Intelligent Internet of Things public safety management and control system and method based on multi-source data fusion

Abstract

The present invention provides an intelligent Internet of Things public safety management and control system and method based on multi-source data fusion, which belongs to the field of Internet of Things security technology. The system includes: by identifying the system operation scenario, loading the corresponding model parameters and response strategies, realizing the unified collection and standardized processing of multi-source data, and building a unified semantic representation structure. The system performs anomaly detection at the edge node, combines the semantic structure with the abnormal feature information, completes the risk scoring and graded alarm judgment, and generates control response instructions based on policy matching. At the same time, it supports the dynamic optimization of the scoring model and identity authentication, behavior auditing and data compliance export during operation. The present invention has the capabilities of intelligent identification, adaptive analysis and closed-loop response, which improves the accuracy and credibility of public safety management in the Internet of Things environment.

Description

Technical Field

[0001] The present invention relates to the field of Internet of Things security technology, and in particular to an intelligent Internet of Things public safety management and control system and method based on multi-source data fusion. Background Art

[0002] With the rapid development of the Internet of Things (IoT), a large number of sensing terminals, video surveillance equipment, network communication nodes, and environmental monitoring units have been deployed in the public security sector to achieve real-time perception and control of key scenarios such as urban operations, industrial sites, transportation, and energy facilities. These sensing devices continuously generate multi-type and multi-structured operational data, prompting public security management and control systems to gradually develop towards multi-source perception, data fusion, and intelligent response.

[0003] Currently, many IoT security systems employ rule-based engines, linear scoring, and threshold-based methods to analyze and score various security indicators during operation. These systems then use the scoring results to determine alerts and respond accordingly. Typical approaches include: setting fixed upper and lower limits for each risk indicator for normalized scoring; setting scoring weights based on expert experience; and employing fixed alert strategies and control command templates for responses. However, existing technologies still have significant limitations in the following areas: They lack a unified processing mechanism for multi-source, heterogeneous data, resulting in limited data fusion capabilities; scoring mechanisms are mostly based on static models, lacking the ability to dynamically adapt to changes in the system environment; weighting methods are sparse and fail to reflect the relative risk sensitivity of each indicator under different operating conditions; there is a lack of effective suppression strategies for abnormal disturbances, making the scoring process highly sensitive to sudden anomalies, impacting the stability of the results; there is a lack of linkage logic between alert levels and control strategies, resulting in delayed responses or inaccurate policy matching; and most scoring models fail to automatically evolve based on feedback, risking accuracy degradation over long-term operation.

[0004] Therefore, the existing Internet of Things public safety management and control system urgently needs a technical solution that can support multi-source data fusion, adaptive scoring mechanism, strong anomaly robustness, and response strategy linkage and model self-optimization capabilities to meet the needs of real-time perception, intelligent judgment and reliable handling of public safety incidents in complex dynamic environments. Summary of the Invention

[0005] In response to the problems existing in the existing Internet of Things public safety management and control system, such as the difficulty in multi-source data fusion, rigid scoring mechanism, insufficient environmental adaptability, disconnected alarm response, inability of model self-evolution, and lack of trusted governance of the system, the present invention provides an intelligent Internet of Things public safety management and control system and method based on multi-source data fusion.

[0006] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0007] The intelligent IoT public safety management and control system based on multi-source data fusion includes:

[0008] The scenario adaptation configuration module is used to identify the current system operation scenario and load the scoring model parameter configuration file and response strategy template corresponding to the operation scenario;

[0009] Adaptive acquisition module, used to uniformly collect and preprocess multi-source data from physical sensors, video surveillance terminals, equipment log systems, environmental monitoring units, and network communication nodes to generate standardized data with a unified structure;

[0010] A semantic fusion module is used to extract entity, attribute and relationship information from the standardized data, convert them into semantic triples, and construct a unified semantic representation structure;

[0011] An edge anomaly processing module is configured to perform a local anomaly detection operation on the standardized data at an edge computing node and output anomaly output information, wherein the anomaly output information includes an anomaly identification mark, an anomaly score value, and characteristic indicator parameters associated with the abnormal event;

[0012] A security scoring module is configured to receive the unified semantic representation structure and the abnormal output information, construct a scoring model based on the scoring indicator system and preset rules, perform fusion analysis based on the scoring model parameter configuration file, and output a risk scoring result;

[0013] An intelligent alarm module is used to perform graded alarm judgment according to the risk scoring results and generate corresponding control response operation instructions based on the response strategy template;

[0014] Model optimization module, used to update scoring model parameters based on historical data and feedback information;

[0015] The trusted governance module is used to implement user identity authentication, operation behavior auditing and structured data export to achieve trusted management of the system operation process.

[0016] As a preferred solution of the present invention, the scene adaptation configuration module includes:

[0017] A scene recognition unit is used to identify the type of operating scene the current system is in through device category identification, deployment parameters or user tag information;

[0018] An indicator preparation loading unit is used to load a scoring model parameter configuration file corresponding to the identified operating scenario type according to the operating scenario type. The scoring model parameters include: the types of risk indicators involved in scoring, and the normalization function type, risk scoring threshold, initial weight and risk sensitivity coefficient corresponding to each risk indicator;

[0019] A policy template matching unit is used to automatically load a corresponding response policy template according to the operation scenario type, wherein the response policy template includes a control action sequence, signal channel parameters and communication protocol adaptation items;

[0020] The ontology structure adaptation unit is used to automatically load the ontology description file that conforms to the OWL or RDF semantic format according to the scenario definition, and update the semantic label and namespace configuration data.

[0021] As a preferred solution of the present invention, the adaptive acquisition module includes:

[0022] The protocol identification unit is used to identify the communication protocol type based on the field characteristics or initial data frame in the initial handshake message of the access data source, and call the corresponding parser from the preset protocol driver library to achieve compatible parsing of MQTT, Modbus, OPC-UA, HTTP and custom protocols;

[0023] The time synchronization unit is used to unify the timestamps of various data sources using the NTP protocol of the master node and perform linear interpolation on data whose sampling interval exceeds a preset threshold through a sliding time window;

[0024] The data structure conversion unit formats and encapsulates the collected image data, text data, and sensor numerical data. The image data uses the YOLOv7 model to extract image entity information, and the text data uses the BERT model to extract keywords, syntactic fragments, and context features. The sensor numerical data is normalized and annotated with signal type labels. The format-encapsulated data forms standardized data with a unified structure.

[0025] As a preferred solution of the present invention, the semantic fusion module includes:

[0026] The device state modeling unit is used to convert the sensor numerical data with the signal type label into the corresponding device state triplet based on the template matching method, including the device identification, attribute category and current value;

[0027] The text semantic extraction unit is used to call the preset large language model to extract semantic entities and identify relationships from keywords, syntactic fragments, and contextual features extracted from text data, and construct behavioral semantic triples, including action subjects, attribute changes, and state identification fields;

[0028] The image semantic extraction unit is used to input image entity information into the pre-trained object detection model, identify key devices or behavioral objects, and map the recognition results into entity nodes in the semantic graph based on the label mapping table to construct visual semantic triples, including visual entities, actions or locations, and spatial objects;

[0029] The knowledge graph reasoning unit is used to uniformly organize the device state triples, behavior semantic triples and visual semantic triples, construct a unified semantic representation structure, and perform entity relationship vector learning based on the graph embedding algorithm TransE to complete missing relationships, and only retain entity relationships with reasoning confidence higher than a set threshold for graph updates.

[0030] As a preferred solution of the present invention, the edge exception processing module includes:

[0031] The data screening unit is used to perform data quality detection on the received standardized data, including calculating the field missing rate, sampling frequency offset and mean fluctuation index, and eliminating data samples with quality below the set threshold;

[0032] An anomaly detection unit, configured to perform differential judgment based on the deviation relationship between the historical statistical mean and the current sampling value within a sliding time window, and filter out possible abnormal data in combination with a preset threshold; the anomaly detection unit further supports the deployment of a lightweight one-dimensional convolutional neural network model at the edge node to identify anomalies in the changing pattern of multivariate data, and the lightweight one-dimensional convolutional neural network model periodically synchronizes parameters through the central node;

[0033] The data annotation unit is used to calculate the anomaly confidence of the detected abnormal data and generate an anomaly score value based on the anomaly confidence value through a nonlinear attenuation function. The anomaly score value is used to mark the quality level of the abnormal data, including high quality, medium quality and low quality.

[0034] As a preferred embodiment of the present invention,

[0035] The security scoring module includes the following units and constructs a scoring model in the following order:

[0036] Weight adjustment unit, used to adjust the system's current environmental parameters Compared with the reference standard value Deviation, dynamically modify the initial weight of each risk indicator , output dynamic weight , the calculation formula is:

[0037] ;

[0038] in: For the The current observed value of an environmental parameter, For the Reference standard values ​​of environmental parameters; For the The environmental parameters The entity in The weight adjustment coefficient of each risk indicator; It is a hyperbolic tangent function, which is used to limit the environmental deviation to an interval and suppress abnormal fluctuations;

[0039] The indicator scoring unit is used to perform nonlinear normalization evaluation on the observed value of each risk indicator, combined with the dynamic weight , output the original score value , the calculation formula is:

[0040] ;

[0041] in: For the The entity in The current observation value of each risk indicator; 、 For the The entity in The upper and lower risk score thresholds for each risk indicator; is the risk sensitivity coefficient; is a very small constant to avoid the denominator being zero;

[0042] Anomaly correction unit, used to correct anomaly scores based on the output of edge anomaly processing module , the original score value Perform nonlinear suppression correction and output the final score value , the calculation formula is:

[0043] ;

[0044] in: For the The abnormal score value of each risk indicator, is the abnormal tolerance coefficient, which indicates the system's tolerance to the The tolerance level for abnormalities of each risk indicator; is the abnormal control weight coefficient, which is used to adjust the suppression strength;

[0045] Total score calculation unit, used to calculate the final score of each risk indicator By dynamic weight Perform weighted summation to generate the final fused risk score result , the formula is:

[0046] ;

[0047] in: is the total number of risk indicators.

[0048] As a preferred solution of the present invention, the intelligent alarm module includes:

[0049] A graded alarm judgment unit is used to judge the risk based on the risk scoring result. The alarm level is divided according to the set risk threshold, among which, when the risk score result When the risk score is less than the first threshold, the system enters the early warning state; when the risk score is less than the first threshold, the system enters the early warning state; Between the first threshold and the second threshold, and the risk score result When the corresponding abnormal score alarm limit is exceeded, the system enters the alarm state; when the risk score result When the value is not lower than the second threshold, the system enters an emergency state;

[0050] A policy matching unit is used to retrieve matching policy parameters from the response policy template based on the current alarm level, including condition-action response rules, control mode fields, and target device mapping parameters, for dynamically triggering system control policies;

[0051] The control instruction sending unit is used to generate a structured control response operation instruction based on the MQTT protocol or Modbus-TCP protocol, including the control type, device identification and action parameter control fields, and send the control response operation instruction to the target device. It determines whether the control response operation instruction is successfully executed based on the feedback information returned by the target device, and generates a control closed-loop record when the execution is successful.

[0052] As a preferred embodiment of the present invention,

[0053] The model optimization module includes:

[0054] Feature attribution unit, used to calculate the input parameters of the scoring model to the original scoring value based on the SHAP algorithm Contribution of

[0055] The incremental learning unit is used to locally update the following parameters of the scoring model by combining small batch data input and gradient update mechanism without re-training the whole model: dynamic weight ; Risk sensitivity coefficient ; Abnormal tolerance coefficient and abnormal control weight coefficient ;

[0056] The gradient update is based on the contribution results output by the feature attribution unit, and the parameters that have a significant impact on the score are adjusted first.

[0057] As a preferred solution of the present invention, the trusted governance module includes:

[0058] Identity authentication unit, used to authenticate access between users and modules based on access control policies and OAuth2.0 protocol;

[0059] Operation audit unit, used to record policy configuration changes, model updates, and control response operation instructions, and add timestamp identification and chain signature information to key operations;

[0060] The operation evidence storage unit is used to calculate the summary of the execution result of the control response operation instruction, generate a summary value based on the hash function, and store the summary result in the distributed ledger through the blockchain node;

[0061] The data export unit is used to provide a RESTful interface to export historical alarm data and policy adjustment records by time interval or event type, and attach corresponding structured evaluation fields and audit metadata.

[0062] The intelligent Internet of Things public safety management and control method based on multi-source data fusion includes the following steps:

[0063] Step 1: Identify the current system's operating scenario, load the scoring model parameter configuration file and response strategy template corresponding to the operating scenario, and provide parameter basis for scoring and control;

[0064] Step 2: Collect multi-source data from physical sensors, video surveillance terminals, device log systems, environmental monitoring units, and network communication nodes. Perform protocol parsing, time synchronization, and format encapsulation on the collected data to generate standardized data with a unified structure.

[0065] Step 3: Perform semantic processing on the standardized data to extract entity, attribute, and relationship information, construct semantic triples, and use the triples to build a unified semantic representation structure as the input of the scoring model;

[0066] Step 4: Input the standardized data into the edge anomaly processing module, perform local anomaly detection, and generate anomaly output information, which includes an anomaly identification mark, anomaly score value, and associated feature indicator parameters;

[0067] Step 5: Parse the scoring model parameter configuration file, extract the scoring indicators, normalization function type, risk threshold, indicator weight, and scoring sensitivity coefficient required for the current running scenario, and build the scoring model;

[0068] Step 6: Input the unified semantic representation structure and abnormal output information into the scoring model, perform fusion analysis, and output the risk score value of each indicator and the overall risk score result of the system;

[0069] Step 7: Based on the risk scoring results, a graded alarm judgment is made, and control response operation instructions are generated according to the response strategy template to achieve linkage execution of the preset control devices;

[0070] Step 8: Use the historical scoring data generated during the operation and the feedback information from manual confirmation to update the indicator parameters in the scoring model parameter configuration file;

[0071] Step 9: Record system operation behaviors, perform user identity authentication and behavior auditing, and export structured data for system management and external interface calls.

[0072] The beneficial effects of the present invention are as follows: through modular architecture design, the present invention opens up the full process path from scene recognition, data collection, semantic modeling, risk analysis, alarm response to behavior governance in the process of public safety management and control of the Internet of Things. The system can accurately load the corresponding scoring model parameters and response strategies according to the current operating environment, realizing scene-oriented adaptive deployment; by integrating a multi-source data acquisition mechanism, it solves the problem of standardized processing of heterogeneous device data; relying on the construction method of semantic triples and unified semantic structures, the system has the ability to understand the association of cross-modal data, providing semantic support for subsequent anomaly detection and risk assessment. After the introduction of the edge anomaly processing module, the system has the ability to identify and label anomalies at the field level, significantly improving the response agility to emergencies and the ability to control border security. In the risk assessment stage, the scoring mechanism designed by the present invention can integrate semantic entity states and abnormal feature indicators, and perform multi-factor fusion analysis on various risk indicators according to preset rules, and output risk scoring results with discrimination and robustness, effectively avoiding the problem that traditional scoring models are overly sensitive to mutation values ​​or single-point anomalies. The scoring results will drive the intelligent alarm module to automatically determine the current risk level, and link the loaded response strategy template to match the control plan to ensure that the system can implement corresponding control responses under different alarm levels, forming an analysis-driven active prevention and control path. During the operation of the system, it also has the ability to optimize the model based on feedback data, and can continuously update and correct the scoring parameters to adapt to environmental changes and data distribution drift. At the same time, the system integrates trusted governance capabilities such as identity authentication, operation auditing and data export to provide compliance and traceability guarantees for the entire management and control process. The management and control system provided by the present invention has significant improvements in integration capabilities, scoring accuracy, response efficiency and evolvability, and can effectively support multi-source perception, intelligent judgment and closed-loop response requirements in complex public safety scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort. Among them:

[0074] Figure 1Schematic diagram of the overall architecture of the system of the present invention;

[0075] Figure 2 This is a safety scoring flow chart in an embodiment of the present invention;

[0076] Figure 3 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0077] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of the present invention.

[0078] To facilitate understanding of the present invention, some of the terms used in this specification are defined below. Unless otherwise specified, those skilled in the art should understand that the meanings of these terms in the present invention are limited to the technical content described in this application:

[0079] Anomaly score: refers to the numerical value obtained by the system through quantitative evaluation of detected abnormal data, reflecting the degree of data anomaly. A higher value indicates a more serious anomaly.

[0080] Risk sensitivity coefficient: used to adjust the response intensity of the scoring function to indicator changes. The larger the value, the more sensitive the scoring function is to abnormal changes.

[0081] Control response operation instructions: refers to the control commands generated by the system based on the risk level, including control type, target device and action parameters, which are used to trigger specific control operations.

[0082] like Figure 1 FIG. 1 is an embodiment of the present invention, which provides an intelligent Internet of Things public safety management and control system based on multi-source data fusion, including:

[0083] (1) Scenario Adaptation Configuration Module

[0084] Used to identify the current system operation scenario and load the scoring model parameter configuration file and response strategy template corresponding to the operation scenario;

[0085] The scene adaptation configuration module includes:

[0086] A scene recognition unit is used to identify the type of operating scene the current system is in through device category identification, deployment parameters or user tag information;

[0087] An indicator preparation loading unit is used to load a scoring model parameter configuration file corresponding to the identified operating scenario type according to the operating scenario type. The scoring model parameters include: the types of risk indicators involved in scoring, and the normalization function type, risk scoring threshold, initial weight and risk sensitivity coefficient corresponding to each risk indicator;

[0088] A policy template matching unit is used to automatically load a corresponding response policy template according to the operation scenario type, wherein the response policy template includes a control action sequence, signal channel parameters and communication protocol adaptation items;

[0089] The ontology structure adaptation unit is used to automatically load the ontology description file that conforms to the OWL or RDF semantic format according to the scenario definition, and update the semantic label and namespace configuration data.

[0090] In an embodiment of the present invention, the scenario adaptation configuration module is used to automatically identify the current scenario and load the corresponding scoring model parameters and response strategy template when the system starts or the operating environment changes, thereby realizing dynamic adaptation of the model and control logic.

[0091] The system identifies the current operating scenario by collecting device identification, deployment parameters and user configuration information, and calls the corresponding model configuration file from the parameter library based on the scenario label, including the scoring indicator set, normalization function type, initial weight value and risk sensitivity coefficient, etc., to drive the scoring module to perform on-demand calculations.

[0092] The system then matches the response strategy template corresponding to the scenario, loading the control instruction sequence, communication parameters, and protocol settings to link the scoring results with the control response. To ensure semantic consistency, the system also calls the semantic ontology file associated with the scenario (in OWL or RDF format) to update the tag namespace and attribute definitions, supporting subsequent data semantic fusion.

[0093] This module has capabilities such as runtime perception, on-demand loading, and adaptive configuration, and can achieve efficient deployment and policy linkage in complex environments. Different from traditional static configuration solutions, it improves scoring accuracy and response adaptability.

[0094] (2) Adaptive acquisition module

[0095] It is used to uniformly collect and preprocess multi-source data from physical sensors, video surveillance terminals, equipment log systems, environmental monitoring units, and network communication nodes to generate standardized data with a unified structure;

[0096] The adaptive acquisition module includes:

[0097] The protocol identification unit is used to identify the communication protocol type based on the field characteristics or initial data frame in the initial handshake message of the access data source, and call the corresponding parser from the preset protocol driver library to achieve compatible parsing of MQTT, Modbus, OPC-UA, HTTP and custom protocols;

[0098] The time synchronization unit is used to unify the timestamps of various data sources using the NTP protocol of the master node and perform linear interpolation on data whose sampling interval exceeds a preset threshold through a sliding time window;

[0099] The data structure conversion unit formats and encapsulates the collected image data, text data, and sensor numerical data. The image data uses the YOLOv7 model to extract image entity information, and the text data uses the BERT model to extract keywords, syntactic fragments, and context features. The sensor numerical data is normalized and annotated with signal type labels. The format-encapsulated data forms standardized data with a unified structure.

[0100] In this embodiment of the present invention, an adaptive acquisition module is used to uniformly collect and preprocess multi-source data from various IoT devices. After the system receives data, it first automatically analyzes the communication protocol type through a protocol recognition mechanism and loads the corresponding decoding module, enabling compatible access to multiple standard and custom protocols.

[0101] Subsequently, the system unifies the timestamps of each data source through the time synchronization unit, and uses a sliding time window for alignment and interpolation to resolve the timing deviation caused by different sampling frequencies and delays.

[0102] Multimodal data is formatted and encapsulated in the structure conversion unit. Target information is extracted from image data, semantic triples are extracted from text data, and sensor data is labeled and normalized. Ultimately, a standardized data structure with time, semantic, and source identifiers is generated for direct use by subsequent modules.

[0103] Through this module, the system realizes heterogeneous protocol recognition, multi-source time alignment and semantic unified structure generation, providing standardized and time-consistent data support for semantic fusion and scoring analysis.

[0104] (3) Semantic fusion module

[0105] Used to extract entity, attribute and relationship information from the standardized data, convert them into semantic triples, and construct a unified semantic representation structure;

[0106] The semantic fusion module includes:

[0107] The device state modeling unit is used to convert the sensor numerical data with the signal type label into the corresponding device state triplet based on the template matching method, including the device identification, attribute category and current value;

[0108] The text semantic extraction unit is used to call the preset large language model to extract semantic entities and identify relationships from keywords, syntactic fragments, and contextual features extracted from text data, and construct behavioral semantic triples, including action subjects, attribute changes, and state identification fields;

[0109] The image semantic extraction unit is used to input image entity information into the pre-trained object detection model, identify key devices or behavioral objects, and map the recognition results into entity nodes in the semantic graph based on the label mapping table to construct visual semantic triples, including visual entities, actions or locations, and spatial objects;

[0110] The knowledge graph reasoning unit is used to uniformly organize the device state triples, behavior semantic triples and visual semantic triples, construct a unified semantic representation structure, and perform entity relationship vector learning based on the graph embedding algorithm TransE to complete missing relationships, and only retain entity relationships with reasoning confidence higher than a set threshold for graph updates.

[0111] In this embodiment, the semantic fusion module converts device identification, attributes, and status information in structured data into semantic triples and constructs a unified semantic representation structure. Based on template rules or model recognition, the system organizes data into a "subject-attribute-value" format and supports semantic mapping for multiple data types.

[0112] Text data is extracted through language models to identify entities and relationships, while image data is identified through detection models to identify key objects. Both are then converted into semantic nodes. The fusion results form a graph structure, and graph reasoning algorithms are used to complete entity relationships, enhancing semantic associations and cross-modal understanding capabilities.

[0113] (4) Edge exception handling module

[0114] Used to perform a local anomaly detection operation on the standardized data at the edge computing node, and output anomaly output information, wherein the anomaly output information includes an anomaly identification mark, an anomaly score value, and characteristic indicator parameters associated with the abnormal event;

[0115] The edge exception processing module includes:

[0116] The data screening unit is used to perform data quality detection on the received standardized data, including calculating the field missing rate, sampling frequency offset and mean fluctuation index, and eliminating data samples with quality below the set threshold;

[0117] An anomaly detection unit, configured to perform differential judgment based on the deviation relationship between the historical statistical mean and the current sampling value within a sliding time window, and filter out possible abnormal data in combination with a preset threshold; the anomaly detection unit further supports the deployment of a lightweight one-dimensional convolutional neural network model at the edge node to identify anomalies in the changing pattern of multivariate data, and the lightweight one-dimensional convolutional neural network model periodically synchronizes parameters through the central node;

[0118] The data annotation unit is used to calculate the anomaly confidence of the detected abnormal data and generate an anomaly score value based on the anomaly confidence value through a nonlinear attenuation function. The anomaly score value is used to mark the quality level of the abnormal data, including high quality, medium quality and low quality.

[0119] In this embodiment, the edge exception processing module is deployed at the edge node to perform real-time quality detection and exception identification on the standardized data at the data acquisition end, thereby reducing system delay and bandwidth usage.

[0120] The module first screens data to determine sampling integrity and fluctuation characteristics, eliminating substandard data samples. It then analyzes the deviation between historical and current values ​​using a sliding time window, combining this with threshold calculations to identify potential anomalies.

[0121] On this basis, the module calls upon a lightweight convolutional neural network model to analyze the changing patterns among multivariate data and further determine the type and characteristics of abnormal events. Model parameters can be periodically updated synchronously by the central node, improving the generalization and adaptability of the edge model.

[0122] The detected abnormal data will be calculated by the annotation unit to have a confidence score, and based on the degree of deviation and the confidence of the model judgment, it will be output as high, medium or low level for reference by the subsequent scoring module.

[0123] (5) Safety scoring module

[0124] The system is used to receive the unified semantic representation structure and the abnormal output information, build a scoring model based on the scoring model parameter configuration file according to the scoring indicator system and preset rules, perform fusion analysis, and output a risk scoring result;

[0125] In this embodiment of the present invention, the security scoring module implements a dynamic quantitative assessment of various risk indicators through a multi-stage scoring mechanism. The module uses standardized structured inputs and abnormal outputs as the basis, combined with scenario-specific parameter configuration files to perform scoring calculations.

[0126] like Figure 2 As shown, the security scoring module includes the following units and executes the scoring model construction in the following order:

[0127] Step S201: Weight adjustment unit

[0128] The weight of each indicator is dynamically adjusted based on the difference between the current value of each risk indicator and the reference standard value.

[0129] By calculating the adjustment factor and optimizing the weight of each indicator, the model can be adjusted in real time according to environmental changes.

[0130] In a specific embodiment, according to the system current environment state parameter Compared with the reference standard value Deviation, dynamically modify the initial weight of each risk indicator , output dynamic weight , the calculation formula is:

[0131] ;

[0132] in: For the The current observed value of the environmental parameter Reference standard values ​​of environmental parameters; The environmental parameters The adjustment coefficient of the weight of each indicator; It is a hyperbolic tangent function, which is used to limit the environmental deviation to an interval and suppress abnormal fluctuations;

[0133] Step S202: Index scoring unit

[0134] Frequency classification of dynamically adjusted risk indicators;

[0135] Each risk indicator is scored based on the adjusted weights. This process converts the data into a unified score, ensuring the final score is highly accurate and comparable.

[0136] In a specific embodiment, a nonlinear normalization evaluation is performed on the observed value of each risk indicator, combined with the dynamic weight , output the original score value , the calculation formula is:

[0137] ;

[0138] in: for The entity in The current observation value of each risk indicator; 、 For the Preset upper and lower thresholds for each indicator; is the risk sensitivity coefficient, which controls the sensitivity of the score to the threshold boundary; is a very small constant to avoid the denominator being zero;

[0139] Step S203: Abnormal correction unit

[0140] Identify and correct scoring biases caused by anomalies;

[0141] Through the anomaly correction model, the impact of extreme values ​​on the scoring results can be further suppressed, thereby improving the robustness of the scoring.

[0142] In a specific embodiment, based on the abnormality score value output by the edge abnormality processing module , the original score value Perform nonlinear suppression correction and output the final score value , the calculation formula is:

[0143] ;

[0144] in: For the The abnormal score value of each indicator, is the abnormal tolerance coefficient, which indicates the system's tolerance to the The tolerance level for abnormality of each indicator; is the abnormal control weight coefficient, which is used to adjust the suppression strength. Used to further strengthen the suppression of outliers;

[0145] Step S204: Total score calculation unit

[0146] Finally, the revised scores of each indicator are aggregated according to their weights to obtain the system's overall risk score. This score reflects the current overall risk status of the system and is an important basis for decision-making and response mechanisms.

[0147] In a specific embodiment, the final score of each indicator is By dynamic weight Weighted summation to generate the overall risk score of the system , the formula is:

[0148] ;

[0149] in: is the total number of risk indicators.

[0150] This embodiment significantly improves the accuracy and responsiveness of the IoT public safety management and control system by comprehensively considering multi-source data and varying environmental conditions, leveraging the collaborative work of multiple modules, including weight adjustment, indicator scoring, anomaly correction, and overall scoring. In particular, it offers greater intelligence and adaptability than existing technologies in terms of dynamic fusion and real-time monitoring of multi-source data. This embodiment enables real-time and accurate comprehensive assessment of potential safety hazards in the environment, and provides effective management and control strategies, effectively reducing the probability of public safety incidents and enhancing public safety levels.

[0151] (6) Intelligent alarm module

[0152] Used to perform graded alarm judgment according to the risk scoring results, and generate corresponding control response operation instructions based on the response strategy template;

[0153] The intelligent alarm module includes:

[0154] A graded alarm judgment unit is used to judge the risk based on the risk scoring result. The alarm level is divided according to the set risk threshold, among which, when the risk score result When the risk score is less than the first threshold, the system enters the early warning state; when the risk score is less than the first threshold, the system enters the early warning state; Between the first threshold and the second threshold, and the risk score result When the corresponding abnormal score alarm limit is exceeded, the system enters the alarm state; when the risk score result When the value is not lower than the second threshold, the system enters an emergency state;

[0155] A policy matching unit is used to retrieve matching policy parameters from the response policy template based on the current alarm level, including condition-action response rules, control mode fields, and target device mapping parameters, for dynamically triggering system control policies;

[0156] The control instruction sending unit is used to generate a structured control response operation instruction based on the MQTT protocol or Modbus-TCP protocol, including the control type, device identification and action parameter control fields, and send the control response operation instruction to the target device. It determines whether the control response operation instruction is successfully executed based on the feedback information returned by the target device, and generates a control closed-loop record when the execution is successful.

[0157] (7) Model optimization module

[0158] Used to update scoring model parameters based on historical data and feedback information;

[0159] The model optimization module includes:

[0160] Feature attribution unit, used to rank the feature importance output by the trained random forest model, or to calculate the input parameters of each scoring model based on the SHAP algorithm, including: the current observation value , risk sensitivity coefficient , preset initial weight The original score value The contribution of is used to identify the key input dimensions that lead to score changes;

[0161] An incremental learning unit, which is used to locally update the nonlinear function parameters and weight correction terms in the scoring model by combining mini-batch data input with a gradient update mechanism without retraining the entire model. The scoring model is a composite structure that performs normalized scoring on each indicator and incorporates a weight correction mechanism.

[0162] A policy adaptive correction unit is used to reconfigure the mapping relationship between the configured alarm level and response action based on the deviation trend of the key indicators involved in the risk scoring results;

[0163] The knowledge graph update unit is used to access newly added semantic triples through the semantic graph interface, perform entity injection operations and conflict triple screening, and update the connection relationship between semantic nodes.

[0164] (8) Trusted Governance Module

[0165] It is used to implement user identity authentication, operation behavior auditing and structured data export to achieve trusted management of the system operation process.

[0166] The trusted governance module includes:

[0167] Identity authentication unit, used to authenticate access between users and modules based on access control policies and OAuth2.0 protocol;

[0168] Operation audit unit, used to record policy configuration changes, model updates, and control response operation instructions, and add timestamp identification and chain signature information to key operations;

[0169] The operation evidence storage unit is used to calculate the summary of the execution result of the control response operation instruction, generate a summary value based on the hash function, and store the summary result in the distributed ledger through the blockchain node;

[0170] The data export unit is used to provide a RESTful interface to export historical alarm data and policy adjustment records by time interval or event type, and attach corresponding structured evaluation fields and audit metadata.

[0171] like Figure 3As shown, the intelligent IoT public security management and control system based on multi-source data fusion described in this paper proposes a comprehensive management and control method for implementing data-driven security situation identification, risk assessment, response control, and compliance management in actual deployment scenarios. This method can be applied to urban transportation, campus security, park management, smart industry, and other fields.

[0172] The method comprises the following steps:

[0173] Step 1: Identify the current system's operating scenario, load the scoring model parameter configuration file and response strategy template corresponding to the operating scenario, and provide parameter basis for scoring and control;

[0174] Step 2: Collect multi-source data from physical sensors, video surveillance terminals, device log systems, environmental monitoring units, and network communication nodes. Perform protocol parsing, time synchronization, and format encapsulation on the collected data to generate standardized data with a unified structure.

[0175] Step 3: Perform semantic processing on the standardized data to extract entity, attribute, and relationship information, construct semantic triples, and use the triples to build a unified semantic representation structure as the input of the scoring model;

[0176] Step 4: Input the standardized data into the edge anomaly processing module, perform local anomaly detection, and generate anomaly output information, which includes an anomaly identification mark, anomaly score value, and associated feature indicator parameters;

[0177] Step 5: Parse the scoring model parameter configuration file, extract the scoring indicators, normalization function type, risk threshold, indicator weight, and scoring sensitivity coefficient required for the current running scenario, and build the scoring model;

[0178] Step 6: Input the unified semantic representation structure and abnormal output information into the scoring model, perform fusion analysis, and output the risk score value of each indicator and the overall risk score result of the system;

[0179] Step 7: Based on the risk scoring results, a graded alarm judgment is made, and control response operation instructions are generated according to the response strategy template to achieve linkage execution of the preset control devices;

[0180] Step 8: Use the historical scoring data generated during the operation and the feedback information from manual confirmation to update the indicator parameters in the scoring model parameter configuration file;

[0181] Step 9: Record system operation behaviors, perform user identity authentication and behavior auditing, and export structured data for system management and external interface calls.

[0182] In summary, the present invention builds a modular and scalable system architecture around the entire process of "data standardization, semantic fusion, anomaly detection, risk scoring, intelligent response and model optimization", which has a high degree of scenario adaptability, evaluation accuracy and response coordination. By introducing a dynamic nonlinear scoring mechanism, environmental perception weight adjustment, anomaly robustness correction and ontology-driven semantic structure mapping, the present invention breaks through the technical bottleneck of existing public safety scoring models that rely on fixed weights, linear scoring and manual rule configuration, and significantly improves the system's ability to identify and respond to public safety risks in complex dynamic scenarios. The system structure of the present invention is clear, the parameters are adjustable, and the algorithm can be implemented. It is suitable for a variety of public safety application scenarios such as urban governance, traffic control, industrial sites, energy systems, etc., has good engineering feasibility and promotion value, and has significant social benefits and industrial application prospects.

[0183] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0184] Any process or method description in the flowchart or otherwise described herein can be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process. The scope of the preferred embodiments of the present application includes additional implementations in which the functions may be performed in a different order than shown or discussed, including in a substantially simultaneous manner or in a reverse order depending on the functions involved.

[0185] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. Intelligent Internet of Things public safety management and control system based on multi-source data fusion, characterized by: include: The scenario adaptation configuration module is used to identify the current system operation scenario and load the scoring model parameter configuration file and response strategy template corresponding to the operation scenario; Adaptive acquisition module, used to uniformly collect and preprocess multi-source data from physical sensors, video surveillance terminals, equipment log systems, environmental monitoring units, and network communication nodes to generate standardized data with a unified structure; A semantic fusion module is used to extract entity, attribute and relationship information from the standardized data, convert them into semantic triples, and construct a unified semantic representation structure; An edge anomaly processing module is configured to perform a local anomaly detection operation on the standardized data at an edge computing node and output anomaly output information, wherein the anomaly output information includes an anomaly identification mark, an anomaly score value, and characteristic indicator parameters associated with the abnormal event; A safety scoring module is configured to receive the unified semantic representation structure and the abnormal output information, and based on the scoring index system and preset rules, perform dynamic weight correction by comparing the deviation between the observed value of the environmental parameter and the corresponding reference standard value in conjunction with the weight adjustment unit, perform nonlinear normalization processing in conjunction with the index scoring unit, and then perform inhibitory correction on the score value through the abnormal correction unit to output the final risk scoring result; The intelligent alarm module is used to compare the risk score with multiple preset score thresholds, trigger the corresponding warning, alarm or emergency status based on the alarm level classification rules, match the control strategy parameters in the response strategy template, construct the control type and control operation instructions of the target device control field, and judge whether the control is successfully executed based on the feedback information of the target device; The model optimization module is used to calculate the contribution of the scoring model input parameters to the scoring results. Without retraining the entire model, it combines small batch input data with the gradient update mechanism to perform incremental updates on local model parameters, including dynamic weights, risk sensitivity coefficients, anomaly tolerance coefficients, and anomaly control weight coefficients. Parameters with significant influence are updated first based on the contribution results. Trusted governance module, used to implement user identity authentication, operation behavior auditing and structured data export.

2. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 1 is characterized in that: The scene adaptation configuration module includes: A scene recognition unit is used to identify the type of operating scene the current system is in through device category identification, deployment parameters or user tag information; An indicator preparation loading unit is used to load a scoring model parameter configuration file corresponding to the identified operating scenario type according to the operating scenario type. The scoring model parameters include: the types of risk indicators involved in scoring, and the normalization function type, risk scoring threshold, initial weight and risk sensitivity coefficient corresponding to each risk indicator; A policy template matching unit is used to automatically load a corresponding response policy template according to the operation scenario type, wherein the response policy template includes a control action sequence, signal channel parameters and communication protocol adaptation items; The ontology structure adaptation unit is used to automatically load the ontology description file that conforms to the OWL or RDF semantic format according to the scenario definition, and update the semantic label and namespace configuration data.

3. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 1 is characterized in that: The adaptive acquisition module includes: The protocol identification unit is used to identify the communication protocol type based on the field characteristics or initial data frame in the initial handshake message of the access data source, and call the corresponding parser from the preset protocol driver library to achieve compatible parsing of MQTT, Modbus, OPC-UA, HTTP and custom protocols; The time synchronization unit is used to unify the timestamps of various data sources using the NTP protocol of the master node and perform linear interpolation on data whose sampling interval exceeds a preset threshold through a sliding time window; The data structure conversion unit formats and encapsulates the collected image data, text data, and sensor numerical data. The image data uses the YOLOv7 model to extract image entity information, and the text data uses the BERT model to extract keywords, syntactic fragments, and context features. The sensor numerical data is normalized and annotated with signal type labels. The format-encapsulated data forms standardized data with a unified structure.

4. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 3 is characterized in that: The semantic fusion module includes: The device state modeling unit is used to convert the sensor numerical data with the signal type label into the corresponding device state triplet based on the template matching method, including the device identification, attribute category and current value; The text semantic extraction unit is used to call the preset large language model to extract semantic entities and identify relationships from keywords, syntactic fragments, and contextual features extracted from text data, and construct behavioral semantic triples, including action subjects, attribute changes, and state identification fields; The image semantic extraction unit is used to input image entity information into the pre-trained object detection model, identify key devices or behavioral objects, and map the recognition results into entity nodes in the semantic graph based on the label mapping table to construct visual semantic triples, including visual entities, actions or locations, and spatial objects; The knowledge graph reasoning unit is used to uniformly organize the device state triples, behavior semantic triples and visual semantic triples, construct a unified semantic representation structure, and perform entity relationship vector learning based on the graph embedding algorithm TransE to complete missing relationships, and only retain entity relationships with reasoning confidence higher than a set threshold for graph updates.

5. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 4 is characterized in that: The edge exception processing module includes: The data screening unit is used to perform data quality detection on the received standardized data, including calculating the field missing rate, sampling frequency offset and mean fluctuation index, and eliminating data samples with quality below the set threshold; An anomaly detection unit, configured to perform differential judgment based on the deviation relationship between the historical statistical mean and the current sampling value within a sliding time window, and filter out possible abnormal data in combination with a preset threshold; the anomaly detection unit further supports the deployment of a lightweight one-dimensional convolutional neural network model at the edge node to identify anomalies in the changing pattern of multivariate data, and the lightweight one-dimensional convolutional neural network model periodically synchronizes parameters through the central node; The data annotation unit is used to calculate the anomaly confidence of the detected abnormal data and generate an anomaly score value based on the anomaly confidence value through a nonlinear attenuation function. The anomaly score value is used to mark the quality level of the abnormal data, including high quality, medium quality and low quality.

6. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 5 is characterized in that: The security scoring module includes the following units and constructs a scoring model in the following order: Weight adjustment unit, used to adjust the system's current environmental parameters Compared with the reference standard value Deviation, dynamically modify the initial weight of each risk indicator , output dynamic weight , the calculation formula is: ； in: For the The current observed value of an environmental parameter, For the Reference standard values ​​of environmental parameters; For the The environmental parameters The entity in The weight adjustment coefficient of each risk indicator; It is a hyperbolic tangent function, which is used to limit the environmental deviation to an interval and suppress abnormal fluctuations; The indicator scoring unit is used to perform nonlinear normalization evaluation on the observed value of each risk indicator and output the original score value , the calculation formula is: ； in: For the The entity in The current observation value of each risk indicator; 、 For the The entity in The upper and lower risk score thresholds for each risk indicator; is the risk sensitivity coefficient; is a very small constant to avoid the denominator being zero; Anomaly correction unit, used to correct anomaly scores based on the output of edge anomaly processing module , the original score value Perform nonlinear suppression correction and output the final score value , the calculation formula is: ； in: For the The abnormal score value of each risk indicator, is the abnormal tolerance coefficient, which indicates the system's tolerance to the The tolerance level for abnormalities of each risk indicator; is the abnormal control weight coefficient, which is used to adjust the suppression strength; Total score calculation unit, used to calculate the final score of each risk indicator By dynamic weight Perform weighted summation to generate the final fused risk score result , the formula is: ； in: is the total number of risk indicators.

7. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 6 is characterized in that: The intelligent alarm module includes: A graded alarm judgment unit is used to judge the risk based on the risk scoring result. The alarm level is divided according to the set risk threshold, among which, when the risk score result When the risk score is less than the first threshold, the system enters the early warning state; when the risk score is less than the first threshold, the system enters the early warning state; Between the first threshold and the second threshold, and the risk score result When the corresponding abnormal score alarm limit is exceeded, the system enters the alarm state; when the risk score result When the value is not lower than the second threshold, the system enters an emergency state; A policy matching unit is used to retrieve matching policy parameters from the response policy template based on the current alarm level, including condition-action response rules, control mode fields, and target device mapping parameters, for dynamically triggering system control policies; The control instruction sending unit is used to generate a structured control response operation instruction based on the MQTT protocol or Modbus-TCP protocol, including the control type, device identification and action parameter control fields, and send the control response operation instruction to the target device. It determines whether the control response operation instruction is successfully executed based on the feedback information returned by the target device, and generates a control closed-loop record when the execution is successful.

8. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 6 is characterized in that: The model optimization module includes: Feature attribution unit, used to calculate the input parameters of the scoring model based on the SHAP algorithm to the original scoring value Contribution of The incremental learning unit is used to locally update the following parameters of the scoring model by combining small batch data input and gradient update mechanism without re-training the whole model: dynamic weight ; Risk sensitivity coefficient ; Abnormal tolerance coefficient and abnormal control weight coefficient ; The gradient update is based on the contribution results output by the feature attribution unit, and the parameters that have a significant impact on the score are adjusted first.

9. The intelligent Internet of Things public safety management and control system based on multi-source data fusion according to claim 1 is characterized in that: The trusted governance module includes: Identity authentication unit, used to authenticate access between users and modules based on access control policies and OAuth2.0 protocol; Operation audit unit, used to record policy configuration changes, model updates, and control response operation instructions, and add timestamp identification and chain signature information to key operations; The operation evidence storage unit is used to calculate the summary of the execution result of the control response operation instruction, generate a summary value based on the hash function, and store the summary result in the distributed ledger through the blockchain node; The data export unit is used to provide a RESTful interface to export historical alarm data and policy adjustment records by time interval or event type, and attach corresponding structured evaluation fields and audit metadata.

10. The method for controlling a public safety control system of an intelligent Internet of Things based on multi-source data fusion according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Identify the current system operation scenario, load the scoring model parameter configuration file and response strategy template corresponding to the operation scenario, and provide parameter basis for scoring and control; Step 2: Collect multi-source data from physical sensors, video surveillance terminals, device log systems, environmental monitoring units, and network communication nodes. Perform protocol parsing, time synchronization, and format encapsulation on the collected data to generate standardized data with a unified structure. Step 3: Perform semantic processing on the standardized data to extract entity, attribute, and relationship information, construct semantic triples, and use the triples to build a unified semantic representation structure as the input of the scoring model; Step 4: Input the standardized data into the edge anomaly processing module, perform local anomaly detection, and generate anomaly output information, which includes an anomaly identification mark, anomaly score value, and associated feature indicator parameters; Step 5: Parse the scoring model parameter configuration file, extract the scoring indicators, normalization function type, risk threshold, indicator weight, and scoring sensitivity coefficient required for the current running scenario, and build the scoring model; Step 6: Input the unified semantic representation structure and abnormal output information into the scoring model, perform fusion analysis, and output the risk score value of each indicator and the overall risk score result of the system; Step 7: Based on the risk scoring results, a graded alarm judgment is made, and control response operation instructions are generated according to the response strategy template to achieve linkage execution of the preset control devices; Step 8: Use the historical scoring data generated during the operation and the feedback information from manual confirmation to update the indicator parameters in the scoring model parameter configuration file; Step 9: Record system operation behaviors, perform user identity authentication and behavior auditing, and export structured data for system management and external interface calls.

Images