Intelligent perception and resource scheduling command system for public area emergency

Abstract

This invention relates to the field of public safety emergency management technology, and discloses an intelligent perception and resource dispatch command system for public area emergencies. It aims to address the technical shortcomings of existing public area emergency management systems, such as single-sensory perception, fragmented data, low level of intelligent decision-making, poor command and coordination, and lack of closed-loop optimization. The system adopts a five-layer architecture combined with full lifecycle closed-loop management, comprising, from top to bottom, a perception layer, a transmission layer, a data layer, an intelligence layer, and an application layer. The perception layer constructs a multi-dimensional, three-dimensional perception network and, combined with a fusion mechanism, achieves accurate collection and filtering of various emergency-related data. This invention realizes a closed-loop handling process for emergencies, from early prediction to post-event analysis and optimization, effectively improving the intelligence and collaboration of public area emergency management. It is adaptable to various public area scenarios, convenient to deploy and maintain, and has broad applicability and practical value.

Description

Technical Field

[0001] This invention relates to the field of public safety emergency management technology, specifically to an intelligent sensing and resource dispatching command system for public area emergencies. Background Technology

[0002] As we all know, public areas are places with frequent human activity and complex scenarios, and are prone to various emergencies. Such events are characterized by their suddenness, urgency, and spread. If they are not handled in a timely manner or are not properly coordinated, they will have an adverse impact on the safety of public life and property and social public order. Therefore, the level of emergency management for public area emergencies is of paramount importance.

[0003] Currently, existing emergency management systems for public areas generally suffer from numerous shortcomings, making it difficult to meet the needs for precise, intelligent, and collaborative emergency response. In terms of perception, existing systems mostly rely on single sensing methods, which cannot comprehensively capture various types of sudden and abnormal information. They also have poor adaptability to complex environments and are prone to problems such as sensing failure and missed alarms. At the same time, the sensing data lacks effective fusion and processing, resulting in a lot of redundant information, insufficient accuracy, and a tendency to generate false alarms, thus affecting the efficiency of emergency response.

[0004] In terms of data management, the existing system's sensing devices, emergency resources, and command platforms often belong to different departments, with inconsistent data formats, making it impossible to achieve interoperability and sharing, forming information silos. This results in a lack of comprehensive and real-time data support for command and decision-making, and problems of information asymmetry and poor coordination when multiple departments coordinate their responses, making it difficult to form a joint force for emergency response.

[0005] In terms of intelligent decision-making and scheduling, existing systems rely heavily on human experience for prediction, identification, and scheduling. Even when some algorithms are introduced, they are mostly single-algorithm applications with poor adaptability. They cannot achieve early prediction, accurate identification, and optimal scheduling of emergencies, resulting in unreasonable allocation of emergency resources, untimely scheduling, and situations of resource redundancy or delayed response.

[0006] In terms of command processes, from event detection and alarm to resource allocation and on-site command, most operations are manually linked, resulting in cumbersome processes, slow responses, and a lack of real-time interaction between on-site rescue and rear command, which easily leads to command disconnect. Furthermore, after an event is handled, there is a lack of a robust debriefing and optimization mechanism, making it impossible to improve the system's subsequent emergency response capabilities through historical data, hindering the continuous upgrading of emergency management capabilities.

[0007] In response to the aforementioned shortcomings of existing technologies, there is an urgent need for an intelligent perception and resource dispatch command system for public area emergencies that can address various pain points, improve the intelligence, collaboration, and precision of emergency management in public areas, and provide reliable protection for public safety. Summary of the Invention

[0008] Technical problems to be solved To address the shortcomings of existing emergency management systems for public areas, such as incomplete perception, lack of data coordination, low intelligence, and poor command efficiency, this invention provides an intelligent perception and resource scheduling command system for public area emergencies.

[0009] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: an intelligent perception and resource scheduling command system for public area emergencies, comprising a perception layer, a transmission layer, a data layer, an intelligence layer, and an application layer, with each layer working collaboratively to achieve closed-loop management of the entire lifecycle, including prediction, perception, alarm, scheduling, command, review, and optimization; the perception layer constructs a multi-dimensional three-dimensional perception network, the intelligence layer embeds a self-developed core algorithm model with deep fusion of multiple algorithms, the data layer builds a unified emergency data platform, the transmission layer adopts a highly reliable hybrid transmission mode, and the application layer provides full-scenario collaborative command functions to achieve fully automated processing of the entire process of proactive prediction, accurate perception, intelligent scheduling, and collaborative command of emergencies.

[0010] Furthermore, the perception layer is a four-dimensional perception network encompassing vision, environment, human body, and location, and it is equipped with a perception fusion mechanism. This perception fusion mechanism employs a triple fusion algorithm architecture of federated learning, attention mechanism, and Bayesian inference to achieve accurate fusion of multi-dimensional perception data. Moreover, the perception fusion mechanism is linked in real time with the prediction algorithm and recognition algorithm of the intelligent layer to form a dynamic closed loop of perception, fusion, and recognition.

[0011] Furthermore, the four-dimensional perception network includes a visual perception unit and a human body state perception unit. The visual perception unit uses an intelligent camera with edge computing capabilities. The intelligent camera embeds a self-developed behavior recognition algorithm to identify ongoing emergencies and predict potential risks. The intelligent camera also supports nighttime infrared imaging and adaptive recognition in severe weather. The human body state perception unit uses a non-contact vital sign sensor. The non-contact vital sign sensor uses millimeter-wave technology to detect vital signs in real time, automatically identifying abnormal signs of sudden illnesses and locating the patient's position. The human body state perception unit can also link with public mobile terminals to proactively report emergencies.

[0012] Furthermore, the four-dimensional sensing network includes an environmental sensing unit and a location sensing unit. The environmental sensing unit is equipped with various types of miniature sensors, which integrate functions for detecting gas, toxic and harmful gases, temperature and humidity, smoke, vibration, and noise. They are flexibly deployed using wireless self-organizing network technology and support adaptive threshold adjustment. When environmental parameters exceed the safe range, they can automatically trigger an early warning and link with the visual sensing unit for image verification. The location sensing unit integrates GPS / BeiDou positioning, UWB ultra-wideband positioning, and WiFi positioning technologies to achieve accurate positioning of personnel, emergency resources, and potential hazards in public areas.

[0013] Furthermore, the transmission layer adopts a hybrid transmission mode of 5G, fiber optic, and satellite backup. Among them, 5G SA standalone networking combined with edge computing nodes realizes the main data transmission, fiber optic transmission serves as a backup for 5G transmission and is deployed in core public areas of the city; portable satellite transmission terminals serve as emergency backups for data transmission in remote public areas or under extreme emergencies; the transmission layer uses the AES-256 encryption algorithm to encrypt the transmitted data end-to-end and sets data transmission priorities, with command instructions, vital sign data, and emergency resource location data set as the highest transmission priority.

[0014] Furthermore, the data layer includes a data access module, a data processing module, a data storage module, and a data sharing module. The data processing module adopts the Hadoop and Spark big data processing framework to perform cleaning, labeling, fusion, and de-identification processing on the accessed data. The data storage module adopts a three-level storage mode of distributed storage, local backup, and cloud backup. The data sharing module has a standardized data sharing interface to achieve bidirectional data sharing with systems of departments such as public security, fire protection, medical care, transportation, and emergency management.

[0015] Furthermore, the core algorithm model of the intelligent layer includes a risk prediction algorithm; the risk prediction algorithm adopts a four-algorithm fusion architecture of LSTM, Transformer, CNN and reinforcement learning, which is used to predict the early risks of sudden events in public areas, and outputs the risk level, prediction probability and risk evolution timeline.

[0016] Furthermore, the core algorithm model of the intelligent layer includes an emergency event identification algorithm and an event level intelligent determination algorithm. The emergency event identification algorithm adopts a fusion architecture of YOLOv8 improved version, Transformer, ViT, and attention mechanism, which is used for the identification of various emergencies in public areas and the location of the event occurrence. The event level intelligent determination algorithm adopts a fusion architecture of "fuzzy comprehensive evaluation method, BP neural network, and random forest", which is used to automatically determine the level of emergency events.

[0017] Furthermore, the core algorithm model of the intelligent layer includes a dynamic path optimization and intelligent scheduling algorithm, and a retrospective analysis and model optimization algorithm. The dynamic path optimization and intelligent scheduling algorithm adopts a fusion architecture of Dijkstra's algorithm, reinforcement learning, ant colony algorithm, and LSTM, which is used for the optimization and dynamic adjustment of emergency resource scheduling paths. The retrospective analysis and model optimization algorithm adopts a fusion architecture of machine learning, Transformer, and genetic algorithm, which is used to realize the autonomous iterative upgrade of the core algorithm model.

[0018] Furthermore, the application layer includes a command center terminal, a rescue personnel terminal, a public terminal, and an operation and maintenance personnel terminal; the command center terminal has the functions of full-domain situation visualization, one-click alarm handling, and multi-party collaborative command; the application layer links the intelligent layer algorithm model and the perception layer data to realize command visualization, intelligent rescue, convenient public participation, and efficient operation and maintenance; the full life cycle closed-loop management includes seven stages: prediction, perception, alarm, scheduling, command, review, and optimization, and the automatic optimization of system performance is achieved through review analysis.

[0019] (III) Beneficial Effects Compared with existing technologies, the present invention provides an intelligent sensing and resource scheduling command system for public area emergencies, which has the following beneficial effects: This public area emergency intelligent perception and resource scheduling command system constructs a multi-dimensional, three-dimensional perception network. Combined with a multi-fusion algorithm architecture, it achieves comprehensive collection and accurate fusion of emergency-related data, effectively filters redundant and invalid information, reduces false alarms and missed alarms, and can overcome the perception limitations in extreme environments to achieve all-weather, all-round accurate perception. This provides reliable data support for subsequent intelligent decision-making and solves the problems of single perception and insufficient accuracy in existing systems.

[0020] By establishing a unified emergency data platform, information barriers between different departments are broken down, enabling centralized management, integrated processing, secure storage, and shared reuse of various emergency-related data. This ensures that command and decision-making receive comprehensive and accurate data support, improves the scientific nature of command and decision-making and the efficiency of multi-departmental collaborative response, and solves the problems of data fragmentation and information silos in existing systems.

[0021] By adopting a multi-algorithm deep fusion architecture and embedding multiple core algorithm models, it realizes the full-process automated processing of emergencies from risk prediction, event identification, level determination to resource scheduling, review and optimization. It replaces the traditional manual operation mode, improves the accuracy of decision-making and response efficiency, and enables early prediction and precise handling of emergencies. It solves the problems of low intelligence level and reliance on manual decision-making in existing systems.

[0022] By adopting a multi-mode hybrid transmission mode, combined with encryption processing and data transmission priority settings, it achieves data transmission with no dead zones, high reliability, and low latency across the entire domain. It can effectively prevent data disconnection problems in extreme scenarios, ensure that command instructions and key data can be transmitted in a timely manner, and guarantee that command work is not interrupted. It solves the problems of insufficient transmission reliability and easy delay of core data in the existing system.

[0023] The construction of a multi-terminal collaborative command platform enabled real-time coordination between rear command, on-site rescue, and relevant departments, breaking down command gaps and improving the orderliness and efficiency of emergency response. Simultaneously, it established channels for public participation, forming a multi-party collaborative and public-participatory emergency response model, further enhancing the timeliness and comprehensiveness of emergency response.

[0024] By constructing a closed-loop management process covering the entire lifecycle and using a self-optimizing algorithm mechanism through retrospective analysis, the system can continuously improve its performance, gradually adapt to ever-changing emergency response needs, and maintain a high level of emergency response over the long term. This solves the problems of existing systems lacking retrospective optimization and being unable to continuously upgrade their emergency response capabilities.

[0025] Adopting a modular design, it can flexibly adapt to various types of emergency response needs in public areas, and supports flexible expansion and upgrades of sensing devices, algorithm models, and application functions without requiring overall system reconstruction, effectively reducing system deployment and maintenance costs and having wide applicability. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall workflow of the system of the present invention; Figure 2 This is a schematic diagram of the data processing flow of the perception layer in this invention; Figure 3 This is a schematic diagram of the core algorithm workflow of the intelligent layer of this invention; Figure 4 This is a schematic diagram of the emergency resource scheduling process of the present invention; Figure 5 This is a schematic diagram of the closed-loop management process for the entire lifecycle of this invention.

[0027] Note: The accompanying drawings are all drawn in a simplified form and are only used to illustrate the basic structure of the present invention. Therefore, they only show the components closely related to the present invention. The shape, size and proportion of each component in the drawings are exemplary and do not constitute a limitation on the scope of protection of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of 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.

[0029] Please see Figures 1 to 5 This invention relates to an intelligent perception and resource scheduling command system for public area emergencies, comprising a perception layer, a transmission layer, a data layer, an intelligence layer, and an application layer. These layers work in synergy to achieve closed-loop management throughout the entire lifecycle, including prediction, perception, alarm, scheduling, command, review, and optimization. The perception layer constructs a multi-dimensional, three-dimensional perception network; the intelligence layer embeds a self-developed core algorithm model with deep fusion of multiple algorithms; the data layer builds a unified emergency data platform; the transmission layer adopts a highly reliable hybrid transmission mode; and the application layer provides full-scenario collaborative command functions, enabling fully automated processing of the entire process of proactive prediction, accurate perception, intelligent scheduling, and collaborative command of emergencies. The system employs a five-layer architecture and a closed-loop design, clearly defining the hierarchical relationship between the five layers (perception layer, transmission layer, data layer, intelligence layer, and application layer). Each layer collaborates through data interaction and command transmission, ensuring the overall continuity of system operation. Specifically, the perception layer collects various emergency-related data, the transmission layer ensures stable data transmission, the data layer centrally manages and shares data, the intelligence layer handles data processing and intelligent decision-making, and the application layer provides operational and command functions. Simultaneously, a closed-loop management system covering the entire lifecycle—prediction, perception, alarm, dispatch, command, review, and optimization—is constructed, linking the functions of each system layer to form a complete process from early risk prediction to event handling and review. Through the seamless coordination of each link, the system achieves fully automated handling of emergencies. Through the collaborative design of the overall architecture, the problem of fragmented modules and disconnected processes in traditional emergency systems is broken down, enabling a shift in emergency response from passive reaction to proactive prediction and precise handling. The five-layer architecture has a clear division of labor, ensuring efficient connection between data collection, transmission, processing, decision-making, and application, thereby improving system operating efficiency. The closed-loop management of the entire lifecycle enables full control over the event handling process, facilitating the timely detection of shortcomings in the handling process, providing support for system optimization, and comprehensively improving the intelligence and collaboration level of emergency management in public areas.

[0030] In this solution, the perception layer is a four-dimensional perception network encompassing vision, environment, human body, and location, and it incorporates a perception fusion mechanism. This mechanism employs a triple-fusion algorithm architecture of federated learning, attention mechanism, and Bayesian inference to achieve accurate fusion of multi-dimensional perception data. Furthermore, the perception fusion mechanism works in real-time with the prediction and recognition algorithms of the intelligent layer, forming a dynamic closed loop of perception, fusion, and recognition. Through the complementarity of the four perception dimensions, it comprehensively covers various perception needs for public area emergencies. The perception fusion mechanism, using a triple-fusion algorithm architecture of federated learning, attention mechanism, and Bayesian inference, enables local data processing and global collaboration among perception units, preventing the leakage of original data while integrating full-domain perception features. The attention mechanism focuses on high-value perception data, filtering redundant and invalid information. Bayesian inference performs probability verification of multi-source data, improving the accuracy of data fusion. This fusion mechanism works in real-time with the prediction and recognition algorithms of the intelligent layer, feeding back the fused accurate data to the intelligent layer, forming a dynamic closed loop of perception, fusion, and recognition, ensuring efficient connection between perception data and intelligent decision-making. The four-dimensional perception network enables comprehensive collection of data related to emergencies, avoiding the limitations of a single perception dimension. It can fully capture sudden anomaly information related to vision, environment, human body, and location. The triple fusion algorithm architecture effectively improves the fusion accuracy of multi-source perception data, reduces invalid data interference, and ensures the accuracy of perception information. The dynamic closed loop of perception, fusion, and recognition achieves seamless connection between perception data and intelligent decision-making, providing accurate data support for the prediction and recognition of subsequent emergencies, reducing data transmission and processing redundancy, and improving the system's sensitivity and accuracy in perceiving emergencies.

[0031] In this solution, the four-dimensional perception network includes a visual perception unit and a human state perception unit. The visual perception unit employs an intelligent camera with edge computing capabilities. This intelligent camera embeds a self-developed behavior recognition algorithm to identify ongoing emergencies and predict potential risks. The intelligent camera also supports nighttime infrared imaging and adaptive recognition in severe weather. The human state perception unit uses a non-contact vital sign sensor. This sensor uses millimeter-wave technology to detect vital signs in real time, automatically identifying abnormal signs of sudden illnesses and locating the patient. Furthermore, the human state perception unit can link with public mobile terminals to proactively report emergencies. The visual perception unit uses an intelligent camera with edge computing capabilities. Edge computing enables preliminary data processing locally, reducing the processing load on the core platform. The embedded self-developed behavior recognition algorithm can analyze the images captured by the camera in real time, identifying ongoing emergencies and predicting potential risks by analyzing human behavior characteristics. Simultaneously, the camera features nighttime infrared imaging and adaptive recognition in severe weather, overcoming environmental limitations and ensuring all-weather perception. The human body status sensing unit employs a non-contact vital sign sensor, utilizing millimeter-wave technology to achieve real-time, non-contact detection of vital signs. It captures key vital signs such as heart rate and respiratory rate without physical contact, enabling the identification of abnormal signs associated with sudden illnesses. Combined with the system's positioning function, it achieves precise patient location. This unit can also be linked to public mobile terminals, providing a proactive reporting channel to supplement passive sensing. The edge computing capabilities of the visual sensing unit enhance data processing efficiency. Its self-developed behavior recognition algorithm enables both proactive prediction and passive identification of emergencies. Its all-weather sensing capability ensures the system's sensing function remains effective even in extreme environments, comprehensively capturing information related to abnormal human behavior in public areas. The non-contact design of the human body status sensing unit improves ease of use, enabling rapid identification of sudden illnesses and precise location to provide clear direction for rescue efforts, shortening response time. The public mobile terminal linkage function broadens the reporting channels for emergencies, forming a dual mode of passive sensing and proactive reporting, further reducing missed reports and improving the system's timeliness in handling human-related emergencies.

[0032] In this solution, the four-dimensional sensing network includes an environmental sensing unit and a location sensing unit. The environmental sensing unit deploys multiple types of miniature sensors, which integrate detection functions for gas, toxic and harmful gases, temperature and humidity, smoke, vibration, and noise. It employs wireless self-organizing network technology for flexible deployment and supports adaptive threshold adjustment. When environmental parameters exceed safe ranges, it automatically triggers an early warning and links with a visual sensing unit for image verification. The location sensing unit integrates GPS / BeiDou positioning, UWB ultra-wideband positioning, and WiFi positioning technologies to achieve accurate positioning of personnel, emergency resources, and potential hazards within public areas. The environmental sensing unit deploys multiple types of miniature sensors, integrating various environmental parameter detection functions, and can comprehensively capture abnormal environmental information such as gas, toxic and harmful gases, temperature, and humidity within public areas. The wireless self-organizing network technology enables flexible sensor deployment without complex wiring, adapting to the deployment needs of different types of public areas. The sensors support adaptive threshold adjustment, flexibly adjusting the safety thresholds of various environmental parameters according to the safety requirements of different scenarios. When a parameter exceeds the threshold, an early warning is automatically triggered, and a visual sensing unit is linked to capture the abnormal area, achieving dual verification of environmental anomalies and avoiding false alarms. The location sensing unit integrates multiple positioning technologies, combining the advantages of different technologies to achieve precise positioning of personnel, emergency resources, and potential hazards within public areas, adapting to the positioning needs of different indoor and outdoor scenarios. The environmental sensing unit's multi-type sensors enable comprehensive collection of environmental anomaly information; wireless self-organizing network technology enhances deployment flexibility; threshold adaptive adjustment function improves the adaptability of environmental early warnings; and the dual verification mechanism effectively reduces false alarms related to environmental anomalies, enabling timely detection of various environmentally related emergencies. The integration of multiple positioning technologies in the location sensing unit achieves precise positioning across the entire indoor and outdoor area, providing accurate location support for personnel evacuation and emergency resource dispatch during emergency response, ensuring that rescue personnel and emergency supplies can quickly reach designated locations, thus improving response efficiency.

[0033] In this solution, the transmission layer adopts a hybrid transmission mode of 5G, fiber optic, and satellite backup. 5G SA (Standalone) networking combined with edge computing nodes enables primary data transmission, while fiber optic transmission serves as a backup for 5G transmission, deployed in core urban public areas. Portable satellite transmission terminals act as an emergency backup for data transmission in remote public areas or during extreme emergencies. The transmission layer employs AES-256 encryption to encrypt transmitted data end-to-end and sets data transmission priorities, with command instructions, vital sign data, and emergency resource location data given the highest priority. The 5G SA networking combined with edge computing nodes as the primary transmission method leverages the high bandwidth and low latency of 5G, combined with the local processing capabilities of edge computing, to ensure efficient transmission of real-time data. Fiber optic transmission, deployed in core urban public areas, serves as a backup for 5G transmission, utilizing the stability of fiber optic transmission to ensure continuous data transmission when 5G signals are interrupted. Portable satellite transmission terminals act as an emergency backup, adaptable to scenarios where both 5G and fiber optic transmission are interrupted due to remote areas or extreme emergencies, enabling data transmission in areas with no network coverage. Meanwhile, the transport layer employs the AES-256 encryption algorithm for end-to-end encryption of transmitted data, ensuring security during data transmission and preventing data leakage. Data transmission priorities are set, with core data such as command instructions and vital sign data given the highest priority, ensuring priority transmission of core data and avoiding delays in core instructions. The hybrid transmission mode combines the advantages of three transmission methods, achieving comprehensive data transmission coverage across public areas without blind spots. This ensures the continuity and stability of data transmission in different scenarios and environments, preventing data disconnection in extreme situations and guaranteeing the real-time transmission of command instructions and sensor data. The AES-256 encryption algorithm effectively enhances data transmission security, preventing the leakage or tampering of emergency data during transmission. The data transmission priority settings ensure the priority transmission of core data and command instructions, avoiding delays in handling core information and improving the timeliness and reliability of emergency response.

[0034] In this solution, the data layer includes a data access module, a data processing module, a data storage module, and a data sharing module. The data processing module uses the Hadoop and Spark big data processing framework to perform cleaning, labeling, fusion, and de-identification processing on the accessed data. The data storage module adopts a three-level storage mode of distributed storage, local backup, and cloud backup. The data sharing module has a standardized data sharing interface to enable bidirectional data sharing with systems in departments such as public security, fire protection, medical care, transportation, and emergency management. The data access module supports access to multi-source data, integrating various relevant data from the perception layer, emergency resources, public services, and public reports. Standardized interfaces ensure rapid access for data of different formats and sources. The data processing module utilizes the Hadoop and Spark big data processing frameworks, leveraging their efficient data processing capabilities to clean, label, merge, and de-identify the accessed data. Invalid and redundant data are removed, key information is labeled, multi-dimensional data is merged, and sensitive data is de-identified to ensure data security and availability. The data storage module employs a three-tiered storage model, combining the fast read speed of distributed storage, the security of local backup, and the large capacity of cloud backup to achieve accurate storage and secure backup of different types of data. The data sharing module establishes standardized data sharing interfaces, enabling bidirectional data sharing with relevant departmental systems and breaking down information barriers between departments. The multi-source data access module enables centralized integration of various emergency-related data, avoiding information silos caused by data fragmentation; the big data processing framework improves the efficiency and quality of data processing, ensuring data accuracy and availability, while de-identification ensures the security of sensitive data; the three-level storage mode achieves secure data storage and efficient data retrieval, balancing the efficiency of real-time data retrieval with the long-term backup needs of historical data; the standardized data sharing interface breaks down information barriers between departments, enabling two-way sharing of emergency data, providing comprehensive and unified data support for multi-departmental collaborative handling of emergencies, and improving the scientific and collaborative nature of command and decision-making.

[0035] In this solution, the core algorithm model of the intelligent layer includes a risk prediction algorithm. This algorithm employs a fusion architecture of LSTM, Transformer, CNN, and reinforcement learning to predict emerging risks of public area emergencies in advance, and outputs risk level, prediction probability, and risk evolution timeline. The four algorithms work collaboratively, each with its own function: the CNN algorithm extracts spatial features from real-time sensing data, capturing the spatial distribution patterns; the LSTM algorithm mines the temporal dimension features of the data, analyzing trends; the Transformer algorithm captures the correlation features between multi-dimensional data, uncovering the inherent connections between different types of data; and the reinforcement learning algorithm, combined with historical emergency data, autonomously learns the risk evolution patterns under different scenarios, dynamically optimizing algorithm parameters to adapt to the differences in different public areas. Through the deep fusion of these four algorithms, the system achieves early prediction of emerging risks of public area emergencies, outputting risk level, prediction probability, and risk evolution timeline, providing a basis for early prevention and control decisions. The four-algorithm fusion architecture combines the advantages of each algorithm, enabling comprehensive mining of spatial, temporal, and related features of perceived data, thus improving the comprehensiveness and accuracy of risk prediction. The enhanced learning algorithm's autonomous learning capability allows it to adapt to different scenarios in various public areas, achieving parameter optimization without human intervention. The risk prediction function can detect early signs of emergencies, providing a clear basis for taking preventive measures in advance, effectively reducing the probability of emergencies, buying sufficient time for subsequent handling, and improving the safety and prevention level of public areas.

[0036] In this solution, the core algorithm model of the intelligent layer includes an emergency event recognition algorithm and an event level intelligent determination algorithm. The emergency event recognition algorithm adopts a fusion architecture of YOLOv8 (improved version), Transformer, ViT, and attention mechanism for the recognition of various emergencies in public areas and the location of the events. The event level intelligent determination algorithm adopts a fusion architecture of fuzzy comprehensive evaluation method, BP neural network, and random forest for the automatic determination of the emergency event level. The emergency event recognition algorithm adopts a multi-algorithm fusion architecture. The YOLOv8 (improved version) algorithm realizes rapid location of the event area and preliminary identification of the event type, improving the recognition response speed. The ViT algorithm deeply mines the fine-grained features of visual data, improves the recognition accuracy in complex scenes, and solves the problem of poor recognition performance in extreme environments. The Transformer algorithm realizes the correlation and fusion of visual, environmental, and human cross-modal data, improving the comprehensiveness of recognition. The attention mechanism focuses on the core features of the event and filters background interference, further improving the recognition accuracy. Through this fusion architecture, accurate recognition and location of various emergencies are achieved. The intelligent event level determination algorithm employs a triple-algorithm fusion architecture. Fuzzy comprehensive evaluation handles fuzzy indicators that cannot be precisely quantified, providing preliminary judgment results. A backpropagation neural network mines the nonlinear correlations between multiple indicators, optimizing judgment accuracy. A random forest algorithm performs multiple rounds of verification on the judgment results, improving reliability and enabling automatic determination of the event level. This multi-algorithm fusion architecture for event identification enables accurate and rapid identification of various emergencies in complex scenarios. It effectively filters background interference, improves recognition performance in extreme environments, and provides precise location positioning for clear direction in response. The fusion architecture of the intelligent event level determination algorithm solves the problem of fuzzy indicators being difficult to quantify, improving the accuracy and reliability of level determination. It automates level determination, quickly providing event levels without manual intervention, providing a basis for subsequent matching of corresponding response procedures and resource scheduling strategies, and improving the targeting and efficiency of response measures.

[0037] In this scheme, the core algorithm model of the intelligent layer includes a dynamic path optimization and intelligent scheduling algorithm, and a retrospective analysis and model optimization algorithm. The dynamic path optimization and intelligent scheduling algorithm adopts a fusion architecture of Dijkstra's algorithm, reinforcement learning, ant colony algorithm, and LSTM, used for the optimization and dynamic adjustment of emergency resource scheduling paths. The retrospective analysis and model optimization algorithm adopts a fusion architecture of machine learning, Transformer, and genetic algorithm, used to achieve autonomous iterative upgrades of the core algorithm model. The dynamic path optimization and intelligent scheduling algorithm adopts a four-algorithm fusion architecture: Dijkstra's algorithm quickly selects multiple potential optimal paths; the ant colony algorithm optimizes the global optimality of paths, avoiding long-term congestion and dangerous areas; the reinforcement learning algorithm combines real-time traffic conditions, weather, and other factors to dynamically adjust path priorities and autonomously learn scheduling experience; the LSTM algorithm predicts changes in traffic conditions over a period of time in the future, avoiding temporary congestion in advance. This architecture realizes the optimization and dynamic adjustment of emergency resource scheduling paths. The debriefing analysis and model optimization algorithms employ a triple-algorithm fusion architecture. Machine learning algorithms deeply mine data from the entire emergency response process, identifying shortcomings and algorithmic weaknesses. The Transformer algorithm captures the correlation between debriefing data and algorithm parameters, pinpointing the core reasons for insufficient accuracy. The genetic algorithm autonomously optimizes the parameters of each core algorithm, enabling autonomous iterative upgrades of the algorithm model without human intervention, thus improving algorithm performance. The fusion architecture of dynamic path optimization and intelligent scheduling algorithms achieves globally optimal and dynamically adjusted emergency resource scheduling paths, effectively avoiding congestion and dangerous areas, ensuring rapid arrival of emergency resources on-site, improving the efficiency and rationality of resource scheduling, supporting multi-resource collaborative scheduling, and avoiding resource conflicts and redundancy. The fusion architecture of debriefing analysis and model optimization algorithms enables autonomous iterative upgrades of the algorithm model, continuously optimizing algorithm performance based on historical response data, constantly improving the system's prediction, identification, and scheduling accuracy, forming a virtuous cycle of response, debriefing, and optimization, and continuously enhancing the system's emergency response capabilities.

[0038] In this solution, the application layer includes a command center terminal, a rescue personnel terminal, a public terminal, and an operation and maintenance personnel terminal. The command center terminal has functions such as full-domain situation visualization, one-click alarm handling, and multi-party collaborative command. The application layer links the intelligent layer algorithm model and the perception layer data to achieve command visualization, intelligent rescue, convenient public participation, and efficient operation and maintenance. The full lifecycle closed-loop management includes seven stages: prediction, perception, alarm, dispatch, command, review, and optimization. Automatic optimization of system performance is achieved through review analysis. The command center terminal, with its full-domain situation visualization, one-click alarm handling, and multi-party collaborative command functions, enables comprehensive control of the emergency situation, rapid issuance of instructions, and collaborative linkage among multiple departments. The rescue personnel terminal, the public terminal, and the operation and maintenance personnel terminal provide dedicated functions for their respective groups, facilitating on-site rescue operations, public participation, and system operation and maintenance. The application layer links the intelligent layer algorithm model and the perception layer data, synchronizing intelligent decision-making results and perception data to each terminal, achieving seamless integration of terminal functions with the system's core modules, and realizing goals such as command visualization and intelligent rescue. The full lifecycle closed-loop management system clearly comprises seven stages, each sequentially linked. The post-event analysis results are used for system performance optimization, achieving continuous improvement in system functionality. The multi-terminal design at the application layer adapts to the needs of different user groups, enabling convenient operation at each stage of emergency response. The collaborative command function at the command center improves the efficiency of multi-departmental coordination, the functional design at the rescue personnel end enhances the convenience and accuracy of on-site response, the public end broadens public participation channels, and the maintenance personnel end ensures stable system operation. The linkage between the application layer and the intelligence and perception layers enables seamless data and command transmission, improving the convenience and efficiency of system operation. The full lifecycle closed-loop management system achieves full controllability in emergency response, continuously improving system performance through post-event optimization, ensuring the system can adapt to ever-changing emergency needs and maintain a high level of emergency response in the long term.

[0039] The perceptual fusion algorithm is used to fuse four-dimensional perceptual data from the perceptual layer: vision, environment, human body, and location. It combines federated learning, attention mechanism, and Bayesian inference to output accurate fused perceptual data. The formula is as follows: The core logic of this formula is as follows: First, an attention mechanism is used to assign weights and filter redundancy from multi-source sensory data. Then, federated learning is used to achieve collaborative fusion of data from multiple sensory units. Finally, Bayesian inference is used to perform probability verification on the fusion result. Combined with the weight coefficients, the final accurately fused data is output, which completely solves the problems of one-sided single-sensory data and redundancy of multi-source data, providing reliable data support for subsequent intelligent decision-making. The detailed explanation of each parameter and operation logic is as follows: The final output of accurate fused perception data is the core data output from the perception layer to the transmission layer and the data layer. It covers the effective features of four types of perception information: vision, environment, human body and location. It is also the basic input data for the intelligent layer algorithm operation and directly determines the accuracy of subsequent event recognition and risk prediction.

[0040] The weighting coefficient of the federated learning fusion result is used to adjust the proportion of the federated learning output in the final fused data. The value is adaptively adjusted by the system according to the perception scenario. Its core function is to highlight the core value of collaborative fusion of multiple perception units and adapt to the differences in perception environment in different public areas.

[0041] The weighting coefficients of the Bayesian inference verification results, and Complementary (the sum of the two is 1) is used to adjust the proportion of probability verification results. Its core function is to correct the deviations that may occur in the federated learning fusion process, improve the reliability of fused data, and reduce the risk of false alarms and false negatives.

[0042] Federated learning fusion operation functions are the core operational logic of the triple fusion architecture, enabling the collaborative fusion of data from multiple sensing units (visual, environmental, human, and position sensing units). Its core principle is to allow each sensing unit to complete preliminary data processing locally (avoiding raw data leakage), uploading only data features to the fusion node. Through global collaborative computation, the effective features of each sensing unit are integrated, outputting a preliminary fusion result. This approach protects data privacy while integrating features from across the entire sensing domain.

[0043] The attention mechanism weighted operation expression is a preprocessing step in federated learning fusion, used to filter redundant data and focus on high-value perceived data.

[0044] The types and quantities of perceived data correspond to the four-dimensional perception network of the perception layer in this invention, namely: These correspond to four categories: visual perception data, environmental perception data, human body state perception data, and location perception data.

[0045] : No. Attention weights for different types of sensory data are used to assign importance to them, and are adaptively adjusted by the system based on the type of event (e.g., human state sensory data in a sudden illness scenario). Higher values ​​are needed; in fire scenarios, environmental perception data... (Higher values ​​are taken), its core function is to filter out invalid and redundant data and focus on high-value data related to emergencies.

[0046] : No. The raw sensory data collected by the sensory unit, among which Corresponding visual perception data (image feature data collected by smart cameras). Corresponding environmental sensing data (environmental parameter data collected by miniature sensors), Corresponding human body state perception data (vital sign data collected by non-contact sensors). Corresponding location sensing data (location coordinate data collected by the positioning terminal).

[0047] The Bayesian inference verification function is used to perform probabilistic verification on the initial fusion results of federated learning. Its core function is to correct fusion bias and improve the accuracy of fused data. Its principle is based on historical sensing data to calculate the reliability probability of the initial fusion results, remove fusion features with low reliability, and output the verified accurate results, reducing fusion bias caused by single sensing unit failures or environmental interference.

[0048] The posterior probability of Bayesian inference represents the probability at the 1st... Class-aware data Given the probability that the preliminary fusion result matches the actual emergency scenario, it is used to judge the credibility of the preliminary fusion result. The higher the probability value, the more reliable the fusion result is; otherwise, the corresponding features need to be removed and the fusion needs to be repeated.

[0049] The risk prediction algorithm combines fused sensing data with historical data to predict potential risks of emergencies in public areas and outputs core risk-related information, as shown in the following formula: By extracting spatial features from fused sensing data using CNN, temporal features using LSTM, and multi-feature correlations using Transformer, the three types of features are fused and input into a reinforcement learning model for risk decision-making. Combined with scene adaptation coefficients, the final risk prediction result is output, enabling early prediction and proactive prevention of emergencies. This addresses the pain point of existing systems being unable to predict in advance and only able to respond passively. A detailed explanation of each parameter and computational logic is as follows: The final risk prediction result is the core basis for the intelligent layer to push early warning information to the application layer. It includes three core contents: risk level, risk type, and risk evolution trend. It directly supports the system's proactive prevention and control function (for example, when the risk of personnel gathering is predicted, early warning information is pushed and prevention and control measures are initiated).

[0050] The reinforcement learning operation function is the core decision-making logic of the risk prediction algorithm. It is used to make risk decisions based on the fused multi-dimensional features and output the risk prediction results. Its core principle is to combine historical emergency data to autonomously learn the risk evolution patterns in different scenarios, dynamically optimize decision parameters, and adapt to the differences in scenarios in different public areas (such as the different risk evolution patterns in train stations, campuses, and shopping malls). It can achieve accurate risk decisions without human intervention.

[0051] CNN (Convolutional Neural Network) feature extraction operations are used to extract spatial features from fused sensor data. It integrates spatial feature data from the sensing data, corresponding to the spatial distribution patterns of the sensing data (e.g., the spatial distribution of people gathering, the location distribution of environmental hazard points, etc.). The core function of CNN is to mine abnormal information in spatial features (such as the abnormal spatial features of densely packed people), providing spatial dimension support for risk prediction.

[0052] LSTM (Long Short-Term Memory) feature extraction operation is used to extract temporal features from fused perceptual data. It integrates temporal feature data from the sensing data, corresponding to the temporal change trend of the sensing data (e.g., the changing trend of population density, the gradual change pattern of environmental parameters, etc.). The core function of LSTM is to discover abnormal changes in the time dimension (such as a sharp increase in population density in a short period of time), solving the problem that traditional algorithms cannot capture dynamic change risks.

[0053] Transformer feature extraction operations are used to extract the correlation between spatial and temporal features. It is a fusion of spatial features extracted by CNN and temporal features extracted by LSTM. The core role of Transformer is to explore the intrinsic relationship between the two types of features (e.g., the relationship between a sharp increase in population density and abnormal environmental parameters, and the relationship between population movement trajectory and risk spread), improve the comprehensiveness and accuracy of risk prediction, and avoid the one-sidedness of prediction based on a single feature.

[0054] Multi-feature fusion expression linearly fuses spatial features, temporal features, and multi-feature correlations to form a multi-dimensional and comprehensive risk feature vector, which serves as input data for reinforcement learning models, ensuring the comprehensiveness of reinforcement learning decisions and avoiding prediction bias caused by single features.

[0055] The scenario adaptation coefficient is used to adjust the adaptability of the risk prediction results to specific public area scenarios. Its value is adaptively adjusted by the system based on the current application scenario (such as a large train station, a small shopping mall, or a campus). Different scenarios have different risk assessment standards (for example, in a campus scenario, a small gathering of students does not constitute a risk, while in a train station scenario, a gathering of the same number of people may constitute a risk). Its core function is to make risk prediction results fit specific scenarios, thereby improving the accuracy and applicability of the predictions.

[0056] The two algorithms mentioned above are the core algorithms of this invention, and they work together: the perceptual fusion algorithm outputs accurate fused data. It is in the risk prediction algorithm (Spatial feature data) and The core input source of (time feature data) is the accuracy of perceptual fusion, which directly determines the effectiveness of risk prediction; the output of the risk prediction algorithm... This will, in turn, optimize the weight coefficients of the perception fusion algorithm. This forms a virtuous cycle of perception fusion, risk prediction, and parameter optimization, which together support the core advantages of the invention, namely "proactive prediction and accurate perception," and solve the core pain points of existing systems.

[0057] Example 1: A large train station scene.

[0058] This embodiment is applied to a large railway station with an average daily passenger flow of 50,000 people, covering areas such as waiting hall, platform, underground passage, catering area, and ticketing area, with an area of ​​approximately 50,000 square meters. The area is densely populated and has high mobility. Potential emergencies mainly include stampedes, sudden illness of passengers, gas leaks in the catering area, fires, and fights on the platform. It is necessary to achieve early prediction, accurate perception, rapid dispatch, and multi-departmental collaborative handling of emergencies.

[0059] Perception Layer: Deploy 80 smart cameras with edge computing capabilities (40 in waiting halls and platforms, and 40 at entrances and exits and underground passages), supporting nighttime infrared imaging and adaptive recognition in severe weather; 25 multi-type miniature sensors (in catering areas and near underground pipelines), integrating gas, smoke, temperature, humidity, and vibration detection; 12 non-contact vital sign sensors (in waiting halls and first aid stations); equip 20 security guards, 5 emergency vehicles, and 3 emergency supply warehouses with positioning terminals, integrating UWB and GPS positioning technologies; and implement federated learning, attention mechanisms, and Bayesian inference perception fusion mechanisms.

[0060] Transport layer: Deploy 3 5G edge nodes, using 5G SA standalone networking as the main transmission method; deploy fiber optic transmission lines as primary backup; equip 2 portable satellite transmission terminals as emergency backup; use AES-256 encryption algorithm, and set emergency instructions and vital signs data as the highest transmission priority.

[0061] Data Layer: An emergency data platform is built to access sensing data, emergency resource data from railway stations, railway police data, data from three surrounding hospitals, real-time traffic conditions, and historical emergency data; data is processed using Hadoop and Spark frameworks, stored in a three-level storage model, and standardized interfaces are built to enable data sharing among multiple departments.

[0062] Intelligent Layer: Embeds a complete set of core algorithm models, optimizes algorithm parameters to adapt to densely populated scenarios, and focuses on improving the efficiency of predicting crowd gatherings, identifying sudden diseases, and scheduling emergency resources.

[0063] Application Layer: A large visual screen is built at the command center, linking railway police and surrounding hospitals; rescue personnel are equipped with mobile phones and smart helmets; a public mini-program is launched to support passenger reporting; and an operation and maintenance management platform is deployed at the operation and maintenance personnel's end, with dedicated operation and maintenance personnel assigned.

[0064] The perception layer collects multi-dimensional data, and the perception fusion mechanism outputs accurate data. The transmission layer encrypts and transmits the data to the data layer, which processes the data and outputs standardized data. The intelligence layer uses a risk prediction algorithm to predict risks such as crowd gathering and issues warnings. If a passenger suddenly falls ill, the human body state perception unit identifies abnormal signs and locates the location. The emergency event identification algorithm confirms the event type, the level determination algorithm determines the level, and the dispatching algorithm optimizes the rescue route. The application layer command center dispatches emergency personnel and vehicles with one click. The rescue personnel receive instructions and navigate to the location. The location perception unit provides real-time feedback on the rescue progress, and the intelligence layer dynamically adjusts the route. After the rescue is completed, the review algorithm organizes the data and optimizes the system parameters before entering the next closed loop.

[0065] Example 2: Small shopping mall scenario.

[0066] This embodiment is applied to a small shopping mall with an area of ​​approximately 8,000 square meters, covering a shopping area, a dining area, and a parking lot. The average daily customer flow is 3,000 people, and the personnel mobility is moderate. Potential emergencies mainly include smoke leakage in the dining area, sudden fainting of customers, minor fights in the shopping area, and conflicts caused by vehicle scratches in the parking lot. The scenario is relatively small in scale, and emergency resources mainly consist of small first aid equipment and security personnel. The system needs to be deployed in a lightweight manner, with accurate perception and rapid response.

[0067] Perception Layer: Deploy 15 smart cameras (8 in the shopping area, 4 in the dining area, and 3 in the parking lot) with basic abnormal behavior recognition capabilities; 8 miniature sensors (5 in the dining area and 3 in the power distribution room) integrating smoke, temperature, humidity, and gas detection; 5 non-contact vital sign sensors (shopping area and dining area); equip 8 security guards and 1 emergency supplies warehouse with positioning terminals, using a hybrid WiFi and GPS positioning system; enable a perception fusion mechanism to simplify redundant parameters.

[0068] Transport layer: 5G public network is used as the main transmission network, with simple fiber optic lines deployed as backup; AES-256 encryption algorithm is used to prioritize the transmission of security dispatch instructions and vital signs data.

[0069] Data Layer: A lightweight emergency data platform is built to access sensor data, shopping mall emergency resource data, and data from a nearby community hospital; a simplified big data processing framework and a two-level storage mode (distributed storage and local backup) are adopted to achieve data sharing with the community hospital and the local police station.

[0070] Intelligent Layer: Embeds core algorithm models, simplifies scheduling algorithms, focuses on scheduling small emergency resources (first aid kits, security guards), and optimizes algorithm parameters for abnormal behavior recognition and sudden disease recognition.

[0071] Application Layer: The command center is equipped with a simple visual terminal (shopping mall security room); rescue personnel (security guards) are equipped with mobile phones; the public mini-program focuses on reporting emergencies and receiving early warnings; the maintenance personnel use simple maintenance tools, which are maintained by shopping mall security personnel on a part-time basis.

[0072] The perception layer collects data from shopping areas, dining areas, and parking lots. The perception fusion mechanism filters redundant data and outputs accurate anomaly information. The transmission layer transmits the data to the data layer for processing. The intelligence layer identifies the type of emergency (such as smoke leakage in the dining area), determines the event level, and the dispatch algorithm assigns nearby security guards to handle the situation. The application layer's security room terminal receives event details, and security guards receive dispatch instructions via their mobile phones. Once the security guards arrive at the scene to handle the situation, they report the handling progress in real time. After the handling is completed, the review algorithm simplifies the review process, optimizes the identification algorithm parameters, and ensures that the system is adaptable to small shopping mall scenarios.

[0073] Example 3: Campus Scene.

[0074] This embodiment is applied to a primary and secondary school with an area of ​​approximately 30,000 square meters, covering teaching buildings, playgrounds, canteens, dormitories, and school gates, with a total of about 2,000 teachers and students, mainly students. Potential emergencies mainly include students' sudden illnesses, school bullying, students leaving school without permission, gas leaks in the canteen, and falls and injuries on the playground. The key points to achieve are student safety protection, rapid response to emergencies, and coordination with the education department and surrounding hospitals.

[0075] Perception Layer: Deploy 30 smart cameras (12 in the teaching building, 6 on the playground, 4 in the canteen, 4 in the dormitory, and 4 at the school gate) to enhance the identification of campus bullying and abnormal student departures; 12 miniature sensors (6 in the canteen, 3 in the power distribution room, and 3 in the dormitory) to integrate gas, smoke, temperature, and humidity detection; 8 non-contact vital sign sensors (teaching building, canteen, and playground); equip 10 security personnel, 2 emergency supply warehouses, and the school gate access control system with positioning terminals, using UWB indoor positioning and GPS outdoor positioning; and activate a perception fusion mechanism to focus on abnormal student data.

[0076] Transport layer: Deploy one 5G edge node, with 5G SA standalone networking as the main transmission method; deploy fiber optic transmission lines to connect the school security office and the education bureau; equip one portable satellite transmission terminal; adopt AES-256 encryption algorithm to prioritize the transmission of student vital signs and abnormal departure warning data.

[0077] Data Layer: Build a dedicated emergency data platform for the campus, integrating sensor data, campus emergency resource data, student and teacher information (anonymized), data from surrounding hospitals, and data from the education bureau; use Hadoop and Spark frameworks to process data, employ a three-level storage model, and achieve data linkage with the education bureau, surrounding hospitals, and parents.

[0078] The intelligent layer embeds core algorithm models to optimize algorithms for identifying school bullying, abnormal student departures, and sudden illnesses. The scheduling algorithm focuses on the scheduling of campus security personnel and school doctors, and is linked to the campus broadcasting system.

[0079] Application Layer: The command center (school security room) is equipped with a large visual screen, linking the education bureau and school clinic; rescue personnel (security and school doctors) are equipped with mobile phones; the public (parents' mini-program) receives student safety alerts; the maintenance personnel are part-time school logistics staff, linking the education bureau's maintenance platform; a new campus broadcast linkage function has been added, which can quickly issue evacuation notices.

[0080] The perception layer collects data from various areas of the campus in real time, focusing on monitoring student behavior and physical condition. The perception fusion mechanism filters abnormal student data (such as bullying or sudden illness). The data is transmitted to the data layer through the transmission layer for processing and desensitization. The intelligence layer predicts potential risks (such as student fights), identifies emergencies, determines their severity, and dispatches school doctors and security personnel. The application layer's security room terminal receives instructions, and school doctors and security personnel navigate to the scene. The campus broadcast system simultaneously issues warnings or evacuation notices. After the incident is resolved, the algorithm is reviewed, data is organized, and the parameters of the student anomaly identification algorithm are optimized. The results are also synchronized to the education bureau, forming a closed-loop management system.

[0081] Comparison table of effects of 3 examples: Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A public area emergency response and resource dispatching command system, characterized in that, It comprises a perception layer, a transmission layer, a data layer, an intelligence layer, and an application layer, with each layer working together to achieve closed-loop management of the entire lifecycle, including prediction, perception, alarm, scheduling, command, review, and optimization. The perception layer constructs a multi-dimensional, three-dimensional perception network; the intelligence layer embeds a self-developed core algorithm model that deeply integrates multiple algorithms; the data layer builds a unified emergency data platform; the transmission layer adopts a highly reliable hybrid transmission mode; and the application layer provides full-scenario collaborative command functions to achieve fully automated processing of the entire process of proactive prediction, accurate perception, intelligent scheduling, and collaborative command of emergencies.

2. The intelligent sensing and resource dispatching command system for public area emergencies according to claim 1, characterized in that, The perception layer is a four-dimensional perception network encompassing vision, environment, human body, and location, and it is equipped with a perception fusion mechanism. This perception fusion mechanism employs a triple fusion algorithm architecture of federated learning, attention mechanism, and Bayesian inference to achieve accurate fusion of multi-dimensional perception data. Furthermore, the perception fusion mechanism is linked in real time with the prediction and recognition algorithms of the intelligent layer, forming a dynamic closed loop of perception, fusion, and recognition.

3. The intelligent sensing and resource dispatching command system for public area emergencies according to claim 1, characterized in that, The four-dimensional perception network includes a visual perception unit and a human state perception unit; the visual perception unit adopts an intelligent camera with edge computing capabilities, the intelligent camera embeds a self-developed behavior recognition algorithm, which is used to identify the sudden events that have occurred and to predict the emerging risks, and the intelligent camera supports night infrared imaging and adaptive recognition in severe weather. The human body status sensing unit uses a non-contact vital sign sensor, which uses millimeter wave technology to detect a person's vital signs in real time. It is used to automatically identify abnormal signs of sudden illness and locate the patient's position. The human body status sensing unit can also be linked with public mobile terminals to enable proactive reporting of emergencies.

4. The intelligent sensing and resource dispatching command system for public area emergencies according to claim 3, characterized in that, The four-dimensional sensing network includes an environmental sensing unit and a location sensing unit. The environmental sensing unit is equipped with various types of miniature sensors, which integrate functions for detecting gas, toxic and harmful gases, temperature and humidity, smoke, vibration, and noise. It is flexibly deployed using wireless self-organizing network technology and supports adaptive threshold adjustment. When environmental parameters exceed the safe range, it can automatically trigger an early warning and link with the visual sensing unit for image verification. The location sensing unit integrates GPS / BeiDou positioning, UWB ultra-wideband positioning, and WiFi positioning technologies to achieve accurate positioning of personnel, emergency resources, and potential hazards in public areas.

5. The intelligent sensing and resource dispatching command system for public area emergencies according to claim 4, characterized in that, The transmission layer adopts a hybrid transmission mode of 5G, fiber optic, and satellite backup. Among them, 5G SA standalone networking combined with edge computing nodes realizes the main data transmission, fiber optic transmission serves as a backup for 5G transmission and is deployed in core public areas of the city; portable satellite transmission terminals serve as emergency backups for data transmission in remote public areas or under extreme emergencies; the transmission layer uses the AES-256 encryption algorithm to encrypt the transmitted data end-to-end and sets data transmission priorities, with command instructions, vital sign data, and emergency resource location data set as the highest transmission priority.

6. The intelligent sensing and resource dispatching command system for public area emergencies according to claim 5, characterized in that, The data layer includes a data access module, a data processing module, a data storage module, and a data sharing module. The data processing module uses the Hadoop and Spark big data processing framework to perform cleaning, labeling, fusion, and de-identification processing on the accessed data. The data storage module adopts a three-level storage mode of distributed storage, local backup, and cloud backup. The data sharing module has a standardized data sharing interface to enable bidirectional data sharing with systems in departments such as public security, fire protection, medical care, transportation, and emergency management.

7. A public area emergency response and resource dispatching command system according to claim 6, characterized in that, The core algorithm model of the intelligent layer includes a risk prediction algorithm. The risk prediction algorithm adopts a four-algorithm fusion architecture of LSTM, Transformer, CNN and reinforcement learning, which is used to predict the early risks of sudden events in public areas and output the risk level, prediction probability and risk evolution timeline.

8. A public area emergency response and resource dispatching command system according to claim 7, characterized in that, The core algorithm model of the intelligent layer includes an emergency event identification algorithm and an event level intelligent determination algorithm. The emergency event identification algorithm adopts a fusion architecture of YOLOv8 improved version, Transformer, ViT, and attention mechanism, which is used to identify various emergencies in public areas and locate the location of the event. The event level intelligent determination algorithm adopts a fusion architecture of "fuzzy comprehensive evaluation method, BP neural network, and random forest", which is used to automatically determine the level of emergency events.

9. A public area emergency response and resource dispatching command system according to claim 7, characterized in that, The core algorithm model of the intelligent layer includes a dynamic path optimization and intelligent scheduling algorithm, and a retrospective analysis and model optimization algorithm. The dynamic path optimization and intelligent scheduling algorithm adopts a fusion architecture of Dijkstra's algorithm, reinforcement learning, ant colony algorithm, and LSTM to optimize and dynamically adjust emergency resource scheduling paths. The retrospective analysis and model optimization algorithm adopts a fusion architecture of machine learning, Transformer, and genetic algorithm to achieve autonomous iterative upgrades of the core algorithm model.

10. A public area emergency response and resource dispatching command system according to claim 1, characterized in that, The application layer includes a command center terminal, a rescue personnel terminal, a public terminal, and an operation and maintenance personnel terminal. The command center terminal has functions such as full-domain situation visualization, one-click alarm handling, and multi-party collaborative command. The application layer links the intelligent layer algorithm model and the perception layer data to realize command visualization, intelligent rescue, convenient public participation, and efficient operation and maintenance. The full life cycle closed-loop management includes seven stages: prediction, perception, alarm, scheduling, command, review, and optimization. Automatic optimization of system performance is achieved through review analysis.

Images