City public security event monitoring and early warning method based on multi-agent cooperation

Abstract

The application relates to the general administrative management technical field, in particular to a city public safety event monitoring and early warning method based on multi-agent cooperation, which is used for a cloud platform. The city public safety event monitoring and early warning method based on multi-agent cooperation comprises the following steps: in the case that preset information of a target sub-region sent by an edge end is received, reading a candidate agent not receiving a monitoring task from the edge end; according to a sub-region where the candidate agent is located, determining a selection strategy of selecting a replacement agent of an offline agent in the preset information; based on the selection strategy, determining the replacement agent of the offline agent from the candidate agent; using the replacement agent to perform replacement processing on the offline agent in the target sub-region, so as to monitor and early warn the target sub-region by using online agents containing the replacement agent.

Description

Technical Field

[0001] This application relates to the field of administrative management technology, and in particular to a method for monitoring and early warning of urban public safety incidents based on multi-agent collaboration. Background Technology

[0002] The multi-agent collaborative urban public safety incident monitoring and early warning system deploys a network of intelligent agents with autonomous perception, decision-making, and collaboration capabilities to fuse and analyze multi-source data, enabling real-time identification, dynamic assessment, and accurate early warning of risks. This, in turn, improves the city's response efficiency and handling accuracy in scenarios such as natural disasters, accidents, and public health emergencies.

[0003] Multi-agent systems used for monitoring urban public safety incidents are mostly mobile intelligent agents such as drones and unmanned patrol vehicles, as well as static intelligent agents such as smart cameras and edge computing units. They mainly carry out monitoring in two ways: wide-area coverage plus close-range perception (drone full-area scanning + unmanned patrol vehicles for close-range monitoring at key locations); and high-density continuous monitoring (multi-sensor collaboration + a small number of unmanned patrol vehicles).

[0004] During monitoring, unstable network conditions can severely impact the operation of intelligent agents. For example, due to network congestion and latency fluctuations, some agents may go offline, preventing them from performing their duties such as situational scanning, close-range monitoring, and supplementary observations, thus seriously undermining the overall integrity of the monitoring layout and the completeness of data collection. Therefore, a superior urban public safety incident monitoring and early warning solution based on multi-agent collaboration is needed to overcome these difficulties. Summary of the Invention

[0005] This application provides a method for monitoring and early warning of urban public safety incidents based on multi-agent collaboration, in order to improve the overall integrity of the monitoring layout and the completeness of data collection.

[0006] Firstly, this application provides a method for monitoring and early warning of urban public safety incidents based on multi-agent collaboration, used on a cloud platform. The method includes: Upon receiving preset information about the target sub-region sent by the edge terminal, candidate intelligent agents that have not received monitoring tasks are read from the edge terminal. The preset information is used to indicate that the number of online mobile intelligent agents is lower than a preset value. Based on the sub-region where the candidate agent is located, a selection strategy is determined to select a replacement agent for the offline agent in the preset information. The offline agent is a mobile agent that is accepting a monitoring task and is in an offline state. Based on the selection strategy, a replacement agent for the offline agent is determined from the candidate agents; The alternative agent is used to replace the offline agent in the target sub-region, so that the online agent containing the alternative agent can be used to monitor and warn of the target sub-region.

[0007] Furthermore, the selection strategy for choosing a replacement agent for the offline agent from the preset information based on the sub-region where the candidate agent is located includes: If the candidate agent includes an agent in the target sub-region, the first strategy is used as the selection strategy, and the first strategy is used to indicate the determination of the alternative agent based on the credibility of the candidate agent in the target sub-region; If the candidate agents do not include agents in the target sub-region, the second strategy is used as the selection strategy. The second strategy is used to indicate the determination of the alternative agent based on the credibility of the mobile agent among the candidate agents and the distance between the candidate agent and the offline agent in the preset information.

[0008] Furthermore, when the second strategy is used as the selection strategy, determining the replacement agent for the offline agent from the candidate agents based on the selection strategy includes: Read historical data, state data, and coordinate data of the mobile agent among the candidate agents from the edge end; The credibility of the mobile agent among the candidate agents is determined based on the historical data and its weight, the state data and its weight. Based on the coordinate data of the mobile agent and the offline agent among the candidate agents, the distance between the mobile agent and the offline agent among the candidate agents is determined; Based on the credibility and the distance, a replacement agent for the offline agent is determined from the mobile agents of the candidate agents.

[0009] Furthermore, before receiving the preset information, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: Based on the administrative planning of the cities to be monitored, the cities to be monitored are divided into regions, resulting in multiple sub-regions; Obtain the risk value in the sub-region used to indicate the scope of risk impact, and obtain the connectivity value in the sub-region used to indicate the road network connectivity status of the sub-region; Based on the first weight corresponding to the risk value and the second weight corresponding to the connectivity value, the risk value and the connectivity value are fused to obtain the security feature value of the sub-region. From the sub-regions of the city to be monitored, target sub-regions whose safety feature values ​​meet the preset range are determined, where the preset range is the range of safety feature values ​​for a preset safety level.

[0010] Furthermore, based on the administrative planning of the city to be monitored, the city is divided into multiple sub-regions, including: Based on the weight values ​​corresponding to different types of security data in a preset sub-region, the security data in the preset sub-region is fused to obtain the monitoring weight of the preset sub-region. If the difference in monitoring weight between two adjacent sub-regions is less than a preset weight difference, the two sub-regions will be merged into a merged sub-region. The monitoring weight of the merged sub-region is determined based on the monitoring weights of two adjacent sub-regions.

[0011] Furthermore, obtaining the risk value in the sub-region used to indicate the scope of risk impact includes: Obtain clustering data of monitoring points in the city to be monitored, wherein the clustering data includes at least the location data of the monitoring points and the weights of the monitoring points; Based on the clustering data, clustering operations are performed on the monitoring points in the city to be monitored to obtain multiple clusters; From the clusters, hotspot clusters with public safety incident risk scores higher than preset scores are selected, and the risk value of the sub-region is determined based on the risk values ​​of the hotspot clusters in the sub-region.

[0012] Furthermore, the clustering operation uses the ST-DBSCAN algorithm. Based on the clustering data, the monitoring points in the city to be monitored are clustered to obtain multiple clusters, including: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, the core parameters of the ST-DBSCAN algorithm are adjusted. The core parameters include the spatial neighborhood radius, the temporal neighborhood radius, and the amount of neighborhood data required by the core point. Based on the clustering data, the ST-DBSCAN algorithm with adjusted core parameters is used to perform clustering operations on the monitoring points in the sub-region to obtain multiple clusters.

[0013] Furthermore, the adjustment of the core parameters of the ST-DBSCAN algorithm based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point includes: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, an adjustment coefficient is constructed to adjust the core parameter; The adjustment coefficients are used to adjust the core parameters of the ST-DBSCAN algorithm.

[0014] Furthermore, before adjusting the core parameters of the ST-DBSCAN algorithm, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: From the edge end, read the status data of the monitoring point in three consecutive target time windows, the last target time window being the current time window, and the status data is used to indicate whether a public safety event has occurred at the monitoring point in the target time window; Based on the status data, determine the status identifier of the monitoring point in the current time window.

[0015] Furthermore, determining the risk value of the sub-region based on the risk value of the hotspot clusters in the sub-region includes: The area covered by the hot spot clusters in the current time window and the area covered by the historical clusters in the historical time windows adjacent to the current time window are respectively subjected to intersection processing and union processing to obtain the first area and the second area. If the ratio of the first area to the second area is greater than or equal to a preset threshold, the degree of overlap of the regions in the current time window is obtained based on the ratio. If the overlap between the current time window and two historical time windows preceding the current time window in the region is greater than or equal to a preset threshold, the hotspot cluster is marked as a stable hotspot cluster. The risk value of the sub-region is obtained based on the risk value of the stable hotspot cluster in the sub-region.

[0016] Secondly, this application provides a multi-agent collaborative urban public safety incident monitoring and early warning device, the multi-agent collaborative urban public safety incident monitoring and early warning device comprising: The candidate module is used to read candidate intelligent agents that have not received monitoring tasks from the edge terminal when receiving preset information of the target sub-region sent by the edge terminal. The preset information is used to indicate that the number of online mobile intelligent agents is lower than a preset value. The strategy module is used to determine the selection strategy for the replacement agent of the offline agent in the preset information based on the sub-region where the candidate agent is located. The offline agent is a mobile agent that is accepting a monitoring task and is in an offline state. The substitution module is used to determine a replacement agent for the offline agent from the candidate agents based on the selection strategy. The monitoring module is used to replace the offline agent in the target sub-region using the alternative agent, so as to use the online agent containing the alternative agent to monitor and warn of the target sub-region.

[0017] Thirdly, embodiments of this application also provide an urban public safety incident monitoring and early warning system based on multi-agent collaboration. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the methods described above.

[0018] Fourthly, this application provides a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the above-described method. The computer-readable storage medium may be volatile or non-volatile.

[0019] Fifthly, this application provides an electronic device, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method when executing instructions stored in the memory.

[0020] Sixthly, this application provides a computer program product including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0021] In the embodiments described in this specification, upon receiving preset information from the edge terminal, the cloud platform reads candidate agents from the edge terminal and selects different selection strategies based on the sub-region where the candidate agents are located. Then, based on the alternative agents determined by the selection strategies, it replaces the offline agents in the target sub-region. This process provides a multi-agent collaborative urban public safety incident monitoring and early warning scheme that matches the sub-region where the candidate agents are located. It can improve the flexibility of urban public safety incident monitoring and early warning methods while using alternative agents to promptly replace offline agents, ensuring the stability and continuity of overall monitoring while achieving efficient utilization of effective monitoring resources. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0023] Figure 1 The diagram illustrates an architecture of an urban public safety incident monitoring and early warning system based on multi-agent collaboration. Figure 2 The diagram illustrates a flowchart of a method for monitoring and early warning of urban public safety incidents based on multi-agent collaboration. Figure 3 The diagram above illustrates a structural schematic of an urban public safety incident monitoring and early warning device based on multi-agent collaboration.

[0024] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0025] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0026] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0027] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0028] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0029] Figure 1 The diagram illustrates a block diagram of an urban public safety incident monitoring and early warning system based on multi-agent collaboration; for example... Figure 1As shown, the system includes a cloud platform and multiple edge terminals. Each edge terminal is responsible for managing (e.g., sending information to the intelligent agent to perform monitoring tasks) and maintaining at least one intelligent agent within a preset geographical area (which may correspond to at least one sub-area), and sending the intelligent agent's data (e.g., real-time monitoring data collected by the intelligent agent, status data such as whether the intelligent agent is offline, and historical task execution status of the intelligent agent) to the cloud platform. In this specification, the intelligent agent is mainly used for collecting images, sounds, videos, etc., from its surroundings.

[0030] The cloud platform, based on cloud computing technology, uses monitoring and real-time status data from intelligent agents at various edge terminals as its core, combined with historical monitoring and status data from these agents, to provide standardized urban public safety incident monitoring and early warning services for monitoring and early warning departments and enterprises based on multi-agent collaboration. This basic service can include selecting replacement agents for offline agents, executing monitoring of online agents, and issuing public safety incident warnings based on monitoring data.

[0031] Figure 2 The flowchart of the urban public safety incident monitoring and early warning method based on multi-agent collaboration is shown. This method can be applied to cloud platforms.

[0032] In some possible implementations, this urban public safety incident monitoring and early warning method based on multi-agent collaboration can be implemented by a processor calling computer-readable instructions stored in memory.

[0033] like Figure 2 As shown, the urban public safety incident monitoring and early warning method based on multi-agent collaboration may include: In step S21, upon receiving preset information about the target sub-region sent by the edge terminal, candidate agents that have not received monitoring tasks are read from the edge terminal.

[0034] The target sub-region is the sub-region from which candidate agents from other sub-regions can be identified as replacement agents. A sub-region is a collection of regions obtained by dividing the city under monitoring; each sub-region represents a unit that executes the subsequent selection strategy. Since sub-regions with low risk profiles (indicating the probability of a public safety incident) often have sufficient agents (especially static agents), eliminating the need for replacement agent identification, the target sub-region is typically a sub-region with a high risk profile.

[0035] Candidate agents are those that have not performed monitoring tasks and are in a candidate state. Replacement agents are those selected from the candidate agents to receive monitoring tasks. Offline agents are those that have received monitoring tasks but are offline due to network issues, while online agents are those that have received monitoring tasks and are currently online.

[0036] Intelligent agents are categorized into mobile agents and static agents based on whether their monitoring locations are fixed. Combining static agents, which provide long-term continuous regional monitoring data, with dynamic agents, which provide localized, fine-grained real-time data, yields comprehensive monitoring data reflecting the regional situation. The data collection frequencies of mobile and static agents can differ. For example, a static agent can upload time-series data such as pedestrian density and abnormal behaviors (e.g., gatherings, fights) within a sub-region every minute; a dynamic agent can upload high-definition video analysis data of a local area every 30 seconds. After data collection, data can be cleaned (removing null values ​​and outliers) and normalized.

[0037] Generally, the candidate agent is a mobile agent. Upon receiving information to replace the offline agent in performing monitoring tasks, it can move to the offline agent's monitoring location and perform the monitoring tasks. If the distance between the candidate agent and the offline agent is less than a preset value, the candidate agent can be a static agent, which can then adjust the camera angle to perform the monitoring tasks in place of the offline agent.

[0038] The preset information indicates that the number of online mobile agents is lower than a preset value. In this case, the monitoring capacity of the target sub-region is insufficient, and alternative agents need to be selected to quickly fill the monitoring gaps. This preset value can be determined based on a preset online ratio (e.g., 50%) and a preset number of mobile agents deployed in the sub-region. For example, the preset value can be obtained by multiplying the online ratio and the preset number of deployed agents.

[0039] Because different sub-regions have different security status levels and importance, preset values ​​for the number of online agents can be determined only for a subset of sub-regions (i.e., target sub-regions). These sub-regions have higher security status and importance, and mobile agents can be allocated from other sub-regions with lower security status levels. The edge device can manage at least one sub-region. If the edge device determines that the number of online mobile agents in the target sub-region it manages is lower than the preset value, it can send preset information to the cloud platform.

[0040] Once the cloud platform receives preset information, it can read relevant information about candidate agents (such as coordinates, historical task execution information, and current state information) from the edge, thereby selecting a replacement agent. To reduce computational load, candidate agents can be read only from the target sub-region and the sub-regions adjacent to the target sub-region.

[0041] In one example, the number of sub-regions adjacent to the target sub-region can be limited, such as to three. Candidate agents read from the target sub-region can include both static and mobile agents, while candidate agents read from agents outside the target sub-region can include only mobile agents.

[0042] It should be understood that, since different edge terminals manage different sub-regions, the cloud platform can use one or more edge terminals to read relevant information about candidate agents. In one example, the sub-regions adjacent to the target sub-region can be determined first based on the location information of the target sub-region, and then the relevant information about candidate agents can be read from the edge terminals managing the sub-regions.

[0043] In step S22, a selection strategy for replacing the offline agent in the preset information is determined based on the sub-region where the candidate agent is located.

[0044] In step S23, based on the selection strategy, a replacement agent for the offline agent is determined from the candidate agents.

[0045] The selection strategy involves determining alternative agents from among the candidate agents. Since there may be multiple candidate agents, a comprehensive consideration of each candidate agent is necessary to ensure faster and better agent replacement and to guarantee stable and continuous monitoring.

[0046] During the process of collecting monitoring data, intelligent agents may be affected by various factors, such as environmental noise, equipment failure, and network problems, which may lead to data distortion. Furthermore, in scenarios of multi-agent collaborative monitoring, the data quality of different agents varies. Therefore, it is necessary to first determine the reliability of the monitoring data of each agent and understand the monitoring status of each agent.

[0047] Furthermore, the positions of different candidate agents and the offline agent are also different. Therefore, different selection strategies can be determined based on position information and / or execution completion rate, etc. In one possible implementation, the step of determining the selection strategy for selecting a replacement agent for the offline agent from the preset information based on the sub-region where the candidate agent is located includes: If the candidate agent includes an agent in the target sub-region, the first strategy is used as the selection strategy, and the first strategy is used to indicate the determination of the alternative agent based on the credibility of the candidate agent in the target sub-region; If the candidate agents do not include agents in the target sub-region, the second strategy is used as the selection strategy. The second strategy is used to indicate the determination of the alternative agent based on the credibility of the mobile agent among the candidate agents and the distance between the candidate agent and the offline agent in the preset information.

[0048] Specifically, different selection strategies can be executed based on whether the candidate agent and the offline agent are in the same sub-region. The first strategy is the selection strategy executed when there is a candidate agent in the target sub-region; the second strategy is the selection strategy executed when all candidate agents are in a sub-region outside the target sub-region.

[0049] Credibility is a measure of the reliability of a candidate agent. It can be determined based on the candidate agent's historical task completion history and current state information. When a candidate agent exists within the target sub-region, determining a replacement agent based on credibility can be used as a first strategy. That is, the higher the credibility of a candidate agent within the target sub-region, the more likely it is to be used as a replacement agent. The process of determining credibility in the first strategy can be found in the implementation example of determining credibility in the subsequent second selection strategy, and will not be elaborated here.

[0050] Since the sub-region is relatively small and mobile agents within the target sub-region are generally online, alternative agents can be selected from redundant static agents (such as backup cameras) that are familiar with the situation in the target sub-region, quickly filling monitoring gaps. It should be understood that if a mobile agent exists in the target sub-region as a candidate agent for various reasons, that mobile agent can also be used as an alternative agent.

[0051] When all candidate agents are located in sub-regions outside the target sub-region, due to the large distance between the candidate agents and the disconnected agents, in addition to considering reliability, the distance between the candidate agents and the disconnected agents in the preset information also needs to be considered to achieve differentiated execution of the selection strategy. In one possible implementation, when the second strategy is used as the selection strategy, determining the replacement agent for the disconnected agent from the candidate agents based on the selection strategy includes: Read historical data, state data, and coordinate data of the mobile agent among the candidate agents from the edge end; The credibility of the mobile agent among the candidate agents is determined based on the historical data and its weight, the state data and its weight. Based on the coordinate data of the mobile agent and the offline agent among the candidate agents, the distance between the mobile agent and the offline agent among the candidate agents is determined; Based on the credibility and the distance, a replacement agent for the offline agent is determined from the mobile agents of the candidate agents.

[0052] Historical data and state data are used to determine credibility, while coordinate data is used to determine the distance between the candidate agent and the offline agent. Historical data may include the candidate agent's historical task completion rate, historical false alarm rate, etc., while state data may include the candidate agent's signal strength, etc. In one example, Formula 1 can be used to determine the credibility of the candidate agent: C i =ω1N i +ω2S i -ω3E i (Formula 1) Where, N i S represents the historical task completion rate. i E represents the signal strength. i Let C be the historical false positive rate, ω1 (which can be 0.4), ω2 (which can be 0.35), and ω3 (which can be 0.25, a negative factor) as weights, and C be the weights. i Let C be the credibility of the candidate agent. After obtaining the historical task completion rate, signal strength, and historical false alarm rate, we can first normalize them to [0,1] and then substitute them into Formula 1 to obtain C. i It can also be normalized to [0,1]. Signal strength normalization can be achieved through S... i =max(0,min(1,( S real -(-100)) / ((-50)-(-100)))) executes, S real This represents the measured signal strength (i.e., the signal strength before normalization). Specifically, when the signal strength is below -100 dBm, S... i =0; above -50dBm, S i =1; Interpolate the intermediate value of the signal strength.

[0053] As can be seen from Formula 1, the candidate agent has a high historical task completion rate, high signal strength, and low historical false alarm rate, and its C i A higher value indicates higher credibility. Credibility in the first strategy can also be determined based on Formula 1, thus allowing the selection of credibility level C. i The largest static agent is used as a substitute agent.

[0054] When borrowing a mobile agent from outside the target sub-region, it is necessary to consider not only credibility but also the distance between the mobile agent and the offline agent. This distance can be directly obtained from the coordinate data of the mobile agent and the offline agent. To simplify the calculation, the center of the region where the mobile agent is located and the center of the region where the offline agent is located can also be used to calculate this distance. In one example, Formula 2 can be used to determine the substitution priority weight of the mobile agent outside the target sub-region based on credibility and distance: Q j =C j ×(1+D ij ) (Formula 2) Among them, Q j C represents the substitution priority weight for mobile agents outside the target sub-region. j D represents the credibility of the j-th mobile agent in a sub-region outside the target sub-region. ij D represents the normalized distance between the target sub-region i (including the offline agent) and the center of the sub-region where the j-th mobile agent is located, with a value range of [0,1]. ij =D1 / D max Where D1 represents the distance between the center of the target sub-region i and the center of the sub-region where the j-th mobile agent is located, and D... max This represents the maximum distance between sub-regions in the city to be monitored.

[0055] As shown in Formula 2, the priority of mobile agents outside the target sub-region is determined by both credibility and distance. The closer the distance and the higher the credibility, the higher the priority. The substitution priority weight Q j The value of Q is larger. We obtain Q using Formula 2. j Then, you can put Q j The value range is normalized to [0, 2]. Furthermore, the substitution priority weight Q can be... j The mobile agent outside the largest target sub-region acts as a substitute agent.

[0056] Because closer proximity reduces data transmission latency and higher reliability ensures higher quality supplementary data, the second strategy ensures the effectiveness and timeliness of supplementary monitoring of the target sub-region by considering the reliability of candidate agents and the distance between candidate agents and offline agents. The first strategy, considering only the reliability of candidate agents, ensures the effectiveness of supplementary monitoring of the target sub-region. These differentiated execution strategies, while ensuring the accuracy of the alternative agent determination process, reduce the computational burden on the cloud platform.

[0057] In step S24, the alternative agent is used to replace the offline agent in the target sub-region, so that the online agent containing the alternative agent can monitor and warn of the target sub-region.

[0058] After identifying a replacement agent, a notification to perform a monitoring task (including monitoring location, monitoring angle, etc.) can be sent to the edge device where the replacement agent is located. The edge device can then forward the monitoring task to the replacement agent, enabling the replacement agent to perform monitoring operations on the target sub-area in place of the offline agent. Furthermore, when a public safety incident is detected through the agent's monitoring data, an early warning message can be issued to relevant departments. This specification does not specify a particular method for determining public safety incidents; it can be determined based on the actual situation (e.g., it can be determined through image / video detection and recognition methods).

[0059] Since the distance between the replacement agent and the offline agent may be relatively large, after determining the replacement agent, it is also necessary to determine the path of the replacement agent and add the path information to the monitoring task notification. In one example, A can be used. * The algorithm combines real-time congestion data from urban traffic platforms to plan optimal routes for alternative intelligent agents. For alternative intelligent agents such as drones, straight flight paths are prioritized, while avoiding no-fly zones.

[0060] It should be understood that if the security situation level of a sub-region changes, the tasks and paths of each agent can be adjusted in real time. For example, when a target sub-region is downgraded from a high-risk situation level to a low-risk situation level, some agents can be redeployed to other high-risk situation level areas. This can meet the monitoring needs for changes in regional risks and special periods, effectively improving the response speed to urban public safety incidents.

[0061] In the embodiments described in this specification, upon receiving preset information from the edge terminal, the cloud platform reads candidate agents from the edge terminal and selects different selection strategies based on the sub-region where the candidate agents are located. Then, based on the alternative agents determined by the selection strategies, it replaces the offline agents in the target sub-region. This process provides a multi-agent collaborative urban public safety incident monitoring and early warning scheme that matches the sub-region where the candidate agents are located. It can improve the flexibility of urban public safety incident monitoring and early warning methods while using alternative agents to promptly replace offline agents, ensuring the stability and continuity of overall monitoring while achieving efficient utilization of effective monitoring resources.

[0062] As mentioned earlier, the target sub-region is a sub-region selected from the sub-regions of the city to be monitored. Therefore, the target sub-region needs to be determined before determining the alternative agent. In one possible implementation, before receiving the preset information, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: Based on the administrative planning of the cities to be monitored, the cities to be monitored are divided into regions, resulting in multiple sub-regions; Obtain the risk value in the sub-region used to indicate the scope of risk impact, and obtain the connectivity value in the sub-region used to indicate the road network connectivity status of the sub-region; Based on the first weight corresponding to the risk value and the second weight corresponding to the connectivity value, the risk value and the connectivity value are fused to obtain the security feature value of the sub-region. From the sub-regions of the city to be monitored, target sub-regions whose safety feature values ​​meet the preset range are determined, where the preset range is the range of safety feature values ​​for a preset safety level.

[0063] The sub-regions are multiple areas obtained by dividing the city to be monitored into regions. Specifically, in the field of urban public safety monitoring, due to the vast territory and complex and diverse situations of cities, different areas have different characteristics such as population density and regional risk, and their regional risks also vary due to differences in regional functions and environments.

[0064] For example, areas such as schools and hospitals have special functional attributes and greater demands for public safety. Therefore, in order to better monitor security incidents based on multi-agent collaboration, it is necessary to first divide the city to be monitored into several sub-regions adapted for public event security monitoring, based on the regional characteristics of the city. Specifically, the city to be monitored can be divided into several sub-regions based on its administrative divisions, using streets as the unit.

[0065] After sub-region division, it is possible that some adjacent sub-regions will have similar indicators for public safety incidents. In this case, sub-regions can be merged. In one possible implementation, based on the administrative planning of the city to be monitored, the city is divided into multiple sub-regions, including: Based on the weight values ​​corresponding to different types of security data in a preset sub-region, the security data in the preset sub-region is fused to obtain the monitoring weight of the preset sub-region. If the difference in monitoring weight between two adjacent sub-regions is less than a preset weight difference, the two sub-regions will be merged into a merged sub-region. The monitoring weight of the merged sub-region is determined based on the monitoring weights of two adjacent sub-regions.

[0066] Safety data can include one or more of the following four types of indicator data: road network connectivity data, public safety incident data, real-time pedestrian density data, and regional risk level data.

[0067] Specifically, the population density value P in the population density data i This could be the ratio of real-time pedestrian traffic in sub-region i to the maximum capacity of that region; or the historical hotspot value H in public safety incident data. i This could be the ratio of the number of events in sub-region i over the past year to the highest number of events in the entire city; or the connectivity value R in road network connectivity data. i It can be the ratio of the number of intersections to the area within a sub-region; or the regional risk value L in the regional risk level data. i This is the mapping value for emergency management grading. Emergency management grading data can include four levels, and the regional risk values ​​for levels 1-4 can be mapped to 0.25, 0.5, 0.75, and 1, respectively. All four indicators can be normalized to [0,1].

[0068] Safety data can be obtained in the following ways: by acquiring road network connectivity data through the city's GIS platform; by acquiring public safety incident data from the past year through the public security historical database; by acquiring real-time pedestrian density data through pedestrian flow statistics cameras; and by acquiring regional risk values ​​from regional risk level data through regional risk grading data from emergency management departments.

[0069] Furthermore, different weights can be assigned to different indicator data. For example, α=0.3 (α is the weight of population density data), β=0.25 (β is the weight of public safety incident data), γ=0.2 (γ is the weight of real-time population density data), and δ=0.25 (δ is the weight of regional risk level data). During special periods such as holidays, α can be manually adjusted to 0.4. Based on this, the comprehensive monitoring weight W for each sub-region can be determined. In one example, the monitoring weight W can be determined according to Formula 3: W i =αP i +βH i +γR i +δL i (Formula 3) Among them, W i P is the monitoring weight of sub-region i. i It is the population density value, H i It is a historical hotspot value, R i It is the connectivity value, L i This represents the regional risk value, with α, β, γ, and δ as weights. After determining the monitoring weights for each sub-region, these weights can be normalized to ensure that W... i The value range is [0,1].

[0070] Formula 3 above enables the fusion processing of different types of security data to obtain the monitoring weights of each sub-region in the city to be monitored (used to indicate the monitoring importance of each sub-region in the city). It should be understood that the weight coefficients and thresholds mentioned in this specification are empirical preset values ​​based on specific urban environments. In practical applications, these values ​​can be calibrated and dynamically adjusted based on historical data training, expert scoring, or simulation.

[0071] Two geographically adjacent sub-regions with similar monitoring weights can be merged. In one example, a preset weight difference can be used as a criterion for determining whether the monitoring weights of two adjacent sub-regions are similar. Specifically, after determining the difference in monitoring weights between two adjacent sub-regions, it can be checked whether this difference is less than the preset weight difference. If the difference is less than the preset weight difference, the two adjacent sub-regions are merged. The above merging operation can be performed in a single iteration; that is, it stops after merging all adjacent region pairs that meet the conditions once, avoiding infinite iteration.

[0072] To prevent the attributes of high-risk areas from being diluted, the monitoring weight W of the merged sub-regions is adjusted. new The monitoring weights (W) of the two sub-regions before merging can be taken. A and W B The maximum value, i.e., W new =max(W A W B The monitoring weights can be used for subsequent clustering operations, which will not be elaborated here. After merging the sub-regions, the data in the cloud platform and edge devices need to be updated based on the latest sub-region division results.

[0073] In the above process, the merging operation of two adjacent sub-regions is performed based on the difference in monitoring weights. By reducing the number of sub-regions in the city to be monitored, the number of times the public safety incident monitoring and early warning method based on sub-regions is executed is reduced, thereby alleviating the computing pressure on the cloud platform.

[0074] A sub-region can contain multiple monitoring points that collect monitoring data. These monitoring points can be either static or dynamic. Static monitoring points use static agents (such as fixed cameras) to collect data, while dynamic monitoring points use dynamic agents (such as drones). The same monitoring point can be converted between static and dynamic monitoring points depending on the type of agent performing the monitoring task.

[0075] Risk values ​​are used to indicate the scope of risk impact. For a sub-region, the risk value of the sub-region can be determined based on the probability of risk occurring at each hotspot (monitoring point where a public safety incident is highly likely to occur). Since hotspots often cluster together, in one example, monitoring points in the city or sub-region to be monitored can be clustered, and the risk value of the sub-region can be determined based on the risk values ​​of the hotspot clusters (i.e., clusters where security incidents, anomalies, and / or risks frequently occur) within the clusters. The specific process of clustering can be referred to in subsequent embodiments, and will not be elaborated here.

[0076] The risk value of the hotspot cluster can be determined based on the average weight of each monitoring point within the cluster and the area proportion of the hotspot cluster within the sub-region. Static monitoring points can be assigned a base weight of 0.7, and dynamic data points can be assigned a base weight of 0.9. The weight of the monitoring points can be adjusted based on the frequency of public safety incidents occurring at the monitoring points and the impact of public safety incidents, etc. Specifically, the risk value of the hotspot cluster can be determined based on Formula 4: F=q0×S / S z (Formula 4) Where F is the risk value of the hotspot cluster, q0 is the average weight of each monitoring point in the hotspot cluster within the sub-region, and S z This represents the area of ​​the sub-region where the hotspot cluster is located, where S is the area covered by the hotspot cluster.

[0077] As shown in Formula 4, the greater the weight of each monitoring point in a hotspot cluster and the larger the area occupied by the hotspot cluster in the sub-region, the greater the risk impact range of the hotspot cluster in the sub-region, and the higher its risk value. After obtaining the risk value of each sub-region in the city to be monitored, the risk value can be normalized so that the risk value ranges from [0,1].

[0078] Connectivity values ​​are used to indicate the connectivity of the road network. For a sub-region, the connectivity value can be obtained by the ratio of the number of intersections to the area of ​​the sub-region. After obtaining the connectivity values ​​of each sub-region in the city to be monitored, the connectivity values ​​can be normalized so that the range of connectivity values ​​is [0,1].

[0079] Security eigenvalues ​​are used to indicate the security status level of a sub-region, and can be determined based on risk and connectivity values. In one example, the security eigenvalue of a sub-region can be calculated using Formula 5: G=mF0+nR i (Formula 5) Where G is the security characteristic value of the sub-region, F0 is the average risk value of each hotspot cluster in the sub-region, and R... i Let m (which can take the value 0.8) and n (which can take the value 0.2) be the connectivity values ​​of the sub-region, and let F0 and R be the connectivity values ​​of the sub-region, respectively. iThe weights are determined by the security feature values ​​of each sub-region. After obtaining these security feature values, normalization can be performed so that the value of G is within the range of [0,1].

[0080] As shown in Formula 5, for a sub-region, the more interconnected the road network (i.e., R... i The larger the value of F0, the faster the security incident spreads (i.e., the larger F0 is), indicating a higher security risk in that sub-region (i.e., the larger G is).

[0081] After fusing risk values ​​and connectivity values ​​to obtain security feature values, sub-regions can be classified according to these security feature values ​​to obtain multiple security status levels. Different levels correspond to different ranges of security feature values. For example, four security status levels can be set: red (G≥0.8, extremely high risk), orange (0.6≤G<0.8, high risk), yellow (0.3≤G<0.6, medium risk), and blue (G<0.3, low risk).

[0082] Because security data (such as the occurrence of public safety incidents) changes over time, the weights of monitoring points will change, thus affecting risk values, security characteristics, and security levels. In one example, the security level of each sub-area can be recalculated every 5 minutes to meet the monitoring needs of regional risk changes and special time periods, thereby effectively improving the response speed to urban public safety incidents.

[0083] Different resource (including agent allocation) standards can be adopted for sub-regions of different levels: Red Zone: Extremely high risk. Allocate candidate agents within the sub-region or mobile agents in adjacent areas to ensure sufficient monitoring coverage of the high-risk area. Orange Zone: Allocate 70% of available mobile agents within the sub-region, giving them a higher level of attention to address the higher risk. Yellow Zone: Allocate 30% of available mobile agents within the sub-region to ensure basic monitoring of the medium-risk area. Blue Zone: Low risk. Only one patrol agent needs to be retained to maintain basic monitoring.

[0084] As can be seen from the above allocation criteria, red-level areas require sufficient monitoring coverage. Therefore, red-level areas can be designated as target sub-regions. When an agent goes offline in a red-level area, agents in that area and adjacent areas need to be invoked (i.e., given priority attention). Other levels of areas, due to the abundance of redundant agents, can directly invoke agents within that area.

[0085] In one example, the security status level of each sub-region can be determined based on its security feature value. Sub-regions meeting the security level requirement (e.g., red-level regions) are then selected as target sub-regions. Sub-regions not selected as target sub-regions can provide redundant agents for the target sub-region. In another example, sub-regions providing candidate agents can also be determined based on security feature values. Sub-regions with higher security feature values ​​are more likely to provide redundant agents for the target sub-region.

[0086] In the embodiments described in this specification, the risk value and connectivity value of the sub-region are first fused to obtain a security feature value that can determine the target sub-region. Then, the target sub-region is determined based on the security feature value. This process, by setting a preset range for the security feature value, realizes the classification of the security level of the sub-region. This makes it easier to tilt the agent towards high-risk areas in the subsequent operation of determining the alternative agent for the target sub-region, avoiding resource waste and realizing differentiated monitoring of "precise coverage of key areas and efficient monitoring of low-risk areas".

[0087] As mentioned earlier, clustering operations can be performed within a sub-region to determine the risk value of the sub-region based on the risk values ​​of the obtained hotspot clusters. In one possible implementation, obtaining the risk value in the sub-region used to indicate the scope of risk impact includes: Obtain clustering data of monitoring points in the city to be monitored, wherein the clustering data includes at least the location data of the monitoring points and the weights of the monitoring points; Based on the clustering data, clustering operations are performed on the monitoring points in the city to be monitored to obtain multiple clusters; From the clusters, hotspot clusters with public safety incident risk scores higher than preset scores are selected, and the risk value of the sub-region is determined based on the risk values ​​of the hotspot clusters in the sub-region.

[0088] Clustering data refers to the data from each monitoring point required for the clustering operation. This specification does not specify the type or quantity of clustering data; it can be selected based on the actual situation. For example, location data of monitoring points and their weights (used to indicate the importance of each monitoring point in the clustering process) can be used as clustering data. In addition to these two types of data, clustering data can be supplemented or deleted as needed.

[0089] In one example, a set {P} with weights and location information can be used. t Let (x,y,w)} represent the set of clustered data of monitoring points in the sub-region, where P t The data is clustered, where t is the timestamp, x and y are the spatial coordinates, and w is the cluster weight of the monitoring points.

[0090] After obtaining the clustering data of monitoring points in the city to be monitored, clustering operations can be performed. This clustering operation can be performed on monitoring points throughout the entire city or on monitoring points within a sub-region. It should be noted that for the former type of clustering operation, there may be cases where the same cluster spans two sub-regions. In this case, when calculating the risk value of a sub-region, only the clusters located within that sub-region need to be considered.

[0091] The clustering operation can be performed using algorithms such as ST-DBSCAN. When using the ST-DBSCAN algorithm for clustering, parameters need to be preset based on the knowledge of the application domain. In one possible implementation, based on the clustering data, the monitoring points in the city to be monitored are clustered to obtain multiple clusters, including: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, the core parameters of the ST-DBSCAN algorithm are adjusted. Based on the clustering data, the ST-DBSCAN algorithm with adjusted core parameters is used to perform clustering operations on the monitoring points in the sub-region to obtain multiple clusters.

[0092] The ST-DBSCAN algorithm requires three core parameters to be preset: spatial neighborhood radius, temporal neighborhood radius, and the amount of neighborhood data required for the core point.

[0093] Status flags are used to indicate whether the data collected by the monitoring point is stable. In one example, the status flag can have three states: stable monitoring point (status flag I). s =1), downgraded monitoring point (status identifier: I) d =1) and ordinary monitoring points (status marked as I) s =0, I d =0).

[0094] The clustering operation based on the ST-DBSCAN algorithm described above can be improved using the scikit-learn extension library in Python. By taking spatiotemporal data points with weights and state labels as input, it can output an initial set of hotspot clusters.

[0095] By employing the ST-DBSCAN algorithm, which is an improvement based on the status identifier of monitoring points within the current time window and the weight of the monitoring points, hotspot cluster identification can be effectively improved, thereby enabling a comprehensive and accurate grasp of the public safety situation in sub-regions and cities under monitoring.

[0096] Based on the above three states, it can be seen that determining the state identifier requires at least two consecutive time windows of state changes. For example, a downgraded monitoring point can only be determined by comparing the state of the current time window with that of the previous time window. In one possible implementation, before adjusting the core parameters of the ST-DBSCAN algorithm, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: From the edge end, read the status data of the monitoring point in three consecutive target time windows, the last target time window being the current time window, and the status data is used to indicate whether a public safety event has occurred at the monitoring point in the target event window; Based on the status data, determine the status identifier of the monitoring point in the current time window.

[0097] The target time window is the time window required to determine the status of the monitoring point within the current time window, while the current time window is the time window corresponding to the moment the preset information is received. In one example, the current time window and the two time windows preceding it can be used as the target time window.

[0098] Based on the status data of the three target time windows, the status identifier of the monitoring point in the current time window can be determined. Specifically, a stable monitoring point is a data point that shows anomalies (such as the occurrence of a public safety incident) in three consecutive target time windows (each time window is 10 minutes). A degraded monitoring point is a data point where the anomalies disappear in the three time windows (no anomalies in the last time window, but anomalies in the first and / or second time windows). A normal monitoring point is a data point that shows no anomalies in any of the three time windows.

[0099] The aforementioned process for determining state identifiers based on state data can effectively improve the accuracy of state identifiers by using state data from multiple time windows, thereby improving the accuracy of hotspot cluster identification.

[0100] In one possible implementation, adjusting the core parameters of the ST-DBSCAN algorithm based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point includes: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, an adjustment coefficient is constructed to adjust the core parameter; The adjustment coefficients are used to adjust the core parameters of the ST-DBSCAN algorithm.

[0101] Specifically, the weights of monitoring points and the state values ​​corresponding to their state identifiers can be used as adjustment coefficients for the core parameters of the ST-DBSCAN algorithm. The adjustment principles can be: adjusting the neighborhood radius through state values ​​results in larger neighborhoods for stable data points, making them easier to cluster; smaller neighborhoods for degraded data points, making them less likely to form clusters; simultaneously, higher weights for monitoring points lead to larger neighborhoods, strengthening the clustering priority of high-confidence data and improving the accuracy of hotspot identification.

[0102] For the ST-DBSCAN algorithm's base spatial radius Eps1 and base temporal radius Eps2, the actual spatial / temporal neighborhood radius of the current data point can be obtained by directly multiplying them by the adjustment coefficients obtained from the key-value pairs. In one example, the spatial / temporal neighborhood radius in the core parameters can be determined based on Equation 6: ε p =ε0×(k+W p ) (Formula 6) Where ε0 is the default spatial / temporal neighborhood radius of the ST-DBSCAN algorithm (spatial radius 50 meters, temporal radius 5 minutes), ε p W represents the spatial / temporal neighborhood radius of the adjusted ST-DBSCAN algorithm, k is the state value corresponding to the state identifier of the monitoring point (1 for stable data points, 0.5 for degraded data points), and W is the value of the monitoring point. p k+W represents the weight of the monitoring points. p To adjust the coefficients. For ordinary data points, ε0 can be used directly.

[0103] To distinguish between stable data points and degraded data points, the spatial / temporal neighborhood radius of the former can be denoted as ε. s The spatial / temporal neighborhood radius of the latter can be denoted as ε. d .

[0104] Besides determining the adjustment coefficient based on the state value, it can also be determined directly based on the state identifier to improve computational simplicity. In one example, the spatial / temporal neighborhood radius in the core parameters can be determined based on Equation 7: ε p =ε0×(aI s +bI d +(1-I s -I d ))×(1+W p ) (Formula 7) Where ε0 is the default spatial / temporal neighborhood radius of the ST-DBSCAN algorithm (spatial radius 50 meters, temporal radius 5 minutes), ε p For the adjusted spatial / temporal neighborhood radius of the ST-DBSCAN algorithm, Is I d W is the identifier value for the status identifier. p For the weights of the monitoring points, (aI) s +bI d +(1-I s -I d ))×(1+W p ε0 is the adjustment factor. For ordinary data points, ε0 can be used directly.

[0105] Besides substituting the status identifier and monitoring point weights into the formula to obtain the adjustment coefficient, we can also construct key-value pairs corresponding to the status identifier, monitoring point weights, and adjustment coefficients. This way, after obtaining the status identifier and monitoring point weights, the adjustment coefficients can be determined based on the key-value pairs. For example, the amount of neighborhood data required for the core points in the core parameters can be determined based on Formula 8. MinPts s =MinPt s ×l (Formula 8) Among them, MinPts s To determine the amount of neighborhood data required for the core points of the adjusted ST-DBSCAN algorithm, MinPt s The amount of neighborhood data required for the default core point of the ST-DBSCAN algorithm is denoted by l, which is an adjustment coefficient determined based on the weight and / or status identifier of the monitoring point (the adjustment coefficient can be obtained from the key-value pair).

[0106] After creating multiple clusters, a risk score can be calculated for each cluster based on multiple dimensions such as the frequency, density, severity, and / or trend of public safety incidents. This risk score is then compared with a preset score, and clusters with risk scores greater than the preset score are designated as hotspot clusters. Furthermore, the average risk value of each hotspot cluster within a sub-region can be used as the risk value of that sub-region.

[0107] The above process determines the risk value of a sub-region by using hotspot clusters obtained through clustering operations. This enables the determination of the risk value of a sub-region on a hotspot cluster basis, eliminating the need to consider the risk value of each monitoring point and thus improving the speed of the risk value determination process.

[0108] The data for some hotspot clusters in a subregion may be unstable, which makes the clustering results inaccurate. For example, non-hotspot clusters (where security incidents are infrequent) may be mistakenly identified as hotspot clusters, which may waste a lot of time and resources on dealing with false hotspots caused by temporary factors.

[0109] Stable hotspot clusters (i.e., hotspot clusters in a stable state) typically cover areas and have relatively stable shapes. Generally, stable hotspot clusters can more accurately reflect real and persistent safety hotspots in urban public safety. These areas are often the focus of public safety management. Without screening, a significant amount of time and resources may be wasted dealing with false hotspots caused by temporary factors. Therefore, it is necessary to screen stable hotspot clusters and take concentrated measures to optimize resource utilization. In one possible implementation, obtaining the risk value of a sub-region based on the risk value of the hotspot clusters within the sub-region includes: The areas covered by hotspot clusters in the current time window and the areas covered by historical clusters in historical time windows adjacent to the current time window are respectively subjected to intersection and union processing to obtain the first area and the second area. If the ratio of the first area to the second area is greater than or equal to a preset threshold, the degree of overlap of the regions in the current time window is obtained based on the ratio. If the overlap between the current time window and two historical time windows preceding the current time window in the region is greater than or equal to a preset threshold, the hotspot cluster is marked as a stable hotspot cluster. The risk value of the sub-region is obtained based on the risk value of the stable hotspot cluster in the sub-region.

[0110] The historical time window refers to the time window preceding the current time window.

[0111] As mentioned earlier, the areas and shapes covered by stable hotspot clusters tend to be relatively stable. Therefore, we can first identify hotspot clusters that cover any area in the current time window and historical hotspot clusters that cover the same area in historical time windows adjacent to the current time window. Then, we can determine whether a hotspot cluster is a stable hotspot cluster based on whether the areas of the hotspot cluster and the historical hotspot cluster are similar.

[0112] Specifically, we can first obtain the areas covered by hotspot clusters and historical hotspot clusters in adjacent time windows (including the current time window and historical time windows adjacent to the current time window), where the hotspot cluster and the historical hotspot cluster cover the same area; then, we can find the intersection and union of the two areas to obtain the first area (corresponding to the intersection) and the second area (corresponding to the union); and determine the first area C. t Second area C t+1 The ratio of these two values ​​gives the overlap O of the current time window (i.e., O = S(C)). t ∩C t+1 ) / S (C t ∪C t+1If the overlap of three consecutive time windows in a region is greater than a preset threshold (e.g., 80%), it indicates that the hotspot clusters corresponding to the region are relatively stable within these 30 minutes (10 minutes per time window). This region is likely to represent a real and continuously existing safe hotspot area. Therefore, the hotspot clusters corresponding to the region can be marked as stable hotspot clusters.

[0113] Furthermore, if the overlap of three consecutive time windows is less than 30%, it indicates that the hotspot cluster changes significantly between different windows. It is very likely that the hotspot is a false hotspot caused by temporary factors (such as the camera being blocked). The hotspot cluster can be marked as a noise cluster and removed to obtain a set of stable hotspot clusters.

[0114] The above process, through the intersection / union of hot spot clusters and historical hot spot clusters, can accurately identify stable hot spot clusters and effectively remove temporary noise clusters, so as to avoid interference from temporary noise clusters to subsequent analysis and improve the accuracy of monitoring and early warning results.

[0115] The present invention also provides an urban public safety incident monitoring and early warning device based on multi-agent collaboration. Figure 3 This diagram illustrates a block diagram of a multi-agent collaborative urban public safety incident monitoring and early warning device according to an embodiment of this specification. The multi-agent collaborative urban public safety incident monitoring and early warning device can be a terminal device, a server, or other processing equipment. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, vehicle-mounted device, wearable device, etc.

[0116] In some possible implementations, this urban public safety incident monitoring and early warning device based on multi-agent collaboration can be implemented by a processor calling computer-readable instructions stored in memory.

[0117] like Figure 3 As shown, the urban public safety incident monitoring and early warning device 30 based on multi-agent collaboration may include: Candidate module 31 is used to read candidate intelligent agents that have not received monitoring tasks from the edge terminal when receiving preset information of the target sub-region sent by the edge terminal. The preset information is used to indicate that the number of online mobile intelligent agents is lower than a preset value. Strategy module 32 is used to determine a selection strategy for selecting a replacement agent for a disconnected agent from the preset information based on the sub-region where the candidate agent is located. The disconnected agent is a mobile agent that is accepting a monitoring task and is in a disconnected state. The substitution module 33 is used to determine a replacement agent for the offline agent from the candidate agents based on the selection strategy. Monitoring module 34 is used to replace the offline agent in the target sub-region using the alternative agent, so as to use the online agent containing the alternative agent to monitor and warn of the target sub-region.

[0118] The present invention also provides an urban public safety incident monitoring and early warning system based on multi-agent collaboration. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the methods described above.

[0119] The present invention also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the above-described method. The computer-readable storage medium may be volatile or non-volatile.

[0120] The present invention also provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0121] The present invention also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0122] In summary, in the embodiments of this specification, when the cloud platform receives preset information sent by the edge terminal, it reads candidate agents from the edge terminal and selects different selection strategies based on the sub-region where the candidate agents are located. Then, based on the alternative agents determined by the selection strategies, it replaces the offline agents in the target sub-region. This process provides a multi-agent collaborative urban public safety incident monitoring and early warning scheme that matches the sub-region where the candidate agents are located. It can improve the flexibility of urban public safety incident monitoring and early warning methods while using alternative agents to promptly replace offline agents, ensuring the stability and continuity of overall monitoring while achieving efficient utilization of effective monitoring resources.

[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring and early warning of urban public safety incidents based on multi-agent collaboration, characterized in that, Used for cloud platforms, including: Upon receiving preset information about the target sub-region sent by the edge terminal, candidate intelligent agents that have not received monitoring tasks are read from the edge terminal. The preset information is used to indicate that the number of online mobile intelligent agents is lower than a preset value. Based on the sub-region where the candidate agent is located, a selection strategy is determined to select a replacement agent for the offline agent in the preset information. The offline agent is a mobile agent that is accepting a monitoring task and is in an offline state. Based on the selection strategy, a replacement agent for the offline agent is determined from the candidate agents; The alternative agent is used to replace the offline agent in the target sub-region, so that the online agent containing the alternative agent can be used to monitor and warn of the target sub-region. The method for determining the selection strategy for the replacement agent of the offline agent in the preset information is as follows: when the candidate agents include the agent in the target sub-region, a first strategy is used as the selection strategy, and the first strategy is used to indicate the determination of the replacement agent based on the credibility of the candidate agents in the target sub-region; when the candidate agents do not include the agent in the target sub-region, a second strategy is used as the selection strategy, and the second strategy is used to indicate the determination of the replacement agent based on the credibility of the mobile agent among the candidate agents and the distance between the candidate agent and the offline agent in the preset information.

2. The urban public safety incident monitoring and early warning method based on multi-agent collaboration as described in claim 1, characterized in that, When the second strategy is used as the selection strategy, determining the replacement agent for the offline agent from the candidate agents based on the selection strategy includes: Read historical data, state data, and coordinate data of the mobile agent among the candidate agents from the edge end; The credibility of the mobile agent among the candidate agents is determined based on the historical data and its weight, the state data and its weight. Based on the coordinate data of the mobile agent and the offline agent among the candidate agents, the distance between the mobile agent and the offline agent among the candidate agents is determined; Based on the credibility and the distance, a replacement agent for the offline agent is determined from the mobile agents of the candidate agents.

3. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 1, characterized in that, Before receiving the preset information, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: Based on the administrative planning of the cities to be monitored, the cities to be monitored are divided into regions, resulting in multiple sub-regions; Obtain the risk value in the sub-region used to indicate the scope of risk impact, and obtain the connectivity value in the sub-region used to indicate the road network connectivity status of the sub-region; Based on the first weight corresponding to the risk value and the second weight corresponding to the connectivity value, the risk value and the connectivity value are fused to obtain the security feature value of the sub-region; From the sub-regions of the city to be monitored, target sub-regions whose safety feature values ​​meet the preset range are determined, where the preset range is the range of safety feature values ​​for a preset safety level.

4. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 3, characterized in that, Based on the administrative planning of the city to be monitored, the city is divided into multiple sub-regions, including: Based on the weight values ​​corresponding to different types of security data in a preset sub-region, the security data in the preset sub-region is fused to obtain the monitoring weight of the preset sub-region. If the difference in monitoring weight between two adjacent sub-regions is less than a preset weight difference, the two sub-regions will be merged into a merged sub-region. The monitoring weight of the merged sub-region is determined based on the monitoring weights of two adjacent sub-regions.

5. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 4, characterized in that, The step of obtaining the risk value in the sub-region used to indicate the scope of risk impact includes: Obtain clustering data of monitoring points in the city to be monitored, wherein the clustering data includes at least the location data of the monitoring points and the weights of the monitoring points; Based on the clustering data, clustering operations are performed on the monitoring points in the city to be monitored to obtain multiple clusters; From the clusters, hotspot clusters with public safety incident risk scores higher than preset scores are selected, and the risk value of the sub-region is determined based on the risk values ​​of the hotspot clusters in the sub-region.

6. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 5, characterized in that, The clustering operation uses the ST-DBSCAN algorithm. Based on the clustering data, the monitoring points in the city to be monitored are clustered to obtain multiple clusters, including: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, the core parameters of the ST-DBSCAN algorithm are adjusted. The core parameters include the spatial neighborhood radius, the temporal neighborhood radius, and the amount of neighborhood data required by the core point. Based on the clustering data, the ST-DBSCAN algorithm with adjusted core parameters is used to perform clustering operations on the monitoring points in the sub-region to obtain multiple clusters.

7. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 6, characterized in that, The adjustment of the core parameters of the ST-DBSCAN algorithm based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point includes: Based on the status identifier of the monitoring point in the current time window and the weight of the monitoring point, an adjustment coefficient is constructed to adjust the core parameter; The adjustment coefficients are used to adjust the core parameters of the ST-DBSCAN algorithm.

8. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 6, characterized in that, Before adjusting the core parameters of the ST-DBSCAN algorithm, the urban public safety incident monitoring and early warning method based on multi-agent collaboration includes: From the edge end, read the status data of the monitoring point in three consecutive target time windows, the last target time window being the current time window, and the status data is used to indicate whether a public safety event has occurred at the monitoring point in the target time window; Based on the status data, determine the status identifier of the monitoring point in the current time window.

9. The urban public safety incident monitoring and early warning method based on multi-agent collaboration according to claim 5, characterized in that, Determining the risk value of a sub-region based on the risk value of hotspot clusters in the sub-region includes: The area covered by the hot spot clusters in the current time window and the area covered by the historical clusters in the historical time windows adjacent to the current time window are respectively subjected to intersection processing and union processing to obtain the first area and the second area. If the ratio of the first area to the second area is greater than or equal to a preset threshold, the degree of overlap of the regions in the current time window is obtained based on the ratio. If the overlap between the current time window and two historical time windows preceding the current time window in the region is greater than or equal to a preset threshold, the hotspot cluster is marked as a stable hotspot cluster. The risk value of the sub-region is obtained based on the risk value of the stable hotspot cluster in the sub-region.

Images