A method, device and equipment for managing and controlling urban governance events and a storage medium

Abstract

This application provides a method, apparatus, equipment, and storage medium for managing urban governance events, relating to the field of urban management technology. The method includes: acquiring original event reports from multiple heterogeneous channels, including text work orders from government service hotlines, structured alarm data reported by IoT devices, video stream summaries captured by urban surveillance cameras, and information reported by citizens via mobile applications containing geographic location, images, and text descriptions; performing spatiotemporal and semantic multidimensional association aggregation on the original event reports to generate a structured unified event object; analyzing the unified event object and outputting an event analysis report to obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, generating a resource scheduling scheme through multi-objective optimization and executing the resource scheduling scheme. This method aims to improve the efficiency and accuracy of managing urban governance events.

Description

Technical Field

[0001] This invention relates to the field of urban management technology, and in particular to a method, apparatus, equipment and storage medium for managing urban governance events. Background Technology

[0002] Currently, under the macro-background of building smart cities and smart communities, urban governance is committed to establishing an event collaborative handling system with process management at its core, in order to achieve efficient and accurate daily management and emergency response.

[0003] In existing technologies, common governance systems primarily rely on static, pre-defined rules and fixed process engines to generate event handling paths, and typically require manual judgment for resource scheduling and task assignment. However, this operating model based on a fixed rule base often exhibits limitations in intelligent judgment and system adaptability when facing constantly evolving and diverse public demands, emerging social risks not covered by existing rules, and unforeseen, hidden governance issues. As a result, the overall handling mechanism lacks sufficient flexibility and intelligence, affecting not only the efficiency of event response but also making it prone to errors due to human scheduling biases, thus failing to support the agile governance and precise service goals pursued by smart cities. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a method, apparatus, equipment, and storage medium for managing urban governance events, which can greatly improve the efficiency and accuracy of managing urban governance events.

[0005] Firstly, it provides a method for managing urban governance incidents, including: Obtain original event reports from multiple heterogeneous channels; perform spatiotemporal and semantic multidimensional association aggregation on the original event reports to generate a structured unified event object. The original event reports from multiple heterogeneous channels include text work orders from government service hotlines, structured alarm data reported by IoT devices, video stream summaries captured by urban surveillance cameras, and information reported by citizens through mobile applications that includes geographical location, images, and text descriptions. Based on a pre-built urban governance knowledge graph and risk transmission model, the unified event object is analyzed and an event analysis report is output; wherein, the event analysis report is a multi-dimensional analysis result of the unified event object; Obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, generate a resource scheduling scheme through multi-objective optimization, and execute the resource scheduling scheme.

[0006] Secondly, a control device for urban governance events is provided, comprising: The aggregation module is used to obtain raw event reports from multiple heterogeneous channels; perform spatiotemporal and semantic multidimensional association aggregation on the raw event reports to generate a structured unified event object; The analysis module is used to analyze the unified event object based on a pre-built urban governance knowledge graph and risk transmission model, and output an event analysis report; wherein the event analysis report is a multi-dimensional analysis result of the unified event object; The scheduling module is used to obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, it generates a resource scheduling scheme through multi-objective optimization and executes the resource scheduling scheme.

[0007] Thirdly, an electronic device is provided, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus; the memory is used to store computer programs; and the processor, when executing the program stored in the memory, implements any of the steps described in the first aspect above.

[0008] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect above.

[0009] The urban governance event management method, device, equipment, and storage medium provided in this application embodiment are based on a pre-built urban governance knowledge graph and risk transmission model. They comprehensively analyze a unified event object, output an event analysis report, and use the event analysis report, the real-time resource status (location, load, professional capabilities) of the personnel handling the event, etc., as constraints. Through a multi-objective optimization algorithm, the optimal resource scheduling scheme is automatically generated, replacing the inefficient manual scheduling mode that relies on personal experience, and greatly improving the efficiency and accuracy of urban governance event management. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a method for managing urban governance events provided by the present invention; Figure 2 A flowchart illustrating another method for managing urban governance events provided by this invention; Figure 3 A schematic diagram of the process for a control device for urban governance events provided by the present invention; Figure 4 An electronic device provided by the present invention. Detailed Implementation

[0011] 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.

[0012] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0013] Example 1: Reference Figure 1 , Figure 1 This is a flowchart illustrating a method for managing urban governance events, provided as an embodiment of this application. Figure 1 As shown, the method may include: Step S101: Obtain the original event reports from multiple heterogeneous channels; perform multi-dimensional spatiotemporal and semantic association aggregation on the original event reports to generate a structured unified event object.

[0014] For example, the executing entity in this embodiment can be an electronic device, a terminal device, a control device or equipment for urban governance events, or other devices or equipment capable of executing this embodiment, and there are no limitations on this. In this embodiment, the executing entity is described as an electronic device.

[0015] In this step, a pre-defined standardized data access interface is used to non-intrusively connect to and acquire raw event reports from various heterogeneous channels of urban governance in real time. These raw event reports come from sources including, but are not limited to: text work orders from government service hotlines; structured alarm data reported by IoT devices (such as surveillance videos, AR gimbals, grid worker handheld terminals, and elderly care wristbands); video stream summaries captured by urban surveillance cameras; and information reported by citizens through mobile applications, including geographic location, images, and text descriptions. To mitigate the heterogeneity of data sources and formats, a pre-defined spatiotemporal correlation and semantic matching algorithm is used to perform multi-dimensional spatiotemporal and semantic correlation aggregation on the raw event reports, generating a structured unified event object.

[0016] Optionally, this embodiment could connect the originally isolated interface data to a data processing center within the electronic device, where the data processing center would perform event identification and unified allocation and scheduling. Alternatively, the electronic device could deploy dedicated data acquisition channels to collect different types of data and perform event identification and unified allocation and scheduling. No limitation is imposed on this approach.

[0017] Step S102: Based on the pre-built urban governance knowledge graph and risk transmission model, analyze the unified event object and output an event analysis report; wherein, the event analysis report is a multi-dimensional analysis result of the unified event object.

[0018] For example, upon receiving a unified event object, intelligent assessment begins based on a pre-built urban governance knowledge graph and an AI model. The AI ​​model can be a risk transmission model, and this is not limited. For instance, based on a pre-built, city-wide urban governance knowledge graph G and a risk transmission model, intelligent assessment is performed on the unified event object, outputting an event assessment report. The event assessment report is a multi-dimensional analysis of the unified event object, including any one or more of the following: event type, risk level, and potential impact area.

[0019] Step S103: Obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, generate a resource scheduling scheme through multi-objective optimization and execute the resource scheduling scheme.

[0020] For example, the system accesses the city's operational resource database in real time to obtain the set of resource units that can be scheduled at the current moment. This allows us to obtain the actual resource status. Each resource unit (i=1, 2…n) have specific attributes, including the current position. Status (idle, en route, working), capability set (Such as having hoisting capabilities, mastering welding techniques, etc.) and operating costs per unit time. Then, based on the event assessment report and the actual resource status, a resource scheduling plan is generated through multi-objective optimization and executed.

[0021] The method provided in this application automatically generates the optimal resource scheduling scheme through a multi-objective optimization algorithm, replacing the inefficient manual scheduling mode that relies on personal experience, and greatly improving the efficiency and accuracy of urban governance event management.

[0022] Example 2: Figure 2 A flowchart illustrating another method for managing urban governance incidents provided in this application, as shown below. Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the method is described in detail below, and the method includes: Step S201: Obtain the original event reports from multiple heterogeneous channels; perform multi-dimensional spatiotemporal and semantic association aggregation on the original event reports to generate a structured unified event object.

[0023] In one example, S201 includes: parsing and mapping each original event report to a preset standard internal event report structure; calculating the comprehensive similarity score between any two original event reports; wherein the comprehensive similarity score is a weighted sum of temporal similarity, spatial similarity and semantic similarity; when the comprehensive similarity score exceeds a preset threshold, determining that the two original event reports belong to the same objective event, and performing multidimensional association clustering to generate a structured unified event object.

[0024] In one example, S201 includes: using a pre-trained deep language model to map the text description of each original event report into a semantic vector; calculating the cosine similarity between any two semantic vectors; and determining a comprehensive similarity score based on the cosine similarity.

[0025] For example, to obtain the original event reports from multiple heterogeneous channels, refer to step S101, which will not be repeated here.

[0026] Optionally, in practical applications, multimodal data from across the entire domain can be aggregated in real time through standardized interfaces. When the system senses the operational status within its jurisdiction, the parameters in the aforementioned original event report are specifically mapped as follows: Data source channel identifier (src): covers grid sensing points distributed in different communities, including AR three-dimensional security panoramic PTZ camera video streams from high points, entry and exit records generated by facial recognition access control panels, and real-time alarms reported by IoT sensors.

[0027] The text description field d is automatically generated or associated based on the data type. For example, "smoke detected" or "fire alarm" reported by IoT devices, and semantic descriptions such as "road repair" or "road flooding" submitted by residents through "one-click request".

[0028] The multimedia information field m corresponds to storing panoramic images captured by the AR gimbal, comparison photos from the feature retrieval database, or video summaries transmitted back by law enforcement recorders.

[0029] Geographic coordinates By utilizing the street base map coordinate system, all scattered sensing points are precisely mapped onto the street geographic base, thereby ensuring that fragmented information from different dimensions can achieve a high degree of consistency in the spatiotemporal dimension, providing a complete and multidimensional event view for subsequent intelligent analysis.

[0030] In the first approach to resolving heterogeneity, to mitigate the heterogeneity of data sources and formats, all data from the original event reports undergoes a parsing and normalization process, being uniformly mapped to a standardized internal event report structure to generate structured, unified event objects. This standardized internal event report structure can be represented as follows: id is the unique identifier of the original event report, t is the timestamp of the event in the original event report, l is the geographical coordinates (latitude and longitude) of the event, d is the text description or alarm content of the event, and m is additional multimedia information (such as pictures, video frames, etc.).

[0031] In the second parallel approach to mitigating heterogeneity, a data fusion and understanding architecture deeply optimized for emergency command scenarios is provided. This architecture first utilizes modality-optimized deep learning models (e.g., using convolutional neural networks optimized for small object detection to process video, and using a "transformer-based bidirectional encoder representation technique" model incorporating emergency domain knowledge to process text) for hierarchical feature extraction. Subsequently, a cross-modal attention fusion network semantically aligns and weights feature vectors from different data sources, automatically learning the relative importance of different data sources within a specific event context to generate a unified and comprehensive semantic representation of the same event—a unified event object.

[0032] Then, a spatiotemporal and semantic multidimensional association aggregation algorithm is employed to actively identify and merge multiple different original event reports pointing to the same objective physical event. This algorithm is used to calculate the value of any two original event reports r. i With r j The overall similarity score between When the overall similarity score exceeds the preset aggregation threshold When the time similarity score is reached, it can be determined that the two events belong to the same event. This comprehensive similarity score is composed of time similarity. Spatial similarity Composed of semantic similarity weights, the mathematical expression is: , These represent the weight coefficients for time, space, and semantic dimensions, respectively. All three are positive and their sum is 1. The specific values ​​of the weighting coefficients can be adaptively adjusted according to the characteristics of different urban governance scenarios (such as traffic congestion and pipeline leakage) to highlight the key correlation dimensions in different scenarios.

[0033] The similarity of each item is calculated as follows: Time similarity This is used to measure the temporal proximity of two original event reports. Considering that the correlation between original event reports decays exponentially with increasing time difference, this embodiment uses an exponential decay function for modeling. Let t... i For the original event report r i timestamp, t j For the original event report r j The timestamp, then the time similarity Defined as: , It is the absolute value of the difference between the timestamps of the two original event reports; It is a preset positive number called the time decay coefficient, which is used to control the decay rate of similarity as the time difference increases; exp is the natural exponential function.

[0034] Spatial similarity This measure is used to assess the geographical proximity of two original event reports. Similar to time, the closer the geographical proximity, the stronger the correlation. Let l i For the original event report r i geographic coordinates, l j For the original event report r j The geographical coordinates, then the spatial similarity Defined as: , function This tool is used to calculate the spherical distance between two latitude and longitude coordinates, and the output unit is meters. This is a preset spatial decay coefficient used to quantify the rate at which similarity decreases with increasing geographical distance. To facilitate the integration of administrative management logic with underlying spatial calculations, a built-in mapping engine is included, capable of automatically converting administrative division nodes (such as labels like "community" and "level-3 grid") into geospatial latitude and longitude coordinates. Its analytical capabilities.

[0035] semantic similarity : Utilize pre-trained deep language models (such as open-source models based on the Transformer architecture) to extract text descriptions from any two original event reports. and These correspond to high-dimensional semantic vectors. and The directions of these two semantic vectors in the vector space represent the deep semantics of the text. Therefore, semantic similarity... It is measured by calculating the cosine similarity between the two semantic vectors: , Represents the dot product of semantic vectors; and represents the Euclidean norm of the semantic vector, which is between -1 and 1 (usually 0 to 1 in this embodiment). The closer the value is to 1, the more semantically similar the two text descriptions (such as "the road manhole cover collapsed" and "a large pothole appeared on the road") are.

[0036] In actual runtime, when a new raw event is reported Upon entering the system, a new original event report will be generated. With all currently "active" unified event objects E k (Each uniform event object is composed of one or more aggregated original event reports) Similarity calculation is performed. During the calculation, E is typically selected. k The first report or a comprehensively generated "representative report" as and Compare the results. If the calculated maximum similarity score is... Higher than the preset aggregation threshold Then Merge into the corresponding unified event object E k In the middle, and update the unified event object E. k Attributes (such as scope of influence, duration). If all similarity scores are below... ,determination This is a completely new, independent event; create a new unified event object for this independent event. It outputs a dynamically updated, deduplicated, and information-rich unified event view. Each view integrates evidence from different sources and includes spatiotemporal scope, multi-dimensional descriptions, and source tracing.

[0037] Step S202: Based on the pre-built urban governance knowledge graph and risk transmission model, analyze the unified event object and output an event analysis report; wherein, the event analysis report is a multi-dimensional analysis result of the unified event object.

[0038] In one example, S202 includes: mapping unified event objects to an initial node set in the urban governance knowledge graph based on a pre-built urban governance knowledge graph; generating risk potential energy for each node in the initial node set based on a risk transmission model of a graph attention network, and simulating the iterative propagation of risk potential energy between nodes in the urban governance knowledge graph; determining the comprehensive risk score of the unified event object based on the risk potential energy of each node after iterative propagation, and outputting an event assessment report based on the comprehensive risk score.

[0039] In one example, an event assessment report includes any one or more of the following: event type, risk level, and potential impact area.

[0040] For example, upon receiving a unified event object, intelligent analysis begins based on a pre-built urban governance knowledge graph and an AI model. The AI ​​model can be a risk transmission model, and this is not limited. The urban governance knowledge graph G is a pre-built graph covering the entire city. It is a vast semantic network. Here, V represents the set of entity nodes in the graph, containing various physical and logical objects in the city, such as infrastructure entities like roads, bridges, pipelines, buildings, and cameras, as well as administrative division entities like streets, communities, and management grids, and responsible unit entities like public security, fire protection, and urban management. R represents the rich set of relationships between entities, such as "located in," "connected," "responsible for," and "dependent on." T is the ontology layer definition containing entity and relationship types.

[0041] Optionally, in specific implementation scenarios, the entity node set V in the aforementioned knowledge graph is further refined into business objects with local governance characteristics, including: The three-tiered administrative management structure, consisting of "street-community-grid," covers all grid units within the jurisdiction; as well as corporate entities in the "one enterprise, one file" database (such as industry category nodes like manufacturing and chemical industries) and special population tag nodes in the "population data warehouse" (such as elderly people and community correction personnel).

[0042] When a "Heli Chemical Fire Point" alarm event is received, which is the result of the above steps, the system automatically retrieves key location nodes around the enterprise by using the "location" and "dependency" relationships between entities in the knowledge graph, identifying the plant as having a "high-risk chemical" attribute. Subsequently, the system dynamically extrapolates the potential impact of the fire on densely populated areas in the surrounding area using the risk transmission paths in the graph.

[0043] When a unified event object E is input, a graph mapping and context enhancement of the event are performed. First, the geographic locations contained in the unified event object E are parsed. The text description d and other attributes, through entity linking technology, locate and activate the initial set of nodes directly related to the events in the unified event object E in the urban governance knowledge graph G. For example, an event described as "a manhole cover on a certain road making an unusual noise near another road" will be mapped to the corresponding road node representing "a certain road", the road node representing "the other road", and a specific "manhole cover" facility node located near the intersection in the graph G.

[0044] After completing the graph mapping, the event chain risk transmission model (hereinafter referred to as the risk transmission model) based on graph attention networks is used to quantitatively assess the potential impact range and severity level of events. Specifically, for each graph node... Define the risk potential that evolves with time step k. At the initial time k=0, for each node in the initial node set V0 Based on the inherent attributes of the unified event object E (such as the authority of the report source, the urgency of the description, etc.), an initial risk potential is assigned. An initial value All other nodes The initial risk potential energy is zero.

[0045] Subsequently, the risk potential energy begins to propagate iteratively within the urban governance knowledge graph G. In the (k+1)th iteration, any node... Risk potential Based on its own potential energy at the previous moment, and from its adjacent nodes The potential energy received is the sum of the two. The mathematical expression is: , Represents nodes The set of directly connected neighbor nodes; It is a preset forgetting factor between 0 and 1, simulating the natural decay of risk over time or distance; This indicates that in the k-th iteration step, any node The potential for risk; This indicates that in the k-th iteration step, any node The potential for risk; The key relationship is the attention transmission coefficient, which determines the risk from the node. Conducted to The efficiency coefficient is dynamically calculated through an attention mechanism to reflect the differences in importance of different relationships in a specific context. The calculation method is as follows: , and These are nodes and nodes The feature vector representations, which are pre-learned by the graph embedding model, contain the attribute information of the nodes. It is a connection and Relationship The eigenvector representation of W and Wi. R It is a trainable weight matrix used to map the features of nodes and relations to a unified semantic space. This indicates a vector concatenation operation. It is a non-linear activation function. The entire formula means that the system will activate the function based on the current node. Based on the characteristics of the relationships between node pairs, a normalized attention score, i.e., the transmission coefficient, is dynamically calculated. For example, when the relationship is "supply electricity to" and When it is a "hospital" node, The value will be significantly higher than when the relationship is "nearby" and This is the case of a "park" node. Represents nodes The set of directly connected neighbor nodes; Indicates the neighboring node n_m of node n_j; This represents the feature vector of the neighbor node n_m; The feature vector represents the relationship between neighboring node n_m and the central node n_j.

[0046] After K rounds of iterative propagation, the algorithm terminates. At this point, the overall risk score I(E) of the entire event is defined as the final risk potential of all nodes in the graph G. The sum: , The overall risk score I(E) quantifies the overall severity of the event, ultimately representing the risk potential. higher nodes This constitutes the potential impact domain of the event. For example, a burst main water supply pipeline, after risk transmission calculation, may significantly increase the risk potential of downstream residential areas, a school, and a hospital.

[0047] Output an event assessment report. This report includes the event's overall risk score I(E), the inferred event type (such as "facility failure", "public safety", etc.), priority, and a visualized impact chain analysis consisting of high-risk nodes.

[0048] Optionally, for event risk assessment, a dynamic evolution model of event risk can be constructed, treating emergency response actions as endogenous variables. This model defines risk as a quantifiable value that changes over time and iteratively calculates it using a recursive function. The input to this function includes not only the inherent attributes of the event (such as the scope of impact and duration, which are not limited) and environmental factors, but also the variable of "response actions already taken." Therefore, this model can calculate in real time the effects of these actions on risk mitigation or catalysis based on the type, quantity, and timing of resource deployment, and, combined with real-time data from multiple sources such as the Internet of Things and video surveillance, continuously predict the evolution curve of the risk value over the next few hours.

[0049] Therefore, transforming vague risk perception into precise risk calculation provides objective data support for task prioritization and resource allocation decisions. By simulating different response plans in this model, the evolution of risks under different decision paths can be predicted through deduction, thereby improving the success rate of decision-making.

[0050] Step S203: Obtain the actual resource status at the current moment.

[0051] For example, this step is described in step S103, and will not be repeated here.

[0052] Step S204: Based on the event analysis report and the actual resource status, generate a resource scheduling scheme through multi-objective optimization and execute the resource scheduling scheme.

[0053] In one example, S204 includes: converting the events in the event assessment report into a set of pending tasks; wherein the set of pending tasks includes multiple tasks, each task being associated with a risk weight based on risk potential; obtaining a set of currently schedulable resource units based on a preset urban operation resource database; the set of resource units includes multiple resource units, each resource unit having multiple preset types of attributes; constructing a multi-objective optimization model based on tasks, risk weights, resource units, and attributes, with the objectives being total response time, total handling cost, and uncovered risks; solving the multi-objective optimization model based on a preset reinforcement learning adaptively guided genetic algorithm, generating a resource scheduling scheme, and executing the resource scheduling scheme.

[0054] For example, after receiving and parsing the event assessment report, the system seamlessly connects to the intelligent scheduling and resource optimization module with multi-objective constraints. This module intelligently selects the benchmark plan with the highest matching degree from the system's built-in emergency plan library (including but not limited to fire safety inspection plans, public health prevention and control plans, and classified management plans for the elderly or people of concern) based on the semantic features and risk level of the event. The fundamental goal of this operation is to inject dynamic parameters such as real-time risk potential, geographical coordinates, and potential impact range from the event assessment report into the selected benchmark plan, thereby generating a comprehensive scheduling instruction set that minimizes response time, reduces handling costs, and maximizes handling effectiveness.

[0055] In the first implementation, the high-risk node set in the event assessment report is first identified, and then combined with a pre-set contingency plan knowledge base to transform the high-risk node set into a specific set of tasks to be done. The to-do list includes multiple tasks, each task T j (j=1, 2…m) includes the task location (i.e., the geographical location of high-risk nodes), task type (such as traffic control, pipeline repair, personnel evacuation, etc.), and the urgency or weight w of the task. j The weight w j This is usually related to the final risk potential energy obtained in the risk transmission calculation of this node. Positive correlation.

[0056] In the specific implementation process, abstract task requirements are dynamically matched with the actual human resources in street-level grid management. Specifically, the system accesses a set of resource units. It covers hundreds of full-time grid workers and over a hundred grid teams that are regularly deployed within the jurisdiction. When assigning tasks, the system considers not only the geographical coordinates of the resource units... It also deeply integrates the unique attribute tags of grassroots governance. For specific governance tasks such as "road repair" and "road flooding" output in the assessment report, the scheduling engine will perform multi-objective optimization based on the historical completion rate and current task load of each grid team, and automatically generate an intelligent instruction set that includes the optimal route, estimated arrival time and targeted handling precautions, so as to ensure that limited grassroots human resources can be accurately and efficiently deployed to high-risk task sites.

[0057] The scheduling problem can be modeled as a complex multi-objective combinatorial optimization problem. A resource scheduling scheme X is defined as a set of task-to-resource assignments and corresponding execution paths. This problem is solved using a reinforcement learning-guided adaptive genetic algorithm, which effectively balances global search and local optimization, dynamically adapting to real-time changes in the problem.

[0058] Specifically, we first define several optimization objective functions for resource scheduling scheme X. The design of these objective functions aims to comprehensively evaluate the merits of a resource scheduling scheme: Minimize total response time The objective function of this optimization aims to ensure that resources reach the mission site as quickly as possible. The mathematical expression is: , in, In resource scheduling scheme X, resources... Assigned to carry out the mission . It is a task The location. Representative Resources From resources Current position Arrive at the mission location The required travel time is not a simple Euclidean distance, but rather the shortest travel time calculated based on the dynamic road network by calling the real-time traffic situation awareness service that is integrated with this system, taking into full account the impact of real-time traffic conditions.

[0059] Minimize total disposal cost The objective function of this optimization focuses on the economy of resource scheduling. Its mathematical expression is: , Representing resources Costs incurred during transit. It is to perform tasks The estimated operating cost, which is related to the complexity of the task and resources. Related to its characteristics.

[0060] Minimize the risk of being uncovered The optimization objective function ensures that the most important tasks are prioritized, maximizing the overall efficiency of the process. Its mathematical expression is: , Indicating in resource scheduling scheme A collection of tasks for which no resources have been assigned. It is a task Risk weights. Minimizing this function is equivalent to maximizing the sum of risk weights for covered tasks.

[0061] In a reinforcement learning-based adaptively guided genetic algorithm, a resource scheduling scheme X is encoded as a "chromosome". The chromosome uses an integer encoding method, its length is equal to the total number of tasks m, and the value of the j-th bit of the chromosome is i, representing the task. Assigned to resources (A value of 0 indicates no assignment). The algorithm's fitness function. The fitness function is a weighted combination of the three objective functions mentioned above, and is as follows: , This represents the normalized objective function value to eliminate the influence of different dimensions. There are three weighting coefficients. The value of the weighting coefficient reflects the decision-maker's preference for trade-offs among response speed, economic cost, and risk control. The weighting coefficient can be a preset value, a user-defined setting, or it can be adaptively calculated according to the actual situation, etc., without any limitation.

[0062] In one implementation method targeting the highly dynamic environment of grassroots governance, the system transforms various governance tasks sensed in real time into a set of pending tasks. For example, the task The condition is classified as "damaged manhole covers on major road sections" (high safety risk). This is a "health warning for elderly and disabled individuals" (high health risk). The task is classified as "community garbage and debris accumulation" (under environmental impact). Based on the intelligent assessment results, the system assigns different risk weights to each task. , such as setting , , At the same time, hundreds of grid workers and grid teams currently online are being mobilized as resource units. .

[0063] When performing scheduling optimization, the scheduling scheme Encoded as chromosomes of length equal to the number of tasks. If the chromosome is encoded as... This indicates the task. Assigned to grid resources ,Task Assign resources , and the task Currently in an unassigned state (belonging to a set) At this point, based on the fitness function, the weighting coefficients are dynamically adjusted according to the current governance focus. For example, during a special safety rectification campaign, decision-makers increase the risk control weight and set... , , Under this weight combination, because Risk of loss due to non-assignment This will be amplified in the fitness calculation, prompting the genetic algorithm to prioritize searching for tasks that can cover more high-weight tasks and have faster response times in subsequent iterations. The shortest possible assignment combination ensures that core governance risks are addressed in a closed-loop manner with the highest priority, even when resources are limited at the grassroots level.

[0064] Genetic algorithms iteratively evolve a population of multiple chromosomes using operators such as selection, crossover, and mutation to find the solution with the highest fitness. The selection of crossover and mutation operators and their parameters are no longer fixed but dynamically controlled by a built-in Q-learning reinforcement learning agent. The state S of this reinforcement learning agent is defined as the statistical characteristics of the genetic algorithm population in the current generation, for example... , It is a measure of the diversity of the t-th generation population. It is the rate of change of average fitness. It represents the number of consecutive, unimproved algebras where the population's optimal solution remains unchanged. The agent's action A is to select a combination from a predefined library of operator policies, for example... CT is the type of crossover operator (such as single-point crossover, uniform crossover). It is the crossover probability, and MT is the type of mutation operator (such as random reset, exchange mutation). It is the mutation probability.

[0065] Each time an action is performed Then, the genetic algorithm evolves to the next generation, and the system calculates a reward value based on the improvement in the fitness of the population. The Q-learning agent updates its Q-table or Q-network based on this reward, and its core update rule is: , It is the learning rate, which controls the extent to which new knowledge covers old knowledge. It is a discount factor, representing the degree of importance attached to future rewards. Through continuous "trial and error" and learning, Q-learning agents can learn to choose more destructive mutation operators (A contains high levels of mutations) when the search gets stuck in a local optimum (manifested as low diversity and fitness stagnation). To escape the trap; and when the population converges rapidly toward a better solution, a crossover operator that preserves superior genes (A contains high-quality genes) is selected. This enables intelligent and adaptive adjustment of the algorithm's search strategy. 'a' represents the next state. Any action that may be taken; This indicates the state at the next time step (i.e., the state at time t+1).

[0066] After the algorithm iteration terminates, the chromosome with the highest fitness in the population is... This is the final selected optimal resource scheduling scheme. Then... The code decodes and generates a structured, standardized intelligent scheduling instruction. This instruction clearly lists each assigned resource unit. The instructions include a list of tasks to be performed, the optimal navigation route planned for each task based on real-time traffic conditions, the estimated time of arrival (ETA), and precautions for task execution. This instruction is automatically distributed to the command systems or mobile terminals of various relevant responsible units through a standard data interface, thereby completing the closed-loop handling of urban incidents.

[0067] Optionally, in one specific implementation based on a street-level grassroots governance platform, the intelligent dispatch scheme supports fully automated system dispatching and provides an enhanced human-machine collaborative decision-making mode to adapt to the interactive needs of the command center's large screen. In this mode, the globally optimal dispatch instruction set generated by the system is pushed to the management terminal in real time. Managers can use the search, filtering, and reset functions provided by the system interface to manually review and intervene in specific tasks across 13 communities and 134 grid points, or in the real-time distribution of 367 grid members. The decision vector is recalculated and updated in real time based on human clicks on the interface (such as reselecting the executor or adjusting task priority), ensuring that the final issued instructions possess both the global optimality of the algorithm and conform to the complex command logic in grassroots governance. The "algorithm pre-playing combined with manual confirmation" mechanism effectively solves the boundary anomaly problems that may arise in fully automated decision-making under high-dynamic environments, while also achieving logical alignment with the pre-set query and interaction functions of the front-end interface.

[0068] In the second implementation, due to the rapidly changing situation at the emergency site, the initially formulated "optimal" plan may become invalid within minutes due to unexpected events such as new incidents, traffic congestion, and resource failures. This embodiment proposes an event-driven rolling optimization scheme for emergency plans. This method continuously monitors the global situational data stream and pre-sets a series of triggering conditions for plan failure, such as the occurrence of new high-priority events, unexpected changes in the status of critical resources, or actual execution deviations exceeding thresholds. Once a triggering condition is triggered, the current global situational awareness (including the latest status of all tasks and the latest location and availability of all resources) is immediately used as the new initial condition, and the global multi-objective optimization scheduling engine is instantly restarted to generate a brand-new, future-oriented, globally optimal scheduling scheme within seconds.

[0069] Therefore, the second approach transforms the emergency command system from the traditional "plan-execution" model to a continuous "perception-planning-action" cycle. It allows for rapid adjustments to deployment based on constantly changing emergency conditions, ensuring that resource allocation strategies at every moment are the optimal solutions given the current information.

[0070] Step S205: Analyze the resource scheduling scheme to obtain the actual execution records of each task; wherein, the resource scheduling scheme includes multiple tasks.

[0071] For example, in generating and issuing optimal intelligent scheduling instructions Next, this step begins with processing the issued optimal intelligent scheduling instructions. Continuous tracking. Through deep integration with the city's IoT platform, front-end law enforcement recorders, vehicle-mounted GPS systems, and structured reports submitted by frontline personnel via mobile applications, massive amounts of multi-source, heterogeneous data on the mission execution process are aggregated in real time. This data is analyzed and integrated to form a detailed "actual execution record." This record is related to the optimal intelligent scheduling instructions. The planned values ​​in the data form a "plan-execution" comparison. For example, regarding the resources allocated in the original plan... Execute the task The entries, It will include the actual departure time and actual arrival location of the resource. time The actual resource costs consumed during task execution And the actual effect evaluation after the task is completed. Actual effect evaluation It is a quantitative indicator with a value ranging from [0,1]. It is determined by a combination of on-site sensor data (such as the return of pipeline pressure to normal or a decrease in the congestion index) or by the on-site commander based on the score given in the standard evaluation manual. This signifies that the expected goals have been perfectly achieved.

[0072] Optionally, multi-dimensional feedback data from grassroots governance can be used to drive continuous iteration of model parameters. Specifically, the system automatically analyzes text and image reports uploaded by frontline personnel, including before-and-after comparison photos, and aggregates macro-governance indicators in real time, such as "incident completion rate," "monthly community incident ranking," and "overall incident trend curve." When monitoring the actual completion time of a certain type of governance task (such as "street flooding" or "illegal road occupation" in a specific area), the system can be used to further refine the data. With the contingency plan and schedule When a significant positive bias occurs, the Bayesian inference correction algorithm will automatically lower the "disposal efficiency parameter" of the corresponding grid team in that area. At the same time, the system combines the quantitative scores of resident evaluations collected in the "One-Click Appeal" module. Dynamically adjust the risk weights for such events. .

[0073] Step S206: Calculate the comprehensive deviation measure between the resource scheduling plan and the actual execution record.

[0074] For example, in order to perform deviation analysis and attribution, adaptive learning is performed based on a preset Bayesian inference-based parameter adaptive correction algorithm, and key parameters in the system knowledge base are dynamically updated. First, a comprehensive scheduling execution deviation metric is defined. It quantifies the gap between the overall performance of a single scheduled task and the expected result. This metric is a weighted average of deviations from multiple dimensions: , These are the weighting coefficients for each deviation, reflecting the decision-maker's focus on controlling time, cost, and effectiveness. The specific calculation methods for each deviation are as follows: Time Deviation : , Here Resources in the original plan Arrive at the mission The estimated travel time, This is the actual travel time. This represents the total number of assigned tasks. This metric measures the accuracy of route planning and time forecasting.

[0075] Cost deviation : , This is the estimated total disposal cost. This is the actual total disposal cost. This item measures the accuracy of the cost budget. This is the total number of tasks assigned.

[0076] Effect deviation : , It is a task The preset risk weights. It refers to the actual quality of the completed product. This represents the set of assigned tasks. This item measures the residual risk that was not completely eliminated due to inadequate execution, and is a direct measure of the effectiveness of the action.

[0077] Step S207: The parameter adaptive correction algorithm based on Bayesian inference updates the preset key efficiency parameters using the comprehensive deviation metric as observational evidence.

[0078] For example, this step performs adaptive adjustments to key efficiency parameters of the drive system. Specifically, it calculates a comprehensive scheduling execution deviation metric. Then, a parameter adaptive correction algorithm based on Bayesian inference is initiated. The goal of this algorithm is to update the key efficiency parameters within the system that lead to inaccurate predictions, thereby updating resource units. Work efficiency parameters For example, this work efficiency parameter Directly affects travel time and operating costs The prediction. The algorithm flow is as follows: Define prior distribution Before any actual execution data is available, the system allocates resources. The efficiency is based on an initial belief, namely the prior distribution. The prior distribution can usually be assumed to have a mean of a pre-defined archival label. variance is Gaussian distribution .variance It represents a measure of the uncertainty of the initial calibration value.

[0079] Constructing the likelihood function When resources are obtained After obtaining the relevant actual execution data, the resources can be calculated. Local deviation caused (For example, considering only the time and cost skewness of the resource). The likelihood function describes the likelihood given a specific efficiency parameter. In the case of observing the current local deviation The possibility of this. A reasonable assumption is that when The closer the predicted efficiency is to the actual efficiency, the smaller the difference between the predicted and actual values ​​should be, i.e., the smaller the local bias. The smaller the value, the lower the likelihood function can be. Therefore, the likelihood function can be modeled as: , in, Based on efficiency parameters Predicted values ​​(such as travel time). This is the actual value (such as travel time). It is the variance of the observation noise.

[0080] Calculate the posterior distribution According to Bayes' theorem, combining the prior distribution with new observational evidence, we can obtain the updated distribution, i.e., the posterior distribution: , In this formula, When local deviations are observed Then, efficiency parameters The posterior probability distribution. It is the likelihood function. It is a prior distribution. , It is the marginal likelihood, which, as a normalization factor, ensures that the integral of the posterior distribution is 1. For the conjugate of the Gaussian prior and the Gaussian likelihood, the posterior distribution will also be a Gaussian distribution, with analytical solutions for its mean and variance, making the computation very efficient.

[0081] Parameter update: The system uses the expected value (i.e., mean) of the posterior distribution. As a resource New, calibrated efficiency parameters The data is then stored in the resource database. Simultaneously, the decrease in the variance of the posterior distribution indicates that the system's estimation of this parameter becomes more certain.

[0082] This self-evolutionary process is not limited to correcting resource efficiency. The same Bayesian framework can be widely applied to other knowledge modules of the system. For example, it can be used to assess the effectiveness of handling certain types of events (such as gas leaks in specific areas). If the overall performance falls short of expectations, the system can use this as evidence to increase the comprehensive risk score for such events in the intelligent assessment module. or its risk weight Prior estimates. If a certain type of resource scheduling scheme (e.g., a scheme that tends to save costs at the expense of time) is found to systematically lead to a large scheduling execution deviation metric D, the system can, in turn, adjust the objective function. Weighting coefficients In future decision-making, greater emphasis will be placed on the importance of time.

[0083] The method provided in this application automatically generates the optimal resource scheduling scheme through a multi-objective optimization algorithm, replacing the inefficient manual scheduling mode that relies on personal experience, and greatly improving the efficiency and accuracy of urban governance event management.

[0084] In one example, this application provides a data-driven, intelligence-driven, full-cycle management system for urban governance events. This system comprises a multi-source heterogeneous data fusion and event aggregation module, an intelligent judgment module based on knowledge graphs and AI models, a multi-objective constrained intelligent scheduling module and resource optimization module, and a closed-loop response and feedback system self-evolution module. The specific functions of these modules are detailed in the above embodiments and will not be repeated here.

[0085] Corresponding to the above method, this application also provides a flowchart of a control device for urban governance events, as shown in the embodiment. Figure 3 As shown, the device includes: Aggregation module 41 is used to obtain original event reports from multiple heterogeneous channels; perform spatiotemporal and semantic multidimensional association aggregation on the original event reports to generate a structured unified event object; The analysis module 42 is used to analyze the unified event object based on a pre-built urban governance knowledge graph and risk transmission model, and output an event analysis report; wherein the event analysis report is a multi-dimensional analysis result of the unified event object; The scheduling module 43 is used to obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, it generates a resource scheduling scheme through multi-objective optimization and executes the resource scheduling scheme.

[0086] The functions of each functional unit of the urban governance event control device provided in the above embodiments of this application can be implemented through the above methods and steps, and will not be repeated here.

[0087] This application also provides an electronic device, such as... Figure 4 As shown, it includes a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540.

[0088] Memory 530 is used to store computer programs; The processor 510 performs the above steps when executing the program stored in the memory 530.

[0089] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the urban governance event management method described in any of the above embodiments.

[0090] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the urban governance event management method described in any of the above embodiments.

[0091] 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 method for managing urban governance incidents, characterized in that, include: The system acquires original event reports from multiple heterogeneous channels, performs spatiotemporal and semantic multidimensional association aggregation on the original event reports, and generates a structured unified event object. The original event reports from multiple heterogeneous channels include text work orders from government service hotlines, structured alarm data reported by IoT devices, video stream summaries captured by urban surveillance cameras, and information reported by citizens through mobile applications that includes geographical location, images, and text descriptions. Based on a pre-built urban governance knowledge graph and risk transmission model, the unified event object is analyzed and an event analysis report is output; wherein, the event analysis report is a multi-dimensional analysis result of the unified event object; Obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, generate a resource scheduling scheme through multi-objective optimization, and execute the resource scheduling scheme.

2. The method according to claim 1, characterized in that, The original event report is subjected to multi-dimensional spatiotemporal and semantic association aggregation to generate a structured unified event object, including: Each raw event report is parsed and mapped to a pre-defined standard internal event report structure; Calculate the overall similarity score between any two original event reports; wherein the overall similarity score is a weighted sum of temporal similarity, spatial similarity, and semantic similarity; When the overall similarity score exceeds a preset threshold, it is determined that the two original event reports belong to the same objective event, and multidimensional association clustering is performed to generate a structured unified event object.

3. The method according to claim 2, characterized in that, Calculate the overall similarity score between any two original event reports, including: Using a pre-trained deep language model, the textual description of each original event report is mapped into a semantic vector; Calculate the cosine similarity between any two semantic vectors, and determine the comprehensive similarity score based on the cosine similarity.

4. The method according to claim 1, characterized in that, Based on a pre-built urban governance knowledge graph and risk transmission model, the unified event object is analyzed, and an event analysis report is output, including: Based on a pre-built urban governance knowledge graph, the unified event object is mapped to the initial node set in the urban governance knowledge graph; For each node in the initial node set, a risk potential energy is generated for each node based on the risk transmission model of the graph attention network, and the iterative propagation of the risk potential energy among the nodes of the urban governance knowledge graph is simulated. Based on the risk potential of each node after iterative propagation, a comprehensive risk score for the unified event object is determined, and an event assessment report is output based on the comprehensive risk score.

5. The method according to claim 1, characterized in that, Based on the event assessment report and the actual resource status, a resource scheduling scheme is generated through multi-objective optimization, and the resource scheduling scheme is executed, including: The events in the event assessment report are transformed into a set of tasks to be done; wherein, the set of tasks to be done includes multiple tasks, and each task is associated with a risk weight based on risk potential. Based on a preset urban operation resource database, obtain the set of currently schedulable resource units; wherein, the set of resource units includes multiple resource units, and each resource unit has multiple preset types of attributes; Based on the task, the risk weight, the resource unit, and the attribute, a multi-objective optimization model is constructed with the total response time, total handling cost, and uncovered risks as the objectives. Based on a pre-defined reinforcement learning adaptively guided genetic algorithm, the multi-objective optimization model is solved to generate a resource scheduling scheme, which is then executed.

6. The method according to claim 1, characterized in that, The method further includes: The resource scheduling scheme is analyzed to obtain the actual execution records of each task; wherein, the resource scheduling scheme includes multiple tasks; Calculate the comprehensive deviation metric between the resource scheduling scheme and the actual execution record; The parameter adaptive correction algorithm based on Bayesian inference updates the preset key efficiency parameters using the comprehensive deviation metric as observational evidence.

7. The method according to any one of claims 1-6, characterized in that, The event analysis report includes any one or more of the following: Event type, risk level, and potential impact area.

8. A control device for urban governance events, characterized in that, The device includes: The aggregation module is used to obtain raw event reports from multiple heterogeneous channels; perform spatiotemporal and semantic multidimensional association aggregation on the raw event reports to generate a structured unified event object; The analysis module is used to analyze the unified event object based on a pre-built urban governance knowledge graph and risk transmission model, and output an event analysis report; wherein the event analysis report is a multi-dimensional analysis result of the unified event object; The scheduling module is used to obtain the actual resource status at the current moment; based on the event analysis report and the actual resource status, it generates a resource scheduling scheme through multi-objective optimization and executes the resource scheduling scheme.

9. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images