Building fire early warning protection system based on Internet platform and deep visual identification

Abstract

The invention provides a building fire early warning and protection system based on an internet platform and deep visual identification, and relates to the technical field of internet fire alarm. Through real-time monitoring and quantitative evaluation of a fatal structure failure risk in a fire, an evacuation decision is established on the basis of building structure safety for the first time; personnel are fundamentally prevented from entering a dangerous area about to collapse; on the basis, a principal component analysis method is adopted to replace traditional fixed weights, objective and dynamic fusion of multi-dimensional risks such as structures, environments and behaviors is achieved, and therefore the really scientific and self-adaptive optimal survival path is planned. And finally, by generating a rescue priority index and adaptively adjusting the guidance intensity, the complex risk data is converted into accurate decision suggestions for firemen and efficient evacuation instructions for the public, intelligent and differentiated linkage control is realized, and the overall rescue and evacuation efficiency is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of Internet fire alarm technology, and in particular to a building fire early warning and protection system based on an Internet platform and deep visual recognition. Background Art

[0002] Large transportation hubs and the commercial buildings within them (such as parking areas, transfer areas, and commercial areas) are characterized by extremely high population density and mobility, complex internal structures, and diverse potential fire sources (including shops and electrical equipment). Traditional fire alarm systems are unable to effectively respond to the rapid spread of fire and smoke in these scenarios, nor can they predict the damage caused by high temperatures to building structures or the panic behavior of people in confined spaces. The static nature of evacuation instructions often leads to crowds rushing into dangerous areas filled with smoke or threatened by structural collapse, resulting in significant casualties. This system aims to address this challenge.

[0003] In the prior art, the announcement number is CN107689133B, and its name is a fire intelligent early warning, disaster reduction and protection system, which includes a background server, an intelligent fire terminal, a smoke detector and a gas detector; the intelligent fire terminal includes an equipment placement cabinet and a monitoring gateway; the monitoring gateway is installed on the equipment placement cabinet; the background server communicates with the monitoring gateway of each intelligent fire terminal via a wireless network; the monitoring gateway of each intelligent fire terminal communicates wirelessly with each smoke detector and each gas detector; the monitoring gateway of the intelligent fire terminal controls the opening and closing of each equipment placement cabinet door; the fire intelligent early warning, disaster reduction and protection system stores fire-fighting equipment for self-rescue in the equipment placement cabinet, and the monitoring gateway controls the opening and closing of the equipment placement cabinet door, which is convenient for users to quickly take out and put away fire-fighting equipment, and at the same time uses the background server to receive the alarm signals of each intelligent fire terminal, and trigger other intelligent fire terminals, so as to achieve the effect of monitoring multiple places and triggering alarms at the same time.

[0004] Despite significant progress, existing technologies still exhibit profound limitations when responding to rapidly changing real-world fire scenarios. These limitations are primarily reflected in the single-dimensional nature of risk assessment, the rigidity of path planning decisions, and the sluggish response of coordinated control.

[0005] First, at the risk assessment level, existing systems primarily focus on the "visible" aspects of danger, specifically the perception of directly observable phenomena like flames, smoke, and crowds. They generally overlook the internal damage fire inflicts on building structures—the degradation of materials and reduced bearing capacity of load-bearing components due to high temperatures. In fires, structural failure is often the root cause of catastrophic consequences. However, existing technologies lack real-time, quantitative monitoring and risk assessment of structural safety status, making it impossible to incorporate this most deadly hidden risk into evacuation decision-making models. Consequently, planned routes may appear unobstructed but in reality lead to dangerous areas on the verge of collapse.

[0006] 2. Secondly, at the path planning level, existing technologies often use a linear weighting method with preset fixed weights when integrating multi-dimensional risk information. This method is highly subjective and cannot adapt to the dynamic dominance of various risk factors (such as structural risk, environmental risk, and behavioral risk) at different stages of a fire. As a result, the "optimal path" it generates may not be the true "safest path" in complex scenarios.

[0007] Finally, at the linkage control level, the existing system's information output model is relatively crude, typically pushing generic evacuation instructions to all users or providing isolated data points (such as fire location and number of trapped people) to rescuers. This lacks in-depth processing and differentiated handling of information. This not only increases the cognitive burden on panicked individuals, but also fails to transform massive amounts of data into high-level intelligence, such as "rescue mission priorities," that can directly guide professional rescuers' decisions. This significantly reduces the system's overall protective effectiveness.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0009] The purpose of the present invention is to provide a building fire early warning and protection system based on an Internet platform and deep visual recognition to solve the problems raised in the above background technology.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] The building fire early warning and protection system based on the Internet platform and deep visual recognition includes:

[0012] Structural status perception module: This module is used to monitor the actual vibration frequency and surface temperature of the load-bearing components corresponding to each escape route on the target building in real time under fire conditions, and analyze and process this data to generate a final structural risk indicator value representing the load-bearing components corresponding to each escape route;

[0013] Environmental and behavior perception module: After receiving the final structural risk indicator value, it analyzes the surveillance video corresponding to each escape route in the target building to generate the local entropy value of the crowd and the safe passage degree of the route;

[0014] Determining a final behavior risk indicator value and a final environment risk indicator value respectively by performing weighted fusion based on the credibility of visual data on the local entropy increment of the crowd and the safe passage degree of the channel;

[0015] Optimal survival path generation module: used to receive a multidimensional risk vector data set consisting of the final structural risk indicator value, the final behavioral risk indicator value, and the final environmental risk indicator value, perform principal component analysis, and extract the first principal component as a dynamic comprehensive impedance representing the comprehensive risk level of each escape route;

[0016] The dynamic comprehensive impedance is assigned as the travel cost to the corresponding path weight in the preset target building three-dimensional topology map, and the A* search algorithm is used to solve in real time the path with the lowest cumulative dynamic comprehensive impedance from any starting point to the safe exit as the optimal survival path;

[0017] Linkage control execution layer: used to execute differentiated information output based on the optimal survival path. The information output includes the rescue priority index sent to the rescue terminal and the output direction instructions of the dynamic adjustment of the public evacuation guidance equipment.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. Achieves in-depth perception and early warning of building structural safety: By real-time monitoring of the vibration frequency and temperature of load-bearing components, the risk of fatal structural failure in a fire is quantified. This overcomes the shortcomings of existing technologies that only focus on visible dangers, enabling evacuation decisions to avoid areas that are about to collapse, fundamentally improving the probability of survival.

[0020] 2. Improved the scientificity and adaptability of evacuation route planning: The principal component analysis (PCA) method is used instead of the traditional fixed weighting method. It can objectively and dynamically integrate multi-dimensional risks and automatically identify the most important contradictions at the current fire scene, thereby planning the truly "safest" path, making decision-making more scientific and reliable.

[0021] 3. Realized intelligent and differentiated linkage of information output: By generating a rescue priority index and adaptively adjusting the guidance intensity, the original data is converted into precise decision-making suggestions for firefighters and efficient evacuation instructions for the public, significantly improving rescue efficiency and evacuation success rate, and achieving refined and intelligent protection for personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a block diagram of the overall system module of the present invention;

[0023] Figure 2 This is a logical diagram of the "structural state perception module" of the present invention;

[0024] Figure 3 It is a logical schematic diagram of the "optimal survival path generation module" of the present invention. DETAILED DESCRIPTION

[0025] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0027] Example 1:

[0028] See also Figures 1 to 3 , the present invention provides a technical solution:

[0029] The building fire early warning and protection system based on the Internet platform and deep visual recognition includes:

[0030] Structural status perception module: This module is used to monitor the actual vibration frequency and surface temperature of the load-bearing components corresponding to each escape route on the target building in real time under fire conditions, and analyze and process this data to generate a final structural risk indicator value representing the load-bearing components corresponding to each escape route;

[0031] Further explanation: 1) The actual vibration frequency is compared with the pre-stored reference vibration frequency to calculate the structural natural vibration frequency offset;

[0032] Based on the comparison relationship between the surface temperature and the sensor operating temperature threshold, a frequency data credibility characterizing the quality of the structural natural vibration frequency offset data on the load-bearing component is generated through a preset nonlinear function;

[0033] When the credibility of the frequency data is lower than a preset compensation trigger threshold, the structural state perception module activates a preset thermal-mechanical coupling finite element model to generate a model-predicted frequency offset;

[0034] The final structural risk indicator value is determined by weighted fusion of the structural natural vibration frequency offset and the model-predicted frequency offset based on the credibility of the frequency data; the specific implementation content includes:

[0035] The reliability of the frequency data is specifically based on the comparison relationship between the surface temperature and the preset operating temperature threshold of the sensor, and is generated by a preset nonlinear function. This embodiment defines various parameters:

[0036] The actual vibration frequency is recorded as ; The surface temperature is recorded as ; The structural natural vibration frequency offset is recorded as ; The reliability of frequency data is recorded as ; The reference vibration frequency is recorded as ; The sensor operating temperature threshold is recorded as ; The model predicts the frequency offset to be ; The compensation trigger threshold is recorded as , the compensation trigger threshold in this embodiment The value range of is selected within the interval (0.2, 0.9), which is determined by the expert group using the fuzzy analytic hierarchy process and will not be described in detail; the slope parameter of the S-type function is recorded as ; The central temperature parameter of the S-type function is recorded as ; The final structural risk indicator value is recorded as ;

[0037] Each escape route is indexed with r, the corresponding load-bearing component is represented by j, and the combination is represented as rj;

[0038] Actual vibration frequency Obtained through acceleration sensors deployed on pre-set load-bearing components;

[0039] The structural state perception module deploys a high-temperature thermocouple next to the acceleration sensor to measure the surface temperature of the load-bearing component in real time. , and according to the Generate a representation of the Frequency of data quality Data credibility ;

[0040] The calculation process of the final structural risk indicator value is broken down as follows:

[0041] 1.1) Real-time Data Collection and Preliminary Calculation: For any load-bearing component j corresponding to any escape route r, the structural state perception module simultaneously collects two data points: the "actual vibration frequency" provided by the acceleration sensor, and the "surface temperature" provided by the associated high-temperature thermocouple;

[0042] Based on the collected "actual vibration frequency" and the pre-stored "reference vibration frequency", the "structural natural vibration frequency offset" is calculated; the calculation method is: take the "actual vibration frequency " and "base vibration frequency "The absolute value of the difference, divided by the "reference vibration frequency ”;

[0043] 1.2) Credibility of frequency data Conduct the following assessments:

[0044] The S-type function (Sigmoid function) is used to calculate the "frequency data credibility ”;

[0045] Frequency data credibility The composition is expressed as: Based on the "surface temperature ","S-type function center temperature parameter ” and “S-shaped function slope parameter "These three inputs; "S-type function center temperature parameter " represents the temperature point at which the sensor performance begins to decline significantly. Its value is determined by experimental calibration or reference to the sensor specification; "S-type function slope parameter "Determines the rate at which the reliability decreases with temperature; the reliability of frequency data The calculation process is: take the "surface temperature " and "S-type function center temperature parameter "The difference is multiplied by the "S-type function slope parameter ", then multiply this product as the power of the natural exponent to calculate the exponential value; finally, "the credibility of the frequency data ” is equal to the result of “1” divided by “1 plus the exponent value calculated above”.

[0046] 1.3) Model compensation calculation: The real-time calculation of "frequency data credibility" ” with a preset “compensation trigger threshold "Compared; when "frequency data credibility "Below the compensation trigger threshold "When the compensation process is started;

[0047] Activate the pre-established finite element model of the corresponding load-bearing component rj, which includes the material thermodynamic properties. " is input into the model as a thermal load, and through thermal-mechanical coupling analysis, the theoretical natural vibration frequency under the influence of this temperature is simulated and calculated, and then the "model predicted frequency offset" is calculated. Its calculation method is the same as "structural natural vibration frequency offset ” is calculated in the same way, except that the input is the “actual vibration frequency ” is replaced by the theoretical natural vibration frequency calculated by the model.

[0048] 1.4) Use weighted average method to calculate the final structural risk indicator value Calculation: It is composed of two parts. The first part is the "structural natural vibration frequency offset " and "Frequency Data Credibility "; the second part is "the model predicts the frequency offset ” and “1 minus the reliability of frequency data ” multiplied by the difference of

[0049] 1.5) For the final structural risk indicator value Explanation of technical effects and logical rationality:

[0050] 1.51) Application level parameter substitution: Set the reference vibration frequency of the jth load-bearing component in escape route r 20Hz, sensor operating temperature threshold Associated S-shaped function center temperature parameter The compensation trigger threshold is 500°C. is 0.5;

[0051] 1.511) Initial stage of fire: surface temperature The surface temperature is 200°C. The temperature parameter far below the center of the S-type function is recorded as 500°C; the reliability of the frequency data calculated by the system Approaching 1, the sensor works normally and the actual vibration frequency is measured The system calculates the structural natural vibration frequency offset is 0.05. Due to the reliability of frequency data Above the compensation trigger threshold 0.5, model compensation is not activated; the final structural risk indicator value The calculation of mainly depends on the measured data, and its value is the structural natural vibration frequency offset Multiply by the reliability of frequency data Plus the model predicted frequency offset Multiply by "1 minus the frequency data credibility At this stage, the second term approaches zero, so the final structural risk indicator value It is approximately equal to 0.05, which accurately reflects minor structural damage;

[0052] 1.512) Fire peak: surface temperature Soars to 800°C. At this time, the surface temperature The temperature parameter far higher than the center temperature of the S-type function is recorded as 500°C, the reliability of the frequency data calculated by the system A significant decrease of 0.1 is observed; this value is below the compensation trigger threshold. 0.5, triggering the compensation module; at this time, the sensor is out of alignment due to high temperature and is likely to output an erroneous actual vibration frequency However, the thermal-mechanical coupling model is based on a surface temperature of 800°C. Input, predict the theoretical natural vibration frequency is 12Hz, calculate the model prediction frequency offset The final structural risk indicator value is 0.4. The weighted fusion calculation is: (0.25 multiplied by 0.1) plus [0.4 multiplied by (1-0.1)], resulting in 0.025 + 0.36 = 0.385. In this example, despite the large error in the sensor data, the final result is mainly dominated by the more reliable model prediction, resulting in a high risk value that is closer to the actual situation.

[0053] 1.513) In this embodiment, it is used to calculate the structural natural vibration frequency offset The calculation logic of is derived from the definition of natural frequency in structural dynamics, and the result is a dimensionless relative rate of change;

[0054] Used to calculate the reliability of frequency data The calculation logic is based on the Sigmoid function, which is a classic function used in mathematics to describe the smooth transition of states. The output is the dimensionless credibility in the interval [0,1].

[0055] Used to calculate the final structural risk indicator value The calculation logic of is derived from the weighted average method in the field of data fusion. , the model predicts the frequency offset and frequency data credibility All are dimensionless, and their linear combination ultimately indicates the structural risk value It is also dimensionless, and the physical dimensions on the left and right sides are consistent;

[0056] 1.6) Final structural risk indicator value The value range of is limited to the range [0,1];

[0057] When the final structural risk indicator value The closer the output is to 0, the safer the system determines the structural state of the load-bearing component j corresponding to the current escape route r is, and the lower the risk of collapse is.

[0058] Reasoning process: Final structural risk indicator value The closer it is to 0, the closer the weighted sum of its two components is to 0; this happens in two cases:

[0059] Case 1: When the frequency data is reliable The closer it is to 1, the greater the deviation of the structural natural vibration frequency. The closer it is to 0, the more reliable the frequency data is. Offset from the structural natural vibration frequency The closer they are to equality, the smaller the damage to the load-bearing component j is.

[0060] Case 2: When the frequency data is reliable The closer it is to 0, the greater the deviation of the structural natural vibration frequency. The closer it is to 0, this indicates that the central temperature parameter based on the S-type function The physical model predicts that the structure is more stable at this temperature. Regardless of whether the sensor is reliable, the lower the final structural risk indicator value All of them point to the precise technical conclusion that “the safer the structure”, proving the rationality of the algorithm design;

[0061] When the final structural risk indicator value The closer the output is to 1, the more serious the damage to the load-bearing component j corresponding to the current escape route r is, and the higher the risk of collapse it faces.

[0062] Reasoning process: Final structural risk indicator value The closer the output is to 1, the closer its weighted sum is to 1. This also happens in two cases:

[0063] Case 1: When the frequency data is reliable The closer it is to 1, the greater the deviation of the structural natural vibration frequency. The closer it is to 1, the higher the actual vibration frequency Compared to the reference vibration frequency The greater the decline, the more serious the structural performance degradation.

[0064] Case 2: When the frequency data is reliable The closer it is to 0, the greater the deviation of the structural natural vibration frequency. The closer it is to 1, the higher the surface temperature. Let the finite element model predict the model-predicted frequency offset This design ensures that regardless of whether sensor data is available, as long as there are real physical conditions that can cause structural failure (frequency fluctuations or extreme high temperatures), the system can output a higher level of risk alert, thereby achieving the technical goal of providing effective early warning under any working conditions.

[0065] Environmental and behavior perception module: After receiving the final structural risk indicator value, it analyzes the surveillance video corresponding to each escape route in the target building to generate the local entropy value of the crowd and the safe passage degree of the route;

[0066] Determining a final behavior risk indicator value and a final environment risk indicator value respectively by performing weighted fusion based on the credibility of visual data on the local entropy increment of the crowd and the safe passage degree of the channel;

[0067] Further explanation: The local entropy increment of the crowd represents the risk level of panic congestion of the crowd in each escape channel;

[0068] The “passage safety degree” is obtained by performing an exponential decay correction based on the final structural risk indicator value on the passable area of ​​the channel identified by the deep vision network;

[0069] The crowd local entropy increment is obtained by calculating the Shannon entropy of the probability distribution of the crowd motion vector in the surveillance video image;

[0070] The visual data credibility is generated by analyzing the smoke concentration of the surveillance video image; when the visual data credibility is lower than the preset threshold, the system will activate the crowd inference model to make compensatory estimates of the crowd status and channel status in the obscured area;

[0071] The specific implementation content of the environment and behavior perception module is as follows:

[0072] 2.1) First, define the following parameters in advance:

[0073] The local entropy increment of the crowd is recorded as ; The channel visual transparency is recorded as ; The structural risk impact factor is recorded as ; The channel safety degree is recorded as ; The credibility of visual data is recorded as ; Let the inferred crowd panic index be ;Record the inferred channel state value as ; The visual compensation trigger threshold is recorded as ; The final behavioral risk indicator value is recorded as ; The final environmental risk indicator value is recorded as ;

[0074] 2.2) The calculation process of the final behavioral risk indicator value and the final environmental risk indicator value is broken down as follows:

[0075] 2.21) Calculate the local entropy gain of the crowd: For surveillance videos of each escape route, use optical flow or target tracking algorithms to extract crowd motion vectors, discretize them into N×M states, calculate the probability of each state, and calculate the "local entropy gain of the crowd" using the Shannon entropy formula. This value is normalized and obtained through real-time video analysis.

[0076] 2.22) Calculating Channel Visual Transparency: Video images are dehazed and a semantic segmentation network is used to identify the traversable ground area. The ratio of this area to the total cross-sectional area of ​​the channel is the "channel visual transparency." This value is obtained through real-time video analysis.

[0077] 2.23) Calculation of channel safety based on the revised “final structural risk indicator value”:

[0078] An exponential decay function is used to modify the "channel visual transparency" using the "final structural risk indicator value" calculated by the structural status perception module, thereby obtaining a comprehensive "channel safe passage";

[0079] The "passageway safety degree" is composed of the "passageway visual transparency," the "final structural risk indicator," and the "structural risk impact factor." The "structural risk impact factor" is a preset constant. In this embodiment, the "structural risk impact factor" is set to 5, which is used to adjust the severity of the impact of structural risk on traffic capacity. The "passageway safety degree" is calculated as follows: "passageway safety degree" is equal to the "passageway visual transparency" multiplied by the natural constant e as the base and the product of the negative structural risk impact factor and the final structural risk indicator as the exponent.

[0080] 2.24) Additional explanation for "The system activates the crowd inference model to make compensatory estimates of the crowd and channel conditions in obscured areas":

[0081] Calculate visual data credibility: Calculate the "visual data credibility" by analyzing the contrast and clarity of the surveillance video image or a dedicated smoke concentration algorithm. The "visual data credibility" value range is [0,1], where 1 indicates complete clarity and 0 indicates complete obscuration. This is obtained through real-time video analysis.

[0082] Further explanation: The crowd flow inference model calculates the crowd flow into all visible areas adjacent to the visually blocked area and performs flow-weighted averaging on the real-time crowd local entropy increments corresponding to each of the inflowing crowd flows to determine the inferred crowd panic index of the visually blocked area;

[0083] The crowd flow inference model updates the estimated total number of people in the visually obscured area based on the law of crowd flow conservation, and dynamically decreases the visual transparency of the last valid channel in the area before the visual obscuration according to the physical space occupied by the estimated total number of people, so as to determine the inferred channel state value of the visually obscured area;

[0084] The triggering and inference of the crowd flow inference model are as follows: when the "visual data credibility" is lower than the preset "visual compensation trigger threshold", it is triggered; in this embodiment, the "visual compensation trigger threshold" is set to 0.4, activating the crowd flow inference model; immediately starting the following inference process, and recording the last valid "channel visual transparency" value before the trigger as the "last valid channel visual transparency" value. ”.

[0085] For the “Inferred Population Panic Index "Determination: The system automatically identifies all cameras that are directly connected to the current visually blocked area and whose cameras are working properly (i.e. ) adjacent regions, which are called “source regions”;

[0086] Collect source data: In each time step, for each "source area" i, the system collects two key data:

[0087] Key data 1: Inflow of people : By analyzing the surveillance video of "source area" i, the number of people moving from that area into the current visual obstruction area per unit time is counted. This is obtained through real-time target tracking of adjacent cameras.

[0088] Key data 2: Local entropy increase of the population :Real-time calculation of the "local entropy value of the crowd" in the "source area" i; the acquisition method is the real-time behavior analysis of adjacent cameras.

[0089] "Inferring the crowd panic index The calculation of " is composed of two parts: the numerator is the sum of the products of the "inflow flow" of all "source areas" and their corresponding "local entropy increase"; the denominator is the sum of the "inflow flow" of all "source areas". The final result is the ratio of these two parts. This calculation ensures that source areas with large flow of people contribute more to the inference results;

[0090] For Inferred Channel State Value Determination of:

[0091] Calculate the estimated total number of people :It needs to comply with the law of “people flow conservation”; at the initial moment of triggering compensation, the system The value of the estimated initial number of people ; represents the initial moment;

[0092] Dynamic update: At each subsequent time step, the system updates the "estimated total number of people" by: the "estimated total number of people" at the current moment is equal to the "estimated total number of people" at the previous moment, plus the total inflow of people from all "source areas", and minus the "outflow of people" from the obscured area to all adjacent visible areas. ”.

[0093] "Infer channel state value " is calculated based on a decreasing correction formula. Its initial value is "the last effective channel visual transparency The correction is calculated by adding the updated "estimated total number of people" to a preset "average individual occupied area". ” (This embodiment =0.5 square meters, which is an empirical constant determined based on ergonomics and evacuation simulation research) to obtain the total occupied area; then divide this total occupied area by the preset "total cross-sectional area of ​​the passage ", and get the occupancy rate. Finally, the "inferred channel state value" is equal to the "last valid channel visual transparency" minus this occupancy rate. To ensure that the value is not negative, the final result takes the larger value between it and zero;

[0094] Inferring the crowd panic index and inferred channel state values The value range of is [0,1];

[0095] When inferring the crowd panic index The closer the output is to 0, the more stable the crowd state in the shielded area is;

[0096] When inferring the crowd panic index The closer the output is to 1, the more unstable the crowd state in the shielded area is;

[0097] When inferring channel state values The closer the output is to 0, the more crowded the escape route is.

[0098] When inferring channel state values The closer the output is to 1, the clearer the escape route is due to the crowd.

[0099] 2.25) Weighted fusion generates final risk indicator:

[0100] Using the weighted average method and based on the "visual data credibility", the visual direct data and model inference data are dynamically integrated to generate the final behavioral risk indicator value and the final environmental risk indicator value;

[0101] The calculation of the "Final Behavior Risk Indicator Value" is the sum of two parts: the first part is the product of the "local entropy increase of the crowd" and the "credibility of the visual data"; the second part is the product of the "inferred crowd panic index" and the difference between "1 minus the credibility of the visual data". The "Final Behavior Risk Indicator Value" is "1 minus this weighted sum";

[0102] The calculation of the "Final Environmental Risk Indicator Value" is the sum of two parts. The first part is the product of the "Channel Safe Passage" and the "Visual Data Credibility". The second part is the product of the "Inferred Channel State Value" and the difference between "1 minus the Visual Data Credibility". The "Final Environmental Risk Indicator Value" is "1 minus this weighted sum".

[0103] 2.3) The application-level parameters for the final behavioral risk indicator value and the final environmental risk indicator value are substituted as follows:

[0104] For the escape route r, its structural risk influencing factor is 5, the visual compensation trigger threshold is 0.4.

[0105] Scenario 1: Slight structural damage, clear vision:

[0106] Final structural risk indicator value The value is 0.1, indicating slight damage. The camera is working properly and the visual data is reliable. The visual transparency of the channel is 0.95. is 0.8;

[0107] Calculate channel safety degree The calculation process is: 0.8 times e to the power of (-5 times 0.1), and the result is approximately 0.8*0.606=0.485. It can be seen that even if the road is visually unobstructed, considering the damage to the underlying structure, its effective traffic capacity has been significantly reduced;

[0108] Because visual data is reliable Above the vision compensation trigger threshold , compensation is not activated; final environmental risk indicator value 1-0.485=0.515, which accurately reflects the actual situation of the escape route being "clear on the surface but with hidden structural risks";

[0109] Scene 2: The structure is intact, but obscured by thick smoke.

[0110] Final structural risk indicator value is 0.01. However, the passage is filled with thick smoke, and the visual data is less reliable. Reduced to 0.2. This value is below the threshold for vision compensation to trigger , trigger compensation;

[0111] At this time, the channel visual transparency Due to the interference of thick smoke, it is wrongly calculated as 0.1; however, the system will activate the compensation model and infer the "inferred crowd panic index" based on the pedestrian flow data of adjacent channels. " is 0.7, "inferred channel state value ” is a conservative 0.1.

[0112] Final behavioral risk indicator value Calculated by weighted fusion: Multiply by 0.2) + (0.7 multiplied by 0.8), the result is dominated by the extrapolated value;

[0113] Final environmental risk indicator value In the calculation of ( The result is also dominated by the conservative extrapolation value, marking the channel as high risk. This avoids the serious mistake of misjudging a dangerous channel as safe due to visual failure.

[0114] Used to calculate the local entropy value of the population The calculation logic of , which comes from Shannon entropy in information theory, is a classic method to measure the disorder of a system, and the result is dimensionless.

[0115] Used to calculate channel safety The calculation logic of the , which is based on the exponential decay model in physics, is used to describe the process of radioactive decay or signal attenuation. This embodiment is used to describe the weakening effect of risk on traffic capacity. and the final structural risk indicator value All are dimensionless, and the combined operation result is the channel safety passability It is also dimensionless and consistent with physical dimensions.

[0116] Used to calculate the final behavioral risk indicator value and final environmental risk indicator value The calculation logic of is derived from the weighted average method in the field of data fusion. All input and output items are dimensionless and logically self-consistent.

[0117] 2.4) Make the passage safe and accessible The value range of is limited to the range [0,1];

[0118] When the passage is safe The closer the output is to 0, the smaller the actual safe passage capacity of the escape passage is determined by the system.

[0119] Reasoning process: channel safety passability The closer it is to 0, the closer its product term is to 0. This is due to two reasons: one is the channel visual transparency. The closer it is to 0, the greater the degree to which the passage is completely blocked by smoke, crowds or obstacles; the second is the final structural risk indicator value The closer it approaches 1, the greater the risk of collapse of the underlying structure. Regardless of the reason, it points to the conclusion that "the escape route is more dangerous." This embodiment design unifies the two completely different risks of congestion and collapse into a single assessment dimension, demonstrating its advanced design.

[0120] When the passage is safe The closer the output is to 1, the greater the actual safe passage capacity of the escape passage determined by the system.

[0121] Reasoning process: channel safety passability The closer it is to 1, the closer the product is to 1. That is, the channel visual transparency The closer it is to 1, the less visual smoke or obstacles there are, and the final structural risk indicator value The closer it is to 0, the safer the escape route is in terms of structure; that is, significant risk in any dimension will lead to a decrease in the safe passage degree of the route. The value decreases, and the channel is considered truly secure only when all dimensions are secure.

[0122] Optimal survival path generation module: used to receive a multidimensional risk vector data set consisting of the final structural risk indicator value, the final behavioral risk indicator value, and the final environmental risk indicator value, perform principal component analysis, and extract the first principal component as a dynamic comprehensive impedance representing the comprehensive risk level of each escape route;

[0123] The dynamic comprehensive impedance is further assigned as a travel cost to the corresponding path weight in the preset target building 3D topology map, and the A* search algorithm is used to solve in real time the path with the lowest cumulative dynamic comprehensive impedance from any starting point to the safe exit as the optimal survival path;

[0124] The specific implementation of the optimal survival path generation module begins with the definition of various parameters:

[0125] The multidimensional risk vector is denoted as ; The dynamic integrated impedance is recorded as ; The actual cumulative cost in the A* algorithm is recorded as ; Let the heuristic estimation cost in the A* algorithm be ; The total path cost in the A* algorithm is recorded as ; Where n represents the current node;

[0126] The calculation process of dynamic integrated impedance is decomposed as follows:

[0127] Step 1: Construct a multidimensional risk vector dataset:

[0128] In each calculation cycle, for each edge in the target building topology graph, the “final structural risk indicator value”, “final behavioral risk indicator value” and “final environmental risk indicator value” are combined into a three-dimensional “multi-dimensional risk vector”; denoted as In this embodiment, “each edge” represents “each escape passage r”;

[0129] The system aggregates the "multidimensional risk vectors" generated by all escape routes over multiple consecutive time steps to form a dynamically updated risk dataset that reflects the actual situation of various risk combinations during the evolution of the fire.

[0130] Step 2: "Dynamic Integrated Impedance Determination of : Principal component analysis (PCA) is used to reduce the three-dimensional multidimensional risk vector into a one-dimensional comprehensive impedance to standardize the multidimensional risk vector; the composition is expressed as follows:

[0131] 1. Data standardization: Standardize each dimension of the multidimensional risk vector data set to make its mean 0 and variance 1;

[0132] 2. Covariance matrix calculation: Based on the standardized data set, calculate the covariance matrix between multidimensional risk vectors;

[0133] 3. Eigenvalue decomposition: Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors;

[0134] 4. Determine the first principal component: Select the eigenvector corresponding to the maximum eigenvalue as the first principal component eigenvector; this eigenvector represents the main direction of data change, and its three components are the optimal combination weights of the three risk indicator values.

[0135] 5. Calculate the dynamic integrated impedance ”: Perform dot product between the standardized “multi-dimensional risk vector” and the selected first principal component eigenvector to obtain the “dynamic integrated impedance Finally, to ensure that its value range is in [0, 1] and conform to the intuition that higher impedance is more dangerous, the dot product result is linearly mapped or normalized by the Sigmoid function.

[0136] Step 3: Generate the optimal survival path:

[0137] Weight assignment: The calculated "dynamic integrated impedance " is the weight of the corresponding edge in the target building topology graph; in this embodiment, the weight represents the "traffic cost";

[0138] Run the A* algorithm: take the location of the people to be evacuated as the starting point and all safe exits as the end point set.

[0139] The "total path cost" of the A* algorithm is equal to the sum of the "actual cumulative cost" and the "heuristic estimated cost";

[0140] Actual cumulative cost : The "dynamic integrated impedance" of all edges on the path from the starting point to the current node n "Sum.

[0141] Heuristic cost estimation : The Euclidean distance from the current node n to the nearest safe exit, multiplied by a preset value that is lower than all "dynamic integrated impedance " value to ensure the admissibility of the heuristic function.

[0142] The A* algorithm searches iteratively to find a path that makes the total cost of the path "The path with the smallest value is the "optimal survival path".

[0143] Dynamic Integrated Impedance The application level parameters of " are substituted into:

[0144] At the preset stage of fire evolution, the data set analysis showed that the final structural risk indicator value The variance of is the largest and is strongly correlated with the other two risk factors. The first principal component eigenvector calculated by PCA is [0.8, 0.4, 0.4]. This shows that at this stage, the final structural risk indicator value is the dominant hazard source;

[0145] Scenario 1, Escape Route A: Multidimensional Risk Vector , that is, the structure is dangerous, but it is temporarily unoccupied and smoke-free;

[0146] After normalization, its dynamic integrated impedance A high value close to 1 will be obtained because its projection in the direction of the principal component is the largest; the A* algorithm will avoid this channel;

[0147] Scenario 2, Escape Route B: Multidimensional Risk Vector , meaning the structure was safe, but the crowds were panicking and the air was filled with smoke.

[0148] Dynamic comprehensive impedance of escape route B Lower than the dynamic comprehensive impedance of escape route A , because the projection of escape route B dominated by structural risk in the principal component direction is smaller;

[0149] Technical Effect: This embodiment not only avoids areas with high risk values, but also identifies the most lethal risk types at the current stage and gives them the highest avoidance priority. If the fire evolves to a later stage and panic among the crowd becomes the main problem, PCA will dynamically update the principal component and adaptively adjust it to [0.3, 0.9, 0.3]. At this time, the system will prioritize avoiding crowded areas such as escape route B.

[0150] All risk indicators involved in PCA calculation are dimensionless. After standardization, linear combination and normalization, the final "dynamic integrated impedance ” is dimensionless, which conforms to the definition of path cost and has the same physical dimensions;

[0151] The dynamic integrated impedance The value range of " is limited to [0,1];

[0152] When the "dynamic integrated impedance The closer the output is to 0, the lower the overall risk level of the escape route is determined by the system, and it is selected as the ideal safe path.

[0153] Reasoning process: "Dynamic integrated impedance The closer it is to 0, the greater the risk vector of the escape route. "The closer the projection in the direction of the principal component is to 0; this requires The lower the overall value of

[0154] When the "dynamic integrated impedance The closer the output is to 1, the higher the overall danger level of the escape route is determined by the system, and the higher the probability of being prohibited from passing through.

[0155] Reasoning process: "Dynamic integrated impedance The closer it is to 1, the greater the risk of the escape route. "The larger the projection value in the direction of the principal component, the higher the final structural risk indicator value. The closer the escape route is to 1, the higher its "dynamic integrated impedance The closer it gets to 1, the more dangerous the impedance. This demonstrates that the PCA method can effectively capture and amplify the primary contradictions in the current scenario, clearly identifying the most dangerous pathways. After receiving this impedance value approaching 1, the A* algorithm avoids it during path planning, achieving the core technical effect of "intelligently avoiding fatal dangers" as outlined in the invention.

[0156] The "linked control execution layer" quantifies "rescue priorities," enabling the system to provide commanders with data-driven, dynamic rescue mission sequencing recommendations. Furthermore, by adjusting "guidance intensity," the system can more effectively direct traffic flow and avoid new congestion at key nodes.

[0157] Linkage control execution layer: used to execute differentiated information output based on the optimal survival path. The information output includes the rescue priority index sent to the rescue terminal and the output direction instructions of the dynamic adjustment of the public evacuation guidance equipment;

[0158] Further explanation: By nonlinearly fusing the final behavioral risk indicator value and the final structural risk indicator value of each escape route, a rescue priority index is generated to characterize the urgency of personnel rescue in each area, and the rescue priority index is combined with the optimal survival path and structural risk thermal dynamics. Figure 1 The information is pushed to the rescue terminal of the fire rescue personnel at the same time; and the linkage control execution layer dynamically adjusts the output direction instructions of the public evacuation guidance equipment based on the real-time final behavior risk indicator value of each escape channel on the optimal survival path, and the output direction instructions achieve the effect of adjusting the visual and / or auditory intensity to achieve adaptive control of the guidance intensity;

[0159] The specific implementation content of the linkage control execution layer is to first define the following parameters:

[0160] The rescue priority index is recorded as ; Let the behavioral risk weight factor be ; Let the structural risk weight factor be ; The guidance intensity adjustment coefficient is recorded as ; The basic guidance strength is recorded as ; The final guidance strength is recorded as ;

[0161] The execution process of the linkage control execution layer is decomposed as follows:

[0162] Step 1: The rescue priority index is designed to quantify the urgency of rescuing people in different escape route areas and provide decision support for fire commanders;

[0163] The calculation logic of the "rescue priority index" is: "behavioral risk weight factor" multiplied by "final behavioral risk indicator value", plus "structural risk weight factor" multiplied by the square of "final structural risk indicator value"; among them, "behavioral risk weight factor" and "structural risk weight factor" are preset coefficients used to adjust the importance of two types of risks, and the sum of the two is 1. The purpose of squaring the structural risk is to amplify the urgency of the high structural risk area and reflect the nonlinear growth of its threat; the initial setting of this embodiment is , ;

[0164] The second step is to build and push the integrated decision view:

[0165] Rescue priority sorting list: sort all escape route areas with personnel present from high to low according to their "rescue priority index" and display them on the rescue terminal interface;

[0166] Optimal survival path: On the 3D model of the target building, the "optimal survival path" leading to each rescue target point is marked with highlighted dynamic lines;

[0167] Structural risk heat map: The load-bearing components in the target building model are rendered in real time in three colors according to their "final structural risk indicator value". In this example, 0-0.3 is green, 0.3-0.7 is yellow, and 0.7-1 is red.

[0168] The system pushes the above fused view to the firefighter's AR helmet or handheld rescue terminal device in real time via wireless network.

[0169] The third step is to explain the dynamic adjustment of the output direction instructions of the public evacuation guidance equipment: the purpose is to dynamically adjust the intensity of the guidance signal according to the congestion and panic level of the channel to attract attention and penetrate the chaotic environment;

[0170] "Ultimate Guidance Strength The calculation logic of " is based on an exponential growth model, specifically multiplying the "basic guidance intensity" by the power of "the natural constant e as the base and the product of the guidance intensity adjustment coefficient and the final behavioral risk indicator value" as the exponent. Among them, the "guidance intensity adjustment coefficient" is recorded as , Is a preset constant. The value 2 is used to control the rate at which the intensity increases with risk;

[0171] In this embodiment, the “basic guidance intensity” refers to the brightness or volume of the device under normal circumstances, which is a preset value;

[0172] The fourth step is to execute the dynamic direction command as follows:

[0173] Instruction parsing: The system parses the "optimal survival path" into specific turning instructions at each node in the target building topology diagram;

[0174] Device linkage: The parsed direction instructions are sent to the public evacuation guidance equipment (such as LED underground lights, electronic signs, and sound and light alarms) at the corresponding location. After receiving the instructions, the equipment will display the correct direction arrow and set its light brightness or alarm volume based on the calculated "final guidance intensity";

[0175] Technical effect: This embodiment quantifies the rescue priority index and final guide strength , upgrading information output from passive presentation to active, intelligent decision-making assistance and behavioral intervention;

[0176] The application-level parameter substitution content is as follows:

[0177] This example sets , , ;

[0178] 1. Rescue priority calculation scenario:

[0179] Escape route A: People are trapped, final behavior risk indicator value Indicates a slight panic, but the final structural risk indicator value of the floor ;

[0180] .

[0181] Escape route B: There are a large number of people gathered, and the final behavioral risk indicator value Characterizes panic congestion, but ultimately indicates structural risk Represents safety.

[0182] .

[0183] Technical Effect: Despite the greater panic in escape route B, the system calculated a higher rescue priority for people in escape route A. This is because the algorithm, by squared-processing the structural risk, correctly determined that the threat to lives in escape route A was more immediate and pressing. This allowed the fire commander to make the informed decision to prioritize rescuing people in area A.

[0184] 2. Guidance intensity adjustment scenario:

[0185] Escape route C: Located on the optimal path, but the crowd is orderly, and the final behavioral risk indicator value .

[0186] The amount of guide strength enhancement is small.

[0187] Escape route D: Also located on the optimal path, but the crowd is crowded and chaotic, and the final behavioral risk indicator value is .

[0188] The guidance strength was significantly enhanced by nearly 5 times, ensuring that its sound and light signals could penetrate the chaotic environment and be noticed by the panicked crowd.

[0189] In this embodiment, the rescue priority index is used to calculate The computational logic of the metric is derived from the multi-attribute utility function in decision theory, a common method for ranking multiple objectives by weighted combination. Nonlinear processing of risk values ​​is a common technique in risk management for assessing the impact of extreme risks. All inputs and outputs are dimensionless and consistent.

[0190] Used to calculate the final guide strength The calculation logic is based on the exponential growth model, which is often used to describe rapid growth caused by positive feedback effects under specific conditions. This example simulates the requirement that "the more chaotic the environment, the stronger the signal." The input is a dimensionless risk value, and the output is a final intensity value with the same dimensions as the base intensity. The logic is self-consistent.

[0191] Rescue Priority Index The value range of is limited to the range [0,1];

[0192] When the rescue priority index The closer the output is to 0, the lower the rescue urgency of the escape route area is determined by the system;

[0193] Reasoning process: Rescue priority index The closer it is to 0, the closer the results of its two weighted items are to 0. This means that the "final behavioral risk indicator value" of the area ” and “Final Structural Risk Indicator Value The smaller the overlay value of ” is.

[0194] When the rescue priority index The closer the output is to 1, the higher the system determines the rescue urgency of the escape route area is, and the easier it is to select it as the current top priority rescue target.

[0195] Reasoning process: Rescue priority index The closer it is to 1, the closer its weighted sum is to 1; due to the " is squared, and when the "final structural risk indicator value "The closer it is to 1, the more dangerous the structure is, even if the final behavioral risk indicator value is Smaller, rescue priority index It will get closer to , and superimpose the final behavioral risk indicator value Conversely, when the final behavioral risk indicator value The closer it is to 1, the more the "final structural risk indicator value" "The closer it is to 0, the higher the rescue priority index will approach This design ensures that the lethality of structural hazards has a higher inherent weight than behavioral hazards, making the rescue priority index It can most accurately reflect the true level of life threat.

[0196] Example 2:

[0197] To verify the effectiveness of the linkage control execution layer described in this invention under real-world disaster scenarios, a digital twin simulation test platform was constructed based on the China World Trade Center Tower A, a mid-rise commercial building in a major transportation hub. This platform pre-loaded the building's complete 3D Building Information Model (BIM) and integrated fire dynamics simulation software (FDS) with the complete system modules described in this invention. The test aimed to compare the performance of the evacuation system described in this invention (test group) with that of an evacuation system using a traditional fixed strategy (control group) under the same disaster scenario.

[0198] Test preparation:

[0199] 1. Hardware environment: Public evacuation guidance equipment is virtually deployed at key nodes of the building model, such as corridor intersections, stairwells, office exits, etc., including underground LED guide light strips with variable brightness and emergency broadcast speakers with variable volume. The brightness is 100 lux and the volume is 60 decibels. Firefighters are equipped with a virtual AR rescue terminal that can receive and display decision-making views.

[0200] 2. Software Environment: The experimental group fully deployed the linkage control execution layer algorithm described in the present invention, including a rescue priority index calculation module based on the formula "rescue priority index equals behavioral risk weight factor multiplied by the final behavioral risk indicator value, plus structural risk weight factor multiplied by the square of the final structural risk indicator value", and a guidance intensity adaptive adjustment module based on the formula "final guidance intensity equals basic guidance intensity multiplied by the product of the natural constant e's 'guidance intensity adjustment coefficient' and the final behavioral risk indicator value' power". Setting parameters: behavioral risk weight factor , structural risk weight factor , guidance strength adjustment coefficient .

[0201] 3. Control Group Settings: The control group system uses a simplified linear risk assessment model. Its rescue priority is determined solely by a linear composite risk equal to 0.5 times the final structural risk indicator value plus 0.5 times the final behavioral risk indicator value. Its public evacuation guidance equipment always operates at a fixed base intensity (100 lux / 60 dB) and does not undergo dynamic adjustment.

[0202] 4. Disaster scenario setting: Using fire dynamics simulation software, an electrical fire source is set in the east storage room on the 4th floor of "China World Trade Center Building A." Once the simulation begins, the fire will gradually spread, generating high temperatures and thick smoke, causing thermal damage to the building's load-bearing structure, and triggering panic and congestion among virtual personnel in different areas. The system collects data in real time through a simulated sensor network and calculates the final structural risk indicator value for each area. and final behavioral risk indicator value .

[0203] After the simulation began, the system entered real-time monitoring. 180 seconds after the fire broke out, when the fire reached its critical stage, the system recorded and analyzed parameters in six representative key areas. These areas were: the west office area on the 3rd floor (far from the fire source but affected by the evacuation flow); the east storage room on the 4th floor (the center of the fire, with severe structural damage); the central corridor on the 5th floor (directly above the fire source, filled with thick smoke and crowded); the server room on the 6th floor (a critical facility, structurally threatened by the fire); the conference center on the 7th floor (the starting point for large-scale evacuation); and the fire escape staircase on the 7th floor (an evacuation bottleneck).

[0204] For the experimental group, the linkage control execution layer performs the following operations at this point:

[0205] 1. For firefighters: Based on the real-time data collected in the six areas above, and The system then uses a nonlinear fusion formula to calculate the rescue priority index for each of the six areas. The system then sorts the six areas from high to low by their rescue priority index values. It then integrates this sorted list, the dynamically planned optimal survival path, and a real-time rendered three-color heat map of structural risk into a fused decision view, which is then pushed to the AR rescue terminal.

[0206] 2. For the public: For passages located on the optimal survival path, such as the central corridor on the 5th floor and the fire stairs on the 7th floor, the linkage control execution layer will The final guidance intensity required for adjustment is calculated using the exponential growth model Then, the system sends control instructions to the LED guide light strips and speakers in these areas to adjust their brightness / volume to the calculated value.

[0207] For the control group, their system also performed analysis at the same time, but their rescue priorities were based on a linear risk model and their guidance device strength remained unchanged. The technical advantages of the present invention were demonstrated by comparing the rescue priority ranking and guidance strength outputs generated by the two systems.

[0208] The performance verification data table of the linkage control execution layer is as follows:

[0209]

[0210] The data of the above embodiments clearly demonstrate the significant advantages and creativity of the linkage control execution layer of the present invention compared with traditional technologies.

[0211] 1. Advantages in rescue decision-making:

[0212] The linear risk model used by the control group ranked the 5th floor central corridor (0.55) as the first priority. This is an intuitive but wrong judgment because it only sees the panic of the crowd. , while ignoring a more deadly threat. If firefighters follow this instruction, they will prioritize an area that is congested but structurally safe.

[0213] In contrast, the rescue priority index RPI of the present invention is calculated by the final structural risk indicator value The square of the value greatly amplifies the risk weight of structural failure. The calculation results show that the rescue priority index RPI value of the "4th floor east side storage room" is as high as 0.526, and it is determined to be the first rescue priority. Although there are fewer people in this area and the panic level is low ( ), but its structure is on the verge of destruction ( ), personnel faced an immediate and fatal collapse threat. The present invention correctly identified this most pressing danger and directed firefighting and rescue forces to where they were most needed, avoiding significant casualties from structural collapse. This demonstrates a profound understanding of the underlying logic of disasters and exceptional decision-making support capabilities. This demonstrates the advanced nature of the present invention's nonlinear fusion algorithm and its immense value in actual rescue operations.

[0214] 2. Advantages in public evacuation guidance:

[0215] The intensity of the guidance equipment in the control group was always fixed at 100 lux / 60 dB. In an environment where the evacuation intensity is 0.9 and 0.8 respectively, this fixed, low guidance intensity is likely to be drowned out by the noise and chaos on site, resulting in the inability to effectively convey the evacuation instructions and low evacuation efficiency.

[0216] The final guide strength of the present invention Adaptive adjustment based on on-site risks has been achieved. In the "5th floor central corridor" where behavioral risk is the highest, the brightness of the guiding light strip has been increased to 605 lux, which is more than 6 times the basic brightness. At the "7th floor fire escape stairs" where there is also congestion, the brightness has also been increased to 495 lux. This exponential enhancement ensures that the guiding signal has sufficient penetration and can be clearly perceived by evacuees in extremely chaotic environments, thereby effectively guiding the flow of people and avoiding delays and secondary congestion caused by unclear instructions. In the "3rd floor west office area" ( ), the guidance intensity was only moderately increased to 149 lux, avoiding unnecessary energy consumption and excessive stimulation. This intelligent adjustment, tailored to local conditions, significantly improved the effectiveness of evacuation guidance and resource utilization.

[0217] In summary, this embodiment fully demonstrates through objective data comparison that the linkage control execution layer of the present invention can provide more scientific rescue decision support and more efficient public evacuation guidance, and has significant innovation and beneficial effects.

[0218] It should be noted that: All calculation formulas in this application document use regression analysis including but not limited to machine learning algorithms to deeply analyze the relevant parameters collected and identify their natural trends and relationships. Use professional software, such as Python's Scikit-learn library or R language, to automatically generate mathematical models that match the data. Then, objectively evaluate the performance of the model through methods such as cross-validation, and combine continuous feedback and optimization to ensure that the created formula truly reflects the inherent laws of the data, thereby ensuring its effectiveness and accuracy. In all calculation formulas in this application, the parameters in each formula are dimensionally non-dimensionalized within a consistent range to ensure that different physical quantities are compared on the same scale; dimensionless technical means include but are not limited to Min-Max-Normalization and Z-Score standardization;

[0219] The technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random-access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0220] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0221] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A building fire early warning and protection system based on an Internet platform and deep visual recognition, characterized in that: Specifically include: Structural status perception module: This module is used to monitor the actual vibration frequency and surface temperature of the load-bearing components corresponding to each escape route on the target building in real time under fire conditions, and analyze and process this data to generate a final structural risk indicator value representing the load-bearing components corresponding to each escape route; Environmental and behavior perception module: After receiving the final structural risk indicator value, it analyzes the surveillance video corresponding to each escape route in the target building to generate the local entropy value of the crowd and the safe passage degree of the route; Determining a final behavior risk indicator value and a final environment risk indicator value respectively by performing weighted fusion based on the credibility of visual data on the local entropy increment of the crowd and the safe passage degree of the channel; Optimal survival path generation module: used to receive a multidimensional risk vector data set consisting of the final structural risk indicator value, the final behavioral risk indicator value, and the final environmental risk indicator value, perform principal component analysis, and extract the first principal component as a dynamic comprehensive impedance representing the comprehensive risk level of each escape route; The dynamic comprehensive impedance is assigned as the travel cost to the corresponding path weight in the preset target building three-dimensional topology map, and the A* search algorithm is used to solve in real time the path with the lowest cumulative dynamic comprehensive impedance from any starting point to the safe exit as the optimal survival path; Linkage control execution layer: used to execute differentiated information output based on the optimal survival path. The information output includes the rescue priority index sent to the rescue terminal and the output direction instructions of the dynamic adjustment of the public evacuation guidance equipment.

2. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 1 is characterized by: Compare the actual vibration frequency with the pre-stored reference vibration frequency to calculate the structural natural vibration frequency offset; Based on the comparison relationship between the surface temperature and the sensor operating temperature threshold, a frequency data credibility characterizing the quality of the structural natural vibration frequency offset data on the load-bearing component is generated through a preset nonlinear function; When the credibility of the frequency data is lower than a preset compensation trigger threshold, the structural state perception module activates a preset thermal-mechanical coupling finite element model to generate a model-predicted frequency offset; The final structural risk indicator value is determined by weighted fusion of the structural natural vibration frequency offset and the model predicted frequency offset based on the credibility of the frequency data; The value range of the final structural risk indicator value is limited to the range of [0,1]; When the final structural risk indicator value output is closer to 0: the system determines that the structural state of the load-bearing components corresponding to the current escape route is safer and the collapse risk is lower; When the final structural risk indicator value output is closer to 1: the system determines that the load-bearing structure corresponding to the current escape passage is more severely damaged and faces a higher risk of collapse.

3. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 2 is characterized by: The local entropy increment of the crowd represents the risk level of panic congestion of the crowd in each escape channel; The "passage safety degree" is obtained by performing an exponential decay correction based on the final structural risk indicator value on the passable area of ​​the channel identified by the deep vision network; The crowd local entropy increment is obtained by calculating the Shannon entropy of the probability distribution of the crowd motion vector in the surveillance video image; The visual data credibility is generated by analyzing the smoke concentration of the surveillance video image; when the visual data credibility is lower than the preset threshold, the system will activate the crowd inference model to make compensatory estimates of the crowd status and channel status of the obscured area.

4. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 3 is characterized by: The crowd flow inference model calculates the crowd flow into all visible areas adjacent to the visually blocked area and performs flow-weighted averaging on the real-time crowd local entropy increments corresponding to each of the inflow crowd flows to determine the inferred crowd panic index of the visually blocked area. The crowd flow inference model updates the estimated total number of people in the visually obscured area based on the law of crowd flow conservation, and dynamically decreases the visual transparency of the last valid channel in the area before the visual obscuration according to the physical space occupied by the estimated total number of people, so as to determine the inferred channel state value of the visually obscured area; The crowd flow inference model is triggered and inferred as follows: when the "visual data credibility" is lower than the preset "visual compensation trigger threshold"; Limit the value range of the channel safety passability to the range of [0,1]; When the channel safety passability output is closer to 0, the system determines that the actual safe passability of the escape channel is smaller; When the channel safety passability output is closer to 1, the system determines that the actual safe passage capacity of the escape channel is greater.

5. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 4 is characterized by: In each calculation cycle, for each edge in the target building topology graph, the "final structural risk indicator value", "final behavioral risk indicator value" and "final environmental risk indicator value" are combined into a three-dimensional "multi-dimensional risk vector"; The system aggregates the "multi-dimensional risk vectors" generated by all escape routes over multiple consecutive time steps to form a dynamically updated risk dataset; Principal component analysis (PCA) is used to reduce the three-dimensional risk vector into a one-dimensional comprehensive impedance to standardize the multidimensional risk vector. Select the eigenvector corresponding to the maximum eigenvalue as the first principal component eigenvector; Perform a dot product between the standardized "multidimensional risk vector" and the selected first principal component eigenvector to obtain the "dynamic integrated impedance"; The calculated "dynamic integrated impedance" is used as the weight of the corresponding edge in the target building topology graph; the value range of "dynamic integrated impedance" is limited to the range of [0,1]; When the "dynamic integrated impedance" output is closer to 0, the system determines that the overall danger level of the escape route is lower, and it is more preferred to select it as an ideal safe path; When the "dynamic comprehensive impedance" output is closer to 1, the system determines that the comprehensive danger level of the escape route is higher, and the probability of being prohibited from passing is higher.

6. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 5 is characterized by: By nonlinearly fusing the final behavioral risk indicator values ​​and final structural risk indicator values ​​of each escape route, a rescue priority index is generated to characterize the urgency of rescuing personnel in each area. This rescue priority index, along with the optimal survival path and structural risk heat map, is then pushed to the rescue terminal of firefighters. The linkage control execution layer dynamically adjusts the output direction instructions of the public evacuation guidance equipment based on the real-time final behavioral risk indicator value of each escape channel on the optimal survival path. The output direction instructions achieve the adjustment effect of visual and / or auditory intensity to achieve adaptive control of guidance intensity.

7. The building fire early warning and protection system based on the Internet platform and deep visual recognition according to claim 6 is characterized by: Limit the value range of the rescue priority index to the range [0,1]; When the rescue priority index output is closer to 0, the system determines that the rescue urgency of the escape passage area is lower; When the rescue priority index output is closer to 1: the system determines that the rescue urgency of the escape passage area is higher, and it is easier to select it as the current top priority rescue target.

Images