Intelligent urban management information visualized comprehensive management method and system

Abstract

The application discloses an intelligent urban management information visualized comprehensive management method and system, and particularly relates to the technical field of urban management, and is used for solving the problem of inaccurate emergency response caused by false association and confused causality in the prior art, and realizes accurate management through multi-modal time sequence data feature alignment, dynamic construction of a causal chain network and visualized decision mapping technology, first extracts multi-dimensional event features of video monitoring, sensors and text work orders and constructs a cross-modal association chain, screens out interference chains through causal strength evaluation and counterfactual verification, generates dynamic correction parameters in combination with space-time weight and feedback priority, finally fuses the causal chain network and a geographic information system to output a visualized guidance scheme, drives decision optimization through causal logic, and significantly improves the accuracy and cross-departmental collaboration efficiency of urban event disposal.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of urban management, more specifically, the present application relates to an intelligent urban management information visualization comprehensive management method and system. BACKGROUND

[0002] In the field of urban management, event recognition and decision support are usually performed by analyzing the relevance of multi-modal data (such as video monitoring, sensor monitoring, and text work orders). Mainly through rule matching or deep learning model based on a single mode to extract event features, and using statistical correlation indicators (such as co-occurrence frequency and time sequence synchronization) to establish cross-modal event correlation. For example, after identifying a traffic accident through video frame analysis, the environmental sensor data fluctuation in the same area is directly associated as the cause and effect. This technical path maps different modal data to a unified feature space, and uses a shallow association network to model the relationship, without fully considering the potential interference of mixed variables between multi-source data.

[0003] The prior art fails to effectively distinguish between spurious correlation and real causal chain in the process of multi-modal implicit event correlation mining, resulting in decision recommendations deviating from actual needs. For example, when multiple events (such as traffic congestion and air quality deterioration) are caused by a common mixed factor (such as road construction), it is easy to misjudge the surface data synchronization as a direct causal relationship, ignoring the implicit interference of mixed variables on the relationship between multi-modal data, which can cause resource scheduling errors (such as incorrectly closing related facilities) or risk prediction failures (such as failing to identify the real source of risk), seriously affecting the emergency response efficiency of urban management. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an intelligent urban management information visualization comprehensive management method and system to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] The intelligent urban management information visualization comprehensive management method comprises the following steps:

[0007] S1, acquiring multi-modal time series data streams of the urban management system and extracting event features respectively, and mapping the event features to a multi-dimensional feature space;

[0008] Among them, the multi-modal time series data stream includes video monitoring data, sensor monitoring data, and text work order data;

[0009] S2, constructing an initial event correlation chain based on the cross-modal time sequence synchronization and co-occurrence frequency of the event features in the multi-dimensional feature space;

[0010] S3, filter out the interfering event chains that are synchronized but have insufficient causal relationships in the initial event association chain to generate a purified association chain;

[0011] S4. Verify the causal consistency of the purification association chain;

[0012] S5. Adjust the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain, and generate dynamic correction weights based on the feedback priority of the text work order data;

[0013] S6. Integrate the dynamic correction weights with the purified association chains verified by causal consistency to construct a cross-modal causal chain network and generate a visual guidance plan.

[0014] In a preferred embodiment, the multimodal time series data stream of the urban management system is obtained and event features are extracted respectively, and the event features are mapped to a multidimensional feature space, including:

[0015] Obtain video surveillance data and extract spatial distribution features in video frames through a pre-trained convolutional neural network;

[0016] Acquire sensor monitoring data and extract the temporal fluctuation characteristics of sensor values ​​through recurrent neural networks;

[0017] Obtain text work order data and extract semantic event label features from the work order text using a natural language processing model;

[0018] The spatial distribution features, temporal fluctuation features and semantic event label features are respectively subjected to dimensionality reduction processing and mapped to a multidimensional feature space with the same dimension. The direction of each feature vector is adjusted so that the cosine similarity calculation results between feature vectors of different modalities can be compared.

[0019] In a preferred embodiment, constructing an initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in a multi-dimensional feature space includes:

[0020] Based on the temporal alignment relationship of the spatial distribution feature sequence of video surveillance data, the temporal fluctuation characteristics of sensor monitoring data, and the semantic event label features in the multidimensional feature space, the cosine similarity of the feature vectors of different modalities in the same time window is calculated as the cross-modal temporal synchronization indicator;

[0021] Count the co-occurrence frequencies of different modal feature vectors in the same geographical area with a probability exceeding a preset threshold. The co-occurrence frequency is calculated based on the ratio of the number of simultaneous multimodal events in the historical data to the total number of events.

[0022] The cross-modal temporal synchronization index and the co-occurrence frequency are weightedly fused to generate an initial event association chain. The initial event association chain contains the association strength value and association direction between event nodes. The association direction is determined by the sequence of feature vectors in the time window.

[0023] In a preferred embodiment, filtering out interfering event chains that are synchronized but have insufficient causal relationships in the initial event correlation chain to generate a purified correlation chain includes:

[0024] The causal strength value between event nodes in the initial event association chain is calculated by the causal strength evaluation method based on time series prediction error. The causal strength value is a quantitative indicator of the influence of the predecessor event node on the subsequent event node.

[0025] Based on counterfactual analysis, the causal relationship between event nodes is verified. If the probability of the occurrence of the predecessor event node decreases below a preset threshold after the subsequent event node is eliminated by intervention, the corresponding association chain is determined to be an interference event chain.

[0026] The filtering rules are generated based on the causal strength value and the counterfactual analysis results to screen out the interfering event chains whose causal strength value is lower than the preset causal strength threshold or whose counterfactual analysis does not meet the conditions, and retain the remaining association chains as the purified association chains.

[0027] In a preferred embodiment, performing causal consistency verification on the purification association chain includes:

[0028] Verify whether the logical triggering sequence of sub-events in the purification association chain complies with the preset urban management business rules. The preset urban management business rules include the logical causal constraints between event types and the triggering time interval range;

[0029] Generate counterfactual time series data corresponding to the interference event chain, simulate causal failure scenarios through time series data reconstruction methods, and extract cross-modal anomaly windows caused by causal failure;

[0030] Calculate the temporal dependency intensity attenuation rate between the temporal fluctuation characteristics of sensor monitoring data and the spatial distribution feature sequence of video surveillance data within the cross-modal anomaly window;

[0031] If the logic triggering sequence complies with the preset urban management business rules and the temporal dependency intensity attenuation rate is within the preset threshold tolerance range, the causal consistency verification is determined to be passed.

[0032] In a preferred embodiment, the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain are adjusted, and the dynamic correction weight is generated in combination with the feedback priority of the text work order data, including:

[0033] The time decay factor and the spatial proximity are calculated based on the occurrence time and the geographical location code of the event nodes in the purification association chain, and a time-space weight parameter is generated;

[0034] The event type distribution density in the same geographical area in the statistical historical data is calculated to generate a coverage density correlation factor, and the event type distribution density is the ratio of the number of occurrences of the event type in a preset time period to the area;

[0035] The feedback priority in the semantic event label feature of the text work order data is extracted, and the feedback priority is calculated according to the weighted average of the emergency degree classification score marked in the work order text and the historical processing time;

[0036] The time-space weight parameter, the coverage density correlation factor and the feedback priority are fused by conditional rules to generate a dynamic correction weight.

[0037] In a preferred embodiment, the time decay factor is calculated according to the interval between the occurrence time of the event and the current time by an exponential function, and the spatial proximity is calculated according to the geographical location code distance between the event nodes.

[0038] In a preferred embodiment, the conditional rule of conditional rule fusion is: when the feedback priority reaches a preset emergency threshold, the dynamic correction weight is determined by the product of the time-space weight parameter and the coverage density correlation factor; when the feedback priority does not reach the emergency threshold, the dynamic correction weight is determined by the ratio of the coverage density correlation factor to the feedback priority.

[0039] In a preferred embodiment, the dynamic correction weight is integrated with the purification association chain verified by the causal consistency to construct a cross-modal causal chain network and generate a visual guidance scheme, including:

[0040] The dynamic correction weight and the verified purification association chain are mapped by nodes and edges to construct a cross-modal causal chain network with a directed weighted graph structure, where the nodes represent event types and the edges represent causal relationships with a dynamic correction weight value;

[0041] Based on the cross-modal causal chain network, a key path is extracted, and the screening condition of the key path is that the total weight of the path exceeds a preset display threshold and the path length meets the minimum event chain node number requirement;

[0042] The key path is superimposed and rendered with the city geographic information system map to generate a multi-dimensional interactive visual guidance scheme;

[0043] The visual guidance scheme is output and a cross-department task priority list is generated synchronously, and the task priority is arranged in descending order according to the weight value of the event node in the key path, and the event nodes with the same weight value are arranged in execution order according to the topological sorting.

[0044] In another aspect, the present application provides an intelligent urban management information visualization comprehensive management system, comprising the following modules:

[0045] A data acquisition module is configured to acquire multi-modal time series data streams of the urban management system and extract event features respectively, and map the event features to a multi-dimensional feature space;

[0046] An association construction module is configured to construct an initial event association chain based on the cross-modal time series synchronization and co-occurrence frequency of the event features in the multi-dimensional feature space;

[0047] An interference screening module is configured to screen out interference event chains that are synchronized but lack causal relationships in the initial event association chain, and generate a purified association chain;

[0048] A causal verification module is configured to verify the causal consistency of the purified association chain;

[0049] A weight adjustment module is configured to adjust the spatiotemporal weight parameters and coverage density association factors of the purified association chain, and generate a dynamic correction weight in combination with the feedback priority of the text work order data;

[0050] A scheme generation module is configured to integrate the dynamic correction weight and the purified association chain that passes the causal consistency verification, construct a cross-modal causal chain network, and generate a visual guidance scheme.

[0051] Compared with the prior art, the present application has the following beneficial effects:

[0052] 1. Through the deep causal association modeling of multi-modal data and the dynamic weight fusion mechanism, the recognition accuracy and decision reliability of urban management events are significantly improved; the cross-modal time series synchronization analysis and counterfactual causal verification technology can effectively identify and screen out false associations caused by mixed variables (such as road construction simultaneously causing traffic congestion and air quality fluctuations), ensuring the authenticity and explainability of the causal chain, thereby avoiding facility misoperation or risk source detection caused by misjudgment; through the dynamic adjustment of the spatiotemporal weight parameters and the coverage density association factors, in combination with the feedback priority of the text work order, the adaptive optimization of resource scheduling strategies is realized, solving the imbalance between local event handling and global resource allocation caused by static weights in traditional methods.

[0053] 2. Through multi-modal causal modeling and dynamic parameter fusion, a closed-loop iterative causal decision chain is formed; not only can causal bias caused by data noise or environmental changes be dynamically corrected, but also the event association strength and handling priority can be intuitively presented through the topological structure and weight heat map of the visual chain network, providing an operable decision guide for cross-departmental collaboration, solving the defects of traditional methods such as reliance on manual experience for parameter adjustment, response lag, and lack of explainability. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1This is a flow chart of the intelligent urban management information visualization integrated management method of the present invention;

[0055] Figure 2 It is a structural diagram of the intelligent urban management information visualization integrated management system of the present invention. DETAILED DESCRIPTION

[0056] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0057] Example 1: Figure 1 The present invention provides a method for comprehensive management of intelligent urban management information visualization, comprising the following steps:

[0058] S1. Obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space;

[0059] Among them, the multimodal time series data stream includes video surveillance data, sensor monitoring data, and text work order data;

[0060] S2. Constructing the initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in the multi-dimensional feature space;

[0061] S3, filter out the interfering event chains that are synchronized but have insufficient causal relationships in the initial event association chain to generate a purified association chain;

[0062] S4. Verify the causal consistency of the purification association chain;

[0063] S5. Adjust the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain, and generate dynamic correction weights based on the feedback priority of the text work order data;

[0064] S6. Integrate the dynamic correction weights with the purified association chains verified by causal consistency to construct a cross-modal causal chain network and generate a visual guidance plan.

[0065] S1. Obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space. The specific implementation is as follows:

[0066] The process of acquiring video monitoring data and extracting spatial distribution features in video frames through a pre-trained convolutional neural network includes: accessing real-time video stream data from the city management system, the video stream data coming from camera devices deployed in various regions of the city, the video stream data being transmitted to the data processing server at a rate of 25 frames per second; dividing the video stream data into video segments according to a fixed time window, each video segment containing continuous multiple frames of images; using a pre-trained convolutional neural network to extract features from each video segment, the training data of the convolutional neural network coming from a publicly available urban traffic monitoring data set, the training task being the positioning and classification of vehicles, pedestrians and road facilities in the video; the convolutional neural network outputs the spatial distribution features of the target objects in each video segment, including the position coordinates, size ratio and category probability distribution of the target objects in the video frames; the spatial distribution features of multiple frames of images in the same video segment are spliced in time sequence to form a spatial distribution feature sequence of the video monitoring data.

[0067] The process of acquiring sensor monitoring data and extracting time series fluctuation features of sensor values through a recurrent neural network includes: accessing sensor devices deployed in urban drainage networks, traffic intersections and environmental monitoring sites, the sensor devices including flow meters, temperature and humidity sensors, and air quality monitors; the sensor devices upload monitoring data at a fixed sampling frequency, the sampling frequency being set according to the type of sensor, the flow meter collecting data once every second, and the temperature and humidity sensor collecting data once every 10 seconds; aligning the sensor monitoring data by device number and time stamp to generate a time series data stream for each sensor; using a recurrent neural network to extract features from the time series data stream, the input layer of the recurrent neural network receiving a time-ordered sequence of sensor values, the hidden layer using a long short-term memory unit structure, and the output layer generating time series fluctuation features of the sensor values; the time series fluctuation features include the mean, variance, peak frequency and differential change between adjacent sampling points of the sensor values within a predetermined time window; for the time series fluctuation features of multiple sensors in the same geographical region, the spatial positions of the sensors are weighted and fused to generate regional-level sensor time series fluctuation features, wherein when the spatial positions of the sensors are weighted and fused, the weight values are dynamically calculated according to the deployment density of the sensor devices in the urban drainage networks and traffic intersections.

[0068] The process of obtaining text work order data and extracting semantic event label features in the work order text through a natural language processing model includes: exporting historical work order data and real-time reported work order data from the city management work order system, where the work order data includes a text description field, a reporting timestamp, and a geographic location code; preprocessing the work order text, which includes removing stop words, word segmentation, part-of-speech tagging, and named entity recognition; inputting the preprocessed work order text into a natural language processing model, which adopts a sequence encoding structure based on an attention mechanism, and the model training task is to identify event types, impact scope, and urgency labels from the work order text; the natural language processing model outputs the semantic event label features of the work order text, which include a probability distribution vector of the event type, a keyword weight list of the impact scope, and a classification score of the urgency; aggregating the semantic event label features of multiple work orders in the same geographic area, and aggregating by taking a weighted average of the probability distribution vectors of the same event type, where the weight is determined by the interval between the work order reporting time and the current time, and the shorter the interval, the higher the weight.

[0069] The spatial distribution features, temporal fluctuation features and semantic event label features are respectively subjected to dimensionality reduction processing, mapped to a multidimensional feature space of the same dimension, and the direction of each feature vector is adjusted so that the cosine similarity calculation results between feature vectors of different modalities are comparable. The process includes: using the principal component analysis method to reduce the dimensionality of the spatial distribution feature sequence of video surveillance data, retaining the principal component with a variance contribution rate of more than 85% as the spatial distribution feature after dimensionality reduction; using the linear discriminant analysis method to reduce the dimensionality of the temporal fluctuation features of regional-level sensors, so that the sensor features of different event categories have the maximum separability in the low-dimensional space; using the non-negative matrix decomposition method to reduce the dimensionality of the aggregated semantic event label features. , the constraint condition is that the value of each dimension of the eigenvector after dimensionality reduction is non-negative; the eigenvectors of the three modalities after dimensionality reduction are uniformly scaled to the same number of dimensions. The number of dimensions is determined based on experimental verification. The specific method is to test the accuracy of cross-modal feature matching under different numbers of dimensions on historical data, and select the number of dimensions with the highest accuracy as the final setting value; the direction of each modal eigenvector is adjusted by normalizing each eigenvector so that its modulus is 1, and rotating the eigenvector so that the cosine value of the angle between different modal eigenvectors in the same event category is maximized; after the direction adjustment, the cosine similarity calculation results between different modal eigenvectors can be directly used for cross-modal correlation measurement.

[0070] In the above implementation, the specific network structure parameters (such as the number of layers and the number of neurons) of the pre-trained convolutional neural network, recurrent neural network, and natural language processing model can be adjusted according to the actual data scale, but the consistency of the feature extraction process of different modalities must be guaranteed; the selection basis of technical parameters such as the variance contribution rate retained in the dimensionality reduction process, the category separability target of the linear discriminant analysis (the category separability target of the linear discriminant analysis is achieved by maximizing the ratio of the inter-class discreteness to the intra-class discreteness), and the constraints of the non-negative matrix decomposition must be verified by historical data to determine the optimal value; in the process of adjusting the direction of the eigenvector, the implementation of the normalization and rotation operations requires the use of a numerical optimization algorithm, such as the gradient descent method to find the optimal rotation matrix, but the specific value of the rotation matrix is ​​not limited to a fixed value. Ensure the comparability of multimodal data in a unified feature space and the feasibility of association analysis, while avoiding causal confusion problems caused by inconsistent feature scales and directions.

[0071] S2. Construct an initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in the multi-dimensional feature space. The specific implementation is as follows:

[0072] Based on the temporal alignment relationship of the spatial distribution feature sequence of video surveillance data, the temporal fluctuation feature of sensor monitoring data and the semantic event label feature in the multidimensional feature space, the process of calculating the cosine similarity of different modal feature vectors in the same time window as the cross-modal temporal synchronization index includes: according to the timestamp information of each modal feature vector in the multidimensional feature space generated in step S1, the different modal data are divided into multiple time segments according to the fixed time window length. The time window length is set according to the typical response cycle of urban management events. For example, the response cycle of traffic congestion events is usually 5 minutes, so the time window length is set to 5 minutes; in each time window, the same The spatial distribution feature sequence of video surveillance data at each moment, the temporal fluctuation features of sensor monitoring data, and the semantic event label features are aligned by interpolating the feature vector of the missing time point into the feature vector of the previous time point. For the different modal feature vectors in each time window, the cosine similarity between each two is calculated. For example, the cosine similarity between the spatial distribution feature sequence of video surveillance data and the temporal fluctuation features of sensor monitoring data is calculated by performing a dot product operation on the two feature vectors and dividing them by the product of their moduli. The cosine similarities of all pairs of modal feature vectors are accumulated and averaged according to the time window to obtain the cross-modal temporal synchronization index.

[0073] The process of counting the co-occurrence frequencies of different modal feature vectors in the same geographic area with a probability exceeding a preset threshold includes: grouping the feature vectors in the multidimensional feature space according to the geographic location code according to the same geographic area division rule defined in step S1; for each multimodal feature vector in the geographic area, counting the number of times the multimodal feature vectors co-occur within the same time window in the historical data; for example, if the spatial distribution feature sequence of the video surveillance data indicates a traffic congestion event, and the temporal fluctuation feature of the sensor monitoring data within the same time window indicates an excessive traffic flow, then this is determined to be a co-occurrence event; when calculating the co-occurrence frequency, the number of co-occurrences is divided by the total number of events in the historical data for the geographic area; for example, if a region has 100 events in the past month, of which the spatial distribution feature sequence of the video surveillance data and the temporal fluctuation feature of the sensor monitoring data co-occur 30 times, then the co-occurrence frequency is 30%; the preset threshold is set based on the minimum co-occurrence significance level of different event types determined through historical data analysis; for example, the co-occurrence frequency threshold for traffic congestion and excessive traffic flow is set at 20%; if the co-occurrence frequency is below this threshold, it is considered that there is no statistical significance.

[0074] The process of weighted fusion of cross-modal temporal synchronization indicators and co-occurrence frequencies to generate the initial event association chain includes: assigning weight coefficients to cross-modal temporal synchronization indicators and co-occurrence frequencies, and setting the weight coefficients based on verifying the contribution of different indicators to the prediction of real event associations through historical data. For example, if the temporal synchronization indicator has an accuracy rate of 60% in historical data, its weight coefficient is set to 0.6, and the co-occurrence frequency weight coefficient is set to 0.4; the weighted fusion is calculated by multiplying the temporal synchronization indicator by the weight coefficient and the co-occurrence frequency by the weight coefficient to generate a comprehensive association strength value; the association direction It is determined by analyzing the order of appearance of feature vectors within the time window. For example, if the spatial distribution feature sequence of the video surveillance data indicates that the time of occurrence of the traffic accident is earlier than the time when the temporal fluctuation feature of the sensor monitoring data indicates a decrease in traffic volume, then the association direction is from the video surveillance event to the sensor monitoring event; the generation rule of the initial event association chain is to connect the node pairs whose comprehensive association strength values ​​exceed the preset strength threshold and have the same association direction to form a directed association chain network. The preset strength threshold is determined through experimental verification, for example, gradually increasing the threshold in the test data until the false alarm rate of the association chain is less than 5%.

[0075] In the above implementation, the setting of the time window length needs to be adjusted in combination with the specific event type. For example, for a pipeline leakage event, the time window can be extended to 10 minutes due to the long response delay of the temporal fluctuation characteristics of the sensor monitoring data. The statistics of the co-occurrence frequency need to exclude abnormal co-occurrence caused by failure of the data acquisition equipment. For example, when the temporal fluctuation characteristics of all sensor monitoring data in the same time window are missing, the time window is not included in the statistics. The weight coefficient of the weighted fusion can be adjusted dynamically. For example, when the accumulated historical data of a certain type of event is insufficient, an equal weight distribution method is adopted. The judgment of the correlation direction needs to take into account the timestamp accuracy of the feature vector. For example, when the timestamp of the spatial distribution feature sequence of the video surveillance data is accurate to the second level, and the temporal fluctuation characteristics of the sensor monitoring data are at the millisecond level, the millisecond timestamp is used as the basis for judging the order.

[0076] S3. Eliminate the interfering event chains that are synchronized but lack causal relationships in the initial event association chain to generate a purified association chain. The specific implementation is as follows:

[0077] The process of calculating the causal strength value between event nodes in the initial event association chain by using a causal strength evaluation method based on time series prediction error includes: extracting event node pairs from the initial event association chain generated in step S2, each event node pair including a predecessor event node and a successor event node; for each event node pair, obtaining the occurrence time series of the predecessor event node in the historical data, and training a time series prediction model based on the time series, wherein the training goal of the time series prediction model is to predict the occurrence time of the successor event node based on the historical occurrence time of the predecessor event node; comparing the actual observed occurrence time of the successor event node with the predicted time, and calculating the absolute value of the prediction error. The causal strength value is calculated by subtracting the absolute value of the prediction error from a preset constant. The setting basis of the preset constant is twice the absolute value of the maximum prediction error in the historical data. For example, if the maximum prediction error in the historical data is 0.5, the preset constant is set to 1.0; the larger the causal strength value, the greater the degree of influence of the predecessor event node on the successor event node.

[0078] The process of verifying the causal relationship between event nodes based on counterfactual analysis includes: performing an intervention operation on each event node pair in the initial event association chain, where the intervention operation is to intervene and eliminate the occurrence record of the subsequent event node in the historical data; re-counting the occurrence probability of the predecessor event node after the intervention and comparing it with the occurrence probability before the intervention; if the decrease in the occurrence probability of the predecessor event node is lower than a preset threshold, the association chain is determined to be an interference event chain. The preset threshold is set based on the lowest value of the decrease in the occurrence probability of the real causal relationship event in the statistical historical data. For example, in the real causal relationship between traffic accidents and traffic volume reduction, the decrease in the probability of traffic accidents after the intervention to eliminate the traffic volume reduction event is 40%, then the preset threshold is set to 30%; the verification process of the counterfactual analysis needs to be repeated multiple times to reduce the impact of random errors. The number of repetitions is set according to the amount of historical data, for example, 10 times when the data volume is greater than 1,000.

[0079] The process of generating filtering rules based on causal strength values ​​and counterfactual analysis results includes: setting a causal strength threshold, which is determined by adjusting a test data set consisting of 80% of the samples in the historical data, and the remaining 20% ​​is used to verify the false positive rate; the adjustment method is to gradually increase the threshold in the test data until the false positive rate of the purified association chain is lower than the preset false positive rate threshold, for example, when the false positive rate threshold is set to 5%, the causal strength threshold is set to 0.7; the association chain with a causal strength value lower than 0.7 in the initial event association chain or a probability decrease of less than 30% in the counterfactual analysis is marked as an interference event chain; the marked interference event chains are screened out, and the remaining association chains are retained as the purified association chains; the purified association chains are stored as a directed graph structure, with nodes representing event types and edges representing causal relationships and association strength values.

[0080] In the above implementation, the specific type of time series prediction model can be selected based on data characteristics. For example, for traffic congestion events with obvious periodicity, an autoregressive integrated moving average model is used, while for pipeline leakage events with strong randomness, a random forest regression model is used. The preset constant of the causal strength value needs to be dynamically updated. For example, the maximum absolute value of the prediction error is recalculated and the constant is adjusted based on new data each month. The intervention operation of the counterfactual analysis must exclude external factors. For example, when the intervention eliminates the record of the subsequent event node, the influence of other related events occurring during the same period must be simultaneously excluded. The setting of the filtering rules must support manual adjustment. For example, when the urban management strategy changes, the causal strength threshold can be manually increased to adapt to the new strategy. This can effectively screen out false association chains and improve the reliability of causal relationships.

[0081] S4. Verify the causal consistency of the purification association chain. The specific implementation is as follows:

[0082] The process of checking whether the logical trigger sequence of the sub-events in the purification association chain conforms to the preset urban management business rules includes: obtaining the preset urban management business rules from the business rule library of the urban management department, the preset urban management business rules including logical causal constraint conditions between event types and trigger time interval ranges, for example, after a traffic accident occurs, there must be a traffic diversion event and the trigger time interval is not more than 10 minutes; extracting the sub-events in the purification association chain in chronological order, and matching the logical causal constraint conditions one by one, if the subsequent event type of a certain sub-event is not within the range allowed by the constraint condition, it is determined that the logical trigger sequence does not conform; for the trigger time interval check, the time difference between the sub-event and its subsequent event is calculated, if the time difference exceeds the preset time interval range, for example, the time difference between the traffic accident and the traffic diversion event is more than 15 minutes, it is determined that it does not conform to the business rules.

[0083] The process of generating counterfactual temporal data corresponding to the interference event chain includes: extracting event node pairs from the interference event chain screened out in step S3, and simulating the causal relationship failure of each event node pair, the simulation method being to randomly disturb or replace the temporal relationship between the predecessor event node and the successor event node in the historical data with the time stamp of an irrelevant event; generating counterfactual temporal data through a temporal data reconstruction method, the specific operation of the temporal data reconstruction method being to retain the original values of the temporal fluctuation characteristics of the sensor monitoring data and the spatial distribution characteristic sequence of the video monitoring data, but rearrange their time sequence so that the causal relationship is invalid; extracting the cross-modal abnormal window caused by the causal relationship failure, the identification method of the cross-modal abnormal window being to detect the time period in which the synchronization of the temporal fluctuation characteristics of the sensor monitoring data and the spatial distribution characteristic sequence of the video monitoring data within the time window is lower than a preset synchronization threshold, for example, when the preset synchronization threshold is 0.6, the time period in which the synchronization is lower than 0.6 is determined as an abnormal window.

[0084] The process of calculating the temporal dependence strength decay rate between the temporal fluctuation characteristics of the sensor monitoring data and the spatial distribution characteristic sequence of the video monitoring data in the cross-modal abnormal window includes: in real data, the calculation process of the temporal dependence strength includes covariance calculation and normalization to 0 to 1 between the temporal fluctuation characteristics of the sensor monitoring data and the spatial distribution characteristic sequence of the video monitoring data; in counterfactual data, the same calculation is performed on the same time window to obtain the counterfactual temporal dependence strength; the calculation method of the decay rate is to divide the counterfactual temporal dependence strength by the real temporal dependence strength, if the decay rate exceeds 1, it indicates that the dependence relationship is enhanced, and if it is lower than 1, it indicates that the dependence relationship is weakened; the preset threshold tolerance range is determined according to the decay rate distribution interval of the normal event association chain in the historical data, for example, the decay rate of the normal event is distributed between 0.85 and 1.15, then the threshold tolerance range is set to 0.8 to 1.2.

[0085] If the logical trigger sequence meets the preset urban management business rules and the time sequence dependency strength decay rate is within the preset threshold tolerance range, the process of determining the consistency verification passing includes: double checking each sub-event chain in the purification association chain, first verifying whether the logical trigger sequence meets the business rules, and second verifying whether the corresponding decay rate is within the tolerance range; if the sub-event chain meets both conditions, it is marked as a verified causal association chain; for the sub-event chain that does not pass the verification, it is returned to step S3 to re-screen out the interference chain and adjust the parameters.

[0086] In the above embodiment, the preset urban management business rule library needs to be updated regularly, for example, when the city traffic management adds night construction regulation rules, the corresponding event type constraint and time interval range need to be supplemented in the business rule library; the timestamp replacement operation in the time sequence data reconstruction method needs to ensure that the geographic position code of the replaced event is consistent with the original event, so as to avoid introducing spatial dimension interference; the preset synchronization threshold of the cross-modal abnormal window is set through historical data analysis, for example, the lowest synchronization of the normal event association chain in the historical data is selected as the threshold benchmark, and the lowest synchronization of the normal event is 0.6, then the threshold is set to 0.6; the statistics of the decay rate distribution interval need to exclude extreme abnormal values, for example, only the data within the 3σ range of the normal distribution is used to calculate the upper and lower limits of the interval, if the mean of the decay rate is 1.0 and the standard deviation is 0.1, then the interval is set to 0.7 to 1.3.

[0087] Step S4 improves the reliability of causal association through a double checking mechanism (logical trigger sequence and counterfactual data verification), traditional methods only rely on statistical correlation or single modal verification, which easily ignores the business logic constraints between events and the risk of dynamic failure of causal relationship; through preset urban management business rule verification, the causal chain is forced to comply with actual management experience (such as traffic accidents must have a diversion event), avoiding logical contradictions; at the same time, counterfactual time sequence data is introduced to simulate causal relationship failure scenarios, and the cross-modal dependency strength decay rate is quantified, which can identify false associations caused by data noise or accidental co-occurrence; compared with the prior art, the fusion of business rule prior knowledge and data-driven verification solves the problems of disconnection between causal association and business logic and weak anti-interference ability, thereby significantly improving the explainability of urban management decision and the accuracy of emergency response.

[0088] S5, adjust the spatiotemporal weight parameters and coverage density correlation factors of the purified association chain, and generate dynamic correction weights combined with the feedback priority of the text work order data, which is specifically implemented as:

[0089] The process of calculating the time decay factor and spatial proximity based on the occurrence time and geographic location code of the event node in the purified association chain includes: extracting the timestamp and geographic location code of the event node from the purified association chain verified in step S4, where the timestamp is accurate to the second level, and the geographic location code adopts the latitude and longitude grid coding rule, and the grid division is consistent with the urban management geographic information system; the time decay factor is calculated by dynamically selecting the decay coefficient according to the event type label, for example, a coefficient of 0.85 is used for traffic accidents and a coefficient of 0.7 is used for pipe network leakage, and the rationality of the coefficient is verified by regression analysis of historical data; the time decay factor is calculated as an exponential function of the interval days, for example, when the interval between traffic accidents is 3 days, the factor is 0.853≈0.614; the spatial proximity is calculated by using the Haversine formula to calculate the geographic spherical distance between event nodes, limiting its application to the city-level geographic range, and normalizing the distance to the range of 0 to 1, where a smaller value indicates a higher proximity.

[0090] The process of generating a coverage density correlation factor by counting the distribution density of event types within the same geographical area in historical data includes: dividing the urban geographical area into square kilometer grids, counting the number of occurrences of a specific event type in each grid within a preset time period, and setting the minimum occurrence threshold to 1 to avoid a too small denominator; the event type distribution density is calculated as (number of occurrences + 1) / (grid area + 1), and sparse data distortion is handled by Laplace smoothing; the coverage density correlation factor automatically classifies historical density data through the K-means clustering algorithm, for example, clustering the density into three categories of low, medium, and high and mapping them to 0.3, 0.6, and 0.9. Threshold generation does not require manual intervention.

[0091] The process of extracting feedback priority from the semantic event tag features of text work order data includes: extracting the urgency classification score (high 1.0, medium 0.5, low 0.2) and the historical processing time score (time ≤ 30 minutes is 1.0, and processing time > 30 minutes is reduced proportionally) from the semantic event tag features generated in step S1; in the cold start phase, equal weights are used to calculate the feedback priority (urgency 50%, processing time 50%), and when the historical data accumulation exceeds 100, it is switched to dynamic weights (the optimal ratio is selected through A / B testing, such as 70% and 30%); the final value of the feedback priority is the weighted score, for example, the traffic accident urgency 1.0×0.7+processing time 1.0×0.3=1.0.

[0092] The process of conditionally fusing the spatiotemporal weight parameter, coverage density correlation factor and feedback priority includes: the preset emergency threshold is dynamically set by the top 15% quantile of historical data (for example, 0.9), and a smooth transition interval of ±5% of the threshold is introduced (0.85-0.95); when the feedback priority is ≥0.95, the dynamic correction weight = spatiotemporal weight parameter × coverage density correlation factor; when the feedback priority is ≤0.85, the dynamic correction weight = coverage density correlation factor / feedback priority; within the transition interval (0.85<priority<0.95), linear interpolation is used to mix the two calculation results, for example, when the priority is 0.90, the weight = 0.5×product + 0.5×ratio; the final weight is normalized to the range of 0-1 to ensure compatibility with the visualization display rule of step S6.

[0093] In the above implementation, the time decay coefficient is dynamically updated monthly through sliding window regression analysis of newly added data; spatial proximity calculation is limited to city-level applications to avoid spherical errors; coverage density clustering is automatically performed every quarter to ensure that the threshold adapts to the latest data; feedback priority weights are optimized through A / B testing every two weeks after the cold start; and smooth transition rules eliminate the risk of conditional jumps.

[0094] Step S5 optimizes the accuracy of urban management decisions through a multidimensional dynamic parameter fusion mechanism. Traditional methods use static weight allocation, which cannot adapt to the spatiotemporal heterogeneity and dynamic changes in the urgency of urban events. This step introduces an adaptive attenuation coefficient for event type (e.g., 0.85 for traffic accidents and 0.7 for pipe leaks). By regressing historical data to quantify the time-dependent attenuation pattern, it addresses the response bias caused by uniform parameters. The Haversine spherical distance calculation and coverage density clustering are combined to improve geographic correlation accuracy. A smooth threshold transition interval is designed in the conditional rule fusion to avoid the weight jump problem caused by traditional hard thresholds. Through dynamic parameter adaptation, precise spatial calculation, and rule smoothing, the real-time and accuracy of the visual guidance solution in resource scheduling and emergency response are significantly improved.

[0095] S6. Integrate the dynamic correction weights with the purified association chain verified by causal consistency to construct a cross-modal causal chain network and generate a visual guidance plan. The specific implementation is as follows:

[0096] The process of constructing the cross-modal causal chain network of the directed weighted graph structure by mapping the dynamic correction weight and the purified association chain verified by the causal consistency includes: obtaining the dynamic correction weight from step S5, the dynamic correction weight is generated by the conditional rule fusion of step S5, and the weight value is a normalized value of 0 to 1; obtaining the purified association chain verified by the causal consistency from step S4, the purified association chain contains the causal relationship and the association direction between the event nodes; defining the node of the directed weighted graph as an event type (such as a traffic accident, a pipeline leakage), and the node attribute includes the event occurrence time and the geographic location code; defining the edge as the causal relationship, the starting point of the edge is the predecessor event node, and the ending point of the edge is the successor event node, and the edge attribute includes the dynamic correction weight value and the association direction; the construction rule is to only keep the edges with a dynamic correction weight greater than 0.3 to exclude low-weight noise interference, and the threshold value 0.3 is determined according to the false positive rate of low-weight edges in historical data.

[0097] The process of extracting the key path based on the cross-modal causal chain network includes: presetting a display threshold value by analyzing the distribution of the path weight sum corresponding to the effective management decision in the historical data, for example, in the traffic accident disposal scene, if 80% of the path weight sum of the effective decision exceeds 2.5, the display threshold value is set to 2.5; the minimum number of event chain nodes is set according to the city management business rules, for example, the traffic accident disposal needs to contain at least three nodes of accident report, traffic diversion and on-site cleaning; all possible paths in the directed weighted graph are traversed, the path weight sum (for example, the path contains three edges with weights of 0.8, 0.7 and 0.6, and the total sum is 2.1) is calculated, and the paths with a total sum greater than 2.5 and a node number greater than or equal to 3 are selected as the key path.

[0098] The process of superimposing and rendering the key path on the city geographic information system map includes: accessing the map layer of the city geographic information system, the map layer contains spatial data of roads, buildings and municipal facilities; mapping the event nodes in the key path to the corresponding coordinate points on the map according to the geographic location code, and the display rule of dynamic marking is: the nodes with a weight value greater than or equal to 0.7 are displayed as red icons, the nodes with a weight value of 0.5 to less than 0.7 are displayed as orange icons, and the nodes with a weight value less than 0.5 are displayed as yellow icons; the color gradient is verified through human-computer interaction experiments to ensure the recognition degree; the flow direction arrow of the causal edge is drawn according to the association direction, and the thickness of the arrow is proportional to the dynamic correction weight value, for example, the edge with a weight of 0.9 is drawn with a 3-pixel thick arrow, and the edge with a weight of 0.5 is drawn with a 1-pixel thin arrow; the weight heat map is generated by superimposing the weight values of the event nodes in the same geographic area, and the color gradient transitions from blue (low weight) to red (high weight), and the color scale division of the heat map is based on the equal frequency binning method.

[0099] The process of outputting a visual guidance plan and simultaneously generating a cross-departmental task priority list includes: outputting the visual guidance plan to the display terminal of the urban management command center, allowing users to view event details through touch interaction (such as clicking an icon to display time, location, and related events); the generation rule of the cross-departmental task priority list is as follows: extracting all event nodes in the critical path and sorting them from high to low according to the dynamically corrected weight value. For example, if the weight of node A is 0.9, the weight of node B is 0.8, and the weight of node C is 0.8, then the priority order is A>B>C; when the weight values ​​are the same, the execution order is determined by topological sorting. The topological sorting rule is to prioritize the execution of nodes with an in-degree of 0 (i.e., nodes without predecessor event constraints). For example, if the weights of nodes B and C are both 0.8 and there is no topological dependency, then they are sorted in order of event occurrence.

[0100] In the above implementation, the construction of a directed weighted graph needs to exclude cyclic dependency paths (such as event A → event B → event A). The detection method for cyclic paths is to mark visited nodes in a depth-first search and automatically remove them when a cycle is detected. The update cycle of the preset display threshold is synchronized with the adjustment of the urban management strategy. For example, the threshold is recalculated every quarter according to the new strategy. The color gradient and size parameters of the visual elements must comply with the human-computer interaction design specifications (such as the color blindness-friendly color scheme uses a blue-yellow gradient instead of a red-green contrast). The output format of the task priority list supports docking with a third-party scheduling system interface. The interface protocol uses JSON format to transmit event type, weight value and execution order fields. The JSON field naming is consistent with the urban management data dictionary. It transforms complex causal relationships into actionable decision-making plans while ensuring the efficiency and consistency of cross-departmental collaborative execution.

[0101] Example 2: Figure 2 The structural diagram of the intelligent urban management information visualization integrated management system of the present invention is given, and the intelligent urban management information visualization integrated management system includes the following modules:

[0102] Data acquisition module: used to obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space;

[0103] Association construction module: used to construct the initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in the multi-dimensional feature space;

[0104] Interference screening module: used to screen out interference event chains that are synchronized but have insufficient causal relationships in the initial event association chain, and generate a purified association chain;

[0105] Causal Verification Module: used to verify the causal consistency of the purification association chain;

[0106] The weight adjustment module is configured to adjust the spatiotemporal weight parameter and the coverage density correlation factor of the purification correlation chain, and generate a dynamic correction weight in combination with the feedback priority of the text work order data.

[0107] The scheme generation module is configured to integrate the dynamic correction weight and the purification correlation chain verified by the causal consistency, construct a cross-modal causal chain network, and generate a visual guidance scheme.

[0108] The above formulas are all dimensionless values, and the formulas are obtained by software simulation of a large amount of data to obtain a formula closest to the actual situation, and the preset parameters and threshold values in the formula are set by a person skilled in the art according to the actual situation.

[0109] It should be noted that the application can be deployed in the device itself to realize embedded application, or run on a PC terminal or other terminal with a user interface, thereby meeting various hardware environments and use requirements.

[0110] The above embodiments can be realized by software, hardware, firmware or any combination thereof, in whole or in part. When realized by software, the above embodiments can be realized in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center and the like containing one or more available medium sets. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0111] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and module can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0112] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the division of the above-described device embodiment is merely an example, and the division of the modules can be different, for example, a plurality of modules or a component can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection can be indirect coupling or communication connection through some interfaces, devices or modules, and can be electrical, mechanical or other forms.

[0113] The modules illustrated as separated components can or can not be physically separated, and the components illustrated as modules can or can not be physical modules, and can be located in one place or distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment.

[0114] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can be physically present alone, or two or more modules can be integrated in one module.

[0115] If the functions are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program codes that can be stored in the medium.

[0116] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0117] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. The intelligent urban management information visualization integrated management method is characterized by: The steps include: S1. Obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space; Among them, the multimodal time series data stream includes video surveillance data, sensor monitoring data, and text work order data; S2. Construct an initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in the multi-dimensional feature space, including: Based on the temporal alignment relationship of the spatial distribution feature sequence of video surveillance data, the temporal fluctuation characteristics of sensor monitoring data, and the semantic event label features in the multidimensional feature space, the cosine similarity of the feature vectors of different modalities in the same time window is calculated as the cross-modal temporal synchronization indicator; Count the co-occurrence frequencies of different modal feature vectors in the same geographical area with a probability exceeding a preset threshold. The co-occurrence frequency is calculated based on the ratio of the number of simultaneous multimodal events in the historical data to the total number of events. The cross-modal temporal synchronization index and the co-occurrence frequency are weightedly fused to generate an initial event association chain. The initial event association chain contains the association strength value and association direction between event nodes. The association direction is determined by the order of the feature vectors in the time window. S3. Eliminate interfering event chains that are synchronized but lack causal relationships in the initial event association chain to generate a purified association chain, including: The causal strength value between event nodes in the initial event association chain is calculated by the causal strength evaluation method based on time series prediction error. The causal strength value is a quantitative indicator of the influence of the predecessor event node on the subsequent event node. Based on counterfactual analysis, the causal relationship between event nodes is verified. If the probability of the occurrence of the predecessor event node decreases below a preset threshold after the subsequent event node is eliminated by intervention, the corresponding association chain is determined to be an interference event chain. Generate filtering rules based on the causal strength value and the counterfactual analysis results to filter out interfering event chains whose causal strength values ​​are lower than the preset causal strength threshold or whose counterfactual analysis does not meet the conditions, and retain the remaining association chains as purified association chains; S4. Verify the causal consistency of the purification association chain; S5. Adjust the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain, and generate dynamic correction weights based on the feedback priority of the text work order data; S6. Integrate the dynamic correction weights with the purified association chains verified by causal consistency to construct a cross-modal causal chain network and generate a visual guidance plan.

2. The intelligent urban management information visualization integrated management method according to claim 1 is characterized in that: Obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space, including: Obtain video surveillance data and extract spatial distribution features in video frames through a pre-trained convolutional neural network; Acquire sensor monitoring data and extract the temporal fluctuation characteristics of sensor values ​​through recurrent neural networks; Obtain text work order data and extract semantic event label features from the work order text using a natural language processing model; The spatial distribution features, temporal fluctuation features and semantic event label features are respectively subjected to dimensionality reduction processing and mapped to a multidimensional feature space with the same dimension. The direction of each feature vector is adjusted so that the cosine similarity calculation results between feature vectors of different modalities can be compared.

3. The intelligent urban management information visualization integrated management method according to claim 1 is characterized in that: Perform causal consistency verification on the purification association chain, including: Verify whether the logical triggering sequence of sub-events in the purification association chain complies with the preset urban management business rules. The preset urban management business rules include the logical causal constraints between event types and the triggering time interval range; Generate counterfactual time series data corresponding to the interference event chain, simulate causal failure scenarios through time series data reconstruction methods, and extract cross-modal anomaly windows caused by causal failure; Calculate the temporal dependency intensity attenuation rate between the temporal fluctuation characteristics of sensor monitoring data and the spatial distribution feature sequence of video surveillance data within the cross-modal anomaly window; If the logic triggering sequence complies with the preset urban management business rules and the temporal dependency intensity attenuation rate is within the preset threshold tolerance range, the causal consistency verification is determined to be passed.

4. The intelligent urban management information visualization integrated management method according to claim 1 is characterized in that: Adjust the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain, and generate dynamic correction weights based on the feedback priority of the text work order data, including: Based on the occurrence time and geographic location coding of the event nodes in the purification association chain, the time decay factor and spatial proximity are calculated to generate the spatiotemporal weight parameters; Calculate the distribution density of event types within the same geographical area in historical data to generate a coverage density correlation factor. The event type distribution density is the ratio of the number of occurrences of the event type within a preset time period to the area of ​​the region. Extract the feedback priority from the semantic event tag features of the text work order data. The feedback priority is calculated based on the weighted average of the urgency classification score annotated in the work order text and the historical processing time. The spatiotemporal weight parameters, coverage density correlation factors and feedback priorities are fused into conditional rules to generate dynamic correction weights.

5. The intelligent urban management information visualization integrated management method according to claim 4 is characterized in that: The time decay factor is calculated based on the exponential decay of the interval between the event occurrence time and the current time, and the spatial proximity is calculated based on the geographic location coding distance between event nodes.

6. The intelligent urban management information visualization integrated management method according to claim 4 is characterized in that: The conditional rules for conditional rule fusion are as follows: when the feedback priority reaches the preset emergency threshold, the dynamic correction weight is determined by the product of the spatiotemporal weight parameter and the coverage density correlation factor; when the feedback priority does not reach the emergency threshold, the dynamic correction weight is determined by the ratio of the coverage density correlation factor to the feedback priority.

7. The intelligent urban management information visualization integrated management method according to claim 1 is characterized in that: Integrate dynamic correction weights with purified association chains verified by causal consistency to construct a cross-modal causal chain network and generate a visual guidance solution, including: The dynamic correction weights are mapped to the verified purified association chains for node and edge weights to construct a cross-modal causal chain network with a directed weighted graph structure. The nodes represent event types, the edges represent causal relationships, and the weight values ​​are the dynamic correction weights. Extract key paths based on the cross-modal causal chain network. The screening conditions for key paths are that the total path weight exceeds the preset display threshold and the path length meets the minimum event chain node number requirement; Overlay and render the critical path with the city's geographic information system map to generate a multi-dimensional interactive visual guidance plan; Output a visual guidance plan and simultaneously generate a cross-departmental task priority list. Task priorities are arranged in descending order according to the weight values ​​of event nodes in the critical path. Event nodes with the same weight values ​​are topologically sorted to determine the execution order.

8. An intelligent urban management information visualization integrated management system, used to implement the intelligent urban management information visualization integrated management method according to any one of claims 1 to 7, characterized in that: Includes the following modules: Data acquisition module: used to obtain the multimodal time series data stream of the urban management system and extract event features respectively, and map the event features to a multidimensional feature space; Association construction module: used to construct the initial event association chain based on the cross-modal temporal synchronization and co-occurrence frequency of event features in the multi-dimensional feature space; Interference screening module: used to screen out interference event chains that are synchronized but have insufficient causal relationships in the initial event association chain, and generate a purified association chain; Causal Verification Module: used to verify the causal consistency of the purification association chain; Weight adjustment module: used to adjust the spatiotemporal weight parameters and coverage density correlation factors of the purification association chain, and generate dynamic correction weights based on the feedback priority of text work order data; Solution generation module: used to integrate dynamic correction weights and purified association chains verified by causal consistency, build a cross-modal causal chain network and generate visual guidance solutions.

Images