Intelligent inference and prediction method and system of agent graph for urban governance

Abstract

This invention provides an intelligent inference and prediction method and system for intelligent agent graphs in urban governance. First, it collects and preprocesses the operational data streams of each intelligent agent node in real time to obtain standardized real-time data. Then, it compares the real-time data of each node with a first-layer threshold to obtain preliminary state labels. Next, it aggregates the preliminary state labels of neighboring nodes and compares them with a second-layer threshold to obtain local area state labels. Then, it aggregates multiple local area state labels and compares them with a third-layer threshold to obtain global state labels. Next, based on the state labels at each level, it identifies abnormal patterns and deduces propagation paths by combining the graph structure, outputting intelligent inference results including event type and impact range. Finally, it inputs the historical state label sequence into a time-series analysis model to predict future states at each level. This invention achieves multi-scale accurate perception and prediction of urban events.

Description

Technical Field

[0001] This invention relates to the field of urban governance, and more specifically, to an intelligent inference and prediction method and system for intelligent agent graphs for urban governance. Background Technology

[0002] With the accelerating pace of urbanization, urban governance faces the technical challenge of fusing massive amounts of multi-source heterogeneous data and making real-time, accurate decisions. Existing urban governance monitoring methods typically use a single threshold to judge isolated data collected by sensors, such as comparing individual traffic flow or air quality values ​​with a fixed threshold to trigger an alarm. This method firstly ignores the spatial correlation and functional coupling between different monitoring points, resulting in a high false alarm rate at single points and difficulty in capturing the overall situation of regional events. Secondly, existing technologies lack a multi-scale, progressive state perception mechanism, failing to gradually aggregate and deduce the local and even global event evolution process from anomalies at micro-nodes, making the identification of abnormal event types and the definition of their impact range often lagging and inaccurate. Thirdly, in the event inference stage, most existing methods rely solely on time-series data while ignoring the graph topology between agents, making it difficult to simulate the spatial propagation path and potential risks of events. Finally, traditional prediction models typically perform time-series analysis based on single-point historical data, failing to incorporate spatial context information into the modeling, resulting in limited accuracy in predicting future states and a lack of spatiotemporal coherence.

[0003] Therefore, there is an urgent need for an urban governance technology that can integrate multi-source data to achieve multi-scale progressive inference and spatiotemporal prediction. Summary of the Invention

[0004] In view of the above problems, the purpose of this invention is to provide an intelligent inference and prediction method and system for intelligent agent graphs for urban governance. By constructing a multi-level comparison system, the system progressively aggregates node-level, local area-level, and global-level states to achieve accurate perception from micro-anomalies to macro-events. Combined with a time-series prediction model, the system outputs future states, thereby improving the level of intelligence in urban governance.

[0005] Specifically, firstly, by rasterizing urban areas into intelligent agent nodes and establishing association edges based on spatial proximity and historical data correlation, an intelligent agent graph containing geographical and functional coupling relationships is constructed. Subsequently, standardized real-time data is obtained by performing missing value imputation, outlier removal, resampling, and normalization on the real-time data stream. Secondly, in the local aggregation process, the local aggregation value is obtained by weighting and summing the first-order neighbor state labels of the central node according to the association edge weights and comparing it with the second-layer threshold, effectively utilizing the graph's neighbor information. Simultaneously, in the global aggregation process, the state labels of each local area are combined with the state labels of its neighboring areas according to preset rules to form a global aggregation feature, which is then compared with the third-layer threshold. Finally, by inputting the state labels of each level and the graph topology into a graph neural network, the final output is an intelligent inference result containing event type, core area, and scope of impact, providing an interpretable decision-making basis for urban governance.

[0006] The first aspect of this invention provides an intelligent inference and prediction method for an intelligent agent graph for urban governance, the method comprising: The operation data streams of each intelligent agent node are collected in real time, and the data streams are preprocessed to obtain standardized real-time data. Based on historical data distribution and governance rules, threshold groups for each level are obtained; Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. The first result is obtained by aggregating the preliminary state labels of neighboring nodes in the graph. The first result is then compared with the second-layer threshold to obtain the local region state label. The second result is obtained by aggregating the state labels of multiple local regions. The global state label is obtained by comparing the second result with the threshold of the third layer. Based on the status labels at each level, and using preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained. By inputting the state labels of each level within a preset historical time period into a pre-trained time series analysis model, the prediction results of the state of each level at future times can be obtained.

[0007] This plan also includes: Based on urban geographic information data, the data is divided into multiple spatiotemporal grid units according to a preset grid size, which serve as intelligent agent nodes; Obtain the spatial coordinates and boundary information of each grid cell, and determine the initial adjacency relationship between adjacent grid cells; Collect historical operational data for each grid cell and calculate the temporal correlation between any two grid cells; Functional association edges are added between grid cells whose temporal correlation exceeds a preset correlation threshold; The initial adjacency relationships and functional association edges are merged to obtain the agent graph.

[0008] In this solution, the preprocessing of the data stream to obtain standardized real-time data specifically includes: Missing values ​​are detected in the real-time collected runtime data stream, and the missing values ​​are filled by interpolation of data from adjacent time points or by filling with the average value of historical data from the same period. Outlier identification is performed on the completed data stream. Outlier data points are removed using box plots or by replacing them with the sliding window mean. The data stream, after being filled and discarded, is resampled at a uniform time frequency, and data of different dimensions are normalized to obtain standardized real-time data.

[0009] In this scheme, the step of aggregating the preliminary state labels of neighboring nodes in the graph to obtain the first result specifically includes: Centered on each agent node, obtain the identifiers and corresponding preliminary state labels of all its first-order neighbor nodes from the agent graph; Based on the weights of the associated edges between neighboring nodes and the central node in the graph, the preliminary state labels of the neighboring nodes are weighted and summed to obtain the local aggregation value of the central node. The local aggregated value is taken as the first result.

[0010] In this scheme, obtaining the second result based on the aggregation of multiple local region state labels specifically includes: Obtain the identifiers of all local regions and their corresponding local region state labels from the agent graph; Based on the adjacency relationships between local regions in the graph, determine the set of neighboring regions for each local region; For each local region, its own status label is combined with the status labels of its neighboring regions according to a preset aggregation rule. The aggregation rule includes counting the number of abnormal regions, calculating the proportion of abnormal regions, or performing a weighted average based on the weight of the associated edges between regions. The aggregated features obtained after combination are used as the second result.

[0011] In this scheme, the step of obtaining intelligent inference results based on the status labels at each level, and based on preset identification anomaly patterns and inference propagation paths, specifically includes: The state labels at each level and the agent graph are input together into a pre-trained graph neural network model to identify the type of abnormal event; Based on the chronological order of the appearance of status labels at each level, and following the rules of reverse chronological tracing, the propagation path of the event is simulated. Based on the spatial distribution and number of nodes along the propagation path, the system outputs intelligent inference results that include event type, core area, and affected range.

[0012] A second aspect of the present invention provides an intelligent inference and prediction system for an intelligent agent graph for urban governance, comprising an intelligent inference and prediction method program for an intelligent agent graph for urban governance. When the intelligent inference and prediction method program for an intelligent agent graph for urban governance is executed by the processor, it implements the following steps: The operation data streams of each intelligent agent node are collected in real time, and the data streams are preprocessed to obtain standardized real-time data. Based on historical data distribution and governance rules, threshold groups for each level are obtained; Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. The first result is obtained by aggregating the preliminary state labels of neighboring nodes in the graph. The first result is then compared with the second-layer threshold to obtain the local region state label. The second result is obtained by aggregating the state labels of multiple local regions. The global state label is obtained by comparing the second result with the threshold of the third layer. Based on the status labels at each level, and using preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained. By inputting the state labels of each level within a preset historical time period into a pre-trained time series analysis model, the prediction results of the state of each level at future times can be obtained.

[0013] A third aspect of the present invention provides a computer-readable storage medium comprising a program for intelligent inference and prediction of an intelligent agent graph for urban governance. When executed by a processor, the program implements the steps of the intelligent inference and prediction method for an intelligent agent graph for urban governance as described in any of the preceding claims.

[0014] This invention provides an intelligent inference and prediction method and system for intelligent agent graphs in urban governance. First, it collects and preprocesses the operational data streams of each intelligent agent node in real time to obtain standardized real-time data. Then, it compares the real-time data of each node with a first-layer threshold to obtain preliminary state labels. Next, it aggregates the preliminary state labels of neighboring nodes and compares them with a second-layer threshold to obtain local area state labels. Then, it aggregates multiple local area state labels and compares them with a third-layer threshold to obtain global state labels. Next, based on the state labels at each level, it identifies abnormal patterns and deduces propagation paths by combining the graph structure, outputting intelligent inference results including event type and impact range. Finally, it inputs the historical state label sequence into a time-series analysis model to predict future states at each level. This invention achieves multi-scale accurate perception and prediction of urban events. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope.

[0016] Figure 1 The flowchart of an intelligent inference and prediction method for an intelligent agent graph for urban governance according to the present invention is shown. Figure 2 The diagram illustrates a scenario of an intelligent agent graph for urban governance according to the present invention. Figure 3 The diagram shows a block diagram of an intelligent inference and prediction system for an intelligent agent graph for urban governance according to the present invention. Detailed Implementation

[0017] 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.

[0018] Unless otherwise defined, all terms (including technical and scientific terms) used in embodiments of this invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in a common dictionary shall be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and not as being interpreted in an idealized or highly formalized sense, unless expressly defined in this embodiment of the invention.

[0019] The terms "first," "second," and similar words used in the embodiments of this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "an," "a," or "the" do not indicate a quantity limitation, but rather indicate the presence of at least one. Similarly, terms such as "including" or "comprising" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The steps preceding or following the steps in the method of the embodiments of this invention are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0020] In addition, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0021] Figure 1 The flowchart of an intelligent inference and prediction method for an intelligent agent graph for urban governance according to the present invention is shown.

[0022] like Figure 1 As shown, the first aspect of this invention discloses an intelligent inference and prediction method for intelligent agent graphs for urban governance, the method comprising: S102, collect the running data stream of each intelligent agent node in real time, and preprocess the data stream to obtain standardized real-time data; S104, based on historical data distribution and governance rules, obtain threshold groups for each level; S106. Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. S108: Based on the preliminary state labels of neighboring nodes in the graph, the first result is obtained by aggregating them. The first result is then compared with the second-layer threshold to obtain the local region state label. S110, the second result is obtained by aggregating the state labels of multiple local regions, and the global state label is obtained by comparing the second result with the threshold of the third layer. S112, Based on the status labels at each level, and based on the preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained; S114. Input the state labels of each level within the preset historical time period into the pre-trained time series analysis model to obtain the prediction results of the state of each level at future time.

[0023] It should be noted that in this embodiment, multiple intelligent agent nodes, such as traffic sensors, environmental monitoring equipment, or crowd counting devices, are first deployed within the urban governance area. The operational data streams of each node are collected in real time through an IoT platform. The received raw data is cleaned, aligned, and normalized by a preprocessing module to generate standardized real-time data, ensuring the comparability of data from different sources. Subsequently, based on the statistical distribution of historical data over different time periods and the business rules of urban governance, the system pre-configures a set of multi-level thresholds, including node-level thresholds for detecting single-point anomalies, region-level thresholds for identifying local clusters, and global-level thresholds for judging the overall situation. Then, the real-time data of each node is compared with its corresponding first-level threshold (node-level threshold) to mark a preliminary state label, such as normal or abnormal. Next, with each node as the center, the preliminary state labels are aggregated from its graph neighbor nodes to form a local aggregated feature. This feature is compared with the second-level threshold (region-level threshold) to obtain a local region state label. Furthermore, all local region state labels are comprehensively aggregated to form a global aggregated feature, which is then compared with the third-level threshold (global-level threshold) to obtain a global state label. By leveraging the spatial and logical relationships between nodes in the agent graph, anomaly detection is expanded from isolated points to regions and even the entire system with contextual semantics. Next, based on state labels at each level and combined with the graph topology, the type, location, and possible propagation path of an event are inferred using pre-defined anomaly pattern recognition rules or graph neural network models. Finally, the state label sequences at each level within historical time periods are input into a pre-trained temporal prediction model, which outputs node-level, region-level, and global-level predicted states for multiple future time steps, providing forward-looking decision support for city managers. This embodiment improves the accuracy of urban event perception and the timeliness of prediction through multi-scale progressive threshold comparison and graph-enhanced spatiotemporal analysis.

[0024] According to an embodiment of the present invention, it further includes: Based on urban geographic information data, the data is divided into multiple spatiotemporal grid units according to a preset grid size, which serve as intelligent agent nodes; Obtain the spatial coordinates and boundary information of each grid cell, and determine the initial adjacency relationship between adjacent grid cells; Collect historical operational data for each grid cell and calculate the temporal correlation between any two grid cells; Functional association edges are added between grid cells whose temporal correlation exceeds a preset correlation threshold; The initial adjacency relationships and functional association edges are merged to obtain the agent graph.

[0025] It should be noted that, in this embodiment, based on urban geographic information data, the urban governance area is divided into multiple spatiotemporal grid units according to a preset grid size, and each grid unit is regarded as an intelligent agent node. The center coordinates and boundary range of each grid unit are obtained, and the initial adjacency relationship between adjacent grid units is determined based on spatial proximity, such as sharing edges or corners, forming a basic spatial topology. At the same time, the operational data of each grid unit in historical time periods are collected, such as traffic flow or air quality index, and the time series correlation coefficient between any two grid units is calculated to measure their functional correlation strength. Functional correlation edges are added between grid units with correlation exceeding a preset threshold. These edges can reflect the node relationship of non-spatial proximity but behavioral linkage, such as the influence of upstream and downstream traffic flow or cross-regional air pollution transmission. Finally, the initial adjacency edges formed by spatial adjacency relationships are merged with the functional correlation edges established based on correlation, and each edge is assigned a corresponding weight, thereby generating an intelligent agent map that includes geographic proximity and functional coupling.

[0026] According to an embodiment of the present invention, the preprocessing of the data stream to obtain standardized real-time data specifically includes: Missing values ​​are detected in the real-time collected runtime data stream, and the missing values ​​are filled by interpolation of data from adjacent time points or by filling with the average value of historical data from the same period. Outlier identification is performed on the completed data stream. Outlier data points are removed using box plots or by replacing them with the sliding window mean. The data stream, after being filled and discarded, is resampled at a uniform time frequency, and data of different dimensions are normalized to obtain standardized real-time data.

[0027] It should be noted that in this embodiment, firstly, missing value detection is performed on the real-time acquired running data stream. If data is missing at a certain time point, it is filled using linear interpolation of adjacent time points or the average of the same historical time period. Next, outlier identification is performed on the filled data stream. Using box plots, based on the interquartile range, it is determined whether the data exceeds the normal fluctuation range. Identified outliers are removed and replaced with the average within a sliding window. Then, the data stream that has undergone filling and removal processing is resampled at a uniform time frequency to align the data of all nodes on the time axis. Finally, the original data of different dimensions (such as speed, concentration, and number of people) are normalized, for example, by mapping them to a uniform numerical range through min-max scaling, to obtain standardized real-time data.

[0028] According to an embodiment of the present invention, the step of aggregating the preliminary state labels of neighboring nodes in the graph to obtain the first result specifically includes: Centered on each agent node, obtain the identifiers and corresponding preliminary state labels of all its first-order neighbor nodes from the agent graph; Based on the weights of the associated edges between neighboring nodes and the central node in the graph, the preliminary state labels of the neighboring nodes are weighted and summed to obtain the local aggregation value of the central node. The local aggregated value is taken as the first result.

[0029] It should be noted that in this embodiment, for each agent node, the identifiers of all its first-order neighbor nodes are first retrieved from the constructed agent graph, and the preliminary state labels of these neighbor nodes at the same time are obtained, for example, -1 for abnormal and 1 for normal. Then, the pre-assigned weights of each neighbor edge in the graph are read; these weights are determined by spatial distance, historical relevance, or business importance. The state label of each neighbor node is multiplied by its corresponding edge weight, and the weighted values ​​of all neighbor nodes are summed to obtain the local aggregation value of the central node. This aggregation value reflects the concentration of abnormal states in the local area where the central node is located and the influence of neighbor nodes. This local aggregation value is used as the first result and compared with a preset second-level threshold. If it exceeds the threshold, the local area where the central node is located is determined to be in an abnormal state. This embodiment enhances or suppresses the abnormality of isolated nodes within the neighbor context, thereby more accurately identifying spatially related local events.

[0030] According to an embodiment of the present invention, obtaining the second result based on the aggregation of multiple local region state labels specifically includes: Obtain the identifiers of all local regions and their corresponding local region state labels from the agent graph; Based on the adjacency relationships between local regions in the graph, determine the set of neighboring regions for each local region; For each local region, its own status label is combined with the status labels of its neighboring regions according to a preset aggregation rule. The aggregation rule includes counting the number of abnormal regions, calculating the proportion of abnormal regions, or performing a weighted average based on the weight of the associated edges between regions. The aggregated features obtained after combination are used as the second result.

[0031] It should be noted that in this embodiment, firstly, all the identifiers of the divided local regions and their corresponding latest local region status labels are obtained from the agent graph, such as whether each local region is marked as abnormal. Simultaneously, based on the spatial adjacency relationships or functional association edges between local regions in the graph, the set of neighboring regions for each local region is determined, thereby constructing a region-level association network. For each local region, its own status label and the status labels of all its neighboring regions are combined according to a preset aggregation rule. This aggregation rule includes counting the number of abnormal regions in the region and its neighboring regions, calculating the proportion of abnormal regions to all neighboring regions, or performing a weighted average of the status labels based on the weights of the association edges between regions. The aggregated feature obtained after combination serves as the second result, which comprehensively reflects the overall situation of a larger area constituted by a certain local region and its surrounding environment. This second result is compared with a preset third-layer threshold; if it exceeds the threshold, the entire system is determined to be in an abnormal state. This embodiment effectively avoids the limitation of misjudging global events due to isolated regional anomalies.

[0032] According to an embodiment of the present invention, the step of obtaining intelligent inference results based on the state labels at each level, and based on a preset identification anomaly pattern and inference propagation path, specifically includes: The state labels at each level and the agent graph are input together into a pre-trained graph neural network model to identify the type of abnormal event; Based on the chronological order of the appearance of status labels at each level, and following the rules of reverse chronological tracing, the propagation path of the event is simulated. Based on the spatial distribution and number of nodes along the propagation path, the system outputs intelligent inference results that include event type, core area, and affected range.

[0033] It should be noted that in this embodiment, firstly, the state labels at each level and the topological structure of the agent graph are input into a pre-trained graph neural network model. The graph neural network model, through multi-layer graph convolution operations, can capture the association patterns between abnormal nodes and their neighbors in a spatial dimension. Combined with the semantic information of the state labels at each level, it identifies the specific type of the current abnormal event, such as a traffic accident, a fire, or a crowd gathering. Subsequently, based on the chronological order of the state labels at each level during the multi-level progressive comparison process, and following the reverse tracing rule, starting from the current global state, it deduces backward along the direction of the associated edges in the graph to gradually locate the initial origin node that triggered the anomaly, thereby simulating the propagation path of the event among the agents. Finally, based on the spatial distribution and number of nodes involved in the propagation path, combined with the regional affiliation relationships in the graph, it determines the core region of the event and the possible affected area, outputting an intelligent inference result containing the event type, core region, and affected area.

[0034] It is worth mentioning that the prediction results of the states at each level at future times specifically include: Extract a sequence of state labels at various levels from a historical database for multiple consecutive time steps. The state labels at each level include node-level, local-level, and global-level labels. The state label sequence is encoded into a multi-dimensional temporal feature vector in chronological order; The multidimensional temporal feature vector is input into a pre-trained spatiotemporal prediction model, which includes a graph convolutional network module and a temporal convolutional network module. Spatial dependency features of the state at each time step are extracted using a graph convolutional network module; The temporal dependency features of state evolution over time are captured through a temporal convolutional network module; The spatial dependency features and temporal dependency features are fused and mapped through a fully connected layer to output the predicted state values ​​of each level for multiple preset time steps in the future.

[0035] It should be noted that in this embodiment, firstly, state label sequences at various levels for multiple consecutive time steps are extracted from the historical database. These sequences fully cover state information at three scales: node level, local region level, and global level. These state label sequences are encoded chronologically and converted into multi-dimensional temporal feature vectors, which serve as input to the prediction model. The multi-dimensional temporal feature vectors and the topological structure of the agent's graph are then input into a pre-trained spatiotemporal prediction model, which integrates a graph convolutional network module and a temporal convolutional network module. The graph convolutional network module is responsible for extracting the spatial dependency features of each node's state at each time step using the adjacency relationships of the graph, i.e., the mutual influence between adjacent nodes. The temporal convolutional network module slides along the time axis, capturing the periodic or trend-like patterns of the state sequence's evolution over time. Subsequently, the extracted spatial dependency features and temporal dependency features are deeply fused, and after nonlinear mapping through a fully connected layer, the predicted state values ​​at various levels for multiple preset time steps are output. These predicted values ​​are reorganized at the node level, local level, and global level to obtain the complete state prediction results for future times.

[0036] It is worth mentioning that, based on historical data distribution and governance rules, the threshold groups for each level are obtained, specifically including: Based on the statistical distribution of historical data in different time periods, the initial values ​​of the thresholds at each level for each preset time period are calculated respectively, wherein the preset time period includes weekdays, rest days, peak hours, and off-peak hours; Using the initial values ​​of the thresholds at each level as a benchmark, the degree of deviation between the data and the real-time data is calculated. If the deviation exceeds the preset deviation threshold, the threshold of the corresponding level will be dynamically corrected based on the preset adjustment rules.

[0037] It should be noted that in this embodiment, firstly, statistical analysis is performed on the operational data in the historical database. Based on the time pattern of urban operation, the statistical distribution characteristics of each level of indicators are calculated under different typical time periods. This is used as a benchmark to determine the initial values ​​of the thresholds for each level corresponding to each time period, forming a dynamic threshold template. During actual operation, the system continuously monitors the global distribution characteristics of real-time data, compares the real-time data distribution of the current time period with the initial threshold values ​​of the corresponding time period, and calculates the degree of deviation between the two. If the degree of deviation exceeds a preset deviation threshold, it indicates that the current urban operation status has changed significantly compared with the historical norm, at which point the dynamic threshold correction mechanism is triggered. According to preset adjustment rules, including scaling according to the deviation ratio or switching to a backup threshold template, the thresholds of the corresponding levels are dynamically corrected in real time, so that the thresholds can adaptively follow the evolution of the urban operation status.

[0038] Figure 3 The diagram shows a block diagram of an intelligent inference and prediction system for an intelligent agent graph for urban governance according to the present invention.

[0039] like Figure 3 As shown, the second aspect of the present invention discloses an intelligent inference and prediction system 3 for intelligent agent graphs for urban governance, including a memory 31 and a processor 32. The memory includes a program for intelligent inference and prediction of intelligent agent graphs for urban governance. When the processor executes the program for intelligent inference and prediction of intelligent agent graphs for urban governance, it performs the following steps: The operation data streams of each intelligent agent node are collected in real time, and the data streams are preprocessed to obtain standardized real-time data. Based on historical data distribution and governance rules, threshold groups for each level are obtained; Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. The first result is obtained by aggregating the preliminary state labels of neighboring nodes in the graph. The first result is then compared with the second-layer threshold to obtain the local region state label. The second result is obtained by aggregating the state labels of multiple local regions. The global state label is obtained by comparing the second result with the threshold of the third layer. Based on the status labels at each level, and using preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained. By inputting the state labels of each level within a preset historical time period into a pre-trained time series analysis model, the prediction results of the state of each level at future times can be obtained.

[0040] It should be noted that in this embodiment, multiple intelligent agent nodes, such as traffic sensors, environmental monitoring equipment, or crowd counting devices, are first deployed within the urban governance area. The operational data streams of each node are collected in real time through an IoT platform. The received raw data is cleaned, aligned, and normalized by a preprocessing module to generate standardized real-time data, ensuring the comparability of data from different sources. Subsequently, based on the statistical distribution of historical data across different time periods (e.g., weekdays and holidays) and the business rules of urban governance, the system pre-configures a set of multi-level thresholds, including node-level thresholds for detecting single-point anomalies, region-level thresholds for identifying local clusters, and global-level thresholds for judging the overall situation. Then, the real-time data of each node is compared with its corresponding first-level threshold (node-level threshold) to mark a preliminary state label, such as normal or abnormal. Next, with each node as the center, the preliminary state labels are aggregated from its graph neighbor nodes to form a local aggregated feature. This feature is compared with the second-level threshold (region-level threshold) to obtain a local region state label. Finally, all local region state labels are comprehensively aggregated to form a global aggregated feature, which is compared with the third-level threshold (global-level threshold) to obtain a global state label. By leveraging the spatial and logical relationships between nodes in the agent graph, anomaly detection is expanded from isolated points to regions and even the entire system with contextual semantics. Next, based on state labels at each level and combined with the graph topology, the type, location, and possible propagation path of an event are inferred using pre-defined anomaly pattern recognition rules or graph neural network models. Finally, the state label sequences at each level within historical time periods are input into a pre-trained temporal prediction model, which outputs node-level, region-level, and global-level predicted states for multiple future time steps, providing forward-looking decision support for city managers. This embodiment improves the accuracy of urban event perception and the timeliness of prediction through multi-scale progressive threshold comparison and graph-enhanced spatiotemporal analysis.

[0041] According to an embodiment of the present invention, it further includes: Based on urban geographic information data, the data is divided into multiple spatiotemporal grid units according to a preset grid size, which serve as intelligent agent nodes; Obtain the spatial coordinates and boundary information of each grid cell, and determine the initial adjacency relationship between adjacent grid cells; Collect historical operational data for each grid cell and calculate the temporal correlation between any two grid cells; Functional association edges are added between grid cells whose temporal correlation exceeds a preset correlation threshold; The initial adjacency relationships and functional association edges are merged to obtain the agent graph.

[0042] It should be noted that, in this embodiment, based on urban geographic information data, the urban governance area is divided into multiple spatiotemporal grid units according to a preset grid size, and each grid unit is regarded as an intelligent agent node. The center coordinates and boundary range of each grid unit are obtained, and the initial adjacency relationship between adjacent grid units is determined based on spatial proximity, such as sharing edges or corners, forming a basic spatial topology. At the same time, the operational data of each grid unit in historical time periods are collected, such as traffic flow or air quality index, and the time series correlation coefficient between any two grid units is calculated to measure their functional correlation strength. Functional correlation edges are added between grid units with correlation exceeding a preset threshold. These edges can reflect the node relationship of non-spatial proximity but behavioral linkage, such as the influence of upstream and downstream traffic flow or cross-regional air pollution transmission. Finally, the initial adjacency edges formed by spatial adjacency relationships are merged with the functional correlation edges established based on correlation, and each edge is assigned a corresponding weight, thereby generating an intelligent agent map that includes geographic proximity and functional coupling.

[0043] According to an embodiment of the present invention, the preprocessing of the data stream to obtain standardized real-time data specifically includes: Missing values ​​are detected in the real-time collected runtime data stream, and the missing values ​​are filled by interpolation of data from adjacent time points or by filling with the average value of historical data from the same period. Outlier identification is performed on the completed data stream. Outlier data points are removed using box plots or by replacing them with the sliding window mean. The data stream, after being filled and discarded, is resampled at a uniform time frequency, and data of different dimensions are normalized to obtain standardized real-time data.

[0044] It should be noted that in this embodiment, firstly, missing value detection is performed on the real-time acquired running data stream. If data is missing at a certain time point, it is filled using linear interpolation of adjacent time points or the average of the same historical time period. Next, outlier identification is performed on the filled data stream. Using box plots, based on the interquartile range, it is determined whether the data exceeds the normal fluctuation range. Identified outliers are removed and replaced with the average within a sliding window. Then, the data stream that has undergone filling and removal processing is resampled at a uniform time frequency to align the data of all nodes on the time axis. Finally, the original data of different dimensions (such as speed, concentration, and number of people) are normalized, for example, by mapping them to a uniform numerical range through min-max scaling, to obtain standardized real-time data.

[0045] According to an embodiment of the present invention, the step of aggregating the preliminary state labels of neighboring nodes in the graph to obtain the first result specifically includes: Centered on each agent node, obtain the identifiers and corresponding preliminary state labels of all its first-order neighbor nodes from the agent graph; Based on the weights of the associated edges between neighboring nodes and the central node in the graph, the preliminary state labels of the neighboring nodes are weighted and summed to obtain the local aggregation value of the central node. The local aggregated value is taken as the first result.

[0046] It should be noted that in this embodiment, for each agent node, the identifiers of all its first-order neighbor nodes are first retrieved from the constructed agent graph, and the preliminary state labels of these neighbor nodes at the same time are obtained, for example, -1 for abnormal and 1 for normal. Then, the pre-assigned weights of each neighbor edge in the graph are read; these weights are determined by spatial distance, historical relevance, or business importance. The state label of each neighbor node is multiplied by its corresponding edge weight, and the weighted values ​​of all neighbor nodes are summed to obtain the local aggregation value of the central node. This aggregation value reflects the concentration of abnormal states in the local area where the central node is located and the influence of neighbor nodes. This local aggregation value is used as the first result and compared with a preset second-level threshold. If it exceeds the threshold, the local area where the central node is located is determined to be in an abnormal state. This embodiment enhances or suppresses the abnormality of isolated nodes within the neighbor context, thereby more accurately identifying spatially related local events.

[0047] According to an embodiment of the present invention, obtaining the second result based on the aggregation of multiple local region state labels specifically includes: Obtain the identifiers of all local regions and their corresponding local region state labels from the agent graph; Based on the adjacency relationships between local regions in the graph, determine the set of neighboring regions for each local region; For each local region, its own status label is combined with the status labels of its neighboring regions according to a preset aggregation rule. The aggregation rule includes counting the number of abnormal regions, calculating the proportion of abnormal regions, or performing a weighted average based on the weight of the associated edges between regions. The aggregated features obtained after combination are used as the second result.

[0048] It should be noted that in this embodiment, firstly, all the identifiers of the divided local regions and their corresponding latest local region status labels are obtained from the agent graph, such as whether each local region is marked as abnormal. Simultaneously, based on the spatial adjacency relationships or functional association edges between local regions in the graph, the set of neighboring regions for each local region is determined, thereby constructing a region-level association network. For each local region, its own status label and the status labels of all its neighboring regions are combined according to a preset aggregation rule. This aggregation rule includes counting the number of abnormal regions in the region and its neighboring regions, calculating the proportion of abnormal regions to all neighboring regions, or performing a weighted average of the status labels based on the weights of the association edges between regions. The aggregated feature obtained after combination serves as the second result, which comprehensively reflects the overall situation of a larger area constituted by a certain local region and its surrounding environment. This second result is compared with a preset third-layer threshold; if it exceeds the threshold, the entire system is determined to be in an abnormal state. This embodiment effectively avoids the limitation of misjudging global events due to isolated regional anomalies.

[0049] According to an embodiment of the present invention, the step of obtaining intelligent inference results based on the state labels at each level, and based on a preset identification anomaly pattern and inference propagation path, specifically includes: The state labels at each level and the agent graph are input together into a pre-trained graph neural network model to identify the type of abnormal event; Based on the chronological order of the appearance of status labels at each level, and following the rules of reverse chronological tracing, the propagation path of the event is simulated. Based on the spatial distribution and number of nodes along the propagation path, the system outputs intelligent inference results that include event type, core area, and affected range.

[0050] It should be noted that in this embodiment, firstly, the state labels at each level and the topological structure of the agent graph are input into a pre-trained graph neural network model. The graph neural network model, through multi-layer graph convolution operations, can capture the association patterns between abnormal nodes and their neighbors in a spatial dimension. Combined with the semantic information of the state labels at each level, it identifies the specific type of the current abnormal event, such as a traffic accident, a fire, or a crowd gathering. Subsequently, based on the chronological order of the state labels at each level during the multi-level progressive comparison process, and following the reverse tracing rule, starting from the current global state, it deduces backward along the direction of the associated edges in the graph to gradually locate the initial origin node that triggered the anomaly, thereby simulating the propagation path of the event among the agents. Finally, based on the spatial distribution and number of nodes involved in the propagation path, combined with the regional affiliation relationships in the graph, it determines the core region of the event and the possible affected area, outputting an intelligent inference result containing the event type, core region, and affected area.

[0051] It is worth mentioning that the prediction results of the states at each level at future times specifically include: Extract a sequence of state labels at various levels from a historical database for multiple consecutive time steps. The state labels at each level include node-level, local-level, and global-level labels. The state label sequence is encoded into a multi-dimensional temporal feature vector in chronological order; The multidimensional temporal feature vector is input into a pre-trained spatiotemporal prediction model, which includes a graph convolutional network module and a temporal convolutional network module. Spatial dependency features of the state at each time step are extracted using a graph convolutional network module; The temporal dependency features of state evolution over time are captured through a temporal convolutional network module; The spatial dependency features and temporal dependency features are fused and mapped through a fully connected layer to output the predicted state values ​​of each level for multiple preset time steps in the future.

[0052] It should be noted that in this embodiment, firstly, state label sequences at various levels for multiple consecutive time steps are extracted from the historical database. These sequences fully cover state information at three scales: node level, local region level, and global level. These state label sequences are encoded chronologically and converted into multi-dimensional temporal feature vectors, which serve as input to the prediction model. The multi-dimensional temporal feature vectors and the topological structure of the agent's graph are then input into a pre-trained spatiotemporal prediction model, which integrates a graph convolutional network module and a temporal convolutional network module. The graph convolutional network module is responsible for extracting the spatial dependency features of each node's state at each time step using the adjacency relationships of the graph, i.e., the mutual influence between adjacent nodes. The temporal convolutional network module slides along the time axis, capturing the periodic or trend-like patterns of the state sequence's evolution over time. Subsequently, the extracted spatial dependency features and temporal dependency features are deeply fused, and after nonlinear mapping through a fully connected layer, the predicted state values ​​at various levels for multiple preset time steps are output. These predicted values ​​are reorganized at the node level, local level, and global level to obtain the complete state prediction results for future times.

[0053] It is worth mentioning that, based on historical data distribution and governance rules, the threshold groups for each level are obtained, specifically including: Based on the statistical distribution of historical data in different time periods, the initial values ​​of the thresholds at each level for each preset time period are calculated respectively, wherein the preset time period includes weekdays, rest days, peak hours, and off-peak hours; Using the initial values ​​of the thresholds at each level as a benchmark, the degree of deviation between the data and the real-time data is calculated. If the deviation exceeds the preset deviation threshold, the threshold of the corresponding level will be dynamically corrected based on the preset adjustment rules.

[0054] It should be noted that in this embodiment, firstly, statistical analysis is performed on the operational data in the historical database. Based on the time pattern of urban operation, the statistical distribution characteristics of each level of indicators are calculated under different typical time periods. This is used as a benchmark to determine the initial values ​​of the thresholds for each level corresponding to each time period, forming a dynamic threshold template. During actual operation, the system continuously monitors the global distribution characteristics of real-time data, compares the real-time data distribution of the current time period with the initial threshold values ​​of the corresponding time period, and calculates the degree of deviation between the two. If the degree of deviation exceeds a preset deviation threshold, it indicates that the current urban operation status has changed significantly compared with the historical norm, at which point the dynamic threshold correction mechanism is triggered. According to preset adjustment rules, including scaling according to the deviation ratio or switching to a backup threshold template, the thresholds of the corresponding levels are dynamically corrected in real time, so that the thresholds can adaptively follow the evolution of the urban operation status.

[0055] A third aspect of the present invention provides a computer-readable storage medium comprising a program for intelligent inference and prediction of an intelligent agent graph for urban governance. When executed by a processor, the program implements the steps of the intelligent inference and prediction method for an intelligent agent graph for urban governance as described in any of the preceding claims.

[0056] In summary, this invention provides an intelligent inference and prediction method and system for intelligent agent graphs in urban governance. First, it collects and preprocesses the operational data streams of each intelligent agent node in real time to obtain standardized real-time data. Then, it compares the real-time data of each node with a first-layer threshold to obtain preliminary state labels. Next, it aggregates the preliminary state labels of neighboring nodes and compares them with a second-layer threshold to obtain local area state labels. Then, it aggregates multiple local area state labels and compares them with a third-layer threshold to obtain global state labels. Next, based on the state labels at each level, it identifies abnormal patterns and deduces propagation paths using the graph structure, outputting intelligent inference results that include event type and impact range. Finally, it inputs the historical state label sequence into a time-series analysis model to predict future states at each level. This invention achieves multi-scale accurate perception and prediction of urban events.

[0057] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent inference and prediction of intelligent agent graphs for urban governance, characterized in that, The method includes: The operation data streams of each intelligent agent node are collected in real time, and the data streams are preprocessed to obtain standardized real-time data. Based on historical data distribution and governance rules, threshold groups for each level are obtained; Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. The first result is obtained by aggregating the preliminary state labels of neighboring nodes in the graph. The first result is then compared with the second-layer threshold to obtain the local region state label. The second result is obtained by aggregating the state labels of multiple local regions. The global state label is obtained by comparing the second result with the threshold of the third layer. Based on the status labels at each level, and using preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained. By inputting the state labels of each level within a preset historical time period into a pre-trained time series analysis model, the prediction results of the state of each level at future times can be obtained.

2. The intelligent inference and prediction method for intelligent agent graphs for urban governance according to claim 1, characterized in that, Also includes: Based on urban geographic information data, the data is divided into multiple spatiotemporal grid units according to a preset grid size, which serve as intelligent agent nodes; Obtain the spatial coordinates and boundary information of each grid cell, and determine the initial adjacency relationship between adjacent grid cells; Collect historical operational data for each grid cell and calculate the temporal correlation between any two grid cells; Functional association edges are added between grid cells whose temporal correlation exceeds a preset correlation threshold; The initial adjacency relationships and functional association edges are merged to obtain the agent graph.

3. The intelligent inference and prediction method for intelligent agent graphs for urban governance according to claim 1, characterized in that, The preprocessing of the data stream to obtain standardized real-time data specifically includes: Missing values ​​are detected in the real-time collected runtime data stream, and the missing values ​​are filled by interpolation of data from adjacent time points or by filling with the average value of historical data from the same period. Outlier identification is performed on the completed data stream. Outlier data points are removed using box plots or by replacing them with the sliding window mean. The data stream, after being filled and discarded, is resampled at a uniform time frequency, and data of different dimensions are normalized to obtain standardized real-time data.

4. The intelligent inference and prediction method for intelligent agent graphs for urban governance according to claim 1, characterized in that, The process of aggregating the preliminary state labels of neighboring nodes in the graph to obtain the first result specifically includes: Centered on each agent node, obtain the identifiers and corresponding preliminary state labels of all its first-order neighbor nodes from the agent graph; Based on the weights of the associated edges between neighboring nodes and the central node in the graph, the preliminary state labels of the neighboring nodes are weighted and summed to obtain the local aggregation value of the central node. The local aggregated value is taken as the first result.

5. The intelligent inference and prediction method for intelligent agent graphs for urban governance according to claim 1, characterized in that, The process of obtaining the second result based on the aggregation of multiple local region state labels specifically includes: Obtain the identifiers of all local regions and their corresponding local region state labels from the agent graph; Based on the adjacency relationships between local regions in the graph, determine the set of neighboring regions for each local region; For each local region, its own status label is combined with the status labels of its neighboring regions according to a preset aggregation rule. The aggregation rule includes counting the number of abnormal regions, calculating the proportion of abnormal regions, or performing a weighted average based on the weight of the associated edges between regions. The aggregated features obtained after combination are used as the second result.

6. The intelligent inference and prediction method for intelligent agent graphs for urban governance according to claim 1, characterized in that, The intelligent inference result obtained based on the status labels at each level, and based on the preset anomaly identification patterns and inference propagation paths, specifically includes: The state labels at each level and the agent graph are input together into a pre-trained graph neural network model to identify the type of abnormal event; Based on the chronological order of the appearance of status labels at each level, and following the rules of reverse chronological tracing, the propagation path of the event is simulated. Based on the spatial distribution and number of nodes along the propagation path, the system outputs intelligent inference results that include event type, core area, and affected range.

7. An intelligent inference and prediction system for intelligent agent graphs for urban governance, characterized in that, The system includes a memory and a processor. The memory includes a program for intelligent inference and prediction of an intelligent agent map for urban governance. When the processor executes the program for intelligent inference and prediction of an intelligent agent map for urban governance, it performs the following steps: The operation data streams of each intelligent agent node are collected in real time, and the data streams are preprocessed to obtain standardized real-time data. Based on historical data distribution and governance rules, threshold groups for each level are obtained; Based on the comparison between the real-time data of each node and the threshold of the first layer, the preliminary state label of each agent node is obtained. The first result is obtained by aggregating the preliminary state labels of neighboring nodes in the graph. The first result is then compared with the second-layer threshold to obtain the local region state label. The second result is obtained by aggregating the state labels of multiple local regions. The global state label is obtained by comparing the second result with the threshold of the third layer. Based on the status labels at each level, and using preset identification anomaly patterns and deduction propagation paths, intelligent inference results are obtained. By inputting the state labels of each level within a preset historical time period into a pre-trained time series analysis model, the prediction results of the state of each level at future times can be obtained.

8. The intelligent inference and prediction system for intelligent agent graphs for urban governance according to claim 7, characterized in that, Also includes: Based on urban geographic information data, the data is divided into multiple spatiotemporal grid units according to a preset grid size, which serve as intelligent agent nodes; Obtain the spatial coordinates and boundary information of each grid cell, and determine the initial adjacency relationship between adjacent grid cells; Collect historical operational data for each grid cell and calculate the temporal correlation between any two grid cells; Functional association edges are added between grid cells whose temporal correlation exceeds a preset correlation threshold; The initial adjacency relationships and functional association edges are merged to obtain the agent graph.

9. The intelligent inference and prediction system for intelligent agent graphs for urban governance according to claim 7, characterized in that, The preprocessing of the data stream to obtain standardized real-time data specifically includes: Missing values ​​are detected in the real-time collected runtime data stream, and the missing values ​​are filled by interpolation of data from adjacent time points or by filling with the average value of historical data from the same period. Outlier identification is performed on the completed data stream. Outlier data points are removed using box plots or by replacing them with the sliding window mean. The data stream, after being filled and discarded, is resampled at a uniform time frequency, and data of different dimensions are normalized to obtain standardized real-time data.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer-readable storage medium includes a program for intelligent inference and prediction of an intelligent agent graph for urban governance. When the program is executed by a processor, it implements the steps of the intelligent inference and prediction method for an intelligent agent graph for urban governance as described in any one of claims 1 to 6.

Images