A fire-fighting facility safety evaluation optimization system and method based on big data

Abstract

The application discloses a fire-fighting facility safety evaluation optimization system and method based on big data, and belongs to the technical field of big data. The application collects image features through a contour detection algorithm, compares feature data of a fire hydrant interface and a water hose interface, and solves the adaptation; when the fire hydrant interface and the water hose interface are adapted, the appearance state is monitored, a water pipe network digital twin model is constructed, and an abnormal position is marked; a fire scene is simulated, the water pipe network distribution and the water source position are combined, and a preset hierarchical supply strategy is triggered; the abnormal position is combined and abstracted as a vertex and an edge of a graph structure, and a weighted directed graph model is constructed; when a certain section of water pipe is determined to be blocked, path search is performed in the graph structure, and a standby water supply path is screened; a reinforcement learning algorithm is used, control instructions are sent to intelligent valves on the standby water supply path, corresponding valves are opened in path order, irrelevant branches are closed, and a closed water supply channel from the water source to the target fire hydrant is formed.

Description

Technical Field

[0001] This invention relates to the field of big data technology, specifically a fire protection facility safety assessment and optimization system and method based on big data. Background Technology

[0002] Fire safety is a core aspect of building safety management. The rapid adaptation of fire hydrants and hoses and the stable water supply of fire water pipe networks are crucial for fire fighting. Existing fire protection systems rely on manual inspection and maintenance. When faced with complex building structures and sudden fires, problems such as interface mismatch, unclear water pipe network status, and blocked water supply paths often delay fire fighting opportunities. There is an urgent need to improve emergency response efficiency through intelligent technology.

[0003] Fire hydrant and hose connections come in various specifications, and current technology relies on firefighters manually identifying and matching them, which is prone to mismatches due to lack of experience or operational errors in emergency situations. Existing systems for monitoring fire hoses mostly use a combination of periodic manual inspections and local sensors, failing to achieve full network coverage. The appearance of key nodes relies on visual observation, which can easily lead to early-stage problems being missed due to insufficient inspection frequency. During a fire, if a section of pipe is deformed or damaged due to high temperatures, obstructing water flow, current technology typically relies on firefighters selecting alternative routes based on their on-site experience, lacking a global analysis of the entire pipe network topology. Summary of the Invention

[0004] The purpose of this invention is to provide a fire protection facility safety assessment and optimization system and method based on big data to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] Firstly, this application provides a method for optimizing the safety assessment of fire protection facilities based on big data, comprising the following steps:

[0007] Image features of fire hydrant interfaces are collected using a contour detection algorithm; feature data of fire hydrant interfaces and hose interfaces are compared, and adaptation solutions are obtained when there is a mismatch.

[0008] When fire hydrant interfaces and hose interfaces are compatible, the appearance of exposed parts of water pipes located at key nodes of the building's fire water pipe network is monitored; the appearance status is summarized to construct a digital twin model of the water pipe network, assess the pressure, flow rate, and unobstructedness of fire water in each area, and mark abnormal locations; based on the digital twin model of the water pipe network, a fire scenario is simulated; based on the simulated fire scenario, combined with the distribution of the water pipe network and the location of water sources, a preset graded replenishment strategy is triggered.

[0009] During the water supply process, in combination with abnormal locations, the key nodes in the fire water pipe network are abstracted as vertices in a graph structure, and the water pipe segments between nodes are abstracted as edges, thus constructing a weighted directed graph model.

[0010] When a section of water pipe is determined to be blocked, a path search is performed in the graph structure, with the available fire hydrants around the fire point as target nodes and the main water source and backup water source as starting nodes. The most efficient path is selected as the backup water supply path. Using a reinforcement learning algorithm, control commands are sent to the smart valves on the backup water supply path to open the corresponding valves in the path sequence and close irrelevant branches, forming a closed water supply channel from the water source to the target fire hydrant.

[0011] In conjunction with the first aspect, in a first embodiment of the first aspect of this application, the step of acquiring image features of the fire hydrant interface through a contour detection algorithm includes:

[0012] The fire hydrant interface is photographed from multiple angles to collect an image dataset containing the interface shape, diameter, thread type, number of clips, and distribution location; based on the image dataset, the image quality is optimized and interference factors are eliminated through image preprocessing algorithms;

[0013] A contour detection algorithm is used to perform a full contour scan of the image, extracting all closed contour regions. Screening conditions are set based on the physical characteristics of the fire hydrant interface to eliminate interfering contours that do not meet the characteristics. The regularity of the contour is judged by the ratio of the contour perimeter to the area, retaining the main contour of the interface with continuous edges. On the main contour of the interface, the edge curve is simplified using the Douglas-Pock algorithm, retaining feature points whose curvature changes meet the conditions, and calculating the core parameters of the interface, including: obtaining the diameter of the contour using the minimum circumcircle algorithm; locating the inner edge contour of the interface and calculating the inner diameter; identifying periodic contour fluctuations formed by threads or clips, and calculating the clip spacing by the distance between adjacent feature points; counting the number of clip-type feature points to determine the clip distribution characteristics of the interface; converting the calculated parameters into a unified unit, calibrating them against a preset image scale, and eliminating outliers caused by image blurring or occlusion, which are then used as image features of the fire hydrant interface.

[0014] In conjunction with the first aspect, in the second embodiment of the first aspect of this application, the step of comparing the feature data of the fire hydrant interface and the hose interface, and obtaining an adaptation solution when an adaptation deviation exists, includes:

[0015] For each feature parameter, an adaptation threshold range is set. According to the set comparison rules, each feature parameter of the fire hydrant interface and the water hose interface is compared item by item. When all parameters are within the threshold range, it is determined to be a complete fit. When there is a parameter that exceeds the threshold range, the specific value of the deviation parameter is recorded and the deviation type is marked. Based on the identified deviation parameters and deviation types, the preset solution database is searched, the matching solution is verified for adaptation, and the adapted solution is output.

[0016] In conjunction with the first aspect, in the third embodiment of the first aspect of this application, the monitoring of the appearance of the exposed portion of the water pipe located at a key node of the building's fire water pipe network when the fire hydrant interface and the water hose interface are compatible includes:

[0017] Key nodes in the building's fire protection water pipe network include branch pipe connections, valve interfaces, and exposed sections of floor risers. Camera shooting parameters are set, and a timed shooting cycle is established. The acquired images of the water pipe appearance are preprocessed. The Canny edge detection algorithm is used to extract features from the preprocessed images, specifically including: identifying color changes on the water pipe surface using the HSV color space model; calculating the texture consistency of the water pipe surface using a gray-level co-occurrence matrix to identify abnormal textures; locating protrusions, depressions, or deformed areas on the water pipe surface using the edge detection algorithm; converting the identified abnormal areas to their actual dimensions using a pixel scale; and evaluating the appearance status of the water pipes at each key node based on the extracted appearance feature parameters and preset judgment criteria.

[0018] In conjunction with the first aspect, in the fourth embodiment of the first aspect of this application, the step of summarizing various appearance states, constructing a digital twin model of the water pipe network, assessing the pressure, flow rate, and unobstructedness of fire-fighting water in each area, and marking abnormal locations includes:

[0019] Based on the fire water pipe network layout in the architectural CAD drawings, the physical structure of the water pipes is mapped, and pipe sections, valves and fire hydrants are transformed into basic components of the digital twin model; the topological relationships between components are defined according to the actual connection relationship, and a unique identifier is bound to each component, corresponding to the ID of the physical node, to construct a digital twin model of the water pipe network; each appearance state is imported into the digital twin model, and appearance state attributes are assigned to the corresponding components.

[0020] The digital twin model is divided into assessment units according to the building's functional zones. Each assessment unit is evaluated as follows: the average pressure value of all pipe sections within the unit is calculated and compared with the design pressure threshold for that area; if the pressure value is lower than the threshold, it is marked as a pressure anomaly. Based on the flow sensor data at each node, the balance between the total inflow and outflow within the unit is analyzed; if the difference exceeds a set threshold, it is marked as a flow anomaly. Combining the appearance, pressure change trend, and flow change trend, the actual pipe diameter is calculated through fluid dynamics simulation; if the diameter reduction rate is greater than a set threshold, it is marked as a blockage. For the assessed abnormal units, the specific abnormal nodes are traced through the topological relationship of the digital twin model, and the abnormal location is marked.

[0021] In conjunction with the first aspect, in the fifth embodiment of the first aspect of this application, the simulation of a fire scenario based on a digital twin model of a water pipe network includes:

[0022] Collect typical fire case data of different types of buildings, extract fire size, fire location characteristics and floor distribution patterns, and establish a fire parameter database; set trigger conditions for each fire type, classifying them into particularly serious fires, major fires, relatively serious fires and general fires; overlay building spatial layout information in the virtual space of the water pipe network digital twin model, mapping the scene variables in the fire parameter database to the corresponding positions in the water pipe network digital twin model; define the interaction logic between the fire scene and the water pipe network, and set the impact of fire smoke diffusion on sensors; start the fire simulation engine to simulate the change of fire intensity over time based on the set initial parameters.

[0023] In conjunction with the first aspect, in the sixth embodiment of the first aspect of this application, during the water supply process, considering abnormal locations, key nodes in the fire hydrant network are abstracted as vertices in a graph structure, and pipe segments between nodes are abstracted as edges, constructing a weighted directed graph model, including:

[0024] Based on abnormal locations and key nodes, each physical node is abstracted as a vertex in a graph structure, and a unique identifier is assigned to each vertex. The water pipe segment connecting two vertices is abstracted as a directed edge in the graph structure, with the edge direction consistent with the normal water flow direction. For each edge, the basic attributes of its corresponding water pipe segment are extracted as initial parameters, including physical attributes and state attributes. The physical attributes include the length, diameter, and material of the water pipe segment. The state attributes are combined with the abnormal location marker. When the water pipe segment is in a normal state, it is assigned the current pressure value and flow rate value. When the water pipe segment is in an abnormal state, the abnormality degree parameter and the corresponding correction coefficient are recorded.

[0025] A weight quantification formula is established based on the initial attribute parameters. The weight values ​​comprehensively reflect the water supply efficiency and reliability of this water pipe section. The calculation formula is as follows: Where R is the basic resistance coefficient, reflecting the water flow resistance caused by the length and diameter of the water pipe section. The larger the value, the greater the resistance. The calculation formula is: Where k is the resistance constant, L is the length of the water pipe segment, and D is the diameter of the water pipe segment; S is the state correction coefficient, which adjusts the weight according to the degree of anomaly of the water pipe segment, and the larger the value, the worse the state; P is the weight adjusted according to the importance of the area where the water pipe segment is located, and the smaller the value, the higher the priority; the weights of vertices, directed edges and edges are entered into the graph structure database to construct a weighted directed graph model.

[0026] In conjunction with the first aspect, in the seventh embodiment of the first aspect of this application, when a section of water pipe is determined to be obstructed, a path search is performed in the graph structure with available fire hydrants around the fire point as target nodes and the main water source and backup water source as starting nodes, to select the most efficient path as the backup water supply path, including:

[0027] When a sensor detects that the water flow in a certain section of the water pipe is obstructed, the digital twin model of the water pipe network is used to locate the graph structure edge corresponding to the obstructed water pipe section and mark it as a failed edge. Based on the location and topological relationship of the failed edge, the marking of related edges that may be affected by the cascading effect is expanded to determine the area in the graph structure that cannot participate in path search. The vertices corresponding to the main water source and the backup water source in the graph structure are included in the starting node set. When the path from the main water source to the obstructed area is partially failed, the backup water source node with normal status is retained first. With the fire point as the center, the available fire hydrants within the radius are delineated, and the vertex status corresponding to the available fire hydrants is confirmed through the digital twin model to form the target node set.

[0028] The Dijkstra algorithm is used to search for all feasible paths in the graph structure, starting from the set of initial nodes and ending at the set of target nodes. Feasible paths do not contain failed edges. For each path, its total weight and estimated water conveyance efficiency are calculated. Paths with total weights exceeding a threshold are removed, and candidate paths are retained. Candidate paths are sorted from highest to lowest estimated water conveyance efficiency, and the candidate path with the highest estimated water conveyance efficiency is selected as the backup water supply path.

[0029] In conjunction with the first aspect, in the eighth embodiment of the first aspect of this application, the step of using a reinforcement learning algorithm to send control commands to the smart valves on the backup water supply path, opening the corresponding valves in the path sequence, closing irrelevant branches, and forming a closed water supply channel from the water source to the target fire hydrant includes:

[0030] The valve control process of the backup water supply path is modeled as a Markov decision process. Environmental elements are defined, including: a state space containing the current state of all smart valves on the backup water supply path, real-time pressure of each pipeline segment, water flow velocity, and water demand of the target fire hydrant; an action space containing the set of operations for each smart valve, including opening, closing, and holding; and an objective function centered on forming a closed water supply channel as quickly as possible, with quantifiable indicators including total valve operation time, path water pressure compliance time, and irrelevant branch closure rate. A deep reinforcement learning algorithm is used to construct a neural network to fit Q, with the state space features as input and the Q-value for each action as output. Training samples are generated using historical fire simulation data and valve operation logs. The system trains a deep reinforcement learning model. After a fire occurs, it acquires the real-time status of the backup water supply path and inputs it into the trained deep reinforcement learning model as the initial state. The model outputs the Q-value of each possible action and selects the action with the largest Q-value as the current optimal operation. It generates a sequence of instructions according to the path: first, it opens the valve at the water source end, then opens the valves in the middle of the path in sequence, and finally closes all branch valves that are not related to the path, ensuring that the water flows along the preset path. It sends control instructions to the smart valves through the industrial bus, including the action type and execution time. When the instruction sequence is completed, it detects the water supply status at the target fire hydrant. When the stable water supply condition is reached, it determines that a closed water supply channel from the water source to the target fire hydrant has been formed.

[0031] Secondly, this application provides a fire protection facility safety assessment and optimization system based on big data, including:

[0032] The adaptation solution generation module includes an image feature acquisition unit and an adaptation solution generation unit. The image feature acquisition unit acquires image features of the fire hydrant interface through a contour detection algorithm. The adaptation solution generation unit compares the feature data of the fire hydrant interface and the hose interface, and obtains an adaptation solution when there is an adaptation deviation.

[0033] The abnormal location marking and supply strategy execution module includes: an appearance status monitoring unit, an abnormal location marking unit, a fire scenario simulation unit, and a supply strategy execution unit. Specifically, the appearance status monitoring unit monitors the appearance of exposed sections of water pipes located at key nodes of the building's fire water pipe network when the fire hydrant interface and hose interface are compatible; the abnormal location marking unit summarizes all appearance statuses, constructs a digital twin model of the water pipe network, assesses the pressure, flow rate, and unobstructedness of fire water in each area, and marks abnormal locations; the fire scenario simulation unit simulates fire scenarios based on the digital twin model of the water pipe network; and the supply strategy execution unit triggers preset tiered supply strategies based on the simulated fire scenario, combined with the water pipe network distribution and water source location.

[0034] Directed graph construction module: includes: directed graph construction unit; wherein, during the water supply process, the directed graph construction unit, combined with abnormal locations, abstracts the key nodes in the fire water pipe network as vertices in the graph structure, and the water pipe segments between nodes as edges, to construct a weighted directed graph model;

[0035] The closed water supply channel generation module includes a backup water supply path screening unit and a closed water supply channel generation unit. The backup water supply path screening unit, when a section of water pipe is determined to be obstructed, uses available fire hydrants near the fire point as target nodes and the main water source and backup water source as starting nodes. It performs a path search in the graph structure to select the most efficient path as the backup water supply path. The closed water supply channel generation unit uses a reinforcement learning algorithm to send control commands to the smart valves on the backup water supply path, opening the corresponding valves in the path sequence and closing irrelevant branches to form a closed water supply channel from the water source to the target fire hydrant.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] 1. This invention uses a machine vision contour detection algorithm to automatically collect image features of fire hydrant and hose interfaces and establish a standardized feature database. During the comparison process, the system can quickly identify adaptation deviations and retrieve the optimal solution from the preset solution library without manual judgment. Even when facing non-standard interfaces, it can match compatible solutions through feature parameters.

[0038] 2. This invention uses machine vision to monitor the appearance of exposed parts of key nodes, and combines pressure and flow sensor data to build a digital twin model that is completely synchronized with the physical pipeline network. The model can quantitatively assess the smoothness of each area and mark abnormal locations in real time.

[0039] 3. This invention abstracts the water pipe network into a weighted directed graph, with the weights comprehensively reflecting the pipe resistance, status, and regional priority. When a section of the water pipe is blocked, the backup path with the minimum total weight is selected by using a path search algorithm, with the fire hydrants around the fire point as the target and the water source as the starting point. Attached Figure Description

[0040] Figure 1 This is a schematic diagram illustrating the steps of a fire protection facility safety assessment and optimization method based on big data according to the present invention;

[0041] Figure 2 This is a system structure diagram of a fire protection facility safety assessment and optimization system based on big data, according to the present invention. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example: Figures 1-2 As shown, the present invention provides a technical solution:

[0044] like Figure 1 As shown, this application provides a method for safety assessment and optimization of fire protection facilities based on big data, including the following steps:

[0045] Step S100: Collect image features of the fire hydrant interface using a contour detection algorithm; compare the feature data of the fire hydrant interface and the hose interface, and obtain an adaptation solution when there is an adaptation deviation.

[0046] Specifically, the fire hydrant interface is photographed from multiple angles to collect an image dataset containing the interface shape, diameter, thread type, number of clips, and their distribution location; based on the image dataset, the image quality is optimized and interference factors are eliminated through image preprocessing algorithms.

[0047] A contour detection algorithm is used to perform a full contour scan of the image, extracting all closed contour regions. Screening conditions are set based on the physical characteristics of the fire hydrant interface to eliminate interfering contours that do not meet the characteristics. The regularity of the contour is judged by the ratio of the contour perimeter to the area, retaining the main contour of the interface with continuous edges. On the main contour of the interface, the edge curve is simplified using the Douglas-Pock algorithm, retaining feature points whose curvature changes meet the conditions, and calculating the core parameters of the interface, including: obtaining the diameter of the contour using the minimum circumcircle algorithm; locating the inner edge contour of the interface and calculating the inner diameter; identifying periodic contour fluctuations formed by threads or clips, and calculating the clip spacing by the distance between adjacent feature points; counting the number of clip-type feature points to determine the clip distribution characteristics of the interface; converting the calculated parameters into a unified unit, calibrating them against a preset image scale, and eliminating outliers caused by image blurring or occlusion, which are then used as image features of the fire hydrant interface.

[0048] Furthermore, an adaptation threshold range is set for each feature parameter. According to the set comparison rules, each feature parameter of the fire hydrant interface and the hose interface is compared item by item. When all parameters are within the threshold range, it is determined to be a complete fit. When there are parameters that exceed the threshold range, the specific value of the deviation parameter is recorded and the deviation type is marked. Based on the identified deviation parameters and deviation types, the preset solution database is searched, the matching solution is verified for adaptation, and the adapted solution is output.

[0049] In one specific embodiment, feature acquisition was performed on a certain type of DN65 fire hydrant interface (nominal diameter 65mm). A 20-megapixel industrial camera was used to take pictures at 8 angles (45° intervals) at a distance of 50cm from the interface, with an ambient light intensity of 500-800 lux.

[0050] Number of images acquired: 8, each with a resolution of 5120×3840 pixels, including the complete outline of the interface and details of threads and buckles; Interference factors before preprocessing: 3 images have slight reflections (reflective area accounts for 5%-8%), and 2 images have dust occlusion at the edges (occlusion area <3%); Effect after preprocessing: Through Gaussian filtering (filter kernel size 5×5) and edge enhancement algorithm, the standard deviation of gray value of reflective area is reduced from 45 to 12 before preprocessing, the dust occlusion area is automatically segmented and removed, and the image signal-to-noise ratio is improved to 38dB.

[0051] Number of closed contours extracted by full contour scanning: an average of 23 closed contours were extracted from a single image (including the main body of the interface, surface stains, background interference, etc.); Filtering conditions: the area threshold was set to 5000-20000 pixels (corresponding to an actual area of ​​10-40cm²), and the contour perimeter to area ratio threshold was 0.01-0.03 (regularity index); Results after filtering: 1-2 main body contours were retained per image, the interference contour removal rate reached 91%, and the edge continuity (pixel level) of the interface main body contour was ≥98%.

[0052] The core parameter calculation results are as follows: Interface outer diameter: Calculated using the minimum circumscribed circle algorithm, the measured values ​​at 8 angles are 64.8mm, 65.1mm, 64.9mm, 65.0mm, 64.7mm, 65.2mm, 64.9mm, and 65.0mm, with an average of 64.96mm, and a deviation from the nominal diameter ≤0.24mm; Interface inner diameter: Calculated after locating the inner ring edge, the measured value is 50.2-50.5mm, with an average of 50.3mm, which conforms to the inner diameter standard of DN65 interface (50mm±0.5mm); Buckle parameters: 6 evenly distributed buckle feature points were identified, and the measured distance between adjacent buckles is 31.4-31.6mm (theoretical value 31.5mm), with a distribution position deviation ≤0.1mm; Outlier removal: One image had an inner diameter measurement of 51.2mm due to local blurring (exceeding the threshold), which was judged as an outlier and removed, resulting in a final effective parameter retention rate of 87.5%.

[0053] The DN65 fire hydrant interface was compared with two types of hose interfaces (adapter type A, incompatible type B). The compatibility thresholds were set as follows: diameter deviation ±0.5mm, clip spacing deviation ±0.3mm, and the number of clips must be consistent. The comparison results with the adapter type A hose interface are as follows: Hose interface parameters: outer diameter 65.0mm, inner diameter 50.4mm, 6 clips, spacing 31.5mm; Item-by-item deviation: outer diameter deviation: 64.96mm-65.0mm=-0.04mm (within ±0.5mm); inner diameter deviation: 50.3mm-50.4mm=-0.1mm (within ±0.5mm); clip spacing deviation: 31.5mm (fire hydrant)-31.5mm (hose)=0mm (within ±0.3mm); Judgment result: fully compatible, output "can be directly connected" signal, comparison time 0.3 seconds. The comparison results with the incompatible type B hose connector are as follows: Hose connector parameters: outer diameter 63.0mm, inner diameter 48.0mm, 5 clips, spacing 33.0mm; Deviation identification: Diameter deviation: 64.96mm-63.0mm=1.96mm (exceeding ±0.5mm threshold); Clip quantity deviation: 6-5=1 (does not meet the quantity consistency condition); Clip spacing deviation: 31.5mm-33.0mm=-1.5mm (exceeding ±0.3mm threshold); Solution matching: The system retrieved "DN65 to DN63 reducer (with adapter clips)", and the simulated adaptation verification showed that the diameter deviation decreased to 0.1mm after connection, and the clip adaptation rate was 100%; Output result: The recommended adapter model is KT-65 / 63, with an installation step diagram. The solution generation time was 0.45 seconds.

[0054] Step S200: When the fire hydrant interface and the hose interface are compatible, monitor the appearance of the exposed parts of the water pipes located at key nodes of the building's fire water pipe network; summarize the appearance conditions, construct a digital twin model of the water pipe network, assess the pressure, flow and unobstructedness of fire water in each area, and mark abnormal locations; simulate fire scenarios based on the digital twin model of the water pipe network; based on the simulated fire scenarios, combined with the distribution of the water pipe network and the location of water sources, trigger the preset graded supply strategy;

[0055] Specifically, key nodes in the building's fire protection water pipe network include branch pipe connections, valve interfaces, and exposed sections of floor risers. Camera shooting parameters are set, and a timed shooting cycle is established. The acquired images of the water pipe appearance are preprocessed. The Canny edge detection algorithm is used to extract features from the preprocessed images, including: identifying color changes on the water pipe surface using the HSV color space model; calculating the texture consistency of the water pipe surface using a gray-level co-occurrence matrix to identify abnormal textures; locating protrusions, depressions, or deformed areas on the water pipe surface using the edge detection algorithm; converting the identified abnormal areas to their actual dimensions using a pixel scale; and evaluating the appearance status of the water pipes at each key node based on the extracted appearance feature parameters and preset judgment criteria.

[0056] Furthermore, based on the fire water pipe network layout in the architectural CAD drawings, the physical structure of the water pipes is mapped, and pipe sections, valves, and fire hydrants are transformed into basic components of the digital twin model; the topological association between components is defined according to the actual connection relationship, and a unique identifier is bound to each component, corresponding to the ID of the physical node, to construct a digital twin model of the water pipe network; each appearance state is imported into the digital twin model, and appearance state attributes are assigned to the corresponding components.

[0057] The digital twin model is divided into assessment units according to the building's functional zones. Each assessment unit is evaluated as follows: the average pressure value of all pipe sections within the unit is calculated and compared with the design pressure threshold for that area; if the pressure value is lower than the threshold, it is marked as a pressure anomaly. Based on the flow sensor data at each node, the balance between the total inflow and outflow within the unit is analyzed; if the difference exceeds a set threshold, it is marked as a flow anomaly. Combining the appearance, pressure change trend, and flow change trend, the actual pipe diameter is calculated through fluid dynamics simulation; if the diameter reduction rate is greater than a set threshold, it is marked as a blockage. For the assessed abnormal units, the specific abnormal nodes are traced through the topological relationship of the digital twin model, and the abnormal location is marked.

[0058] Furthermore, typical fire case data of different types of buildings are collected to extract fire size, ignition location characteristics, and floor distribution patterns, and a fire parameter database is established. Trigger conditions are set for each type of fire, classifying them into particularly serious fires, major fires, relatively serious fires, and general fires. In the virtual space of the digital twin model of the water pipe network, building spatial layout information is overlaid, mapping the scene variables in the fire parameter database to the corresponding positions in the digital twin model of the water pipe network. The interaction logic between the fire scene and the water pipe network is defined, and the impact of fire smoke diffusion on sensors is set. The fire simulation engine is started to simulate the change of fire intensity over time based on the set initial parameters.

[0059] In one specific embodiment, the fire water pipe network of a 10-story office building is taken as the monitoring object. The key nodes include 20 branch pipe connections, 15 valve interfaces and 10 exposed sections of floor risers. A 16-megapixel industrial camera is used, and the shooting parameters are set as shutter speed 1 / 200s, ISO400, timed shooting cycle once every 10 minutes, and ambient light intensity 300-1000 lux.

[0060] Image preprocessing and feature extraction were performed. Rust area ratio <5% was classified as "slight corrosion", 5%-10% as "moderate corrosion"; crack length >5mm was classified as "severe defect"; among 20 branch pipe connections, 1 was classified as "moderate corrosion" (rust area 8%), and the remaining 19 were "normal"; among 15 valve interfaces, 2 had "slight protrusions" (height <3mm); among 10 exposed sections of floor risers, 1 was detected as "slight crack" (length 3mm), which was judged as "slight abnormality".

[0061] A digital twin model of the water pipe network was constructed based on the office building's CAD drawings, including a total pipe length of 1200m and pipe diameters of DN100-DN150, linked to the appearance status data of each node and real-time data from 100 pressure / flow sensors. The building was divided into 10 assessment units (floors 1-10). The average pressure in the 8th-floor unit was 0.28MPa, lower than the design threshold of 0.3MPa, marked as "pressure anomaly," traced back to insufficient pressure in the 8th-floor booster pump (ID:P8). The 5th-floor unit had a total inflow of 2.8L / s and an outflow of 2.5L / s, a difference of 10.7% > 5%, marked as "flow anomaly," and combined with the appearance status, a slight leak was identified at the 5th-floor branch pipe interface (ID:B5-3). The 3rd-floor riser had a diameter reduction rate of 35% > 30% due to "moderate corrosion," marked as "obstructed flow." In the digital twin model, the 8th-floor booster pump, the 5th-floor branch pipe interface, and the 3rd-floor riser were highlighted in red, with a marking accuracy of 100%.

[0062] A fire scenario in a four-story office area (coordinates X=25m, Y=15m) was simulated in a digital twin model, with initial conditions set based on a fire parameter library. Simulation parameters were set: a general fire triggers "local replenishment" (activating the water source on this floor), while a larger fire triggers "area replenishment" (linking water sources on adjacent floors). When the fire reaches the "larger fire" threshold after 5 minutes, the system automatically triggers the "area replenishment" strategy. This instructs the main water source (underground reservoir) to increase its supply pressure to 0.5MPa, opens the backup pipeline valves on floors 3-5, and closes 50% of the branch flow in non-fire areas (floors 1-2 and 6-10), ensuring the total flow in the fire area increases from 2.0L / s to 3.5L / s to meet firefighting requirements.

[0063] Step S300: During the water supply process, based on the abnormal locations, the key nodes in the fire water pipe network are abstracted as vertices in the graph structure, and the water pipe segments between nodes are abstracted as edges, thus constructing a weighted directed graph model.

[0064] Specifically, based on abnormal locations and key nodes, each physical node is abstracted as a vertex in a graph structure, and a unique identifier is assigned to each vertex; the water pipe segment connecting two vertices is abstracted as a directed edge in the graph structure, with the edge direction consistent with the normal flow direction of the water; for each edge, the basic attributes of its corresponding water pipe segment are extracted as initial parameters, including physical attributes and state attributes. The physical attributes include the length, diameter, and material of the water pipe segment, and the state attributes are combined with the abnormal location marker. When the water pipe segment is in a normal state, it is assigned the current pressure value and flow rate value; when the water pipe segment is in an abnormal state, the abnormality degree parameter and the corresponding correction coefficient are recorded.

[0065] A weight quantification formula is established based on the initial attribute parameters. The weight values ​​comprehensively reflect the water supply efficiency and reliability of the water pipe section. The calculation formula is as follows: Where R is the basic resistance coefficient, reflecting the water flow resistance caused by the length and diameter of the water pipe section. The larger the value, the greater the resistance. The calculation formula is: Where k is the resistance constant, L is the length of the water pipe segment, and D is the diameter of the water pipe segment; S is the state correction coefficient, which adjusts the weight according to the degree of anomaly of the water pipe segment, and the larger the value, the worse the state; P is the weight adjusted according to the importance of the area where the water pipe segment is located, and the smaller the value, the higher the priority; the weights of vertices, directed edges and edges are entered into the graph structure database to construct a weighted directed graph model.

[0066] In one specific embodiment, the key nodes include 20 core nodes such as the main water source (underground water storage tank), backup water source (rooftop water tank), 8-layer booster pump (ID: P8), 5-layer branch pipe interface (ID: B5-3), 3-layer riser (ID: L3), and 4-layer fire hydrant (ID: H4-1). Each node is abstracted as a vertex of a graph structure and assigned a unique identifier, where the main water source is V0, the backup water source is V1, the 8-layer booster pump is V8, the 5-layer branch pipe interface is V12, and the 3-layer riser is V12. The riser is designated V5, and the fire hydrant in the fire zone on the 4th floor is designated V15. The 18 water pipe segments connecting the vertices are abstracted as directed edges, with the edge direction consistent with the water flow direction. For example, V0→V8 represents water flowing from the main water source to the booster pump on the 8th floor, and V1→V5 represents water flowing from the backup water source to the riser on the 3rd floor. Each edge is assigned a unique ID, such as E0-8 corresponding to the V0→V8 water pipe segment, which is a DN150 steel pipe with a length of 25m; E1-5 corresponding to the V1→V5 water pipe segment, which is a DN120 steel pipe with a length of 18m. The 20 vertices, 18 edges, and edge weights are entered into the graph database to form a weighted directed graph.

[0067] Step S400: When a section of water pipe is determined to be blocked, take the available fire hydrants around the fire point as target nodes and the main water source and backup water source as starting nodes, perform path search in the graph structure, and select the most efficient path as the backup water supply path; use reinforcement learning algorithm to send control commands to the smart valves on the backup water supply path, open the corresponding valves in the path sequence, close irrelevant branches, and form a closed water supply channel from the water source to the target fire hydrant.

[0068] Specifically, when a sensor detects that the water flow in a certain section of the water pipe is obstructed, the graph structure edge corresponding to the obstructed water pipe section is located using the digital twin model of the water pipe network and marked as a failed edge. Based on the location and topological relationship of the failed edge, the marking of related edges that may be affected by the cascading effect is expanded to determine the area in the graph structure that cannot participate in path search. The vertices corresponding to the main water source and the backup water source in the graph structure are included in the starting node set. When the path from the main water source to the obstructed area has been partially failed, the backup water source node with normal status is retained first. With the fire point as the center, the available fire hydrants within the radius are delineated, and the vertex status corresponding to the available fire hydrants is confirmed through the digital twin model to form the target node set.

[0069] The Dijkstra algorithm is used to search for all feasible paths in the graph structure, starting from the set of initial nodes and ending at the set of target nodes. Feasible paths do not contain failed edges. For each path, its total weight and estimated water conveyance efficiency are calculated. Paths with total weights exceeding a threshold are removed, and candidate paths are retained. Candidate paths are sorted from highest to lowest estimated water conveyance efficiency, and the candidate path with the highest estimated water conveyance efficiency is selected as the backup water supply path.

[0070] Furthermore, the valve control process of the backup water supply path is modeled as a Markov decision process, defining environmental elements including: a state space containing the current state of all smart valves on the backup water supply path, real-time pressure of each pipeline segment, water flow velocity, and water demand of the target fire hydrant; an action space containing the set of operations for each smart valve, including opening, closing, and holding; and an objective function centered on forming a closed water supply channel as quickly as possible, with quantifiable indicators including total valve operation time, path water pressure compliance time, and irrelevant branch closure rate. A deep reinforcement learning algorithm is used to construct a neural network to fit Q, with the state space features as input and the Q value for each action as output. Training is generated using historical fire simulation data and valve operation logs. The system uses samples to train a deep reinforcement learning model. After a fire occurs, the real-time status of the backup water supply path is acquired and used as the initial state input to the trained deep reinforcement learning model. The model outputs the Q-value of each possible action and selects the action with the largest Q-value as the current optimal operation. A sequence of instructions is generated according to the path order: the valve at the water source is opened first, then the valves in the middle of the path are opened in sequence, and finally all branch valves unrelated to the path are closed to ensure that the water flows along the preset path. Control instructions, including action type and execution time, are sent to the smart valves through the industrial bus. When the instruction sequence is completed, the water supply status at the target fire hydrant is detected. When a stable water supply condition is reached, it is determined that a closed water supply channel from the water source to the target fire hydrant has been formed.

[0071] In one specific embodiment, the sensor detected that the flow rate of E8-15 suddenly dropped from 2.2 L / s to 0 and the pressure dropped from 0.3 MPa to 0, which determined that the water flow was completely blocked and marked as a "failed edge" in the graph structure. Based on the topology, the adjacent edges of E8-15, E0-8 (from the main water source to the 8th floor booster pump) and E8-14 (from the 8th floor booster pump to another fire hydrant), were marked as "high-risk associated edges" due to pressure fluctuations (a drop of 0.1 MPa) and included in the unusable area. The remaining 16 edges (such as E1-5, E5-12, etc.) were in normal status and allowed to participate in path search.

[0072] The main water source V0 was excluded because the associated edge E0-8 was a high-risk edge. Only the backup water source V1 (rooftop water tank) was retained as the starting node, with a current pressure of 0.5MPa and a flow rate of 3.0L / s. Taking the fire ignition point on the 4th floor (X=25m, Y=15m) as the center, a radius of 30m was defined. The fire hydrant V15 on the 4th floor (requiring 2.5L / s water) and the fire hydrant V16 on the 5th floor (requiring 2.0L / s water) were identified as usable target nodes. Neither of them was directly affected by the fire.

[0073] Dijkstra's algorithm is used to search for feasible paths from V1 to V15 and V16, resulting in 3 valid paths: Path 1: V1→V5→V12→V15 (including edges E1-5, E5-12, and E12-15); Path 2: V1→V5→V6→V15 (including edges E1-5, E5-6, and E6-15); Path 3: V1→V5→V12→V16 (including edges E1-5, E5-12, and E12-16).

[0074] The total weight of path 1 is 30+15.36+18=63.36, path 2 is 30+22+20=72, and path 3 is 30+15.36+16=61.36. Based on pipe diameter and pressure calculations, path 1 is 2.8L / s, path 2 is 2.5L / s, and path 3 is 2.3L / s. The total weight threshold is set to 70, and path 2 is eliminated. Since the water conveyance efficiency of path 1 is higher than that of path 3, path 1 is selected as the backup water supply path.

[0075] For the backup water supply path V1→V5→V12→V15, a deep reinforcement learning algorithm is used to control three smart valves (valve F1 at V1, valve F5 at V5, and valve F12 at V12) and two unrelated branch valves (F7 and F9) on the path. State space features: Includes initial states F1 (closed), F5 (closed), and F12 (closed); pressures of pipe segments E1-5: 0.4 MPa; E5-12: 0.35 MPa; E12-15: 0.3 MPa; target fire hydrant V15: water demand 2.5 L / s. The input layer contains 12 features (3 valve states + 3 pressure values ​​+ 1 flow value + 5 auxiliary parameters), the hidden layer consists of 2 layers with 64 neurons, and the output layer is the Q-value corresponding to the 3 actions (open / close / hold) of each valve. Using 1000 sets of historical data (including operation records of different paths and valve states), after training for 500 epochs, the model achieved an average reward of 85 points (out of 100) on the validation set, with an action prediction accuracy of 92%.

[0076] The model receives F1, F5, and F12 as all in the off state, and the pressure and flow data are as described above. It outputs the Q value of each action, with F1 having the highest Q value (35) for the "on" action (30), followed by F5 (30), and F12 having a Q value of 28.

[0077] The instruction sequence is as follows:

[0078] t=0s: Sends an "on" command to F1, requiring full on within 3 seconds;

[0079] t=3s: F1 feedback is activated, E1-5 pressure rises to 0.45MPa, model output F5 "activation" action Q value is the highest (42), and command is sent;

[0080] t=6s: F5 feedback is activated, E5-12 pressure rises to 0.4MPa, model output F12 "activation" action Q value is the highest (38), and command is sent;

[0081] t=9s: F12 feedback is activated, E12-15 pressure rises to 0.38MPa, and the model outputs F7 and F9 "close" commands (Q values ​​are 25 and 26 respectively).

[0082] Execution result: F7 and F9 were closed at t=12s, and the closure rate of irrelevant branches was 100%.

[0083] Status detection: At t=15s, the pressure at the target fire hydrant V15 stabilized at 0.35MPa, and the flow rate was 2.6L / s, meeting the water demand of 2.5L / s for 3 seconds; the total valve operation time was 12 seconds, and the time for the path water pressure to reach the standard (from the first opening to the pressure ≥0.3MPa) was 8 seconds, both of which were better than the preset thresholds (20 seconds and 15 seconds respectively); final judgment: the closed water supply channel from the backup water source V1 to the target fire hydrant V15 was successfully formed, and the water supply stability reached 98% (pressure fluctuation ≤0.02MPa).

[0084] like Figure 2 As shown, this application provides a fire protection facility safety assessment and optimization system based on big data, including:

[0085] The adaptation solution generation module includes an image feature acquisition unit and an adaptation solution generation unit. The image feature acquisition unit acquires image features of the fire hydrant interface through a contour detection algorithm. The adaptation solution generation unit compares the feature data of the fire hydrant interface and the hose interface, and obtains an adaptation solution when there is an adaptation deviation.

[0086] The abnormal location marking and supply strategy execution module includes: an appearance status monitoring unit, an abnormal location marking unit, a fire scenario simulation unit, and a supply strategy execution unit. Specifically, the appearance status monitoring unit monitors the appearance of exposed sections of water pipes located at key nodes of the building's fire water pipe network when the fire hydrant interface and hose interface are compatible; the abnormal location marking unit summarizes all appearance statuses, constructs a digital twin model of the water pipe network, assesses the pressure, flow rate, and unobstructedness of fire water in each area, and marks abnormal locations; the fire scenario simulation unit simulates fire scenarios based on the digital twin model of the water pipe network; and the supply strategy execution unit triggers preset tiered supply strategies based on the simulated fire scenario, combined with the water pipe network distribution and water source location.

[0087] Directed graph construction module: includes: directed graph construction unit; wherein, during the water supply process, the directed graph construction unit, combined with abnormal locations, abstracts the key nodes in the fire water pipe network as vertices in the graph structure, and the water pipe segments between nodes as edges, to construct a weighted directed graph model;

[0088] The closed water supply channel generation module includes a backup water supply path screening unit and a closed water supply channel generation unit. The backup water supply path screening unit, when a section of water pipe is determined to be obstructed, uses available fire hydrants near the fire point as target nodes and the main water source and backup water source as starting nodes. It performs a path search in the graph structure to select the most efficient path as the backup water supply path. The closed water supply channel generation unit uses a reinforcement learning algorithm to send control commands to the smart valves on the backup water supply path, opening the corresponding valves in the path sequence and closing irrelevant branches to form a closed water supply channel from the water source to the target fire hydrant.

[0089] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for safety assessment and optimization of fire protection facilities based on big data, characterized in that, Includes the following steps: Image features of fire hydrant interfaces are collected using a contour detection algorithm; feature data of fire hydrant interfaces and hose interfaces are compared, and adaptation solutions are obtained when there is a mismatch. When fire hydrant interfaces and hose interfaces are compatible, the appearance of exposed parts of water pipes located at key nodes of the building's fire water pipe network is monitored; the appearance status is summarized to construct a digital twin model of the water pipe network, assess the pressure, flow rate, and unobstructedness of fire water in each area, and mark abnormal locations; based on the digital twin model of the water pipe network, a fire scenario is simulated; based on the simulated fire scenario, combined with the distribution of the water pipe network and the location of water sources, a preset graded replenishment strategy is triggered. During the water supply process, in combination with abnormal locations, the key nodes in the fire water pipe network are abstracted as vertices in a graph structure, and the water pipe segments between nodes are abstracted as edges, thus constructing a weighted directed graph model. Specifically, based on abnormal locations and key nodes, each physical node is abstracted as a vertex in a graph structure, and a unique identifier is assigned to each vertex; the water pipe segment connecting two vertices is abstracted as a directed edge in the graph structure, with the edge direction consistent with the normal flow direction of the water; for each edge, the basic attributes of its corresponding water pipe segment are extracted as initial parameters, including physical attributes and state attributes. The physical attributes include the length, diameter, and material of the water pipe segment, and the state attributes are combined with the abnormal location marker. When the water pipe segment is in a normal state, it is assigned the current pressure value and flow rate value; when the water pipe segment is in an abnormal state, the abnormality degree parameter and the corresponding correction coefficient are recorded. A weight quantification formula is established based on the initial attribute parameters. The weight values ​​comprehensively reflect the water supply efficiency and reliability of the water pipe section. The calculation formula is as follows: Where R is the basic resistance coefficient, reflecting the water flow resistance caused by the length and diameter of the water pipe section. The larger the value, the greater the resistance. The calculation formula is: Where k is the resistance constant, L is the length of the water pipe segment, and D is the diameter of the water pipe segment; S is the state correction coefficient, which adjusts the weight according to the degree of anomaly of the water pipe segment. The larger the value, the worse the state; P is the weight adjusted according to the importance of the area where the water pipe segment is located. The smaller the value, the higher the priority; the weights of vertices, directed edges, and edges are entered into the graph structure database to construct a weighted directed graph model. When a section of water pipe is determined to be blocked, a path search is performed in the graph structure, with the available fire hydrants around the fire point as target nodes and the main water source and backup water source as starting nodes. The most efficient path is selected as the backup water supply path. Using a reinforcement learning algorithm, control commands are sent to the smart valves on the backup water supply path to open the corresponding valves in the path sequence and close irrelevant branches, forming a closed water supply channel from the water source to the target fire hydrant.

2. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, The acquisition of image features of the fire hydrant interface through a contour detection algorithm includes: The fire hydrant interface is photographed from multiple angles to collect an image dataset containing the interface shape, diameter, thread type, number of clips, and distribution location; based on the image dataset, the image quality is optimized and interference factors are eliminated through image preprocessing algorithms; A contour detection algorithm is used to perform a full contour scan of the image, extracting all closed contour regions. Screening conditions are set based on the physical characteristics of the fire hydrant interface to eliminate interfering contours that do not meet the characteristics. The regularity of the contour is judged by the ratio of the contour perimeter to the area, retaining the main contour of the interface with continuous edges. On the main contour of the interface, the edge curve is simplified using the Douglas-Pock algorithm, retaining feature points whose curvature changes meet the conditions, and calculating the core parameters of the interface, including: obtaining the diameter of the contour using the minimum circumcircle algorithm; locating the inner edge contour of the interface and calculating the inner diameter; identifying periodic contour fluctuations formed by threads or clips, and calculating the clip spacing by the distance between adjacent feature points; counting the number of clip-type feature points to determine the clip distribution characteristics of the interface; converting the calculated parameters into a unified unit, calibrating them against a preset image scale, and eliminating outliers caused by image blurring or occlusion, which are then used as image features of the fire hydrant interface.

3. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, The feature data of the fire hydrant interface and the hose interface are compared, and when there is a mismatch, an adaptation solution is obtained, including: For each feature parameter, an adaptation threshold range is set. According to the set comparison rules, each feature parameter of the fire hydrant interface and the water hose interface is compared item by item. When all parameters are within the threshold range, it is determined to be a complete fit. When there is a parameter that exceeds the threshold range, the specific value of the deviation parameter is recorded and the deviation type is marked. Based on the identified deviation parameters and deviation types, the preset solution database is searched, the matching solution is verified for adaptation, and the adapted solution is output.

4. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, When the fire hydrant interface and the hose interface are compatible, the appearance of the exposed part of the water pipe located at a key node of the building's fire water pipe network is monitored, including: Key nodes in the building's fire protection water pipe network include branch pipe connections, valve interfaces, and exposed sections of floor risers. Camera shooting parameters are set, and a timed shooting cycle is established. The acquired images of the water pipe appearance are preprocessed. The Canny edge detection algorithm is used to extract features from the preprocessed images, specifically including: identifying color changes on the water pipe surface using the HSV color space model; calculating the texture consistency of the water pipe surface using a gray-level co-occurrence matrix to identify abnormal textures; locating protrusions, depressions, or deformed areas on the water pipe surface using the edge detection algorithm; converting the identified abnormal areas to their actual dimensions using a pixel scale; and evaluating the appearance status of the water pipes at each key node based on the extracted appearance feature parameters and preset judgment criteria.

5. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, The above summarizes the various appearance conditions, constructs a digital twin model of the water pipe network, assesses the pressure, flow rate, and unobstructedness of fire-fighting water in each area, and marks abnormal locations, including: Based on the fire water pipe network layout in the architectural CAD drawings, the physical structure of the water pipes is mapped, and pipe sections, valves and fire hydrants are transformed into basic components of the digital twin model; the topological relationships between components are defined according to the actual connection relationship, and a unique identifier is bound to each component, corresponding to the ID of the physical node, to construct a digital twin model of the water pipe network; each appearance state is imported into the digital twin model, and appearance state attributes are assigned to the corresponding components. The digital twin model is divided into assessment units according to the building's functional zones. Each assessment unit is evaluated as follows: the average pressure value of all pipe sections within the unit is calculated and compared with the design pressure threshold for that area; if the pressure value is lower than the threshold, it is marked as a pressure anomaly. Based on the flow sensor data at each node, the balance between the total inflow and outflow within the unit is analyzed; if the difference exceeds a set threshold, it is marked as a flow anomaly. Combining the appearance, pressure change trend, and flow change trend, the actual pipe diameter is calculated through fluid dynamics simulation; if the diameter reduction rate is greater than a set threshold, it is marked as a blockage. For the assessed abnormal units, the specific abnormal nodes are traced through the topological relationship of the digital twin model, and the abnormal location is marked.

6. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, The digital twin model based on the water pipe network, simulating a fire scenario, includes: Collect typical fire case data of different types of buildings, extract fire size, fire location characteristics and floor distribution patterns, and establish a fire parameter database; set trigger conditions for each fire type, classifying them into particularly serious fires, major fires, relatively serious fires and general fires; overlay building spatial layout information in the virtual space of the water pipe network digital twin model, mapping the scene variables in the fire parameter database to the corresponding positions in the water pipe network digital twin model; define the interaction logic between the fire scene and the water pipe network, and set the impact of fire smoke diffusion on sensors; start the fire simulation engine to simulate the change of fire intensity over time based on the set initial parameters.

7. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, When a section of water pipe is determined to be obstructed, a path search is performed in the graph structure, using available fire hydrants near the fire ignition point as target nodes and the main water source and backup water source as starting nodes, to select the most efficient path as the backup water supply path, including: When a sensor detects that the water flow in a certain section of the water pipe is obstructed, the digital twin model of the water pipe network is used to locate the graph structure edge corresponding to the obstructed water pipe section and mark it as a failed edge. Based on the location and topological relationship of the failed edge, the marking of related edges that may be affected by the cascading effect is expanded to determine the area in the graph structure that cannot participate in path search. The vertices corresponding to the main water source and the backup water source in the graph structure are included in the starting node set. When the path from the main water source to the obstructed area is partially failed, the backup water source node with normal status is retained first. With the fire point as the center, the available fire hydrants within the radius are delineated, and the vertex status corresponding to the available fire hydrants is confirmed through the digital twin model to form the target node set. The Dijkstra algorithm is used to search for all feasible paths in the graph structure, starting from the set of initial nodes and ending at the set of target nodes. Feasible paths do not contain failed edges. For each path, its total weight and estimated water conveyance efficiency are calculated. Paths with total weights exceeding a threshold are removed, and candidate paths are retained. Candidate paths are sorted from highest to lowest estimated water conveyance efficiency, and the candidate path with the highest estimated water conveyance efficiency is selected as the backup water supply path.

8. The method for safety assessment and optimization of fire protection facilities based on big data according to claim 1, characterized in that, The method of using reinforcement learning algorithms to send control commands to smart valves on the backup water supply path, opening corresponding valves in sequence according to the path, closing irrelevant branches, and forming a closed water supply channel from the water source to the target fire hydrant includes: The valve control process of the backup water supply path is modeled as a Markov decision process. Environmental elements are defined, including: a state space containing the current state of all smart valves on the backup water supply path, real-time pressure of each pipeline segment, water flow velocity, and water demand of the target fire hydrant; an action space containing the set of operations for each smart valve, including opening, closing, and holding; and an objective function centered on forming a closed water supply channel as quickly as possible, with quantifiable indicators including total valve operation time, path water pressure compliance time, and irrelevant branch closure rate. A deep reinforcement learning algorithm is used to construct a neural network to fit Q, with the state space features as input and the Q-value for each action as output. Training samples are generated using historical fire simulation data and valve operation logs. The system trains a deep reinforcement learning model. After a fire occurs, it acquires the real-time status of the backup water supply path and inputs it into the trained deep reinforcement learning model as the initial state. The model outputs the Q-value of each possible action and selects the action with the largest Q-value as the current optimal operation. It generates a sequence of instructions according to the path: first, it opens the valve at the water source end, then opens the valves in the middle of the path in sequence, and finally closes all branch valves that are not related to the path, ensuring that the water flows along the preset path. It sends control instructions to the smart valves through the industrial bus, including the action type and execution time. When the instruction sequence is completed, it detects the water supply status at the target fire hydrant. When the stable water supply condition is reached, it determines that a closed water supply channel from the water source to the target fire hydrant has been formed.

9. A fire protection facility safety assessment and optimization system based on big data, using the fire protection facility safety assessment and optimization method based on big data as described in any one of claims 1-8, characterized in that, include: The adaptation solution generation module includes an image feature acquisition unit and an adaptation solution generation unit. The image feature acquisition unit acquires image features of the fire hydrant interface through a contour detection algorithm. The adaptation solution generation unit compares the feature data of the fire hydrant interface and the hose interface, and obtains an adaptation solution when there is an adaptation deviation. The abnormal location marking and supply strategy execution module includes: an appearance status monitoring unit, an abnormal location marking unit, a fire scenario simulation unit, and a supply strategy execution unit. Specifically, the appearance status monitoring unit monitors the appearance of exposed sections of water pipes located at key nodes of the building's fire water pipe network when the fire hydrant interface and hose interface are compatible; the abnormal location marking unit summarizes all appearance statuses, constructs a digital twin model of the water pipe network, assesses the pressure, flow rate, and unobstructedness of fire water in each area, and marks abnormal locations; the fire scenario simulation unit simulates fire scenarios based on the digital twin model of the water pipe network; and the supply strategy execution unit triggers preset tiered supply strategies based on the simulated fire scenario, combined with the water pipe network distribution and water source location. Directed graph construction module: includes: directed graph construction unit; wherein, during the water supply process, the directed graph construction unit, combined with abnormal locations, abstracts the key nodes in the fire water pipe network as vertices in the graph structure, and the water pipe segments between nodes as edges, to construct a weighted directed graph model; The closed water supply channel generation module includes a backup water supply path screening unit and a closed water supply channel generation unit. The backup water supply path screening unit, when a section of water pipe is determined to be obstructed, uses available fire hydrants near the fire point as target nodes and the main water source and backup water source as starting nodes. It performs a path search in the graph structure to select the most efficient path as the backup water supply path. The closed water supply channel generation unit uses a reinforcement learning algorithm to send control commands to the smart valves on the backup water supply path, opening the corresponding valves in the path sequence and closing irrelevant branches to form a closed water supply channel from the water source to the target fire hydrant.

Images