A fire-fighting auditing process optimization system based on a knowledge graph and AI diagnosis

Abstract

The present application relates to the technical field of wisdom fire and building engineering digital supervision, in particular to a fire inspection and verification process optimization system based on knowledge graph and AI diagnosis, comprising: a data receiving step: obtaining basic attributes such as building age, height and purpose; a theoretical graph generation step: constructing a theoretical compliance graph based on attribute filtering regulation meta-graph; a fact graph construction step: converting structured Internet of Things sensor data stream and unstructured field image data to generate field fact graph; a difference quantization diagnosis step: comparing graph differences and calculating compliance deviation; a decision closed-loop feedback step: outputting compliance signal or rectification path data packet according to deviation; the present application greatly improves the applicability of the fire inspection and verification system in urban renewal and stock building reconstruction projects.

Description

A fire safety inspection process optimization system based on knowledge graph and AI diagnosis Technical Field

[0001] This invention relates to the field of smart fire protection and digital supervision technology of building engineering, specifically a fire protection inspection process optimization system based on knowledge graphs and AI diagnosis. Background Technology

[0002] In the current environment of fire protection administrative review and compliance supervision, with the increasing complexity of building functions, the review process involves a large amount of heterogeneous data, including basic attributes such as building age, spatial geometry, IoT sensor data streams, and on-site inspection images.

[0003] To determine the compliance of this multi-source data, existing solutions generally rely on manual verification against textual regulations or static comparison based on a single data source. While such solutions are applicable to some conventional new construction projects, they often lead to deadlocks in complex scenarios such as the renovation of existing buildings due to logical incompatibilities caused by the transition between old and new regulations, where simply applying the new regulations often results in renovation deadlocks. Furthermore, the lack of cross-modal logical consistency verification between structured sensor values ​​and unstructured on-site images makes it difficult to effectively identify data paradoxes caused by falsified physical conditions or equipment malfunctions. In addition, existing evaluation systems often use discrete state identifiers with a veto system, failing to quantify the overall compliance deviation of the building or automatically generate rectification paths based on cost optimization principles. This results in a lack of quantitative basis for regulatory decisions, rigid review processes, and inefficient allocation of rectification resources. Therefore, how to achieve accurate mapping between multi-source heterogeneous data and complex regulatory logic, conduct differential quantitative diagnosis while ensuring data credibility, and output dynamically adapted compliance optimization decisions has become an urgent technical problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a fire safety inspection process optimization system based on knowledge graphs and AI diagnostics. Specifically, the technical solution of this invention includes:

[0005] A processor and a memory, the memory storing a computer program; the processor, when executing the program, performs the following steps:

[0006] Data receiving steps: Obtain the basic attribute parameters of the target building, including the building's age, building height, and usage classification code;

[0007] Theoretical graph generation steps: Call the preset regulatory meta-knowledge graph, filter and activate the entity nodes in the regulatory meta-knowledge graph based on the basic attribute parameters, and construct a theoretical compliance graph adapted to the target building;

[0008] The steps for constructing a fact map are as follows: receive structured IoT sensor data streams and unstructured field image data, use data mapping to convert the structured IoT sensor data streams into node attribute values, and use image recognition to convert the image data into entity state labels, thereby constructing a field fact map.

[0009] Difference quantification diagnostic steps: Perform subgraph isomorphic matching between the theoretical compliance map and the on-site fact map to identify differences, and calculate the compliance deviation value based on the preset difference type weight;

[0010] Decision-making closed-loop feedback steps: If the compliance deviation value is zero, output a compliance signal; if it is greater than zero, perform path search in the regulatory meta-knowledge graph, generate a rectification path data package that makes the deviation value zero, and output it.

[0011] Optionally, the theoretical map generation steps specifically include:

[0012] Traverse all rule nodes in the aforementioned regulatory meta-knowledge graph;

[0013] Determine whether the building age value of the target building is less than a preset existing building judgment threshold;

[0014] If the value is less than the existing building determination threshold, then the nodes in the regulatory meta-knowledge graph marked with renovation clauses will be set to active state, and the nodes marked with new standards and mutually exclusive with the renovation clauses will be set to inactive state.

[0015] If the value is greater than or equal to the existing building determination threshold, then the node in the regulatory meta-knowledge graph marked with a new standard is set to an active state.

[0016] Extract all active nodes and logical relationship edges to generate the theoretical compliance graph.

[0017] Optionally, the fact graph construction step includes cross-modal data consistency verification logic, specifically including:

[0018] Read the water pressure or flow rate values ​​from the structured IoT sensor data stream and define them as a first state feature;

[0019] The unstructured field image data is processed using an image recognition algorithm to identify the opening and closing state image features of the valve assembly and define them as second state features.

[0020] Query the preset physical linkage logic table to obtain the theoretical valve state corresponding to the first state feature;

[0021] Obtain the current system operating condition variables. If there is a linkage test mode, determine whether the second state characteristic is consistent with the theoretical valve state.

[0022] If there is a discrepancy, an abnormal conflict node will be generated in the on-site fact map, and the confidence attribute of the node will be marked as a suspected forgery state, while triggering an alarm signal.

[0023] If they match, the first state feature and the second state feature are fused and written into the corresponding entity node of the field fact map.

[0024] Optionally, in the difference quantification diagnostic step, the specific logic for calculating the compliance deviation value is as follows:

[0025] Establish a mapping relationship between entity nodes in the on-site fact map and corresponding nodes in the theoretical compliance map;

[0026] The missing edges or nodes in the aforementioned field fact map are defined as missing class differences;

[0027] Nodes in the field fact map that exist but whose attribute values ​​exceed the limits of the theoretical compliance map are defined as parameter-type difference items;

[0028] Query the difference type weight table to obtain the first weight coefficient corresponding to the missing class difference item and the second weight coefficient corresponding to the parameter class difference item;

[0029] Multiply the number of the missing class difference items by the first weight coefficient to obtain the first product, and multiply the number of the parameter class difference items by the second weight coefficient to obtain the second product;

[0030] Calculate the sum of the first product and the second product, and define the sum as the compliance deviation value.

[0031] Optionally, in the decision-making closed-loop feedback step, the specific logic for generating the rectification path data package is as follows:

[0032] Identify all target discrepancy nodes that cause the compliance deviation value to be greater than zero;

[0033] In the regulatory meta-knowledge graph, starting from the target difference node, a set of all candidate nodes that can replace the target difference node and satisfy compliance constraints are searched.

[0034] For each candidate node in the candidate node set, read the preset construction and renovation cost weight attribute value of that candidate node;

[0035] Select the candidate nodes with the smallest cumulative construction and renovation cost weight attribute value as the rectification plan;

[0036] The rectification plan is converted into a sequence containing specific equipment model replacement instructions or parameter adjustment instructions, and the rectification path data packet is generated.

[0037] Optional data structures for the regulatory meta-knowledge graph include:

[0038] The entity node layer stores fire protection facility names and spatial names parsed from fire protection laws and regulations texts;

[0039] The logical relationship layer stores directed edges connecting the entity nodes. Each directed edge has an attribute label, which clearly defines the logical dependency, spatial position relationship, or functional mutual exclusion relationship between the connected nodes.

[0040] The rule constraint layer stores a set of numerical ranges attached to the entity node, which is used to limit the range of legal attributes of the entity node.

[0041] Optionally, the image recognition algorithm specifically performs the following steps:

[0042] Receive the unstructured scene image data, perform grayscale processing and edge detection, and extract the geometric contour features of the image;

[0043] The geometric contour features are input into a pre-trained convolutional neural network model;

[0044] Output the valve handle angle value and component integrity probability value in the image;

[0045] The identification result is confirmed to be valid only when the component integrity probability value is greater than the preset confidence threshold; otherwise, a retake instruction is generated and returned to the data receiving end.

[0046] Optionally, the basic attribute parameters also include the fire compartment area value;

[0047] In the theoretical map generation step, the system obtains a preset standard value of the protection area of ​​a single detector, and calculates the minimum number of detectors required by dividing the fire compartment area value by the standard value of the protection area of ​​a single detector using the rounding logic.

[0048] In the theoretical compliance map, a corresponding number of detector occupant nodes are generated, and the inclusion relationship edges between the occupant nodes and the fire compartment nodes are established.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. This system effectively solves the logical mutual exclusion problem between old and new regulations in the renovation of existing buildings by constructing a dynamic graph pruning strategy based on the time dimension; the system intelligently identifies applicable clauses based on the building age, automatically activates appropriately relaxed renovation standards and shields conflicting new construction constraints, avoiding compliance deadlock caused by mechanically applying new regulations; this mechanism takes into account both principles and flexibility, greatly improving the applicability of the fire protection inspection system in urban renewal and existing building renovation projects;

[0051] 2. This system introduces cross-modal data consistency verification logic. By integrating structured IoT sensor data streams with on-site image features, it uses physical causality to identify data forgery and logical paradoxes. The system can automatically compare the logical matching degree between valve opening and closing status images and pipeline water pressure values. Once a conflict between the physical state and the sensor value is detected, an alarm is triggered. This mechanism effectively eliminates the risk of a single data source being easily forged or falsely reported, and significantly improves the credibility of the verified data and the ability to reconstruct the scene.

[0052] 3. This system establishes a multi-dimensional weighted compliance quantitative evaluation model, breaking through the limitations of the traditional veto system. By distinguishing between missing-class differences and parameter-class differences and assigning different weights, the system can calculate the compliance deviation value that reflects the overall safety risk exposure. This refined measurement method provides an objective mathematical basis for administrative penalty discretion and rectification period setting, helping regulatory authorities to accurately allocate enforcement resources based on risk levels and prioritize the handling of high-risk non-compliant projects.

[0053] 4. This system realizes a closed-loop rectification decision based on the principle of cost optimization, and can automatically generate executable repair paths. After identifying violations, the system uses a graph search algorithm to traverse alternative nodes that meet compliance constraints in the knowledge graph, and selects the optimal solution by combining the construction and renovation cost weights. This technology transforms passive violation notification into proactive solution recommendation, which greatly reduces the professional threshold and economic cost of rectification plan formulation and improves the service efficiency of the review process. Attached Figure Description

[0054] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0055] Figure 1 is a structural diagram of the system of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0057] Example 1:

[0058] Please refer to Figure 1. A fire safety inspection process optimization system based on knowledge graphs and AI diagnostics includes a processor and a memory, wherein the memory stores a computer program; when the processor executes the program, it performs the following steps:

[0059] Data receiving steps: Obtain the basic attribute parameters of the target building, including the building age, building height and usage classification code; at the same time, obtain the spatial geometric structure data of the building's interior through the BIM model interface or laser point cloud data interface, including the measured value of fire separation distance, the three-dimensional dimensions of evacuation routes and the spatial boundary box information of equipment installation points.

[0060] Theoretical graph generation steps: Call the preset regulatory meta-knowledge graph, filter and activate the entity nodes in the regulatory meta-knowledge graph based on the basic attribute parameters, and construct a theoretical compliance graph adapted to the target building;

[0061] The steps for constructing a fact map are as follows: receive structured IoT sensor data streams and unstructured field image data, use data mapping to convert the structured IoT sensor data streams into node attribute values, and use image recognition to convert the image data into entity state labels, thereby constructing a field fact map.

[0062] Difference quantification diagnostic steps: Perform subgraph isomorphic matching between the theoretical compliance map and the on-site fact map to identify differences, and calculate the compliance deviation value based on the preset difference type weight;

[0063] Decision-making closed-loop feedback steps: If the compliance deviation value is zero, output a compliance signal; if it is greater than zero, perform path search in the regulatory meta-knowledge graph, generate a rectification path data package that makes the deviation value zero, and output it.

[0064] This embodiment discloses a fire safety inspection process optimization system based on knowledge graphs and AI diagnostics. The system executes instructions in memory through a processor at the hardware level to achieve the following rigorous logical process:

[0065] In the data receiving step, the system receives the target building's basic attribute parameters in JSON format via a RESTful API interface; the processor executes the data logic, uniformly converting the building's age values ​​into YYYY format timestamp integers. Convert building height values ​​to floating-point values. The system then maps the usage classification codes to enumeration variables defined internally, for example, mapping commercial complexes to enumeration values. Establish a standardized input vector;

[0066] In the theoretical graph generation step, the system calls the regulatory meta-knowledge graph stored in the graph database; the processor executes the conditional traversal algorithm, using the input vector... , and To filter the rules, the system iterates through each rule node in the graph. For each node, the system checks the applicable scope field in its attributes. If this field contains the feature value of the input vector, the node is marked as active; otherwise, it is marked as dormant. The system extracts only all active nodes and their associated edges to construct a theoretical compliance graph for that specific building. ;

[0067] In the fact graph construction step, the system performs multi-source data fusion. For structured IoT sensor data streams, the system loads a pre-built device-to-fact graph mapping table, which contains... The system establishes a key-value pair mapping relationship; the processor parses the sensor data packets, looks up the mapping table based on the DeviceID, and directly writes the values ​​into the PropertyKey field of the corresponding NodeID; for unstructured field image data, the system calls a convolutional neural network model to output entity state labels, such as "valve: open," and attaches the label to the corresponding entity node, thereby constructing a field fact map. ;

[0068] In the differential quantification diagnostic step, the system employs an improved VF2 subgraph isomorphism algorithm to... Mapped to ;all Unable to map to The corresponding node's item, or the mapped attribute value, does not satisfy... Items within the constrained interval are stored in a set of differences; the system calls a weight table of difference types, such as a HashMap structure. Calculate the weighted compliance deviation value;

[0069] In the decision-making closed-loop feedback step, if the compliance deviation value is greater than zero, the system starts from the difference node and executes the A search algorithm in the regulatory meta-knowledge graph. The heuristic function of the search is set as the sum of the construction and modification cost weights of the nodes on the path, and the goal is to find a path that can replace the current difference node and satisfy the following conditions. The minimum cost path under all constraints is used to generate a data packet containing a specific sequence of rectification instructions, which is then sent to the user terminal.

[0070] Example 2:

[0071] The specific steps for generating the theoretical map include:

[0072] Traverse all rule nodes in the regulatory meta-knowledge graph;

[0073] Determine whether the building age of the target building is less than the preset threshold for existing buildings;

[0074] If the value is less than the threshold for existing buildings, nodes marked with renovation clauses in the regulatory meta-knowledge graph will be set to active status, and nodes marked with new standards and mutually exclusive with renovation clauses will be set to inactive status.

[0075] If the threshold for judging existing buildings is greater than or equal to the threshold for judging existing buildings, then the nodes in the regulatory meta-knowledge graph marked with the new standard are set to the active state; all active nodes and logical relationship edges are extracted to generate a theoretical compliance graph.

[0076] This embodiment further details the special processing logic for existing building renovation scenarios in the theoretical map generation step;

[0077] The system traverses all rule nodes in the legal meta-knowledge graph;

[0078] Determine the building's age using a comparator. Is it less than the preset threshold for existing buildings? For example, it can be set to 2000, or based on the effective date of a specific regulation;

[0079] When the judgment result is less than the existing building judgment threshold, that is When the system identifies the building as an existing building, in order to avoid the renovation being impossible due to the application of overly strict new regulations, the system executes a special graph activation strategy: nodes in the regulatory meta-knowledge graph marked with renovation clauses, such as the original stair width tolerance clause, are set to the active state.

[0080] Meanwhile, in order to eliminate logical conflicts, nodes marked with new standards and mutually exclusive with renovation clauses, such as mandatory standards for the width of staircases in new buildings, are set to an inactive state; conversely, if the threshold for existing building judgment is greater than or equal to the threshold for new standards, the system strictly follows the new standards and sets nodes marked with new standards to an active state. The system extracts all active nodes and logical relationship edges, and after topological consistency checks, generates a theoretical compliance graph.

[0081] This embodiment enables the system to handle the complex administrative issue of the transition between old and new regulations through logical branches in the time dimension; it ensures that the review standards do not violate the bottom line of the law, and conform to the actual solution path of historical legacy issues, which greatly improves the applicability and flexibility of the system in existing building renovation projects.

[0082] This embodiment details a time-based dynamic graph pruning strategy. Addressing the pain point of existing building renovation, the system uses building age as a logical gating signal to intelligently switch activation paths in the regulatory meta-knowledge graph. In response to buildings older than the existing building threshold, the system automatically blocks strongly constrained nodes of newly constructed buildings and activates renovation clause marker nodes with moderately relaxed attributes. This mechanism effectively solves the logical mutual exclusion problem between old and new regulations in the same physical space, avoiding the deadlock state of old buildings failing to comply with regulations due to mechanically applying new regulations. It provides a digital solution that combines principle and flexibility for fire safety inspections during urban renewal.

[0083] Example 3:

[0084] The fact graph construction steps include cross-modal data consistency verification logic, specifically including: reading water pressure or flow rate values ​​from structured IoT sensor data streams and defining them as first state features;

[0085] Unstructured field image data is processed using image recognition algorithms to identify the opening and closing state image features of valve components and define them as second state features. A preset physical linkage logic table is queried to obtain the theoretical valve state corresponding to the first state feature. The current system operating condition variables are obtained. If there is a linkage test mode, it is determined whether the second state feature is consistent with the theoretical valve state. If they are inconsistent, an abnormal conflict node is generated in the field fact map, and the confidence attribute of the node is marked as a suspected forgery state, while triggering an alarm signal. If they are consistent, the first state feature and the second state feature are merged and written into the corresponding entity node in the field fact map.

[0086] This embodiment elaborates on the cross-modal data consistency verification logic in the fact graph construction step, which aims to identify data fraud through physical causality.

[0087] The system reads numerical values ​​from structured IoT sensor data streams, such as the pressure values ​​of end-of-pipe water testing devices. Define it as the first state feature ;

[0088] Simultaneously, the system processes unstructured field image data; specifically, the system calls an image recognition algorithm to obtain the valve handle angle value. And apply a piecewise function to discretize the state: if Define the state as Open; if The state is defined as Closed; this discretized state is defined as the second state feature. ;

[0089] The system queries a preset physical linkage logic table; this table is stored as a rule base in key-value pair format, such as rules. The system is based on the current Matching rule conditions to obtain the mandatory theoretical valve state For example, the Open state; the system obtains the current operating status bit of the building's fire protection system by reading the ingress communication interface data of the fire alarm controller in real time, and introduces system operating context variables accordingly. Only when In the linked testing mode, a strong correlation derivation of the stress state is performed; if In static standby mode, the valve status is decoupled from the static pressure value of the pipeline network, and no forced theoretical valve status is output. This avoids logical misjudgments caused by the pressure maintaining pump.

[0090] The system performs a Boolean comparison operation: If the theoretical valve state is not obtained In a static interrupt mode, the comparison is skipped and a match is assumed; if a match is obtained... Then perform a Boolean comparison operation: ;like If the value is False, for example, the sensor shows pressure, but the image shows the valve is closed, the system determines that there is a physical paradox, generates an abnormal conflict node in the on-site fact map, and assigns the confidence attribute to it. Set to 0 and trigger an alarm; if If True, the system will Numerical and The tags are also written into the attribute fields of the graph nodes to complete the reliable fusion of data.

[0091] Example 4:

[0092] In the differential quantification diagnostic step, the specific logic for calculating the compliance deviation value is as follows:

[0093] Establish a mapping relationship between entity nodes in the field fact graph and corresponding nodes in the theoretical compliance graph; define missing edges or nodes in the field fact graph as missing-class differences; define nodes in the field fact graph that exist but whose attribute values ​​exceed the limits of the theoretical compliance graph as parameter-class differences.

[0094] Query the difference type weight table to obtain the first weight coefficient corresponding to the missing class difference item and the second weight coefficient corresponding to the parameter class difference item;

[0095] Multiply the number of missing class differences by the first weighting coefficient to obtain the first product;

[0096] The second product is obtained by multiplying the number of parameter class differences by the second weighting coefficient.

[0097] Calculate the sum of the first product and the second product, and define this sum as the compliance deviation value.

[0098] This embodiment details the calculation of compliance deviation values ​​in the differential quantification diagnostic process. The specific algorithm;

[0099] The system establishes a mapping relationship between entity nodes in the on-site factual graph and corresponding nodes in the theoretical compliance graph using a graph matching algorithm. During the comparison process, the system defines two types of differences: missing difference items. This refers to missing edges or nodes in the on-site factual map. For example, the theoretical map requires emergency lighting, but it is not found in the on-site map.

[0100] Parameter Class Differences This refers to nodes in the actual field map that exist but whose attribute values ​​exceed the limits of the theoretical compliance map. For example, the measured evacuation width is 0.9m, while the theoretical requirement is 1.1m.

[0101] The system queries the difference type weight table to assign weights to issues of different severity levels; it then retrieves the first weight coefficient corresponding to the missing difference item. For example, setting it to 10 represents a severe missing item, and the second weight coefficient corresponding to the parameter class difference item. For example, setting it to 5 indicates that the parameter does not meet the standard;

[0102] The calculation logic is as follows: Multiply the number of missing class differences by the first weighting coefficient to obtain the first product; multiply the number of parameter class differences by the second weighting coefficient to obtain the second product; calculate the sum of the first and second products, and define this sum as the compliance deviation value.

[0103] To reflect the varying degrees of criticality of different facilities to the safety system, a risk threshold coefficient is introduced. ,in Corresponding difference items, Corresponding parameter class differences; for example, critical components are set to 10.0, general components are set to 1.0, and parameters deviate from normalized values. The calculation logic for the normalized parameter value is as follows: In the formula These are the measured attribute values ​​of nodes in the on-site fact map. The standard attribute value or boundary threshold corresponding to the theoretical compliance map is calculated using the following formula:

[0104]

[0105] This embodiment establishes a weighted evaluation system; by distinguishing the severity of non-compliance and non-compliance, the system can more objectively reflect the overall fire safety level of the building, providing a quantitative basis for determining administrative penalties or rectification periods;

[0106] This embodiment constructs a multi-dimensional weighted quantitative evaluation model for compliance. Unlike the traditional veto system, this model distinguishes between missing difference items (functional deficiencies) and parameter difference items (performance indicator deviations), and assigns them first and second weight coefficients respectively, thus achieving a refined measurement of the degree of non-compliance. The calculated compliance deviation value is not only a simple status indicator, but also a continuous variable reflecting the overall safety risk exposure of the building. This value provides mathematical support for subsequent risk classification and control, enabling regulatory authorities to accurately allocate enforcement resources based on the magnitude of the deviation and prioritize the handling of high-risk, seriously non-compliant projects.

[0107] Example 5:

[0108] In the decision-making closed-loop feedback step, the specific logic for generating the rectification path data package is as follows:

[0109] Identify all target discrepancy nodes that cause the compliance deviation value to be greater than zero; in the regulatory meta-knowledge graph, starting from the target discrepancy node, search for all candidate nodes that can replace the target discrepancy node and satisfy the compliance constraints.

[0110] For each candidate node in the candidate node set, read the preset construction and renovation cost weight attribute value of that candidate node; select the group of candidate nodes with the smallest cumulative construction and renovation cost weight attribute value as the rectification plan;

[0111] The rectification plan is transformed into a sequence containing specific equipment model replacement instructions or parameter adjustment instructions, generating a rectification path data package.

[0112] This embodiment details the rectification path generation algorithm based on the principle of optimal cost;

[0113] The system iterates through the set of discrepancies, identifying all target discrepancy nodes that cause compliance deviation values ​​to be greater than zero, such as non-compliant Class B fire doors. To find alternative solutions, the system reads the functional classification code attribute of the target discrepancy node. For example, F_05: fire-resistant partition components;

[0114] The system performs index queries at the entity node layer of the regulatory meta-knowledge graph to retrieve all entities that have the same... However, nodes with different performance levels, such as Class A nodes, are filtered out, and nodes that are unusable under the current spatial constraints, such as devices that are too large to be installed, are filtered out, thus forming a set of candidate nodes.

[0115] The system performs cost evaluation on the candidate node set; for each candidate node in the set... The system reads its preset static attributes, including the construction and renovation cost weight attribute value. This value is based on the market average material and labor costs; specifically, the system pre-assesses and establishes a cost correction mapping table synchronized with the local engineering cost information database, converting equipment unit prices and standard working hours into standardized weighted values ​​to ensure the timeliness of cost calculations; the system constructs an objective function, where... For nodes Independent renovation costs, For nodes With nodes The engineering coupling cost between them is positive if there is a construction conflict, and negative if there is a collaborative construction advantage:

[0116]

[0117] in, This represents the set of candidate rectification paths consisting of candidate nodes. and For a specific rectification node in this set;

[0118] The algorithm uses dynamic planning to accumulate path costs, and combines this with a graph cut algorithm to handle mutually exclusive edge relationships. Pruning is performed on the conflict terms with positive infinity, so that the set of nodes that minimizes the total cost is selected as the optimal rectification solution from the candidate set.

[0119] The system maps the solution into natural language commands, such as replacing the Class B fire door at ID_xx with a Class A fire door of model ID_yy, generating a rectification path data packet and outputting it.

[0120] Example 6:

[0121] The data structure of the regulatory meta-knowledge graph includes:

[0122] The entity node layer stores fire protection facility names and spatial names parsed from fire protection laws and regulations texts;

[0123] The logical relationship layer stores directed edges that connect entity nodes. These directed edges have attribute labels, which clearly define the logical dependencies, spatial relationships, or functional exclusions between connected nodes.

[0124] The rule constraint layer stores a set of numerical ranges attached to entity nodes, which are used to limit the range of legal attributes of the entity node.

[0125] This embodiment specifically describes the data structure of the regulatory meta-knowledge graph that supports the above-mentioned reasoning functions; the graph adopts a multi-layer architecture design:

[0126] The entity node layer serves as the skeleton of the graph, storing fire protection facility and spatial terms parsed from fire protection laws and regulations, such as fire doors, evacuation routes, and smoke exhaust fans. Each node has a unique identifier (ID).

[0127] The logical relationship layer is the lifeblood of the graph, storing directed edges connecting entity nodes. These directed edges have explicit attribute labels used to define the complex relationships between nodes:

[0128] Logical dependency: For example, node A, i.e., the sprinkler pump, points to node B, i.e., the wet alarm valve, indicating that the function of A depends on B;

[0129] Spatial location relationship: For example, node A, i.e. the detector, is located in node B, i.e., inside the room;

[0130] Functional mutual exclusion relationship: For example, the mutual exclusion logic between node A, i.e., the natural smoke exhaust window, and node B, i.e., the mechanical pressurized air supply outlet, within the same smoke control zone;

[0131] The rule constraint layer is the criterion of the graph, storing a set of numerical ranges attached to entity nodes; for example, attaching the constraint condition Width>=1.1m to an evacuation corridor node is used to limit the range of legal attributes of the entity node.

[0132] This embodiment transforms unstructured natural language regulations into a directed graph structure that can be traversed and computed by computers, making the implicit logic between legal provisions explicit and supporting subsequent automated reasoning and path search.

[0133] This embodiment defines a layered graph architecture specifically for the digitization of fire protection regulations. By deconstructing the regulatory text into an entity node layer, a logical relationship layer, and a rule constraint layer, the system successfully transforms legal provisions described in natural language into a directed graph structure that can be understood by computers. In particular, the design of attribute tags in the logical relationship layer makes the complex logical dependencies and functional mutual exclusion relationships between fire protection facilities explicit, such as the mutual exclusion logic between mechanical smoke exhaust and natural smoke exhaust. This structured storage method overcomes the shortcomings of traditional relational databases in expressing complex network constraints, providing an efficient data index and query foundation for subsequent subgraph isomorphic matching and logical reasoning.

[0134] Example 7:

[0135] The image recognition algorithm specifically performs the following steps:

[0136] Receive unstructured field image data, perform grayscale processing and edge detection, and extract the geometric contour features of the image;

[0137] Input geometric contour features into a pre-trained convolutional neural network model;

[0138] Output the angle value of the valve handle and the component integrity probability value in the image; only when the component integrity probability value is greater than the preset confidence threshold is the recognition result confirmed to be valid, otherwise a retake instruction is generated and returned to the data receiving end.

[0139] This embodiment details the image recognition algorithm process for extracting entity state labels;

[0140] The system receives unstructured field image data in RGB format, performs grayscale processing using a weighted average method, calculates the image gradient using the Canny operator, extracts the edge lines of the valve assembly, and forms a binary geometric contour feature map.

[0141] The system inputs this FeatureMap into a pre-trained Convolutional Neural Network (CNN) model; this model uses ResNet-50 as its backbone network and is trained on a large number of samples labeled with valve angles and damage states; the model's output layer contains two neuron branches:

[0142] Regression branch: Outputs normalized valve handle angle values. The range is 0-90 degrees;

[0143] Classification branch: Outputs the component integrity probability value after Softmax normalization. The range is 0-1.

[0144] The system introduces quality gating logic: reads a preset confidence threshold. For example, set it to 0.85; the judgment logic is: if If the image is clear and its components are intact, then the angle value is output. For use in subsequent logic; if If the image is not clear, the recognition is deemed invalid, possibly due to occlusion or blurriness. The system generates a retake command signal and sends it back to the acquisition terminal.

[0145] Example 8:

[0146] Basic attribute parameters also include the area value of the fire compartment;

[0147] In the theoretical map generation step, the system obtains the preset standard value of the protection area of ​​a single detector, and calculates the minimum number of detectors required based on the result of dividing the fire compartment area value by the standard value of the protection area of ​​a single detector and using the rounding logic.

[0148] Generate a corresponding number of detector occupant nodes in the theoretical compliance graph, and establish containment relationships between the occupant nodes and the fire compartment nodes.

[0149] This embodiment supplements the automatic placement calculation logic for the specific facility of fire detectors;

[0150] The basic attribute parameters received by the system also include the area value of the fire compartment. ;

[0151] In the theoretical map generation step, the system retrieves the preset standard value of the protection area of ​​a single detector from the knowledge base. This value is dynamically obtained by looking up a table based on the detector type and room height, such as 60 square meters;

[0152] The system executes the following calculation logic: Based on the result of dividing the fire compartment area by the standard protection area of ​​a single detector, the minimum number of detectors required is calculated using the rounding logic CeilingFunction. : Based on this calculation result, the system generates the corresponding number in the theoretical compliance map, namely... Each detector occupancy node is identified, and an inclusion relationship is established between the occupancy node and the fire compartment node.

[0153] This embodiment transforms the computational clauses in the specification into a topological structure in the graph; this enables the system to automatically verify whether the number of detectors installed on-site meets the minimum standard: it only needs to count the number of actual detector nodes in the on-site factual graph and compare it with the number of placeholder nodes in the theoretical compliant graph;

[0154] This embodiment achieves automated mapping from numerical calculations to topological graphs. For the tedious calculation-based requirement of determining the number of fire detectors, the system automatically deduces the minimum required number of detectors by reading the area values ​​of fire compartments and the standard protection area of ​​a single detector. More importantly, the system dynamically generates corresponding detector occupant nodes in the theoretical compliance graph and establishes explicit inclusion edges. This approach concretizes abstract mathematical requirements into node existence constraints in the graph, eliminating the need for repeated calculations in subsequent compliance checks. Instead, it simplifies the verification logic for quantity-based compliance indicators by comparing the number of nodes in the actual on-site graph with those in the theoretical graph.

[0155] 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.

Claims

1. A fire safety inspection process optimization system based on knowledge graphs and AI diagnostics, characterized in that, Includes a processor and a memory, the memory storing a computer program; when the processor executes the program, it performs the following steps: Data receiving step: obtaining basic attribute parameters of the target building, the basic attribute parameters including building age, building height and usage classification code; Theoretical graph generation steps: Call the preset regulatory meta-knowledge graph, filter and activate the entity nodes in the regulatory meta-knowledge graph based on the basic attribute parameters, and construct a theoretical compliance graph adapted to the target building; The steps for constructing a fact map are as follows: receive structured IoT sensor data streams and unstructured field image data, use data mapping to convert the structured IoT sensor data streams into node attribute values, and use image recognition to convert the image data into entity state labels, thereby constructing a field fact map. Difference Quantification Diagnosis Step: Perform subgraph isomorphic matching between the theoretical compliance graph and the on-site fact graph to identify differences, and calculate the compliance deviation value based on the preset difference type weight; Decision Closed-Loop Feedback Step: If the compliance deviation value is zero, output a compliance signal; if it is greater than zero, perform path search in the regulatory meta-knowledge graph, generate a rectification path data package that makes the deviation value zero, and output it; The fact graph construction step includes cross-modal data consistency verification logic, specifically including: reading water pressure or flow rate values ​​from the structured IoT sensor data stream and defining them as a first state feature; processing the unstructured field image data using an image recognition algorithm to identify the opening and closing state image features of valve components and defining them as a second state feature; querying a preset physical linkage logic table to obtain the theoretical valve state corresponding to the first state feature; obtaining the current system operating condition variable, and if there is a linkage test mode, determining whether the second state feature is consistent with the theoretical valve state; if inconsistent, generating an abnormal conflict node in the field fact graph and marking the confidence attribute of the node as a suspected forgery state, while triggering an alarm signal; if consistent, merging the first state feature and the second state feature and writing them into the corresponding entity node of the field fact graph.

2. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, The theoretical graph generation steps specifically include: traversing all rule nodes in the regulatory meta-knowledge graph; determining whether the building age value of the target building is less than a preset existing building judgment threshold; if it is less than the existing building judgment threshold, setting the nodes in the regulatory meta-knowledge graph marked with renovation clauses to an active state, and setting the nodes marked with new standards and mutually exclusive with the renovation clauses to an inactive state; if it is greater than or equal to the existing building judgment threshold, setting the nodes in the regulatory meta-knowledge graph marked with new standards to an active state; extracting all active nodes and logical relationship edges to generate the theoretical compliance graph.

3. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, In the difference quantification diagnosis step, the specific logic for calculating the compliance deviation value is as follows: establish a mapping relationship between the entity nodes in the on-site fact map and the corresponding nodes in the theoretical compliance map; The missing edges or nodes in the aforementioned field fact map are defined as missing class differences; Nodes in the on-site factual map that exist but whose attribute values ​​exceed the limits of the theoretical compliance map are defined as parameter-type difference items; the difference type weight table is queried to obtain the first weight coefficient corresponding to the missing difference item and the second weight coefficient corresponding to the parameter-type difference item; the number of missing difference items is multiplied by the first weight coefficient to obtain a first product, and the number of parameter-type difference items is multiplied by the second weight coefficient to obtain a second product; the sum of the first product and the second product is calculated, and the sum is defined as the compliance deviation value.

4. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, In the decision-making closed-loop feedback step, the specific logic for generating the rectification path data package is as follows: Identify all target difference nodes that cause the compliance deviation value to be greater than zero; in the regulatory meta-knowledge graph, starting from the target difference node, search for a set of all candidate nodes that can replace the target difference node and satisfy compliance constraints; for each candidate node in the candidate node set, read the preset construction and modification cost weight attribute value of the candidate node; select the group of candidate nodes with the smallest cumulative construction and modification cost weight attribute value as the rectification plan; convert the rectification plan into a sequence containing specific equipment model replacement instructions or parameter adjustment instructions, and generate the rectification path data package.

5. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, The data structure of the regulatory meta-knowledge graph includes: an entity node layer, which stores fire protection facility names and spatial names parsed from fire protection laws and regulations; a logical relationship layer, which stores directed edges connecting the entity nodes, each directed edge having an attribute label that clearly defines the logical dependency, spatial location, or functional mutual exclusion relationship between connected nodes; and a rule constraint layer, which stores a set of numerical ranges attached to the entity nodes to limit the range of legal attributes of the entity node.

6. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, The image recognition algorithm specifically performs the following steps: receiving the unstructured on-site image data, performing grayscale processing and edge detection, and extracting the geometric contour features of the image; inputting the geometric contour features into a pre-trained convolutional neural network model; outputting the angle value of the valve handle and the component integrity probability value in the image; only when the component integrity probability value is greater than the preset confidence threshold, confirming the recognition result as valid, otherwise generating a retake instruction and returning it to the data receiving end.

7. The fire safety inspection process optimization system based on knowledge graph and AI diagnosis according to claim 1, characterized in that, The basic attribute parameters also include the area value of the fire compartment; in the theoretical map generation step, the system obtains the preset standard value of the protection area of ​​a single detector, and calculates the minimum number of detectors required by dividing the area value of the fire compartment by the standard value of the protection area of ​​a single detector using the rounding up logic. In the theoretical compliance map, a corresponding number of detector occupant nodes are generated, and the inclusion relationship edges between the occupant nodes and the fire compartment nodes are established.

Images