status. The temporal stability entropy of the transmission link selected in this scheme is based on information entropy theory and couples three dynamic dimensions: the temporal distribution of link transmission latency, the coefficient of variation of data packet arrival intervals, and the probability of link bandwidth fluctuations. This allows for a comprehensive and accurate quantification of the dynamic instability and congestion of the link. If the temporal stability entropy of the park data transmission link is greater than the preset primary entropy threshold, only the park data transmission priority index exceeding the preset threshold will be considered. Emergency transmission threshold data is transmitted via a 5G private network, while real-time uploading of other park data is suspended and converted to edge local caching. If the temporal stability entropy of the park data transmission link is not greater than the preset first-level entropy threshold and not less than the preset second-level entropy threshold, then park data with a data transmission priority index greater than the preset regular grading threshold will be transmitted according to a dynamic grading strategy, while other park data will be uploaded in batches based on deep compression. If the temporal stability entropy of the park data transmission link is less than the preset second-level entropy threshold, then a dynamic grading transmission strategy will be implemented. This helps to reduce the delay in core risk warnings due to link congestion in highly congested link states. For moderately congested link states, it can alleviate link congestion while meeting the basic data needs of park security monitoring. For normal link states, it makes full use of network resources, ensuring both the real-time requirements of park security monitoring and improving the utilization efficiency of network resources. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform provided in an embodiment of the present invention;

[0015] Figure 2 This is a general overview of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform, provided in this embodiment of the invention. Figure 1 ;

[0016] Figure 3 This is a general overview of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform, provided in this embodiment of the invention. Figure 2 ;

[0017] Figure 4This is a diagram of the neural network model architecture of the smart park security operation status real-time monitoring system based on a unified perception base provided in this embodiment of the invention;

[0018] Figure 5 This is a comparison chart of the performance indicators of the security risk prediction algorithm of the real-time monitoring system for the safe operation of a smart park based on a unified perception platform provided in this embodiment of the invention;

[0019] Figure 6 This is a comparison chart of the system early warning response time under different risk scenarios of the real-time monitoring system for the safe operation of a smart park based on a unified perception platform, provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0022] Example 1: This embodiment of the invention provides a real-time monitoring system for the safe operation of a smart park based on a unified sensing platform, such as... Figure 1The diagram shows the structure of a real-time monitoring system for the safe operation of a smart park based on a unified sensing platform. This system includes: an edge sensing timing requirement monitoring and scheduling optimization module, a park data transmission hierarchical monitoring module, and a park security risk early warning and timeliness assessment module. The edge sensing timing requirement monitoring and scheduling optimization module initiates an edge sensing timing requirement verification during park safety operation monitoring. This verification quantifies the matching between the timing of data acquisition from multi-source heterogeneous sensing terminals and the safety monitoring requirements. Based on the verification results, it determines whether to adjust the edge sensing timing and acquisition frequency requirements. Edge sensing timing requirement adjustment includes collaborative timing calibration and acquisition frequency scheduling. Collaborative timing calibration allocates standardized time bases to multi-source heterogeneous sensing terminals under the unified sensing platform, reducing timing deviations caused by independent timing synchronization of different sensing terminals. Acquisition frequency scheduling dynamically adjusts the acquisition frequency of each sensing terminal within the unified sensing platform, ensuring precise matching between the acquisition frequency and the smart park safety monitoring requirements. The sensing terminal represents an integrated high-precision timing hardware foundation and protocol adaptation capability (e.g., integrating a BeiDou timing module and supporting IEEE). The multi-source heterogeneous security sensing equipment in the park (such as the 1588 PTP high-precision time synchronization protocol, etc.) includes video surveillance cameras, environmental sensors (such as temperature and humidity sensors, smoke sensors), equipment status monitoring sensors (such as voltage sensors, current sensors), personnel positioning terminals, etc. By monitoring the edge sensing timing requirement verification results, it helps to ensure the timing consistency and demand adaptability of multi-source sensing data from the data collection source, and reduce the waste of resources caused by invalid data collection.

[0023] The park data transmission hierarchical monitoring module is used to determine the priority of park data transmission after the edge sensing timing requirement verification is completed. This is done to quantify the urgency of park data transmission and classify transmission priorities. Based on the hierarchical determination results, it decides whether to activate the dynamic hierarchical transmission strategy. The dynamic hierarchical transmission strategy is used to match differentiated transmission links, compression methods, and upload times for park data with different urgency levels, ensuring low latency and high reliability transmission of park data, while optimizing network resource utilization. Park data represents multi-source heterogeneous sensing data used to support the smart park's safe operation situational awareness, risk assessment, and emergency response. This includes environmental sensing data (such as temperature and humidity, smoke concentration, and light intensity), personnel and target sensing data (such as personnel density, movement speed, and unauthorized intrusion indicators) collected by various sensing terminals under the unified sensing base, equipment operation data (such as equipment voltage, current, operating temperature, and vibration frequency), and link transmission status data (such as transmission latency and bandwidth fluctuation). By monitoring the priority classification of park data transmission, it helps to accurately allocate network resources, ensure the real-time transmission of high-urgency security data, reduce the bandwidth occupation of low-priority data, and alleviate link congestion.

[0024] The park security risk early warning and timeliness assessment module is used to sequentially extract core features of multimodal park data and perform multimodal feature extraction and fusion of the park after the priority classification of park data transmission is completed, and to assess the timeliness of park security risk early warning to quantify the timeliness of park security risk early warning. By monitoring the timeliness of park security risk early warning, it helps to improve the comprehensiveness and accuracy of smart park security risk identification and ensure the real-time and reliability of security monitoring.

[0025] It should be understood that the implementation of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform provided in this application embodiment relies on a pre-built unified sensing platform database. This database includes core judgment parameters verified by park safety management experts, such as preset acquisition frequency, preset synchronization deviation threshold, and preset matching degree threshold. Simultaneously, the unified sensing platform database also includes standardized pre-processed historical data of park safety monitoring (covering time-series data of sensing terminals, transmission link status data, and corresponding hierarchical transmission and early warning handling results records under different park scenarios and different safety risk types) and model iteration optimization data (such as multi-round multimodal feature fusion). (Integrated effect evaluation, artificial intelligence model); This unified perception base database adopts a domain-based heterogeneous storage architecture, and at the same time utilizes a non-relational database to store unstructured data, such as environmental waveform sequences collected by raw sensors, target feature images captured by cameras, and link status fluctuation time series data. Operation and maintenance personnel can conduct periodic verification and dynamic iterative adjustment of various preset parameters stored in the data system based on the real-time collected park perception data and security early warning performance feedback, thereby ensuring that the unified perception base database can continuously adapt to the complex and ever-changing security monitoring scenarios of smart parks, such as peak traffic periods, extreme weather environments, and intensive equipment operation phases.

[0026] In this embodiment, the edge-aware time-series demand monitoring and scheduling optimization module, the park data transmission hierarchical monitoring module, and the park security risk early warning and timeliness assessment module help to achieve intelligent, accurate, and efficient monitoring of the safe operation status of the smart park. Specifically, the edge-aware time-series demand monitoring and scheduling optimization module provides a high-quality park data foundation for subsequent stages, the park data transmission hierarchical monitoring module ensures efficient flow of park data, and the park security risk early warning and timeliness assessment module completes park risk identification and timeliness verification based on the output of the preceding stages. The three modules are connected and work together, which not only improves the independent efficiency of each stage, but also strengthens the overall system's security monitoring capabilities through link collaboration.

[0027] like Figure 2 The flowchart shown is an overview of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform. Figure 1 ,Depend on Figure 2 The process involves: verifying edge-aware timing requirements and acquiring timing synchronization deviation and scene data requirement matching indicators. It determines whether the timing synchronization deviation is greater than a preset synchronization deviation threshold. If not, a requirement adaptability assessment is performed; otherwise, collaborative timing calibration is initiated. After collaborative timing calibration, the timing synchronization deviation is reacquired, and its value is checked against the preset threshold. If it is, a collaborative timing calibration failure message is sent; otherwise, a requirement adaptability assessment is performed, specifically checking whether the scene data requirement matching indicator is less than a preset matching threshold. If not, a priority classification of data transmission within the park is determined; otherwise, data acquisition frequency scheduling is initiated. After data acquisition frequency scheduling, the scene data requirement matching indicator is reacquired, and its value is checked against the preset threshold. If it is, a data acquisition frequency scheduling failure message is sent; otherwise, a priority classification of data transmission within the park is performed.

[0028] Preferably, the specific process for edge sensing timing requirement verification is as follows: A timing synchronization deviation index is calculated based on the absolute difference between timestamps of sensing terminals to evaluate the consistency of multi-source terminal acquisition actions. Specifically, the calculation process involves: calculating the absolute difference between the timestamps of data acquired by any two sensing terminals within the same monitoring period; then summing all the absolute differences to obtain the timing synchronization deviation index. For example, selecting the acquisition timestamps of three types of sensing terminals—AI cameras, infrared sensors, and RFID positioning tags—within the same monitoring period, and sequentially calculating the absolute differences between the timestamps of the camera and infrared sensor, the camera and RFID tag, and the infrared sensor and RFID tag, then summing these three sets of differences to obtain the timing synchronization deviation index. A scenario data requirement matching degree index is calculated based on the actual acquisition frequency of the sensing terminals to quantify the degree of fit between the existing acquisition strategy and monitoring requirements. Specifically, the ratio of the actual acquisition frequency of each sensing terminal monitored by the edge gateway to the preset acquisition frequency is used as the scenario data requirement matching degree index. According to the demand matching index, the closer the ratio is to 1, the better the adaptability of the scene data collection frequency. The preset collection frequency is set in advance by preset personnel. Timing consistency is judged based on the timing synchronization deviation index, and demand adaptability is judged based on the scene data demand matching index. The timing consistency judgment determines whether the timing synchronization deviation index is greater than the preset synchronization deviation threshold. If so, it indicates that the timing deviation of the multi-source sensing terminals is too large and cannot meet the collaborative monitoring requirements, so collaborative timing calibration is initiated. Otherwise, demand adaptability is judged. The preset synchronization deviation threshold is represented by the average value of the timing synchronization deviation index over a historical time period. The demand adaptability judgment determines whether the scene data demand matching index is less than the preset matching threshold. If so, it indicates that the collection frequency of each sensing terminal does not match the regional security requirements, resulting in insufficient data coverage. In this case, collection frequency scheduling is initiated. Otherwise, priority classification of park data transmission is determined. The preset matching threshold is represented by the average value of the scene data demand matching index over a historical time period.

[0029] In this embodiment, edge sensing timing requirement verification helps to effectively solve the problems of misaligned acquisition timing and lack of spatiotemporal correlation of data caused by independent timing of multiple terminals in traditional park security monitoring. It reduces the waste of edge computing power and network bandwidth resources caused by excessively high acquisition frequency of sensing terminals, or the defects of insufficient coverage of key security data caused by excessively low acquisition frequency. It not only ensures the consistency and comparability of multi-source sensing data in the time dimension, laying a high-quality data foundation for subsequent multimodal data fusion analysis, but also ensures that the acquisition strategy of sensing terminals can dynamically adapt to the security monitoring needs of different areas and time periods in the park, thus improving the acquisition quality and adaptability of edge sensing data.

[0030] Preferably, the specific process of collaborative timing calibration is as follows: Based on the edge gateway device of the unified sensing base, sensing terminals with high-precision timing access capabilities are selected as reference timing nodes. The specific selection process is as follows: Step 1.1: Through the edge gateway, call the hardware configuration parameter interface of each sensing terminal to collect the timing module type (such as Beidou timing module, NTP network timing module, PTP precision clock synchronization module, etc.), timing protocol support type (such as SNTP protocol, CDMA timing protocol, etc.), historical timing deviation fluctuation value, and historical operation without timing fault duration; Step 1.2: Obtain the selection criteria, including the historical timing deviation fluctuation value being less than the preset fluctuation threshold, and the historical operation without timing fault duration being greater than the preset stable duration. Among them, the preset fluctuation threshold and the preset stable duration are... The duration is preset by designated personnel; the historical time synchronization deviation fluctuation value is represented by the arithmetic mean of the time synchronization deviation fluctuation values ​​over a historical time period. The time synchronization deviation fluctuation value is obtained by subtracting the absolute value of the average absolute value of the time synchronization deviation of the sensing terminal over the historical time period from the absolute value of the difference between the actual time synchronization time of the sensing terminal at a certain sampling moment and the standard time synchronization time (such as the Beidou satellite time synchronization signal or the national time synchronization center network time synchronization signal). That is: Time synchronization deviation fluctuation value = |actual time synchronization time at a certain moment - standard time synchronization time| - |historical average time synchronization deviation|. The historical time synchronization deviation fluctuation value is used to characterize the stability of the timing accuracy of the sensing terminal; No timing faults in historical operation means that the sensing terminal has not recorded any faults such as timing signal loss, timing deviation exceeding the threshold alarm, or abnormal restart of the timing module within the preset historical statistical time period; Step 1.3. For sensing terminals that meet the above screening criteria (representing various sensing terminals connected to the unified sensing base, including AI cameras, infrared sensors, RFID positioning tags, temperature and humidity sensors, etc.), further determine whether the signal coverage strength (such as satellite signals, network timing signals, etc.) at the deployment location of the sensing terminal is greater than a preset signal strength threshold. If so, mark the corresponding sensing terminal as a qualified candidate for benchmark timing; otherwise, mark it as an unqualified candidate for benchmark timing. If multiple qualified candidates exist, select the sensing terminal with the smallest timing deviation fluctuation value over a historical period as the qualified benchmark timing sensing terminal. If no qualified candidates exist, send a benchmark timing node screening failure message. The preset signal strength threshold is represented by the average signal coverage strength at the deployment location of the sensing terminal over a historical period; the signal coverage strength is represented by the signal received power (unit: dBm). The process involves the sensing terminal collecting real-time timing signal power values. For example, if the coverage strength of the BeiDou satellite timing signal exceeds a preset BeiDou signal strength threshold, or the network timing signal coverage strength exceeds a preset network signal strength threshold, the signal coverage strength is deemed to meet the requirements, and the corresponding sensing terminal is marked as a qualified candidate sensing terminal for reference timing. A standard time reference signal is then transmitted to all sensing terminals within the unified sensing base based on a timing protocol (such as SNTP or CDMA). After receiving the standard time reference signal, each sensing terminal automatically calibrates its local clock and generates a data acquisition timestamp according to the calibrated time reference. The calibrated timestamp data of each sensing terminal is re-collected, and the timing synchronization deviation index of the multi-source sensing terminals is calculated to verify the calibration effect. If the calibrated timing synchronization deviation index is not greater than a preset synchronization deviation threshold, a demand adaptability assessment is performed. If the timing synchronization deviation index is still greater than the preset synchronization deviation threshold, a collaborative timing calibration failure notification is sent.

[0031] In this embodiment, collaborative timing calibration helps to build a standardized time reference system for multi-source heterogeneous sensing terminals under a unified sensing base, breaking the discrete state of independent timing of different sensing terminals, reducing the risk of data spatiotemporal correlation breakage caused by misalignment of data collection timing of multiple terminals, improving the synergy and utilization efficiency of multi-source sensing data, and providing reliable underlying data support for real-time monitoring of the safe operation status of smart parks.

[0032] Preferably, the specific process of frequency acquisition scheduling is as follows: Step 2.1: Collect security requirement level data and real-time scene status data of each monitoring area in the park through the edge gateway of the unified sensing base; the security requirement level data represents the preset area level label, such as Level 1 key area: perimeter entrances and exits, computer room; Level 2 regular area: office corridor; Level 3 non-key area: green area; real-time scene status data includes the current area personnel density, equipment operating status and frequency of historical abnormal events; Step 2.2: Input the security requirement level data, real-time scene status data and unified sensing base CPU utilization rate into the preset acquisition frequency adjustment mapping table, and output the acquisition frequency adjustment coefficient corresponding to the deployment area of ​​each sensing terminal, which is used to characterize the adjustment range of the terminal's baseline acquisition frequency; the unified sensing base CPU utilization rate is represented by the result of the ratio of the sum of the CPU time of all edge computing nodes in working state to the total available CPU time during the monitoring period; Step 2.3: For each sensing terminal, the base The target acquisition frequency of the sensing terminal is obtained based on the acquisition frequency adjustment coefficient. The target acquisition frequency is represented by the product of the baseline acquisition frequency corresponding to the deployment area of ​​the sensing terminal and the acquisition frequency adjustment coefficient (e.g., for an AI camera deployed at the perimeter entrance of a Level 1 key area, the baseline acquisition frequency is preset to 30fps. When a real-time intrusion alarm signal is triggered in this area, the acquisition frequency adjustment coefficient matched from the preset acquisition frequency adjustment mapping table is 200%, then the target acquisition frequency of the camera = 30fps × 200% = 60fps, etc.). The baseline acquisition frequency is the basic acquisition frequency of the terminal in each area preset based on the security requirement level (e.g., the baseline acquisition frequency of Level 1 key area: camera 30fps, temperature sensor 2 times / second; the baseline acquisition frequency of Level 2 regular area: camera 20fps, temperature sensor 1 time / second; the baseline acquisition frequency of Level 3 non-key area: camera 10fps, temperature sensor 1 time / 5 seconds). The acquisition frequency of the sensing terminal is adjusted based on the target acquisition frequency.

[0033] It should be understood that the preset acquisition frequency adjustment mapping table on which the acquisition frequency scheduling optimization in this application embodiment depends is pre-built by park security operation and maintenance experts in combination with massive historical data of park monitoring scenarios and stored in the edge gateway database of the unified perception base. When the system performs acquisition frequency scheduling optimization, it can directly import the input security requirement level data, real-time scene status data and unified perception base CPU utilization into the mapping table, quickly match and output the acquisition frequency adjustment coefficient of the corresponding perception terminal deployment area, and ensure that the acquisition frequency adjustment is accurately adapted to the actual security monitoring needs of the park and the operating load of the equipment.

[0034] Specifically, the preset acquisition frequency adjustment mapping table adopts a three-dimensional relational table structure. The horizontal dimension of the table indicates the security requirement level, the vertical dimension indicates the real-time scene status type, and the depth dimension indicates the CPU (Central Processing Unit) occupancy range of the unified sensing base. The data of the three-dimensional cross nodes in the table are the acquisition frequency adjustment coefficients under the corresponding parameter combinations. The construction process of the mapping table fully relies on multi-dimensional historical data resources, covering the acquisition parameter combinations of sensing terminals under different security requirement levels, different scene statuses, and different device loads. Each set of parameters is given a comprehensive quantitative score based on the validity of the acquired data, device energy consumption, and network bandwidth usage. For example, when the security requirement level is a first-level key area and the real-time scene is a high population density in the current area, a larger acquisition frequency adjustment coefficient is matched to increase the data acquisition density. At the same time, the actual application effect data of the acquisition frequency adjustment coefficient in each historical scene is recorded. Through correlation analysis methods (such as Kendall correlation analysis), abnormal parameter combinations caused by instantaneous failure of sensing terminals, data transmission interference, etc. are eliminated, and the statistically significant parameter correspondences are retained to ensure the stability and scene adaptability of the mapping table output results.

[0035] Specifically, the process of adjusting the acquisition frequency of the sensing terminals is as follows: The edge gateway sends a target acquisition frequency adjustment command to each sensing terminal. This command indicates that the acquisition frequency of the current sensing terminal is adjusted step-by-step towards the target acquisition frequency, with a step size representing the adjustment range. The step size represents the result of a ratio calculation between the target acquisition frequency and a preset number of adjustments, which is pre-set by designated personnel. After receiving the target acquisition frequency adjustment command, each sensing terminal adjusts its current acquisition frequency to the target acquisition frequency and sends a signal indicating that the frequency adjustment is complete to the edge gateway. After receiving the feedback from each sensing terminal indicating that the acquisition frequency adjustment is complete, the edge gateway re-acquires the scene data demand matching index to verify the scheduling optimization effect. If the scene data demand matching index is not less than the preset matching threshold, the system enters the park data transmission priority grading determination stage; otherwise, it sends a acquisition frequency scheduling failure prompt.

[0036] In this embodiment, by scheduling the collection frequency, it is helpful to dynamically adjust the collection frequency of each sensing terminal in the unified sensing base according to the different security monitoring needs of different areas and time periods in the smart park. This reduces the generation of redundant data, reduces the resource occupation pressure on edge gateway computing power and network transmission bandwidth caused by invalid data collection, and avoids problems such as excessive system load and data transmission congestion caused by high-frequency collection of all terminals. At the same time, it can also reduce the energy consumption of sensing terminals, extend the service life of equipment, and ensure that the collected data can fully cover the key security monitoring points in the park without data redundancy caused by excessive collection. This provides a high-quality data source for subsequent priority classification and security risk warning of park data transmission.

[0037] like Figure 3 The flowchart shown is an overview of the real-time monitoring system for the safe operation of a smart park based on a unified sensing platform. Figure 2 ,Depend on Figure 3 It is known that: In the process of determining the priority of data transmission within the park, a priority index is obtained. If the priority index is greater than the preset first-level priority threshold, transmission is prioritized via the 5G private network. If the priority index is not greater than the preset first-level priority threshold but is greater than the preset second-level priority threshold, data is lightly compressed using the unified perception base edge gateway and uploaded periodically via a regular communication link. If the priority index is not greater than the preset second-level priority threshold, data is deeply compressed using the unified perception base edge gateway, batch-cached, and uploaded during off-peak network periods. After the priority determination is completed, park data is continuously collected and transmitted to the park security monitoring center for feature extraction and fusion. After feature extraction and fusion, the timeliness of park security risk warnings is assessed, and a warning response timeliness index is obtained. It is then determined whether the warning response timeliness index is less than the preset response threshold. If so, real-time monitoring of the park's security operation status continues; otherwise, a warning of insufficient timeliness is sent.

[0038] Preferably, the specific process for determining the priority of data transmission in the park is as follows: Obtain a park data transmission priority index to quantify the urgency of data transmission output by the unified sensing base station; a larger value indicates a higher urgency. The park data transmission priority index is represented by the product of the park data volume deviation rate and the park data timeliness value. The park data volume deviation rate is represented by the ratio of the park data volume deviation value to the preset maximum number of bytes per packet, used to quantify the inverse impact of data volume on data transmission priority. The preset maximum number of bytes per packet is set in advance by designated personnel. The park data volume deviation value is calculated by comparing the preset maximum number of bytes per packet with the actual number of bytes per packet monitored by the edge gateway's data packet byte counter. The result of the difference operation on the number of bytes is represented; the number of bytes per packet represents the actual number of bytes of data stored in a single data packet in the unified perception base, which is a basic attribute parameter of the data packet that can be directly collected by the edge gateway; the timeliness value of the park data is represented by the result of the ratio operation of the deviation value of the waiting time of the park data to be transmitted to the preset maximum waiting time of the data, which is used to quantify the reverse impact of the waiting time of the data after generation on the transmission priority, wherein the preset maximum waiting time of the data is set in advance by preset personnel; the deviation value of the waiting time of the park data is represented by the result of the difference operation on the preset maximum waiting time of the data to be transmitted to the actual waiting time of the data monitored by the edge side data cache queue duration monitor; a dynamic hierarchical transmission strategy is executed based on the park data transmission priority index.

[0039] Specifically, the dynamic hierarchical transmission strategy proceeds as follows: If the park data transmission priority index is greater than the preset first-level priority threshold, it indicates that the data volume is small and the transmission time after generation is extremely short, with the highest urgency. Therefore, first-level park data transmission is performed to ensure data integrity and timely transmission. The preset first-level priority threshold is set in advance by designated personnel. If the park data transmission priority index is not greater than the preset first-level priority threshold, but is greater than the preset second-level priority threshold, it indicates that the data volume or transmission time is moderate, with a moderate urgency. Therefore, second-level park data transmission is performed. If the park data transmission priority index is not greater than the preset second-level priority threshold, it indicates that the data volume is large or the transmission time after generation is long, with the lowest urgency. Therefore, third-level park data transmission is performed. The preset second-level priority threshold is represented by the average value of the park data transmission priority index over a historical period. The preset first-level priority threshold is greater than the preset second-level priority threshold. After the dynamic hierarchical transmission strategy ends, park data collection continues, and the collected park data is transmitted to the park security monitoring center for multimodal feature extraction and fusion. First-level park data transmission indicates prioritizing park data based on the 5G private network. Transmission is performed without data compression. Secondary transmission of park data involves lightweight compression of park data via the unified sensing base edge gateway, followed by periodic uploads via regular communication links. Tertiary transmission involves deep compression of park data via the unified sensing base edge gateway, batch caching, and uploads during off-peak network periods. Batch caching involves temporarily storing the deeply compressed park data in a local cache at the edge of the unified sensing base, integrating and packaging the data according to a preset data volume to form standardized batch data packets. Off-peak network periods are pre-set by designated personnel, such as 00:00-06:00 daily, with the preset data volume set in advance. Lightweight compression uses lossless compression algorithms to compress park data, reducing data transmission volume while ensuring no information loss after decompression. Deep compression uses lossy compression algorithms to compress non-core data, significantly reducing data transmission volume while only losing non-critical redundant information, adapting to the transmission needs of low-priority data. Periodic uploads via regular communication links involve uploading the lightweight compressed park data in chunks at preset fixed time intervals via the park's regular IP communication links during preset periods of stable network load.

[0040] In this embodiment, by using a graded classification and dynamic graded transmission strategy for data transmission in the park, the limitations of indiscriminate scheduling of traditional park data transmission can be overcome. This reduces the risks of high latency and packet loss rate in high-priority secure data transmission caused by non-core data vying for bandwidth resources. It also reduces the delay of critical security early warning information caused by link congestion and reduces the ineffective occupation of edge gateway computing power and network bandwidth by redundant data transmission. This improves the overall efficiency, reliability and resource utilization of park data transmission, achieves optimal transmission scheme matching for data of different priorities, and enhances the system's adaptability and anti-interference capability in complex park scenarios.

[0041] Preferably, the specific process of multimodal feature extraction and fusion in the park is as follows: Based on data fusion methods, such as weighted average fusion algorithms and deep learning feature fusion, feature extraction and fusion are performed to obtain multimodal features of the park; the multimodal features of the park are input into a preset artificial intelligence model, such as a convolutional neural network, a temporal prediction model, or a graph neural network situation assessment model, to output park security risk early warning results, such as perimeter intrusion early warning, equipment failure early warning, overcrowding early warning, and fire hazard early warning. Simultaneously, the actual inference time of the output park security risk early warning results is obtained based on a high-precision timing interface, and the timeliness of the park security risk early warning is evaluated based on the actual inference time. The specific training process of the artificial intelligence model is as follows: First, a comprehensive... The dataset contains labeled datasets for various security scenarios within the park. These datasets include multimodal historical data collected by the unified perception base, such as video frame features, environmental sensor time-series data, and equipment operating parameters, along with corresponding security risk type annotations and early warning timeliness labels. The dataset is then divided into training, validation, and test sets according to a pre-defined ratio. A pre-defined model is iteratively trained on the training set, with the weight parameters continuously optimized using a backpropagation algorithm. Finally, the validation set is used to monitor overfitting during model training and adjust the model structure and hyperparameters. The test set is then used to validate the model's performance, ensuring that the model possesses high-precision risk identification capabilities and efficient inference speed across various security scenarios. The pre-defined ratios are set in advance by designated personnel.

[0042] The assessment of the timeliness of park security risk early warnings involves determining whether the early warning response timeliness index is less than the preset response threshold. If so, the timeliness of the park security risk early warning is deemed to meet the park's security monitoring needs, and real-time monitoring of the park's security operation status continues. Conversely, if the timeliness is less than the preset threshold, the early warning response timeliness is deemed to be insufficient, a warning timeliness deficiency alert is sent, and key information about the insufficient timeliness event is recorded, including the early warning type, actual inference time, the corresponding data transmission link timing stability entropy value, and transmission strategy execution details. This key information is uploaded to the park security monitoring center for subsequent optimization. The preset response threshold is represented by the average value of the early warning response timeliness index over a historical time period. The early warning response timeliness index is represented by the ratio of the actual inference time of the park security risk early warning result to the maximum allowable inference time of the preset model, used to quantify the timeliness level of the early warning result generation. The smaller the ratio, the stronger the early warning timeliness. The maximum allowable inference time of the preset model is set in advance by preset personnel.

[0043] In this embodiment, by extracting and fusing multimodal features of the park and assessing the timeliness of park security risk early warning, it helps to break down information silos of different types of sensing data, reduce monitoring blind spots and judgment biases existing in a single data dimension, improve the accuracy, comprehensiveness and timeliness of park security risk early warning, and strengthen the closed-loop management capability of the real-time monitoring system for the overall security operation of the smart park.

[0044] Example 2, as an alternative to the priority classification of campus data transmission in Example 1, indicates that network congestion has occurred when the temporal stability entropy of the transmission link from the edge of the unified sensing base to the cloud exceeds the preset secondary entropy threshold. In the full data upload mode, the data transmission latency from the edge to the cloud will increase significantly, failing to meet the real-time requirements of campus security monitoring. Therefore, a second scheme for priority classification of campus data transmission is required. The specific process for priority classification of campus data transmission is as follows: Obtain the temporal stability entropy of the campus data transmission link to quantify the dynamic fluctuation characteristics of the transmission link from the edge to the cloud. The temporal stability entropy of the campus data transmission link is a link state quantification index based on information entropy theory. It is obtained by weighting and summing the temporal distribution entropy of the link transmission latency within a preset time window, the coefficient of variation entropy of the data packet arrival interval, and the probability entropy of the link bandwidth fluctuation with the corresponding temporal stability impact parameters of the campus data transmission link. The temporal stability impact parameters include the temporal distribution entropy influence coefficient, which reflects the degree of influence of the temporal distribution entropy on the temporal stability entropy of the campus data transmission link, and the coefficient of variation entropy, which reflects the influence of the coefficient of variation entropy on the temporal stability entropy of the campus data transmission link. The coefficient of variation of the influence of entropy and the coefficient of probability entropy, which reflect the influence of probability entropy on the temporal stability of the data transmission link in the park; the temporal distribution entropy of the link transmission delay, which characterizes the disorder of the data transmission delay distribution in the park (the higher the entropy value, the more chaotic the delay distribution, and the more unstable the link), are obtained by: dividing the transmission delay time sequence of each frame of park data packets within a preset time window into several discrete intervals (such as interval A, interval B, interval C, etc.) based on a preset delay interval division; and using frequency statistics methods (such as histogram statistics, etc.). Methods such as interval counting statistics are used to count the frequency of transmission delay data within each discrete interval, calculate the probability distribution of each interval, and quantify the temporal distribution entropy based on the probability distribution of each interval. The transmission delay time series sequence represents a one-dimensional time series data sequence formed by orderly arranging the end-to-end transmission delay of each data packet frame according to the actual transmission order of data packets within a preset time window. Each data node in the sequence uniquely corresponds to the transmission delay value of a single data packet frame, and the node order is completely consistent with the time sequence of data packet transmission. For example, the transmission delay time series sequence is [t1, t2, ...[t1, tn], where t1 represents the end-to-end transmission delay of the first completed transmission of a campus data packet within a preset time window, and tn represents the end-to-end transmission delay of the nth completed transmission of a campus data packet within a preset time window. The preset time window and the preset delay intervals are pre-set by preset personnel. The quantitative calculation of the temporal distribution entropy is based on the information entropy formula to transform the disorder of the delay distribution of the link transmission delay. The transmission delay is represented by the difference between the timestamp of the campus data packet being sent from the edge gateway of the unified sensing base and the timestamp of the cloud receiving the packet. The probability distribution of each interval is represented by the ratio of the frequency of occurrence of transmission delay data in a single interval to the total number of occurrences of transmission delay data.

[0045] Specifically, the information entropy formula is expressed as follows:

[0046]

[0047] Where H represents the temporal distribution entropy of the link transmission delay, i represents the index of the discrete interval, i=1,2,3,...,m, m represents the total number of discrete intervals, and P... i This represents the probability distribution value corresponding to the i-th discrete interval.

[0048] The coefficient of variation (COV) entropy of data packet arrival intervals is used to characterize the regularity of data transmission timing within the park (the higher the entropy value, the greater the fluctuation in arrival intervals and the more severe the link timing instability). It is obtained through the following method: First, a cloud-based data receiving timestamp collector monitors the arrival timestamps of two adjacent data packets within a preset time window, calculates the difference between adjacent timestamps to obtain the data packet arrival interval timing sequence, divides the COV into multiple discrete levels based on a preset COV level classification standard, and calculates the probability of occurrence of each level using frequency statistics. The COV entropy is then quantified based on the probability of occurrence of each level. The preset COV level classification standard is pre-set by designated personnel. The quantification of COV entropy represents the quantification of the fluctuation degree of data packet arrival intervals based on the information entropy formula. The COV is calculated by subtracting the standard deviation of the data packet arrival interval timing sequence from the mean value of the sequence. The result of the ratio operation is used to quantify the relative degree of sequence fluctuation; the probability entropy of link bandwidth fluctuation is used to characterize the fluctuation characteristics of bandwidth resource supply (the higher the entropy value, the more unstable the bandwidth supply, and the higher the risk of data transmission lag). It is obtained by: the instantaneous bandwidth within a preset time window monitored by the edge gateway link bandwidth real-time sampler, generating a bandwidth fluctuation time series, performing differential processing on the bandwidth fluctuation time series to obtain bandwidth fluctuation amplitude data, dividing the fluctuation amplitude into preset fluctuation intervals based on a preset fluctuation amplitude threshold, and calculating the proportion probability of each interval based on the frequency statistics method, and performing probability entropy quantification based on the proportion probability of each interval; the quantification calculation of probability entropy represents the quantitative transformation of bandwidth resource supply fluctuation characteristics based on the information entropy formula; and a transmission priority control strategy is implemented based on the time series stability entropy of the campus data transmission link.

[0049] Specifically, the transmission priority control strategy works as follows: If the temporal stability entropy of the campus data transmission link is greater than the preset first-level entropy threshold, indicating that the network is in a state of severe congestion, then only campus data with a data transmission priority index greater than the preset emergency transmission threshold will be transmitted through the 5G private network, while the real-time upload of other campus data will be suspended and converted to edge local caching. The preset first-level entropy threshold and the preset emergency transmission threshold are both pre-set by pre-defined personnel. If the temporal stability entropy of the campus data transmission link is not greater than the preset first-level entropy threshold, and the temporal stability entropy of the campus data transmission link is not less than the preset second-level entropy threshold, indicating that the network is in a state of moderate congestion, then campus data with a data transmission priority index greater than the preset regular grading threshold will be transmitted according to the dynamic grading strategy, while other campus data will be processed based on deep compression. Batch uploads are performed in several ways. The preset regular tiered threshold is set in advance by designated personnel. If the temporal stability entropy of the park's data transmission link is less than the preset secondary entropy threshold, the network is considered to be in normal condition, and a dynamic tiered transmission strategy is executed. The preset secondary entropy threshold is represented by the average temporal stability entropy of the park's data transmission link over a historical period, and the preset primary entropy threshold is greater than the preset secondary entropy threshold. Batch uploads based on deep compression involve deep compression of park data through the unified perception base edge gateway, followed by temporary storage in the edge local cache module. When the temporal stability entropy of the park's data transmission link is less than the preset secondary entropy threshold, the park data is batch-packaged and uploaded. The preset emergency transmission threshold is greater than the preset primary priority threshold, which in turn is greater than the preset regular tiered threshold, and the preset regular tiered threshold is greater than the preset secondary priority threshold.

[0050] It should be understood that the timing stability impact parameter system on which the priority classification of data transmission in the embodiment of this application depends is pre-constructed by the park network operation and maintenance experts in combination with historical data of massive link operation scenarios and stored in the unified perception base data. This provides the core basis for the weighted coupling calculation of the timing distribution entropy of link transmission delay, the coefficient of variation of data packet arrival interval, the probability entropy of link bandwidth fluctuation and the timing stability entropy of park data transmission link, and directly determines the accuracy of the quantitative assessment of link status.

[0051] Specifically, the construction process of this time-series stability impact parameter system first requires collecting a large amount of historical link operation data, covering different park scenarios such as peak personnel hours, concentrated equipment operation periods, extreme weather environments, and three types of sub-entropy data (time-series distribution entropy, coefficient of variation entropy, and probability entropy) under different link load levels (light load / medium load / heavy load), as well as corresponding actual link operation status records (e.g., stable / slightly unstable / severely unstable) and data transmission quality feedback (e.g., latency compliance rate, packet loss rate). Simultaneously, complete sample data of different sub-entropy parameter combinations and the actual values ​​of the time-series stability entropy of the park's data transmission links are compiled. Each sub-entropy parameter combination is assigned a weighted quantitative value based on its impact on the link stability assessment results. For example, when the time-series distribution entropy of the link transmission latency exceeds the preset benchmark range, it leads to impaired core security data transmission timeliness. When risks increase significantly, a higher temporal distribution entropy influence coefficient is matched to strengthen its weight in the coupling calculation. The actual effective values ​​of the temporal distribution entropy influence coefficient, coefficient of variation entropy influence coefficient, and probability entropy influence coefficient under each historical scenario are recorded simultaneously. Then, through correlation analysis (such as Spearman rank correlation coefficient), abnormal correlation data caused by instantaneous equipment anomalies, such as temporary edge gateway lag, sensor signal drift, external sudden interference such as short-term electromagnetic interference, and network peak impact are eliminated. The parameter correspondence with statistical stability is retained. Finally, all effective data are integrated to form a complete temporal stability influence parameter system. When the system performs temporal stability entropy calculation of the park data transmission link, various influence coefficients that match the current link operation scenario can be quickly retrieved from this system to ensure the accuracy and efficiency of weighted coupling calculation.

[0052] In this embodiment, by using a graded classification of data transmission priorities within the campus and corresponding transmission priority control strategies, the limitations of traditional fixed-level models in dealing with link congestion fluctuations can be overcome. This enables accurate assessment of the congestion status of the transmission link from the edge to the cloud, effectively reducing risks such as excessive latency and packet loss rate in core security data transmission under congested conditions, as well as delays in emergency response caused by non-critical data occupying limited bandwidth resources. At the same time, it reduces the impact of link fluctuations on overall data transmission efficiency, minimizes network congestion caused by blindly uploading all data, improves the anti-interference capability, resource utilization, and reliability of campus data transmission, and ensures that all types of campus data can obtain the optimal transmission solution in complex link environments.

[0053] like Figure 4The diagram shows the architecture of a graph neural network model for real-time monitoring of smart park safety operation based on a unified perception platform. This graph neural network model takes multimodal safety features of the park (including data on perimeter perception, equipment operation, personnel distribution, and fire status) as input. First, it achieves unified encoding of multi-source heterogeneous features through node representation of the input layer. The core processing layer sequentially completes feature processing through a graph structure estimation layer (dynamically learning the association graph structure of park safety features based on the temporal correlation capability of LSTM), a dynamic neural network layer (capturing the spatiotemporal correlation evolution law of different safety elements with graph structure as constraint), and a safety risk feature fusion layer (integrating multi-dimensional features to achieve complementary enhancement of cross-modal and cross-temporal information). Finally, the output layer outputs safety risk warning results for multiple scenarios such as perimeter intrusion, equipment failure, excessive personnel density, and fire hazards. Through the end-to-end process from multimodal feature encoding to graph structure dynamic learning to spatiotemporal feature fusion to risk warning output, intelligent and accurate early warning of park safety risks is achieved.

[0054] like Figure 5 The chart showing the performance comparison of security risk prediction algorithms in a real-time monitoring system for the safe operation of a smart park based on a unified perception platform demonstrates the comparison of different security risk prediction algorithms in terms of precision, recall, and F1 score. The horizontal axis represents the different algorithms used, including: traditional rule matching, CNN (Convolutional Neural Network) (image features only), LSTM (Long Short-Term Memory Network) (temporal features only), and ST-GCN (Spatial Temporal Graph Convolutional Network). (Spatiotemporal Graph Convolutional Network), the vertical axis represents the algorithm's performance metrics, specifically precision, recall, and F1 score. Precision represents the proportion of samples predicted as positive that are actually positive; recall represents the proportion of all true positive samples correctly predicted as positive; and the F1 score is the harmonic mean of precision and recall, used to comprehensively evaluate the model's accuracy. The bar charts below each algorithm show its specific performance on these three metrics, with numerical values ​​labeled on each bar. Error bars represent the fluctuation range across multiple experiments. Values ​​in the charts (e.g., 0.72, 0.81, etc.) are directly marked, clearly showing the specific results of different algorithms under each performance metric. According to the results in the charts, ST-GCN performs best across all metrics, demonstrating the model's superiority in multi-task learning.

[0055] like Figure 6The chart showing the comparison of system early warning response times under different risk scenarios in the real-time monitoring system for the safe operation of a smart park based on a unified perception platform illustrates the distribution of system early warning response times under different security risk scenarios. The horizontal axis represents different security risk scenarios, including: perimeter intrusion, equipment failure, personnel gathering, fire hazard, and gas leak. The vertical axis represents the system response time (in seconds). The box plot in the figure shows the statistical distribution of response time under each scenario: the red line in the middle of the box represents the median response time, and the upper and lower boundaries of the box represent the 25th and 75th percentiles (i.e., quartiles) of the response time, reflecting the concentrated distribution of the data. The "whiskers" of the box plot represent the maximum and minimum values ​​of the data. After removing outliers, the outliers (blue dots in the figure) represent individual data with significantly higher response times. The red dashed line represents the set upper limit of response time (5 seconds), which serves as a performance benchmark. Responses exceeding this time are considered unqualified. As can be seen from the figure, the response times of some scenarios (such as perimeter intrusion and equipment failure) are mostly concentrated below the upper limit, while the response times of fire hazard and gas leak scenarios have a wider distribution.

[0056] The above description is only an optional embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A real-time monitoring system for the safe operation of a smart park based on a unified perception platform, including: Edge-aware timing demand monitoring and scheduling optimization module, campus data transmission hierarchical monitoring module, and campus security risk early warning and timeliness assessment module: The edge sensing timing requirement monitoring and scheduling optimization module is used to initiate edge sensing timing requirement verification to quantify the matching of the acquisition timing of multi-source heterogeneous sensing terminals with security monitoring requirements, and to decide whether to adjust the edge sensing timing requirements based on the verification results. The park data transmission hierarchical monitoring module is used to determine the priority of park data transmission after the edge-aware timing requirement verification is completed, and to decide whether to start the dynamic hierarchical transmission strategy based on the hierarchical determination result. The park security risk early warning and timeliness assessment module is used to sequentially extract and fuse multimodal features of the park and assess the timeliness of park security risk early warning after the priority classification of park data transmission is completed. The specific process of edge-aware temporal requirement verification is as follows: Calculate the timing synchronization deviation index based on the absolute difference of timestamps between sensing terminals; The matching degree index of scenario data demand is calculated based on the actual collection frequency of the sensing terminal. Timing consistency is assessed based on timing synchronization deviation index, and demand adaptability is assessed based on scenario data demand matching degree index. The timing consistency assessment means determining whether the timing synchronization deviation index is greater than the preset synchronization deviation threshold. If it is, it means that the timing deviation of the multi-source sensing terminal is too large and cannot meet the collaborative monitoring requirements. In this case, collaborative timing calibration is initiated. Otherwise, a requirement adaptability assessment is performed. The demand adaptability assessment means determining whether the scenario data demand matching degree index is less than the preset matching degree threshold. If it is, the collection frequency scheduling is started; otherwise, the priority classification of park data transmission is determined. The specific process of the collaborative timing calibration is as follows: Based on the edge gateway device with a unified sensing platform, sensing terminals with high-precision time synchronization access capabilities are selected as reference time synchronization nodes. The specific selection process is as follows: By calling the hardware configuration parameter interface of each sensing terminal through the edge gateway, the timing module type, timing protocol supported type, historical timing deviation fluctuation value, and historical running time without timing failure of the sensing terminal are collected. The selection criteria include historical time synchronization deviation fluctuation values ​​being less than the preset fluctuation threshold, and the duration of historical operation without time synchronization failures being greater than the preset stable duration. For sensing terminals that meet the above screening criteria, it is further determined whether the signal coverage strength of the sensing terminal deployment location is greater than the preset signal strength threshold. If so, the corresponding sensing terminal is marked as a qualified sensing terminal for reference timing; otherwise, the corresponding sensing terminal is marked as an unqualified sensing terminal for reference timing. Based on the timing protocol, a standard time reference signal is sent to all sensing terminals within the unified sensing base. After receiving the standard time reference signal, each sensing terminal automatically calibrates its local clock and generates a data acquisition timestamp according to the calibrated time reference. Re-collect the timestamp data of each sensing terminal after calibration, calculate the timing synchronization deviation index of the multi-source sensing terminals, and verify the calibration effect; If the calibrated timing synchronization deviation index is not greater than the preset synchronization deviation threshold, then a demand adaptability assessment is performed. If the timing synchronization deviation index is still greater than the preset synchronization deviation threshold, then a collaborative timing calibration failure prompt is sent.

2. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 1, characterized in that, The specific process of frequency scheduling is as follows: Collect security requirement level data and real-time scene status data for each monitored area in the park; The security requirement level data represents a preset area level label; Input the security requirement level data, real-time scene status data and unified perception base CPU utilization into the preset acquisition frequency adjustment mapping table, output the acquisition frequency adjustment coefficient corresponding to the deployment area of ​​each perception terminal, and obtain the target acquisition frequency of each perception terminal based on the acquisition frequency adjustment coefficient. Adjust the acquisition frequency of the sensing terminal based on the target acquisition frequency.

3. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 2, characterized in that, The specific process for adjusting the sampling frequency of the sensing terminal is as follows: The edge gateway sends a target acquisition frequency adjustment command to each sensing terminal. The target acquisition frequency adjustment command means adjusting the current sensing terminal's acquisition frequency step by step in the direction of approaching the target acquisition frequency, with the adjustment range being a stepped step size. After receiving the target acquisition frequency adjustment instruction, each sensing terminal adjusts its current acquisition frequency to the target acquisition frequency and sends a frequency adjustment completion signal back to the edge gateway. After receiving feedback from each sensing terminal that the collection frequency adjustment has been completed, the edge gateway re-acquires the scene data demand matching index to verify the scheduling optimization effect. If the matching degree index of the scenario data demand is not less than the preset matching degree threshold, the data transmission priority classification judgment in the park will be initiated; otherwise, a collection frequency scheduling failure prompt will be sent.

4. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 3, characterized in that, The specific process for determining the priority level of data transmission in the park is as follows: Obtain the priority indicators for data transmission within the park; The priority index for data transmission in the park is represented by the product of the park data volume deviation rate and the park data timeliness value. A dynamic hierarchical transmission strategy is implemented based on the campus data transmission priority index. The specific process of the dynamic hierarchical transmission strategy is as follows: If the priority index of data transmission in the park is greater than the preset first-level priority threshold, then the first-level data transmission in the park will be carried out to ensure data integrity and transmission timeliness. If the priority index of data transmission in the park is not greater than the preset first-level priority threshold, and the priority index of data transmission in the park is greater than the preset second-level priority threshold, then second-level data transmission in the park will be performed. If the priority index of data transmission in the park is not greater than the preset secondary priority threshold, then the park data will be transmitted at the third level. After the dynamic hierarchical transmission strategy ends, park data will continue to be collected and transmitted to the park security monitoring center for multimodal feature extraction and fusion.

5. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 4, characterized in that, The term "primary data transmission in the park" means that park data will be transmitted preferentially based on the 5G private network without data compression. The secondary transmission of park data refers to the periodic uploading of park data through a conventional communication link after lightweight compression of the park data based on the edge gateway of the unified perception base. The three-level transmission of park data refers to the deep compression of park data based on the unified perception base edge gateway, followed by batch caching and uploading during off-peak network periods. The lightweight compression refers to the compression of park data based on a lossless compression algorithm; The deep compression refers to the compression of non-core data based on a lossy compression algorithm; The scheduled upload via the conventional communication link refers to the uploading of lightly compressed park data in segments at preset fixed time intervals during a preset period of stable network load.

6. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 4, characterized in that, The specific process of multimodal feature extraction and fusion in the park is as follows: Feature extraction and fusion are performed based on data fusion methods to obtain multimodal features of the park; The multimodal features of the park are input into a preset artificial intelligence model, which outputs park security risk warning results. At the same time, the actual inference time of the output park security risk warning results is obtained, and the timeliness of the park security risk warning is evaluated based on the actual inference time. The timeliness assessment of the park's safety risk early warning means determining whether the early warning response timeliness index is less than the preset response threshold. If so, real-time monitoring of the park's safety operation status is carried out; otherwise, an early warning timeliness insufficient prompt is sent.

7. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 3, characterized in that, The specific process for determining the priority level of data transmission in the park is as follows: Obtain the temporal stability entropy of the data transmission link in the park; The temporal stability entropy of the campus data transmission link is a link state quantification index based on information entropy theory. It is the result of weighted summation of the temporal distribution entropy of link transmission delay within a preset time window, the coefficient of variation entropy of data packet arrival interval, and the probability entropy of link bandwidth fluctuation with the corresponding temporal stability impact parameters of the campus data transmission link. The temporal distribution entropy of the link transmission delay is obtained in the following way: The transmission delay time sequence, composed of the transmission delay of each frame of campus data packets within the preset time window, is divided into several discrete intervals based on the preset delay division interval. The frequency of occurrence of transmission delay data in each discrete interval is counted, the probability distribution of each interval is calculated, and the temporal distribution entropy is quantitatively calculated based on the probability distribution of each interval. The coefficient of variation entropy of the data packet arrival interval is obtained in the following way: First, the arrival timestamps of two adjacent data packets within a preset time window are collected. The difference between the adjacent timestamps is calculated to obtain the data packet arrival interval time sequence. Based on the preset coefficient of variation level classification standard, the coefficient of variation is divided into multiple discrete levels and the occurrence probability of each level is counted. Based on the occurrence probability of each level, the coefficient of variation entropy is quantitatively calculated. The probability entropy of the link bandwidth fluctuation is obtained in the following way: Instantaneous bandwidth within a preset time window is collected to generate a bandwidth fluctuation time series. The bandwidth fluctuation time series is differentially processed to obtain bandwidth fluctuation amplitude data. Based on a preset fluctuation amplitude threshold, the fluctuation amplitude is divided into preset fluctuation intervals and the proportion probability of each interval is calculated. Based on the proportion probability of each interval, the probability entropy is quantitatively calculated. Transmission priority control strategy based on the temporal stability entropy of the data transmission link in the park.

8. The real-time monitoring system for the safe operation of a smart park based on a unified sensing platform as described in claim 7, characterized in that, The specific process of the transmission priority control strategy is as follows: If the temporal stability entropy of the data transmission link in the park is greater than the preset first-level entropy threshold, only the park data whose data transmission priority index is greater than the preset emergency transmission threshold will be transmitted through the 5G private network, and the real-time upload of the rest of the park data will be suspended. If the temporal stability entropy of the data transmission link in the park is not greater than the preset first-level entropy threshold and the temporal stability entropy of the data transmission link in the park is not less than the preset second-level entropy threshold, then the park data with a data transmission priority index greater than the preset conventional grading threshold will be transmitted according to the dynamic grading transmission strategy, and the remaining park data will be uploaded in batches based on deep compression. If the temporal stability entropy of the data transmission link in the park is less than the preset secondary entropy threshold, then a dynamic hierarchical transmission strategy will be implemented.

Images