A method for generating a park safety risk warning and response plan

Abstract

This invention discloses a method for generating early warning and response plans for safety risks in industrial parks, relating to the field of knowledge graph technology. The method includes: collecting and preprocessing multi-source data; constructing a knowledge graph to generate a triple structure of nodes, edges, and attributes; integrating the triple structure and preprocessed multi-source data through path queries to obtain risk paths; evaluating the type, level, and probability of risk paths using a semantic reasoning rule base and a random forest model to generate a risk assessment report; analyzing the interaction effects of multiple risk paths using a graph neural network model to obtain interaction risk probabilities; outputting priority adjustment suggestions based on interaction risk probabilities, preprocessed multi-source data, and priority rules; allocating rescue resources; and generating an optimized plan. This invention, by constructing a knowledge graph and generating a triple structure of nodes, edges, and attributes, achieves accurate modeling of complex risk factors and their interrelationships within industrial parks.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, and in particular to a method for generating early warning and response plans for park security risks. Background Technology

[0002] With the integration of information and intelligent sensing technologies, park safety risk early warning has gradually developed into a multi-level linkage architecture. In terms of data collection, modern IoT achieves real-time acquisition of multi-source data through distributed nodes and effectively eliminates cross-source interference through data cleaning. In the field of knowledge modeling, the focus is on building a structured risk association network, integrating equipment topology relationships, safety specification documents, and historical event records, ultimately forming a risk knowledge base with semantic attributes. In the risk assessment stage, ensemble learning models are widely used for risk probability prediction, combined with path search algorithms to identify key risk paths. However, there are still limitations in the dynamic coupling analysis of risk paths. Most models adopt static association models, which make it difficult to fully consider the dynamic interaction between equipment, environment, and personnel factors, resulting in a weak real-time perception capability for the evolution of complex risks.

[0003] Existing technologies still face challenges in early warning of safety risks in industrial parks: the analysis of complex interactions between risk paths usually relies on traditional statistical methods or simple logical correlation models; for example, regression models are built to assess the correlation between risk factors, or machine learning algorithms such as decision trees are used for classification and prediction; these methods often only provide an understanding of linear relationships, but are difficult to capture complex nonlinear interactions and deep implicit patterns. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for generating early warning and response plans for park safety risks to solve the problem of analyzing the complex interactive effects between risk paths.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for generating early warning and response plans for safety risks in industrial parks. The method includes: collecting and preprocessing multi-source data, including environmental data, personnel activity data, and equipment status data; constructing a knowledge graph to generate a triple structure of nodes, edges, and attributes; integrating the triple structure and preprocessed multi-source data through path query to obtain risk paths; evaluating the type, level, and probability of risk paths using a semantic reasoning rule base and a random forest model to generate a risk assessment report; analyzing the interaction effects of multiple risk paths using a graph neural network model to obtain interaction risk probabilities; outputting priority adjustment suggestions based on interaction risk probabilities, preprocessed multi-source data, and priority rules; allocating rescue resources; generating an optimized plan; distributing the optimized plan to terminal devices through an emergency management platform; displaying the plan content using a visual interface; further optimizing the plan based on the semantic reasoning rule base; and generating an operation log.

[0008] As a preferred embodiment of the method for generating park safety risk early warning and response plans described in this invention, the construction of a knowledge graph to generate a triple structure of nodes, edges, and attributes refers to collecting risk information from historical accident data and safety guidelines, extracting nodes and edges from the risk information using natural language processing technology, determining the causal relationship weights of edges and node risk thresholds as attributes through statistical analysis, importing nodes, edges, and attributes into a graph database, verifying the accuracy of node risk thresholds and edge weights by comparing with historical accident data, and forming a structured knowledge representation that can be stored and queried.

[0009] As a preferred embodiment of the method for generating early warning and response plans for park safety risks according to the present invention, wherein: the node is an entity type;

[0010] The edges represent relationships between entity types;

[0011] The attributes are the node risk threshold and the causal weight of the edge.

[0012] As a preferred embodiment of the method for generating early warning and response plans for park safety risks according to the present invention, the entity types include risk type, environmental factors, equipment status, personnel density, resources, and spatial location.

[0013] The edges include causal relationships, spatial relationships, and state dependencies.

[0014] As a preferred embodiment of the method for generating park safety risk early warning and response plans described in this invention, the step of obtaining risk paths by integrating triplet structures and multi-source data through path query refers to activating entity types that meet the node risk thresholds based on the nodes, edges, attributes, and multi-source data of the knowledge graph, and querying the causal relationship paths connected by edges through the path query function of the graph database to identify risk paths.

[0015] As a preferred embodiment of the park safety risk early warning and response plan generation method described in this invention, the semantic reasoning rule base refers to generating reasoning rules based on the causal relationship of historical accident data and the safety threshold of safety criteria, comparing the numerical values ​​of multi-source data with the reasoning rules, and determining the type and level of risk paths;

[0016] The random forest model refers to processing numerical values ​​from multiple sources of data and attributes from knowledge graphs using multiple decision trees, statistically analyzing the decision tree results, and predicting the risk probability of risky paths using a voting method.

[0017] As a preferred embodiment of the method for generating park safety risk early warning and response plans described in this invention, the step of analyzing the interactive influence of multiple risk paths through a graph neural network model to obtain the interactive risk probability refers to mapping the risk paths into a graph structure, using node embedding and message passing mechanisms to quantify the interaction between multiple risk paths, and generating the interactive risk probability.

[0018] As a preferred embodiment of the method for generating park safety risk early warning and response plans according to the present invention, the priority rule is based on historical accident data and a priority threshold is set. When the probability of interaction risk exceeds the priority threshold, a high-risk interaction scenario is identified and the priority rule is triggered.

[0019] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the park security risk early warning and response plan generation method as described in the first aspect of the present invention.

[0020] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for generating park security risk early warning and response plans as described in the first aspect of the present invention.

[0021] The beneficial effects of this invention are as follows: By constructing a knowledge graph and generating a triple structure of nodes, edges, and attributes, this invention achieves accurate modeling of complex risk factors and their interrelationships within the park. It not only integrates historical accident data with real-time monitoring information but also accurately captures the causal relationships between different risk factors. By utilizing a semantic reasoning rule base combined with a random forest model to evaluate the risk type, level, and probability of these paths, it improves the accuracy and reliability of risk prediction, thereby generating a detailed risk assessment report. It not only overcomes the limitations of traditional single-dimensional analysis but also makes emergency response measures more reasonable, enabling dynamic adjustments to plans, optimization of resource allocation, and improved speed and efficiency in responding to emergencies, ultimately achieving the goal of reducing park safety risks and protecting the safety of personnel and property. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart for generating early warning and response plans for safety risks in the park.

[0024] Figure 2 The flowchart is for the core method processing.

[0025] Figure 3 Build flowcharts for knowledge graphs.

[0026] Figure 4 This is a flowchart for risk path analysis. Detailed Implementation

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0030] Reference Figures 1-4 This embodiment provides a method for generating early warning and response plans for park security risks, including the following steps:

[0031] S1. Collect data from multiple sources and preprocess it.

[0032] Furthermore, IoT devices are deployed in chemical warehouses in industrial parks (such as Area A). These IoT devices include infrared temperature sensors, methane gas detectors, 4K cameras, and industrial sensors to collect multi-source data, including environmental data, personnel activity data, and equipment status data.

[0033] Infrared temperature sensors collect temperature and humidity data from the environment every second, covering the top and floor of the warehouse storage units;

[0034] Methane gas detectors collect data on the concentration of combustible gases in the environment every second and are deployed at the ventilation openings and entrances of warehouse storage units;

[0035] The 4K camera collects data on the number of people and their location per second (through face detection and coordinate positioning) and the valve opening / closing status in the equipment status data (through image recognition).

[0036] Industrial sensors collect pump operating power and motor vibration frequency data from equipment status data every second, and are installed on the pumps and motors in the warehouse;

[0037] All IoT devices use the MQTT protocol to package environmental data, device status data, and personnel activity data into multi-source data and transmit it to the edge computing node, where it is temporarily stored in the memory of the edge computing node in JSON format.

[0038] Multi-source data is filtered through edge computing nodes. The filtering operations include deduplication and removal of invalid data.

[0039] Deduplication is achieved by comparing timestamps and data types from multiple sources to remove duplicate records.

[0040] Invalid data is discarded based on preset judgment thresholds. For example, records with temperatures exceeding -20°C to 100°C, humidity exceeding 0% to 100%, gas concentration exceeding 0% to 5%, pump power exceeding 0 to 50 kW, vibration frequency exceeding 0 to 100 Hz, valve open / close status not 0 or 1, number of people exceeding 0 to 100, and location records exceeding the area of ​​warehouse A (0 ≤ x ≤ 100 meters, 0 ≤ y ≤ 50 meters) are considered invalid and discarded directly. The preset judgment thresholds are set based on the Regulations on the Safety Management of Hazardous Chemicals.

[0041] The output is filtered multi-source data, which is stored in the local cache of the edge computing node.

[0042] Filtered multi-source data is obtained from the local cache of the edge computing node, and missing values ​​are filled in using the mean method;

[0043] The mean method examines each filtered multi-source data item by item to identify missing records where the numerical fields are empty or marked as invalid. For example, a temperature sensor may have no data recorded or a camera may have an empty number of people recorded.

[0044] For each missing record (filtered multi-source data), the mean method extracts normal records (non-empty and not marked as invalid) of the same type and location from the previous 5 minutes based on the multi-source data type (e.g., temperature) and location (e.g., warehouse A area). The mean is then used as the imputation value. For example, if a temperature sensor is missing temperature data for a period of time, and the temperature sensor at the same location generated 5 normal records in the previous 5 minutes (50°C, 51°C, 49°C, 50.5°C, and 50.2°C), the imputation value is 50.14°C. If there are no normal records in the previous 5 minutes, the numerical field is left empty.

[0045] Due to the discreteness of personnel activity data, the number of people and locations are not filled in. Missing records are left with empty numerical fields. For example, if a camera records no number of people, the numerical value is left empty.

[0046] The output is the padded multi-source data, which is stored in the local cache of the edge computing node.

[0047] A sliding window method with a 5-second time window was used to calculate moving averages for temperature, humidity, and gas concentration in the filled environmental data, and pump power and vibration frequency in the filled equipment status data. For example, gas concentration records were 0.11%, 0.12%, 0.13%, 0.10%, and 0.14%, with a moving average of 0.12%. The 5-second window was selected based on the second-by-second acquisition frequency of the filled multi-source data, the second-level noise characteristics of the environmental data and equipment status data, and the performance of the edge computing nodes. Experiments verified that 3-second, 5-second, and 10-second windows were tested, and the smoothing effect (percentage reduction in standard deviation) and calculation latency were compared, confirming that 5 seconds was the optimal window.

[0048] Due to the non-smoothness of dispersion, the original values ​​of the number of people and locations in the personnel activity data are retained.

[0049] The output is smoothed multi-source data, preserved in JSON format, containing time, location, data type, value, and unit, and stored in the local cache of the edge computing node.

[0050] Smoothed multi-source data is obtained from the local cache of edge computing nodes, and then formatted, time synchronized (time stamps are calibrated based on the NTP protocol), and data grouped (by location, such as warehouse A area and data type (temperature, gas concentration, etc.)) to generate standardized multi-source data;

[0051] The standardized multi-source data is uploaded to a cloud database (MongoDB) via a 5G network and stored as a JSON document containing time, location, data type, value, and unit.

[0052] S2. Construct a knowledge graph to generate triples of nodes, edges, and attributes. Integrate the triples structure and preprocessed multi-source data through path query to obtain risk paths. Use a semantic reasoning rule base and random forest model to evaluate the type, level, and probability of risk paths and generate a risk assessment report.

[0053] Furthermore, a knowledge graph is constructed based on the graph database (Neo4j), as follows:

[0054] Obtain historical accident data from the internal accident records of chemical warehouses in industrial parks;

[0055] Risk information is extracted from historical accident data and industry safety guidelines. This risk information includes fire (caused by high temperatures and flammable gases), explosion (gas leaks), equipment failure (pump overload), and personnel safety (excessive density) risks.

[0056] Nodes, edges, and attributes are extracted from risk information to construct a triplet structure for the knowledge graph, as follows:

[0057] Nodes represent entity types, including risk types (such as fire, explosion, and equipment failure), environmental factors (such as high temperature and excessively high concentration of combustible gases), equipment status (such as pump overload and abnormal vibration frequency), personnel density (e.g., exceeding the safety threshold, >30 people), resources (such as fire hydrants and emergency exits), and spatial location (such as warehouse A area and warehouse B area). Among these, environmental factors, equipment status, resources, and spatial location are extended content of the nodes, which are reasonably derived based on the triggering conditions of risk information, historical accident data, industry safety guidelines (such as regulations on the safety management of hazardous chemicals and fire protection guidelines for building design), and application scenario requirements. This extension is reasonable and is intended to support risk analysis and emergency response.

[0058] Edges represent relationships between entity types, including causal relationships (such as high temperature causing a fire and excessively high concentration of combustible gas causing an explosion), spatial relationships (such as warehouse A being adjacent to warehouse B), and state dependencies (such as pump failure accompanied by abnormal vibration frequency).

[0059] The attributes include the risk threshold of the node (e.g., the fire risk threshold for high temperature: temperature > 50°C, duration ≥ 5 minutes, set according to fire risk control guidelines for the chemical storage environment in warehouse A area to prevent fire caused by high temperature) and the causal relationship weight of the edge (e.g., the probability of a fire caused by high temperature is 0.8, based on historical accident data statistics).

[0060] Natural language processing technology is used to parse text data containing risk information and extract nodes and edges. For example, in parsing "high temperature caused fire" in historical accident reports, the nodes are extracted as "high temperature" and "fire", and the edge is "caused". By statistically analyzing the weight of causal relationships, attributes are extracted. For example, if 100 fire accidents are analyzed and 80 of them are related to high temperature, then the weight of high temperature causing fire is set to 0.8.

[0061] The extracted nodes, edges, and attributes are imported into a graph database to generate an initial knowledge graph. For example, the node "high temperature" is connected to the node "fire" through the edge "cause", and the edge attribute is "causal relationship with weight 0.8".

[0062] The verification process ensures the accuracy of the node risk thresholds and edge weights of the initial knowledge graph. The verification process is as follows: compare the node risk thresholds (e.g., temperature > 50°C) and edge weights (e.g., 0.8 for fires caused by high temperature) of the initial knowledge graph with historical accident data and calculate the percentage deviation.

[0063] For example, if the initial edge weight of the knowledge graph is 0.8, and the historical accident data statistics are 0.78, then the percentage deviation is expressed as:

[0064] (0.8-0.78) / 0.78×100%=2.56%;

[0065] Among them, only when the deviation percentage is less than 5% (based on industry error tolerance) can the node risk threshold and edge weight pass the verification.

[0066] After the verification process is completed, the output is the constructed knowledge graph, which is stored in the graph database.

[0067] Retrieve recent standardized multi-source data from a cloud database and load the knowledge graph into a graph database;

[0068] Nodes, edges, attributes, and risk thresholds are retrieved from the graph database. JSON structured fields (time, location, data type, value, and unit) are extracted from standardized multi-source data. Each JSON structured field is then matched against the entity type and risk threshold in the knowledge graph, as detailed below:

[0069] If the flammable gas concentration is ≥0.1% (based on chemical safe storage guidelines), the entity matching the environmental factor is: if the flammable gas concentration is too high, for example, the gas concentration is 0.12%, then the entity activating the environmental factor is;

[0070] If the pump power is ≥40 kW (based on 80% of the rated power in the equipment manual), match the entity for the equipment status: pump overload. For example, if the pump power is 42 kW, then activate the entity for the equipment status.

[0071] If the number of people is ≥30 and the spatial location is warehouse A (based on personnel evacuation criteria and warehouse area), the entity matching personnel density is: if the number of people is >30, for example, if the number of people is 35, then the entity matching personnel density is activated.

[0072] Using the path query function of the graph database, we can find the causal relationship between active entities triggered by matching JSON structured fields, forming potential risk paths. For example, "high temperature" connects to "fire" through the edge "trigger", or "excessive concentration of combustible gas" and "high temperature" together trigger "explosion", forming a risk path: high temperature + excessive concentration of combustible gas → explosion.

[0073] The risk path is evaluated using a semantic reasoning rule base, as detailed below:

[0074] Extract causal relationships from historical accident data of risk information, such as high temperature causing fire; obtain safety thresholds from chemical safety storage guidelines, such as temperature > 50°C and gas concentration ≥ 0.1%, based on the chemical characteristics and historical accident data of warehouse A area;

[0075] Based on causal relationships and security thresholds, and combined with the nodes, edges and attributes of the knowledge graph, inference rules are generated and stored in the semantic inference rule base.

[0076] The semantic reasoning rule library loads predefined reasoning rules;

[0077] Extract JSON structured fields from standardized multi-source data, such as temperature 51°C and gas concentration 0.12%;

[0078] Retrieve the numerical value from the JSON field, such as 51°C, and use it as the field value;

[0079] Compare the field values ​​with the conditions of the inference rule (e.g., temperature > 50°C);

[0080] By comparison, it can be confirmed whether a risk path has been triggered, such as "high temperature → fire";

[0081] If the field value meets the conditions, the conclusion of the inference rule is output, such as the risk type being fire and the level being high; if the conditions are not met, the conclusion of the inference rule is not output.

[0082] A random forest model is trained based on historical accident data to predict the risk probability of risky paths, as detailed below:

[0083] Extract JSON structured fields and field values ​​from standardized multi-source data, and obtain attributes from knowledge graphs;

[0084] Normalize the field values ​​of the JSON structured fields and the attributes of the knowledge graph, map the field values ​​and attributes to the [0, 1] interval, and generate risk feature vectors with unified dimensions to represent the conditions of risk paths;

[0085] Load the random forest model trained on historical accident data. The training process is as follows: Based on the random forest algorithm, use historical accident data to build multiple decision trees (e.g., 100 trees). Each tree obtains a subset from the historical accident data through random sampling with replacement. The subset includes features (temperature, gas concentration, and edge weights) and historical risk path labels (determined based on the event outcome of historical accident data, e.g., "high temperature → fire" is 1 if it occurs, and 0 if it does not occur). Use the subset to train the decision tree, generate node thresholds and branch paths, and form a decision tree structure to predict the binary classification result.

[0086] The risk feature vector is input into the random forest model, where each decision tree of the random forest model predicts whether a risk path will occur. By comparing the risk feature vector with the node threshold generated during training, a binary classification result is output (1 indicates occurrence, 0 indicates non-occurrence, for example, "high temperature → fire" is predicted as 1).

[0087] The binary classification results of each decision tree are statistically analyzed, and the proportion of risk occurrence is calculated by voting to generate the risk probability. The voting method is as follows: count the number of decision trees that predict 1 in multiple decision trees, divide by the total number of decision trees to obtain the risk probability. For example, if there are 100 decision trees in total, and 85 decision trees predict "high temperature → fire" as 1, then the risk probability is 85%.

[0088] The reasoning conclusions and risk probabilities are directly integrated into risk assessment data. The fields of the reasoning conclusions, the fields of the risk probabilities, and the fields of the standardized multi-source data are merged into a JSON document, which includes time, location, data type, value, unit, risk type, risk level, risk probability, priority, scope of impact, and risk path, and is stored in the local cache of the cloud computing node.

[0089] Risk assessment data is retrieved from the local cache of cloud computing nodes, and high-risk records are aggregated according to risk level (high, medium and low) and risk probability (e.g., <30% is low risk, 0%-60% is medium risk, and >60% is high risk) to generate aggregated results;

[0090] A risk assessment report is generated based on the aggregated results; the risk assessment report includes risk type, risk level, risk probability, priority, scope of impact, and risk path;

[0091] The risk assessment report is uploaded to a cloud database via a 5G network and stored as a JSON document.

[0092] S3. Analyze the interaction impact of multiple risk paths using a graph neural network model to obtain the interaction risk probability. Based on the interaction risk probability, preprocessed multi-source data, and priority rules, output priority adjustment suggestions, allocate rescue resources, and generate an optimized plan.

[0093] Furthermore, risk assessment reports are loaded from a cloud database to filter out high-risk records;

[0094] Based on historical accident data, risk contingency plan templates are generated and stored in a graph database as an extension of the knowledge graph.

[0095] The risk contingency plan template includes a command structure (e.g., emergency command team), resource allocation (e.g., fire trucks and fire extinguishers), evacuation routes (e.g., emergency exits on the north side of warehouse A area), and communication procedures (e.g., notifying the fire department).

[0096] The fields of high-risk records are compared with risk plan templates to match the applicable risk plan templates for high-risk situations. For example, for fire risk (such as a risk probability of 85%, which is high risk), the risk plan template of "high-level fire risk plan" is matched to generate initial response measures (such as activating fire trucks and evacuating personnel) and resource allocation (such as 2 fire trucks and 10 fire extinguishers).

[0097] Resource entities and spatial location entities are obtained from the knowledge graph to support resource allocation and evacuation route planning for risk contingency plan templates. For example, the fire truck is located "near the fire station (2 kilometers away from warehouse A)" and the emergency exit is located "north of warehouse A".

[0098] Extract multiple risk paths from the risk assessment report;

[0099] By using a pre-trained graph neural network model, the interactive impact of multiple risk paths is analyzed, and an optimized contingency plan is generated, as follows:

[0100] Multiple risk paths are extracted from the risk assessment report. The path query function of the knowledge graph is used to obtain the corresponding activation entities, adjacency relationships and edge weights from the knowledge graph. The risk paths are then mapped into a graph structure, where activation entities are nodes, adjacency relationships are edges, and edge weights are the attributes of the edges, thus forming a graph representation of the risk paths.

[0101] Node embedding is performed on the nodes in the graph representation, mapping each node to a low-dimensional vector to represent entity features. For example, "high temperature" is mapped to a low-dimensional vector to represent the feature of "high temperature". Then, through a message passing mechanism, information is exchanged using the low-dimensional vectors generated by node embedding. Each node receives the low-dimensional vectors of its neighbors and performs weighted aggregation based on edge weights. For example, the low-dimensional vector of "high temperature" is passed to "fire" through the edge "trigger" with an edge weight of 0.8. After receiving the vector, "fire" updates the low-dimensional vector of the node.

[0102] After multiple rounds of message passing, node information is aggregated, and the Sigmoid activation function is used to map the aggregated node information into interaction risk probabilities. The interaction risk probability is a quantification of the interaction impact of multiple risk paths, representing the probability that two risk paths will occur simultaneously or aggravate each other in the current scenario. For example, the interaction risk probability of "fire" and "explosion" is 90%, which means that the probability of the two occurring simultaneously or aggravating each other in warehouse A area is 90%.

[0103] Based on historical accident data, a priority threshold (e.g., 85%) is set. When the probability of interaction risk exceeds the priority threshold, a high-risk interaction scenario is identified, and priority rules are triggered. Priority rules include priority evacuation and priority allocation of rescue resources.

[0104] Based on priority rules and interaction risk probabilities, priority adjustment suggestions are generated. For example, if the interaction risk probability of "fire + explosion" is 90%, it meets the priority rules, and it is recommended to "prioritize the evacuation of explosion risk areas".

[0105] After acquiring multi-source data that has undergone real-time preprocessing, based on optimization rules, multi-source data, interaction risk probability, and priority adjustment suggestions, rescue resources are allocated to confirm priority types (such as priority evacuation) and an optimized plan is generated.

[0106] Using natural language generation technology, the optimized contingency plan is transformed into clear instructions, which include response measures (based on priority rule types, such as prioritizing the evacuation of explosion-risk areas), resource allocation (such as deploying fire trucks and explosion-proof equipment), priorities, map data (based on spatial location entities in a knowledge graph and standardized multi-source data to generate regional map information), and resource locations (based on real-time multi-source data of resource entities in a knowledge graph to record the location information of rescue resources); and the instructions are uploaded to a cloud database via 5G.

[0107] S4. Distribute the optimized contingency plan to terminal devices through the emergency management platform, display the content of the contingency plan using a visual interface, further optimize the contingency plan based on the semantic reasoning rule base, and generate operation logs.

[0108] Furthermore, the optimized contingency plan JSON document is extracted from the cloud database and converted into a response contingency plan document, retaining the instructions, resource allocation, priority, map data and resource location fields;

[0109] The emergency response plan document was distributed via 5G network to managers' mobile devices, fire department command equipment, and broadcast equipment in Warehouse Area A. The distribution process employed an encrypted communication protocol (AES-256) to ensure secure data transmission.

[0110] The manager's mobile device receives the response plan document, displays instructions (such as "prioritize evacuation of explosion risk areas") and resource allocation (such as 3 fire trucks and 10 emergency personnel), and supports the manager in confirming or adjusting, such as adjusting resource allocation;

[0111] The fire department's command equipment receives response plan documents, displays resource allocation and map data (such as warehouse area A and a 100-meter area), and supports the fire department in coordinating rescue operations;

[0112] The broadcasting equipment in Warehouse A receives instructions from the emergency response plan document and automatically plays voice instructions, such as "Please evacuate personnel within 100 meters east of Warehouse A along the north emergency exit." The voice broadcast frequency is once every 30 seconds for 5 minutes (based on emergency broadcasting guidelines).

[0113] The emergency management platform generates a visual interface based on contingency plan documents, displaying the risk location, impact range, resource deployment, and execution steps, as detailed below:

[0114] Risk location: Warehouse A, marked with the center coordinates of the chemical warehouse;

[0115] Impact range: 100-meter area, a circular area centered on warehouse A (based on map data from the response plan document).

[0116] Resource deployment: Fire truck locations (e.g., near a fire station, 2 km away) and emergency personnel locations (e.g., north entrance of warehouse A area) are marked with icons;

[0117] Execution steps: The sequence of command execution is displayed in timeline format. For example, start: broadcast evacuation order; 2 minutes later: fire truck departs; 5 minutes later: fire truck arrives;

[0118] The visual interface is deployed on the mobile devices of managers and the command equipment of the fire department, and supports real-time updates. For example, when the location of a fire truck changes, the visual interface updates its coordinates synchronously.

[0119] The visual interface integrates a semantic reasoning rule base and dynamically optimizes response plans based on real-time multi-source data, as detailed below:

[0120] Retrieve real-time multi-source data from cloud databases, extract JSON structured fields, call semantic reasoning rule library, input the JSON structured fields of real-time multi-source data and the fields of contingency plan documents into semantic reasoning rule library, generate optimization recommendations, such as "It is recommended to add 1 fire truck and adjust the resource allocation to 4 trucks";

[0121] The optimized recommendations are displayed on the visualization interface for managers to confirm or modify, such as confirming "add 1 fire truck" or adjusting it to "add 2 fire trucks"; the optimized plan document is stored in JSON format, uploaded to the cloud database, and redistributed to receiving terminals;

[0122] The emergency management platform records all operations related to distribution and display, generating operation logs that include the sending time, receiving terminal (manager's mobile device, fire department's command equipment, and broadcasting equipment in warehouse A area), adjustment records (e.g., the manager adjusts resource allocation, adding 2 emergency personnel, increasing the total from 10 to 12), and optimization recommendation records. The operation logs are stored in a cloud database in JSON format.

[0123] This embodiment also provides a computer device applicable to the method for generating park security risk early warning and response plans, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for generating park security risk early warning and response plans as proposed in the above embodiment.

[0124] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0125] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the method for generating park security risk early warning and response plans as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0126] In summary, this invention achieves accurate modeling of complex risk factors and their interrelationships within a park by constructing a knowledge graph and generating a triple structure of nodes, edges, and attributes. It not only integrates historical accident data with real-time monitoring information but also accurately captures the causal relationships between different risk factors. By utilizing a semantic reasoning rule base combined with a random forest model to evaluate the risk type, level, and probability of these paths, it improves the accuracy and reliability of risk prediction, thereby generating a detailed risk assessment report. It not only overcomes the limitations of traditional single-dimensional analysis but also makes emergency response measures more rational, enabling dynamic adjustments to plans, optimization of resource allocation, and improved speed and efficiency in responding to emergencies, ultimately achieving the goal of reducing park safety risks and protecting the safety of personnel and property.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating early warning and response plans for park safety risks, characterized in that: include, Collect and preprocess multi-source data, including environmental data, personnel activity data, and equipment status data; Construct a knowledge graph to generate triples of nodes, edges, and attributes. Integrate the triples structure and preprocessed multi-source data through path query to obtain risk paths. Use a semantic reasoning rule base and random forest model to evaluate the type, level, and probability of risk paths and generate a risk assessment report. By analyzing the interactive impact of multiple risk paths using a graph neural network model, the probability of interactive risks is obtained. Based on the probability of interactive risks, preprocessed multi-source data, and priority rules, priority adjustment suggestions are output, rescue resources are allocated, and an optimized plan is generated. The emergency management platform distributes and optimizes the contingency plan to terminal devices, displays the plan content using a visual interface, further optimizes the plan based on a semantic reasoning rule base, and generates operation logs. A random forest model is trained based on historical accident data to predict the risk probability of risky paths, as detailed below: Extract JSON structured fields and field values ​​from standardized multi-source data, and obtain attributes from knowledge graphs; Normalize the field values ​​of JSON structured fields and the attributes of knowledge graphs, mapping the field values ​​and attributes to the [0, 1] interval, and generate risk feature vectors with unified dimensions; Load the random forest model trained based on historical accident data. The training process is as follows: Based on the random forest algorithm, use historical accident data to build multiple decision trees. Each tree obtains a subset from the historical accident data through random sampling with replacement. The subset includes features and historical risk path labels. Use the subset to train the decision tree, generate node thresholds and branch paths, and form a decision tree structure. The risk feature vector is input into the random forest model. Each decision tree in the random forest model predicts whether a risk path will occur. By comparing the risk feature vector with the node threshold generated during training, a binary classification result is output. The binary classification results of each decision tree are statistically analyzed, and the proportion of risk occurrence is calculated using a voting method to generate the risk probability; The inference conclusions and risk probabilities are integrated into risk assessment data, and the fields of the inference conclusions, the fields of the risk probabilities, and the fields of the standardized multi-source data are merged into a JSON document and stored in the local cache of the cloud computing node. Risk assessment data is retrieved from the local cache of cloud computing nodes, high-risk records are aggregated according to risk level and risk probability, and aggregated results are generated. A risk assessment report is generated based on the aggregated results; the risk assessment report includes risk type, risk level, risk probability, priority, scope of impact, and risk path.

2. The method for generating park safety risk early warning and response plans as described in claim 1, characterized in that: The construction of a knowledge graph to generate a triple structure of nodes, edges, and attributes refers to collecting risk information from historical accident data and safety guidelines, extracting nodes and edges from the risk information using natural language processing technology, determining the causal weights of edges and node risk thresholds as attributes through statistical analysis, importing nodes, edges, and attributes into a graph database, verifying the accuracy of node risk thresholds and edge weights by comparing with historical accident data, and forming a structured knowledge representation that can be stored and queried.

3. The method for generating park safety risk early warning and response plans as described in claim 2, characterized in that: The node is an entity type; The edges represent relationships between entity types; The attributes are the node risk threshold and the causal weight of the edge.

4. The method for generating early warning and response plans for park safety risks as described in claim 3, characterized in that: The entity types include risk type, environmental factors, equipment status, personnel density, resources, and spatial location; The edges include causal relationships, spatial relationships, and state dependencies.

5. The method for generating park safety risk early warning and response plans as described in claim 1, characterized in that: The method of obtaining risk paths by integrating triplet structures and multi-source data through path query refers to activating entity types that meet the node risk threshold based on the nodes, edges, attributes, and multi-source data of the knowledge graph, and then querying the causal relationship paths connected by edges through the path query function of the graph database to identify risk paths.

6. The method for generating park safety risk early warning and response plans as described in claim 1, characterized in that: The semantic reasoning rule base refers to generating reasoning rules based on the causal relationships of historical accident data and the safety thresholds of safety criteria. The numerical values ​​of multi-source data are compared with the reasoning rules to determine the type and level of risk paths.

7. The method for generating park safety risk early warning and response plans as described in claim 1, characterized in that: The method of analyzing the interaction effects of multiple risk paths using a graph neural network model to obtain the interaction risk probability refers to mapping the risk paths into a graph structure, using node embedding and message passing mechanisms to quantify the interaction between multiple risk paths and generate the interaction risk probability.

8. The method for generating early warning and response plans for park safety risks as described in claim 1, characterized in that: The priority rule is based on historical accident data and sets a priority threshold. When the probability of interaction risk exceeds the priority threshold, a high-risk interaction scenario is identified and the priority rule is triggered.

Images