Highway tunnel fire multi-source fusion detection method and system

Abstract

The invention discloses a highway tunnel fire multi-source fusion detection method and system, and relates to the technical field of tunnel fire detection, and the method comprises the steps: obtaining the real-time monitoring data of various types of fire alarm equipment in a tunnel; based on the real-time data and the historical verification data, calculating the comprehensive score of the health degree of each device to quantify the real-time reliability of the device; dividing dynamic parameter partitions according to a tunnel geographic structure and environmental characteristics, and configuring an independent fire detection basic parameter set for each partition; after receiving alarm data, determining a partition to which an event belongs, calling a corresponding parameter set, and dynamically correcting the reference weight according to the equipment health degree score to obtain a real-time expected weight; and calculating a comprehensive probability value of fire occurrence, comparing the comprehensive probability value with a partition judgment threshold value, and outputting a fire judgment result, and the system comprises a data access module and the like. According to the invention, collaborative management and control of equipment reliability and scene difference are realized, and robustness and accuracy of fire detection are improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel fire detection technology, specifically to a multi-source fusion detection method and system for highway tunnel fires. Background Technology

[0002] Highway tunnels are crucial nodes in modern transportation networks. Their relatively enclosed interiors and complex structures mean that fires can spread rapidly, posing a serious threat to lives and causing significant property damage. Therefore, early and accurate fire detection is the primary line of defense for ensuring safe tunnel operation. With technological advancements, single heat, smoke, or flame detectors are no longer sufficient to meet high reliability requirements. Employing multi-sensor data fusion technology, which integrates the advantages of various detection methods, has become the mainstream approach to improving the performance of tunnel fire detection systems.

[0003] Several studies and applications of multi-source fusion detection for tunnel fires exist in existing technologies. For example, invention patent CN103136893B discloses a tunnel fire early warning and control method and system based on multi-sensor data fusion technology. This technology deploys primary and secondary sensor groups and uses a composite neural network model to analyze historical and current environmental data, dynamically calculating and updating the fire judgment threshold to reduce false alarms and missed alarms caused by changes in the overall tunnel environment (such as diurnal temperature differences and seasonal temperature and humidity changes). Another related paper proposes a multi-sensor gradient data fusion model. This model sets up an intermediate station between the sensors and the fusion center, first preprocessing and performing local fusion based on least squares method on similar sensor data, and then using DS evidence theory for global fusion decision-making, aiming to reduce dependence on single sensor data and improve the overall accuracy of system identification.

[0004] However, the aforementioned existing technical solutions still suffer from a critical common problem when dealing with the special scenario of tunnels, which operate for long periods and face harsh environments: the systems lack the ability to coordinate and refine the time-varying degradation of the performance of sensing terminals (i.e., various fire alarm sensors) and the heterogeneity of the tunnel's internal spatial environment. Specifically, firstly, existing dynamic optimization strategies (such as neural networks adjusting thresholds) mainly focus on the impact of changes in macroscopic environmental parameters on "software decision-making," but neglect the irreversible degradation or sudden changes in the core performance of sensor hardware, such as detection sensitivity, response speed, and data reliability, under long-term exposure to exhaust gas pollution, dust adhesion, and moisture corrosion. A sensor with severely degraded performance will continue to contaminate the fusion results, even if its data is included in the fusion calculation. Existing systems cannot automatically identify and reduce the decision weight of such "unreliable data sources," leading to a decline in the overall reliability of the system over time. Secondly, the microenvironment (airflow, background temperature, interference sources) varies significantly between different sections inside the tunnel (such as the entrance section, the middle stable section, the exit section, the high-velocity section near the fan, and areas with fixed heat sources). Existing fusion parameters that are globally uniform or only slowly adjusted over time (such as fixed sensor weights or a single dynamic threshold) cannot adaptively match this spatial heterogeneity, which can easily lead to false alarms in complex environmental sections and delayed alarms in stable environmental sections due to insufficient sensitivity.

[0005] In summary, existing multi-source fusion detection technologies for tunnel fires, whether focusing on dynamically adjusting judgment thresholds through algorithms or optimizing the fusion method of multi-source data, are limited to the data and time dimensions in their core optimization. They fail to establish a closed-loop adaptive system that can simultaneously and proactively manage the reliability of front-end sensing devices and accurately adapt to the differences in back-end decision-making scenarios. This has become a key technical bottleneck restricting the long-term robustness, high accuracy, and strong environmental adaptability of tunnel fire detection systems. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a multi-source fusion detection method and system for highway tunnel fires. By constructing a comprehensive equipment health assessment mechanism to quantify the real-time reliability of equipment, combining tunnel structure and environmental characteristics to divide dynamic parameter zones and configure independent detection parameters, and dynamically adjusting weights based on equipment health and calculating the comprehensive probability of fire, this method can accurately adapt to the heterogeneity of the tunnel environment, avoid interference from equipment with degraded performance, achieve accurate early fire detection, and improve the long-term stability and reliability of the system.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, a multi-source fusion detection method for highway tunnel fires, the method comprising the following steps: Step 1: Obtain real-time monitoring data from various types of fire alarm devices deployed within the tunnel; Step 2: Based on the real-time monitoring data and historical verification data, calculate the comprehensive health score of each fire alarm device. The comprehensive health score is used to quantitatively characterize the real-time reliability of the device. Step 3: Based on the geographical structure and environmental characteristics of the tunnel, the tunnel is divided into multiple dynamic parameter zones, and an independent set of basic fire detection parameters is configured for each dynamic parameter zone. The basic parameter set includes the benchmark weights corresponding to the types of fire alarm devices and the fire determination probability threshold. Step 4: When fire alarm data is received, determine the dynamic parameter partition to which the alarm event belongs and call the corresponding basic parameter set; Step 5: Dynamically adjust the benchmark weights in the basic parameter set based on the real-time health comprehensive score of the alarm devices to obtain the real-time expected weights used for the current fusion calculation. Step 6: Calculate the comprehensive probability value of the current fire occurrence based on the real-time expected weights of all online alarm devices; Step 7: Compare the comprehensive probability value with the fire determination probability threshold corresponding to the dynamic parameter partition, and determine whether a fire has occurred based on the comparison result; Step 8: Output the fire determination result.

[0008] Furthermore, in step two, the overall health score of each fire alarm device is calculated, specifically using the following formula:

[0009] in, Indicates the first The overall health score of the fire alarm equipment in Taiwan. Indicates the first The detection accuracy rate of the equipment within the statistical period. Indicates the first The false alarm rate of the equipment within the statistical period. Indicates the first The average response delay of the devices within the statistical period. Indicates the first The data quality score for the equipment is based on a comprehensive evaluation of signal stability, data packet loss rate, and the proportion of invalid data output. , , , These are the weight coefficients of the corresponding indicators, and they satisfy... ; Among them, the detection accuracy pass The formula is used for calculation, where For the first The number of times a device triggered an alarm within the statistical period and was ultimately verified as a real fire. For the first The total number of alarms for the device within the statistical period; The false alarm rate pass The formula is used for calculation, where For the first The number of times a device alarms within the statistical period but is ultimately verified to be a non-fire event; The average response delay pass The formula is used for calculation, where This represents the total number of verified actual fire incidents within the statistical period. In the first The first fire incident Alarm time of the device In the first The earliest alarm time among all alarm devices in this fire incident.

[0010] Furthermore, in step two, after calculating the comprehensive health score, the method further includes implementing tiered management of the fire alarm equipment based on the comprehensive health score. This tiered management includes: Set a first health score and a high health score; If the overall health score of a device is greater than or equal to the high health score, then the baseline weight of the device is maintained or increased for participation in the fusion calculation. When the overall health score of a device falls between the first health score and the high health score, the baseline weight of the device is reduced proportionally. If the overall health score of a device is lower than the first health score, a progressive elimination operation is performed. The progressive elimination operation includes linearly reducing the effective weight of the device in the fusion calculation to zero over multiple consecutive detection cycles, and marking the device as an invalid device after the overall health score of the device is lower than the set failure risk score.

[0011] Furthermore, in step three, the division of dynamic parameter zones is based on tunnel structure zones, environmental risk zones, and fire safety zones. The tunnel structure is divided into three sections: entrance section, middle section, and exit section. Environmental risk zones are defined based on dust concentration distribution, background temperature distribution, and ventilation airflow intensity. Fire safety zoning is based on the fire protection unit division in the tunnel fire protection design code.

[0012] Furthermore, in step five, the baseline weights are dynamically adjusted based on the real-time health score of the alarm devices to obtain the real-time expected weights. In step six, the comprehensive probability value of a fire is calculated, specifically including: Determine the dynamic parameter partition to which the alarm event belongs, and obtain the baseline weight vector for that partition. Fire detection probability threshold ; Obtain the overall health score of all currently online alarm devices, and find the corresponding weight decay coefficient based on the overall health score; The benchmark weight vector Each component in the vector is multiplied by the weight decay coefficient of the corresponding device to generate the real-time expected weight vector. ; Calculate the overall probability value of a fire occurring Overall probability value The ratio of the sum of the real-time expected weights of all devices in alarm state to the sum of the real-time expected weights of all online devices is calculated using the following formula:

[0013] in, This indicates the real-time expected weight of the alarm status device. This represents the real-time expected weight of all online status devices.

[0014] On the other hand, a multi-source fusion detection system for highway tunnel fires, applicable to a multi-source fusion detection method for highway tunnel fires, comprises the following components: The data access module is used to receive and cache monitoring data streams uploaded by various types of fire alarm devices in real time; The equipment health assessment engine is used to continuously calculate and update the comprehensive health score of each fire alarm device based on the monitoring data stream and historical event database, and generate the equipment weight attenuation coefficient based on the score. The scene partitioning management engine is used to store and manage the partitioning information of tunnel dynamic parameter partitions, as well as the fire detection basic parameter set corresponding to each partition; The dynamic fusion decision core is used to obtain the basic parameter set corresponding to the event from the scene partition management engine after receiving alarm data, obtain the weight attenuation coefficient of the relevant equipment from the equipment health assessment engine, synthesize the real-time expected weight and calculate the comprehensive probability value of fire occurrence, and make a fire judgment by comparing the probability value with the threshold. The results output and feedback module is used to output the fire judgment results and related information, and to feed back the finally verified event results to the historical event database for updating the equipment health assessment.

[0015] Furthermore, the device health assessment engine includes a performance indicator tracking unit and a health calculation unit; The performance tracking unit maintains an independent performance profile for each device, recording detection accuracy, false alarm rate, response latency, and data quality indicators in real time. The health calculation unit calls the performance file at a preset period, calculates the health score based on the comprehensive health score formula, and outputs the corresponding device status label and weight decay coefficient according to the preset health interval mapping table.

[0016] Furthermore, the scene partitioning management engine includes a partitioning model database and a parameter set storage; The partition model database stores dynamic parameter partition digital models constructed based on tunnel construction information models, historical environmental monitoring data, and fluid dynamics simulation results; The parameter set storage stores a base parameter set that can be independently configured and updated for each partition. The base parameter set includes at least an equipment type baseline weight table and a partition fire determination probability threshold.

[0017] Furthermore, the dynamic fusion decision core includes an event locator, a parameter synthesizer, and a probability decision maker; The event locator queries the scene partition management engine based on the device physical location information of the alarm data to determine the dynamic parameter partition to which the event belongs; The parameter synthesizer obtains the basic parameter set according to the partition number and receives the weight decay coefficient sent by the device health assessment engine, and generates a real-time expected weight vector through weighted calculation. The probability decision-maker uses a synthesized real-time expected weight vector to calculate the comprehensive probability value according to the formula, and compares it with the fire determination probability threshold of the partition to generate the final fire determination instruction.

[0018] Furthermore, the result output and feedback module includes an instruction output interface and a learning feedback loop; The command output interface encapsulates the fire judgment command, the list of triggering devices and the event location partition information into a standard message and sends it to the tunnel monitoring platform and the emergency linkage control subsystem. The learning feedback loop receives the manual confirmation results from the monitoring platform or the final verification conclusion of the fire event provided by other verification subsystems. After associating the conclusion with the original alarm data, it archives it into the historical event database as the data basis for the equipment health assessment engine to perform the next round of performance index calculation.

[0019] Compared with existing technologies, this multi-source fusion detection method and system for highway tunnel fires has the following advantages: I. This invention establishes a comprehensive equipment health assessment mechanism, which quantitatively characterizes the real-time reliability of each fire alarm device by combining key indicators such as detection accuracy, false alarm rate, response delay, and data quality. At the same time, it divides dynamic parameter zones according to tunnel structural characteristics, environmental risks, and fire safety requirements, and configures independent benchmark weights and fire judgment thresholds for each zone. This achieves collaborative and refined control of the reliability management of front-end sensing devices and the differences in back-end decision-making scenarios, effectively avoids the interference of performance-degraded devices on the fusion detection results, solves the problem that globally unified parameters are difficult to adapt to the heterogeneity of the tunnel's internal spatial environment, and significantly improves the long-term stability and judgment accuracy of tunnel fire detection.

[0020] Second, this invention constructs a closed-loop operation system of data collection, health assessment, dynamic fusion, and result feedback. The final verification results of fire events are fed back into the historical event database to continuously optimize the calculation of equipment performance indicators and the configuration of zoning parameters, enabling the system to have adaptive iteration capabilities. At the same time, the system has a clear modular design, with each unit functioning independently and working together efficiently. It supports the dynamic access and flexible parameter adjustment of various fire alarm devices, without the need for large-scale renovation of existing tunnel facilities, reducing engineering implementation costs and thus adapting to highway tunnels of different sizes and environmental conditions.

[0021] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

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

[0023] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system architecture diagram of the present invention; Figure 3 This is a schematic diagram of the tunnel dynamic parameter zoning and system linkage of the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below. Example

[0025] like Figures 1 to 3 As shown, this embodiment uses a 1200-meter-long, two-way, four-lane highway tunnel as the application scenario. The tunnel has a stable geological structure and an average daily traffic flow of approximately 8000 vehicles. Various types of fire alarm equipment, including heat detectors, smoke detectors, flame detectors, and gas detectors, have been deployed inside the tunnel, and data transmission and power supply capabilities are available. The following detailed description of a multi-source fusion detection method and system for highway tunnel fires, based on specific embodiments, further illustrates this invention.

[0026] The specific implementation steps of the multi-source fusion detection method for highway tunnel fires in this embodiment are as follows: Step 1: Obtain real-time monitoring data from various types of fire alarm devices within the tunnel: During tunnel operation, the data access module continuously receives real-time monitoring data uploaded by various fire alarm devices deployed within the tunnel. In this embodiment, the fire alarm devices include the following types and monitoring parameters: Temperature sensors: A combination of distributed fiber optic temperature sensors and point-type temperature sensors is used. The distributed fiber optic temperature sensors are laid along the centerline of the tunnel top with a sampling interval of 1 meter to monitor the ambient temperature inside the tunnel. The measurement range is -20℃ to 120℃, and the data output resolution is 0.1℃. The point-type temperature sensors are laid at intervals of 50 meters to monitor the ambient temperature at the installation points. The measurement range is 0℃ to 100℃, and the data output resolution is 0.5℃.

[0027] Smoke detectors: Photoelectric smoke detectors are used, which are alternately deployed with point-type heat detectors at 50-meter intervals to monitor the smoke concentration in the tunnel. The optical density of smoke (OBS) is used as the monitoring parameter, with a measurement range of 0.01dB / m to 20dB / m and a data output resolution of 0.01dB / m.

[0028] Flame detectors: Four infrared flame detectors are deployed at the tunnel entrance, middle section and exit section to monitor flame radiation signals in the tunnel. The response wavelength range is 4.3μm, the detection angle is 120° and the monitoring distance is 0 to 50 meters.

[0029] Gas detectors: Carbon monoxide (CO) detectors are used, deployed at 100-meter intervals to monitor CO gas concentration in the tunnel. The measurement range is 0 ppm to 1000 ppm, and the data output resolution is 1 ppm.

[0030] All fire alarm devices establish communication connections with the data access module via industrial Ethernet, using the Modbus TCP protocol. Data uploads are completed in 1-second intervals to ensure real-time monitoring data. The data access module performs preliminary verification of the received monitoring data, including checking the data format validity and measurement range validity. Obviously abnormal data is removed, and valid data is cached locally for 7 days for subsequent comparison and analysis with historical data.

[0031] Step 2: Calculate the overall health score of each fire alarm device and implement tiered management: The core of this step is to evaluate the real-time reliability of the equipment by quantifying its detection accuracy, false alarm rate, average response latency, and data quality, thus providing a basis for subsequent weight adjustments.

[0032] In this embodiment, the statistical period for the overall equipment health score is set to 30 days, meaning that the equipment performance indicators are summarized and calculated every 30 days, and the overall health score is updated. This statistical period takes into account both the probability of tunnel fires and the timeliness of equipment performance changes, ensuring sufficient sample data to support indicator calculations while also reflecting the decline in equipment performance in a timely manner.

[0033] The overall health score of the equipment is calculated using the following formula:

[0034] in, : No. The overall health score of the fire alarm equipment ranges from 0 to 100. A higher score indicates stronger real-time reliability of the equipment, while a lower score indicates poorer equipment performance or a risk of failure.

[0035] : No. The normalized value of the detection accuracy of the equipment within the statistical period, ranging from 0 to 100. Detection accuracy is a representation of the equipment's ability to correctly identify real fire events; the original detection accuracy... Calculated using the following formula:

[0036] in For the first The number of times a device alarms within the statistical period and is ultimately verified as a real fire; For the first The total number of alarms for each device within the statistical period. The normalization method is as follows: ,make The value range is consistent with that of other indicators.

[0037] : No. The normalized value of the false alarm rate of the equipment within the statistical period, ranging from 0 to 100. The false alarm rate is a probability representation of equipment misclassifying a non-fire event as a fire; the original false alarm rate... Calculated using the following formula:

[0038] in For the first This refers to the number of times a device triggered an alarm within the statistical period but was ultimately verified as a non-fire event. Non-fire events include situations such as a sudden increase in dust concentration inside the tunnel, abnormal vehicle exhaust emissions, and accidental triggering of equipment malfunctions. The normalization method is as follows: .

[0039] : No. The normalized value of the average response delay of the equipment within the statistical period, ranging from 0 to 100. Average response delay is the time it takes for the equipment to issue an alarm signal after a fire occurs, reflecting the equipment's response speed. The original average response delay... Calculated using the following formula:

[0040] in, This represents the total number of verified actual fire incidents within the statistical period. If no actual fire incidents occurred within the statistical period, then... ,at this time Take the design response delay value of the device; For the first The first fire incident The alarm time of the device is the timestamp after the device detects that the fire characteristic parameters have reached the alarm threshold and issues an alarm signal. For the first The earliest alarm time among all alarm devices in a fire incident, i.e. the timestamp of the first alarm signal issued by all devices involved in the detection, is used to measure the speed of a single device's alarm response relative to the overall alarm response.

[0041] The normalization method is ,in The maximum allowable response delay for the equipment design is set at 10 seconds, taking into account the requirements of tunnel fire rescue. seconds, Follow Increases linearly; when seconds, The value is fixed at 100. At this point, the device response delay exceeds the safety requirements, resulting in the greatest negative impact on the health score.

[0042] : No. The data quality score for each device ranges from 0 to 100, with lower scores indicating better data quality. The score is based on a comprehensive evaluation of signal stability, data packet loss rate, and the proportion of invalid data output. The specific calculation method is as follows:

[0043] in, Signal stability is scored, with a value ranging from 0 to 100, calculated based on the fluctuation amplitude of the device's output signal. The fluctuation amplitude formula is as follows: ( This refers to real-time monitoring data of the equipment within the statistical period. (This is the average of the data). When hour ;when hour ;when hour ;when hour .

[0044] The data packet loss rate is scored, with a value ranging from 0 to 100. The packet loss rate formula is as follows: ( (This represents the number of data frames lost during the statistical period, and the total number of data frames the device should have sent during the statistical period.) hour ;when hour ;when hour ;when hour .

[0045] The invalid data output ratio score ranges from 0 to 100. The formula for the invalid data ratio is as follows: ( This refers to the number of invalid data frames output within the statistical period, such as data exceeding the device's measurement range, logical error data, etc. hour ;when hour ;when hour ;when hour .

[0046] , , , The weighting coefficients of the corresponding indicators are used to adjust the degree of influence of each indicator on the overall health score, to meet the following requirements. In this embodiment, based on the importance analysis of each performance indicator in the tunnel fire detection scenario, the value is set to... The values ​​are determined as follows: detection accuracy directly determines whether the equipment can effectively identify fires and is a core indicator of equipment reliability, therefore it is given the highest weight; excessively high false alarm rates lead to a waste of emergency resources, average response delay affects the efficiency of early fire response, and data quality affects the accuracy of fusion calculations; all three are secondary indicators and are given lower and equal weights. Those skilled in the art can adjust this weighting coefficient according to the actual tunnel environment and equipment type, as long as the sum is equal to 1.

[0047] Example of calculating overall health score: Taking the smoke detector numbered S01 in the tunnel as an example, the performance parameters within the statistical period (30 days) are as follows: Total number of alarms Second-rate; Actual number of fire alarms This was confirmed through manual verification and video playback. False alarm count The second incident was caused by a sudden increase in dust levels at the tunnel entrance. Actual fire incidents occurring within the statistical period Second-rate: In the first fire, the alarm time of S01 Earliest alarm time The response delay is 2 seconds; In the second fire, the alarm time of S01 Earliest alarm time The response delay is 1 second; Data quality related parameters: Signal fluctuation amplitude ; Data packet loss rate ; invalid data ratio .

[0048] Calculated based on the above parameters: Detection accuracy After normalization ; False alarm rate After normalization ; Average response delay Seconds, after normalization ; Data quality score: ; ; ; ; Overall health score: ; Tiered management based on comprehensive health scores: In this embodiment, the health score standard is set as follows: High health score: 85; First health score: 60; Failure risk score: 30.

[0049] The specific implementation method of hierarchical management is as follows: When the overall health score of a device is greater than or equal to 85, it indicates that the device is in optimal performance and has extremely high reliability. The device's baseline weight is maintained for participation in the fusion calculation. If the device's score is ≥90 for three consecutive statistical periods, its baseline weight can be increased by 10% to further highlight its role in the fusion calculation.

[0050] When the overall health score of the equipment is between 60 and 85, it indicates a slight decline in equipment performance, but it still meets basic testing requirements. The baseline weight of the equipment is then reduced proportionally. The weight reduction coefficient is calculated as follows: In other words, the attenuation coefficient is linearly positively correlated with the health score; the closer the score is to 85, the closer the attenuation coefficient is to 1, and the smaller the attenuation magnitude of the baseline weight. For example, the health score of the S01 detector mentioned above is 73, and the attenuation coefficient... Its base weight will be decayed according to this coefficient.

[0051] When a device's overall health score falls below 60, it indicates severe performance degradation or a malfunction, triggering a gradual removal process. Specifically, over three consecutive monitoring cycles (each cycle lasting one day), the device's effective weight in the fusion calculation is linearly reduced to zero. In the first monitoring cycle, the effective weight is the baseline weight × 0.6; in the second cycle, it's the baseline weight × 0.3; in the third cycle, it's the baseline weight × 0.1; and after the third cycle, the effective weight drops to zero. If the device's overall health score is below 30 (the failure risk score), it is directly marked as invalid and removed from the fusion calculation. A repair or replacement notification is then sent to management personnel via the tunnel monitoring platform.

[0052] Step 3: Divide the tunnel dynamic parameter partitions and configure the basic parameter set: Based on the tunnel's geographical structure and environmental characteristics, this step divides the tunnel into multiple dynamic parameter zones, with each zone configured with an independent set of basic fire detection parameters to adapt to the environmental heterogeneity of different sections.

[0053] The dynamic parameter zoning is based on tunnel structure zoning, environmental risk zoning, and fire safety zoning. These three are integrated to form the final dynamic parameter zoning, and the specific implementation is as follows: Tunnel structural zoning: Based on the tunnel's physical structure, the 1200-meter-long tunnel is divided into three sections: Entrance section: 0 meters to 250 meters. This section has frequent vehicle traffic, intense airflow disturbances, and large variations in environmental parameters. Middle section: 250 meters to 950 meters. This section is far from the entrance and exit, with a relatively stable environment and gentle airflow. It is the core passage section of the tunnel. Exit section: 950 meters to 1200 meters. Similar to the entrance section, there is significant airflow disturbance and large fluctuations in environmental parameters when vehicles exit.

[0054] Environmental risk zoning: Based on the distribution of dust concentration, background temperature, and ventilation airflow intensity within the tunnel, and through analysis of long-term environmental monitoring data (one year of history), environmental risk levels are classified. High-risk areas: entrance section (0 meters to 250 meters) and exit section (950 meters to 1200 meters). The average dust concentration in these areas is ≥3mg / m³, the background temperature fluctuation range is ≥8℃, the ventilation airflow intensity is ≥4m / s, and fire detection is easily affected by environmental interference. Medium-risk area: The middle section near the wind turbine (450 meters to 550 meters, 750 meters to 850 meters). In this area, the airflow intensity is ≥3m / s due to the operation of the wind turbine, and the background temperature fluctuates within a range of ≥5℃, which poses a certain risk of interference. Low-risk zone: The remaining areas in the middle section (250m to 450m, 550m to 750m, 850m to 950m), where the average dust concentration is ≤1.5mg / m³, the background temperature fluctuation range is ≤3℃, the ventilation airflow intensity is ≤2m / s, the environment is stable, and there is little interference.

[0055] Environmental parameters are monitored using the following equipment: Dust concentration monitoring: A laser dust sensor is used, with one unit deployed every 200 meters to monitor PM10 concentration; Background temperature monitoring: Utilize historical data from distributed fiber optic temperature sensors to statistically analyze the temperature fluctuation range of each section; Ventilation airflow intensity monitoring: Wind speed sensors are used, with one sensor installed every 300 meters at the top of the tunnel to monitor airflow speed.

[0056] Fire safety zones: According to the fire protection unit division requirements in the "Specifications for Design of Highway Tunnels" (JTGD70 / 2-2014), each 500 meters is divided into one fire protection unit. This tunnel is divided into three fire protection units: First fire protection unit: 0 meters to 500 meters; Second fire protection unit: 500 meters to 1000 meters; Third fire protection unit: 1000 meters to 1200 meters.

[0057] Final dynamic parameter partitioning: Combining the results of the three partitioning methods described above, the tunnel is ultimately divided into 5 dynamic parameter partitions: Zone 1: 0 meters to 250 meters, entrance section + high-risk zone + first fire protection unit; Zone 2: 250 meters to 500 meters, low-risk zone in the middle section + first fire protection unit; Zone 3: 500 meters to 850 meters, intermediate section, medium-risk zone + second fire protection unit; Zone 4: 850 meters to 1000 meters, low-risk zone in the middle section + second fire protection unit; Zone 5: 1000 meters to 1200 meters, exit section + high-risk area + third fire protection unit.

[0058] Configuration of the basic parameter set: Each dynamic parameter partition is configured with an independent set of basic fire detection parameters. The basic parameter set includes an equipment type baseline weight table and a fire determination probability threshold. The specific configuration is as follows: Equipment Type Baseline Weight Table: Baseline weights are determined based on the environmental characteristics and equipment detection performance of different zones, with a total weight of 1. Specific configuration: Zone 1 (Entrance Section + High-Risk Area): Heat Detector 0.3, Smoke Detector 0.25, Flame Detector 0.3, Gas Detector 0.15. This zone has a high dust concentration, making smoke detectors susceptible to interference; therefore, their weight is reduced. Flame detectors are less affected by dust and are assigned a higher weight.

[0059] Zone 2 (Middle Section, Low-Risk Area): Heat Detector 0.25, Smoke Detector 0.35, Flame Detector 0.25, Gas Detector 0.15. This area has a stable environment, and the smoke detector has the best detection efficiency, thus it is assigned the highest weight.

[0060] Zone 3 (Medium-risk area): Heat detector 0.3, smoke detector 0.3, flame detector 0.25, gas detector 0.15. Airflow intensity is high in this area; the effectiveness of heat and smoke detectors is comparable, and their weights are equally distributed.

[0061] Zone 4 (low-risk zone in the middle section): The configuration is the same as Zone 2, namely, heat detector 0.25, smoke detector 0.35, flame detector 0.25, and gas detector 0.15.

[0062] Zone 5 (Exit Section + High-Risk Area): The configuration is the same as Zone 1, namely, 0.3 heat detectors, 0.25 smoke detectors, 0.3 flame detectors, and 0.15 gas detectors.

[0063] The fire detection probability threshold is determined based on the fire risk level of the zone. The higher the risk, the lower the threshold, in order to improve fire detection sensitivity; the lower the risk, the higher the threshold, in order to reduce the false alarm rate. Zone 1 (High-risk area): 0.6; Zone 2 (Low-risk zone): 0.8; Zone 3 (Medium-risk area): 0.7; Zone 4 (Low-risk area): 0.8; Zone 5 (High-risk area): 0.6.

[0064] The basic parameter set is stored in the parameter set memory of the scene partition management engine, which supports administrators to manually update it according to changes in the tunnel environment, or to automatically optimize and adjust it through environmental monitoring data.

[0065] Step 4: Determine the dynamic parameter partition to which the alarm event belongs and call the basic parameter set: When the data access module receives alarm data from any fire alarm device, the event locator of the dynamic fusion decision core immediately parses the physical location information of the device in the alarm data. This information has been pre-stored in the device file, including the tunnel mileage marker where the device is installed.

[0066] Taking the flame detector numbered F03 as an example, its pre-stored installation location is 180 meters in the tunnel. The event locator queries the partition model database of the scene partition management engine and determines that 180 meters belongs to partition 1 (0 meters to 250 meters). Subsequently, the scene partition management engine returns the basic parameter set corresponding to partition 1 to the dynamic fusion decision core, including the benchmark weight table (temperature 0.3, smoke 0.25, flame 0.3, gas 0.15) and the fire judgment probability threshold of 0.6.

[0067] Step 5: Dynamically adjust the baseline weights to obtain the real-time expected weights: After receiving the basic parameter set of partition 1, the parameter synthesizer of the dynamic fusion decision core obtains the comprehensive health score and corresponding weight decay coefficient of all current online alarm devices from the device health assessment engine.

[0068] In this embodiment, when an alarm event is triggered, the online fire alarm devices in zone 1 include: 3 heat detectors (T01, T02, T03), 3 smoke detectors (S01, S02, S03), 2 flame detectors (F03, F04), and 1 gas detector (G01). The health score and attenuation coefficient of each device are as follows: T01: Health score 86, decay coefficient 1.0 (because ≥85); T02: Health score 78, decay coefficient 78 / 85≈0.918; T03: Health score 82, decay coefficient 82 / 85≈0.965; S01: Health score 73, decay coefficient 73 / 85≈0.859 (example calculation above); S02: Health score 80, decay coefficient 80 / 85≈0.941; S03: Health score 76, decay coefficient 76 / 85≈0.894; F03: Health score 88, decay coefficient 1.0; F04: Health score 83, decay coefficient 83 / 85≈0.976; G01: Health score 79, decay coefficient 79 / 85≈0.929.

[0069] The parameter synthesizer multiplies the baseline weights in the basic parameter set with the attenuation coefficients of the corresponding devices to generate the real-time expected weights for each device. T01: 0.3 × 1.0 = 0.3; T02: 0.3 × 0.918 ≈ 0.275; T03: 0.3 × 0.965 ≈ 0.290; S01: 0.25 × 0.859 ≈ 0.215; SO2: 0.25 × 0.941 ≈ 0.235; S03: 0.25 × 0.894 ≈ 0.224; F03: 0.3 × 1.0 = 0.3; F04: 0.3 × 0.976 ≈ 0.293; G01: 0.15 × 0.929 ≈ 0.139.

[0070] Therefore, the current real-time expected weight vector for partition 1 is: [0.3,0.275,0.290,0.215,0.235,0.224,0.3,0.293,0.139] (corresponding to the device sequence T01, T02, T03, S01, S02, S03, F03, F04, G01).

[0071] Step 6: Calculate the overall probability of the current fire occurring. The overall probability of a fire occurring is calculated as the ratio of the sum of the real-time expected weights of the alarm devices to the sum of the real-time expected weights of all online devices, using the following formula:

[0072] in, The sum of the expected weights of all devices in alarm state in real time; : The sum of the real-time expected weights of all online status devices.

[0073] In this embodiment, when an alarm event is triggered, the devices in alarm state within partition 1 are T01, S01, and F03, with corresponding real-time expected weights of 0.3, 0.215, and 0.3, respectively. Therefore:

[0074] The sum of the real-time expected weights of all online devices is:

[0075] Therefore, the overall probability of the current fire occurring is:

[0076] Step 7, Fire Determination: The probabilistic decision maker of the dynamic fusion decision core calculates the comprehensive probability value. The alarm was compared with the fire detection probability threshold of 0.6 corresponding to zone 1. Since 0.359 < 0.6, it was determined that no fire had occurred and the alarm was a false alarm.

[0077] If in another scenario, within partition 1, T01, T02, S01, S02, F03, and F04 simultaneously trigger alarms, the sum of the expected real-time weights of these alarm devices is approximately 0.3 + 0.275 + 0.215 + 0.235 + 0.3 + 0.293 ≈ 1.618, and the sum of the expected real-time weights of online devices remains 2.271. Then the overall probability value is... Since 0.713 > 0.6, a fire is determined to have occurred.

[0078] Step 8: Output the fire determination result: The command output interface of the results output and feedback module encapsulates the fire judgment result, the list of devices that triggered the alarm, and the information of the zone to which the event belongs into a standard message. The message adopts JSON format, and a specific example is as follows: { "decision":"fire", "zone":"Partition 1", "zone_range":"0-250m", "alarming_devices":[ {"id":"T01","position":"50m"}, {"id":"T02","position":"100m"}, {"id":"S01","position":"150m"}, {"id":"S02","position":"200m"}, {"id":"F03","position":"180m"}, {"id":"F04","position":"220m"} ], "probability": 0.713, "threshold":0.6, "timestamp":"2024-05-2010:05:30" } The standard message is sent to the tunnel monitoring platform and the emergency response control subsystem via industrial Ethernet. After receiving the message, the tunnel monitoring platform displays the fire alarm information, and the emergency response control subsystem activates the corresponding fire plan based on the zoning information.

[0079] Simultaneously, the learning feedback loop of the result output and feedback module receives manual confirmation results from the tunnel monitoring platform. This confirmation result is then correlated with the original alarm data, equipment health scores, and other information, and archived into the historical event database as the data basis for the next round of performance index calculations by the equipment health assessment engine. For example, if the current assessment determines it to be a real fire, and the alarm of device T01 is verified as valid, then device T01... The count is incremented by 1 and used for the detection accuracy calculation in the next statistical period.

[0080] The multi-source fusion detection system for highway tunnel fires in this embodiment is applicable to the above-described method embodiments. The system consists of a data access module, an equipment health assessment engine, a scene partitioning management engine, a dynamic fusion decision core, and a result output and feedback module. The specific implementation methods of each module are as follows: The data access module uses an industrial control computer with an Intel Core i7-12700 CPU, 16GB of memory, and a 1TB hard drive as its hardware platform. It is equipped with four Gigabit Ethernet ports, supporting simultaneous connection of up to 64 fire alarm devices. The module has built-in data receiving and caching software developed in C++ and implemented using the Qt framework.

[0081] Data Reception Process: The data access module establishes a long connection with each fire alarm device via the Modbus TCP protocol, sending a data request command every second. Upon response, the device returns monitoring data. The module performs format and range verification on the received data. Data that passes verification is marked as valid and stored in the local cache; data that fails verification is marked as invalid, and the device number, data content, and reason for failure are recorded for subsequent troubleshooting.

[0082] Caching Strategy: The module employs a circular caching mechanism, with the local cache capacity set to the total monitoring data for 7 days. When the cached data reaches its capacity limit, the oldest cached data is automatically overwritten. Cached data is stored categorized by device ID and timestamp, supporting queries of historical data by device ID and time range, providing data support for the device health assessment engine.

[0083] The device health assessment engine consists of a performance indicator tracking unit and a health calculation unit. The hardware is a server with an Intel Xeon E5-2680 CPU, 32GB of memory, and a 4TB hard drive. The software is developed using Python and implemented based on the Django framework.

[0084] Performance Tracking Unit: This unit establishes an independent performance profile for each fire alarm device. The performance profile includes basic device information and performance data. The performance data is updated in real time, including: Detection accuracy and false alarm rate: Updated after each alarm data is received and manually confirmed. , and It also calculates the detection accuracy and false alarm rate in real time for the current statistical period; Average response delay: After each actual fire event, the alarm time of all alarm devices is collected, the response delay of a single device is calculated, and the average response delay within the statistical period is updated. Data quality score: The signal fluctuation amplitude, data packet loss rate and invalid data ratio of the device are counted every hour, and the data quality score is calculated and updated.

[0085] The performance profile is stored in a relational database (MySQL 8.0), which supports CRUD operations on the data, making it easy for administrators to query the historical performance of the equipment.

[0086] Health Calculation Unit: This unit automatically retrieves performance profile data for each device according to a preset 30-day statistical period and calculates the overall health score for each device based on the comprehensive health score formula. After calculation, it queries a preset health interval mapping table and outputs the corresponding device status label ("Normal", "Decreased", "Failed") and weight decay coefficient. Health score ≥ 85: Status label "Normal", decay coefficient 1.0; 60≤HealthScore<85: Status label "Decrease", Decrease coefficient =HealthScore / 85; Health score <60: Status label "Invalid", decay coefficient dynamically adjusted according to progressive elimination rules.

[0087] The health calculation unit feeds back the calculation results to the performance profile, updates the device's health score, status label and attenuation coefficient, and synchronizes them to the dynamic fusion decision core to provide a basis for weight correction.

[0088] The scene partition management engine consists of a partition model database and a parameter set storage. The hardware is a server, and the software is developed in Java and implemented based on the Spring Boot framework.

[0089] Zoning Model Database: This database stores dynamic parameter zoning digital models constructed based on tunnel building information modeling (BIM), historical environmental monitoring data, and fluid dynamics simulation results. The tunnel BIM model includes information such as the tunnel's three-dimensional geometry, equipment installation locations, and fire protection unit divisions; the historical environmental monitoring data includes dust concentration, background temperature, and ventilation airflow intensity data for the past year; the fluid dynamics simulation uses ANSYS Fluent software to simulate airflow patterns within the tunnel, providing a theoretical basis for environmental risk zoning.

[0090] The partition model database uses PostgreSQL, which supports spatial data storage and querying. The event locator can quickly query the dynamic parameter partition to which it belongs by using the spatial coordinates of the device's installation location.

[0091] Parameter Set Storage: This storage partition stores an independent set of basic parameters for each dynamic parameter, including an equipment type baseline weight table and a fire determination probability threshold. The parameter sets are stored in XML format for easy reading and modification. An example of the storage structure is shown below: <zoneid="1"name="Zone 1"range="0-250m"> <base_weights> <thermo> 0.3< / thermo> <smoke> 0.25< / smoke> <flame> 0.3< / flame> <gas> 0.15< / gas> < / base_weights> <threshold> 0.6< / threshold> The parameter set memory supports two update methods: one is manual update, where administrators modify the base weights or decision thresholds through the parameter configuration interface of the tunnel monitoring platform; the other is automatic update. The system analyzes the environmental monitoring data every three months. If the average value of the environmental parameters in a certain zone changes by more than 20%, the base weights and decision thresholds of that zone will be automatically adjusted, and the adjustment range does not exceed ±0.1.

[0092] The dynamic fusion decision core is the core processing unit of the system, which consists of an event locator, a parameter synthesizer, and a probability discriminator. The hardware carrier is an industrial control computer, and the software is developed using the C++ language and implemented based on the Boost framework to ensure the real-time nature of data processing.

[0093] The core function of the event locator is to determine the dynamic parameter zone to which the alarm event belongs. When receiving alarm data, the event locator first analyzes the device number in the data, queries the installation location of the device from the performance profile of the device health assessment engine, and then matches the station number with the dynamic parameter zone range in the zone model database to determine the belonging zone and send the zone number to the parameter synthesizer.

[0094] The location response time of the event locator ≤ 100 milliseconds, ensuring the quick invocation of the corresponding basic parameter set without affecting the real-time nature of subsequent fusion calculations.

[0095] The parameter synthesizer obtains the base weight table of the zone from the parameter set memory of the scenario zone management engine according to the zone number sent by the event locator, and at the same time obtains the weight decay coefficients of all currently online devices from the device health assessment engine. The parameter synthesizer matches the base weights with the decay coefficients according to the device type, generates the real-time expected weight of each device through multiplication operation, and forms a real-time expected weight vector.

[0096] The parameter synthesizer supports the dynamic increase and decrease of online devices. When a device goes offline, it is automatically excluded from the real-time expected weight calculation; when the device comes back online, it is automatically included in the calculation.

[0097] The probability decision unit receives the real-time expected weight vector generated by the parameter synthesizer, calculates the sum of the real-time expected weights of devices in alarm state and the sum of the real-time expected weights of all online devices, and substitutes these sums into the comprehensive probability value formula to obtain the result. Subsequently, the probability decision maker retrieves the fire determination probability threshold for that partition from the parameter set memory, and... The fire determination command is generated by comparing the result with a threshold.

[0098] The probability decision-maker has a calculation response time of ≤200 milliseconds, ensuring that a rapid judgment can be made in the event of a fire, thus buying time for emergency response.

[0099] The result output and feedback module consists of an instruction output interface and a learning feedback loop. The hardware carrier is an industrial control computer, and the software is developed in C++ and implemented based on the Qt framework.

[0100] The command output interface encapsulates the fire judgment command generated by the probability decision unit, the list of devices triggering the alarm, the event's zone information, and the comprehensive probability value into a standard JSON message, which is then sent to the tunnel monitoring platform and the emergency response control subsystem via industrial Ethernet. The interface supports the TCP / IP protocol, with a data transmission rate of ≥100Mbps, ensuring fast and stable message transmission.

[0101] Meanwhile, the command output interface will log the judgment results locally. The log content includes the judgment time, partition information, alarm device list, comprehensive probability value, threshold, etc. The log is retained for 1 year to facilitate subsequent event tracing.

[0102] The learning feedback loop receives manual confirmation results from the tunnel monitoring platform or automatic verification results from other verification subsystems in real time, and associates these results with information such as original alarm data, equipment health scores, and comprehensive probability values ​​to generate event feedback records.

[0103] Event feedback records are archived to a historical event database (using MySQL 8.0), serving as the data foundation for the next round of performance metric calculations by the device health assessment engine. For example, if a manual verification confirms an alarm as a real fire, the corresponding device is updated. Count the numbers; if a false alarm is confirmed, update the corresponding device. Counting enables the system to achieve closed-loop self-learning.

[0104] This embodiment of the multi-source fusion detection method and system for highway tunnel fires achieves dynamic management of the reliability of front-end sensing equipment through equipment health assessment. By dynamically adapting parameters to the environmental heterogeneity within the tunnel, it effectively solves the problem of decreased detection reliability caused by neglecting equipment performance degradation and environmental differences in existing technologies. Practical application shows that this system can reduce the false alarm rate and false negative rate of tunnel fire detection, improve the long-term stability and accuracy of the system, and provide reliable protection for safe tunnel operation. Furthermore, the system's modules function independently yet work collaboratively, with a clear structure, easy deployment and maintenance, and good engineering practicality.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-source fusion detection method for highway tunnel fires, characterized in that, The method includes the following steps: Step 1: Obtain real-time monitoring data from various types of fire alarm devices deployed within the tunnel; Step 2: Based on the real-time monitoring data and historical verification data, calculate the comprehensive health score of each fire alarm device. The comprehensive health score is used to quantitatively characterize the real-time reliability of the device. Step 3: Based on the geographical structure and environmental characteristics of the tunnel, the tunnel is divided into multiple dynamic parameter zones, and an independent set of basic fire detection parameters is configured for each dynamic parameter zone. The basic parameter set includes the benchmark weights corresponding to the types of fire alarm devices and the fire determination probability threshold. Step 4: When fire alarm data is received, determine the dynamic parameter partition to which the alarm event belongs and call the corresponding basic parameter set; Step 5: Dynamically adjust the benchmark weights in the basic parameter set based on the real-time health comprehensive score of the alarm devices to obtain the real-time expected weights used for the current fusion calculation. Step 6: Calculate the comprehensive probability value of the current fire occurrence based on the real-time expected weights of all online alarm devices; Step 7: Compare the comprehensive probability value with the fire determination probability threshold corresponding to the dynamic parameter partition, and determine whether a fire has occurred based on the comparison result; Step 8: Output the fire determination result.

2. The multi-source fusion detection method for highway tunnel fires according to claim 1, characterized in that, In step two, the overall health score of each fire alarm device is calculated, specifically using the following formula: ； in, Indicates the first The overall health score of the fire alarm equipment in Taiwan. Indicates the first The detection accuracy rate of the equipment within the statistical period. Indicates the first The false alarm rate of the equipment within the statistical period. Indicates the first The average response delay of the devices within the statistical period. Indicates the first The data quality score for the equipment is based on a comprehensive evaluation of signal stability, data packet loss rate, and the proportion of invalid data output. , , , These are the weight coefficients of the corresponding indicators, and they satisfy... ; Among them, the detection accuracy pass The formula is used for calculation, where For the first The number of times a device triggered an alarm within the statistical period and was ultimately verified as a real fire. For the first The total number of alarms for each device within the statistical period; The false alarm rate pass The formula is used for calculation, where For the first The number of times a device alarms within the statistical period but is ultimately verified to be a non-fire event; The average response delay pass The formula is used for calculation, where This represents the total number of verified actual fire incidents within the statistical period. In the first The first fire incident Alarm time of the device In the first The earliest alarm time among all alarm devices in this fire incident.

3. The multi-source fusion detection method for highway tunnel fires according to claim 1, characterized in that, In step two, after calculating the comprehensive health score, the method further includes implementing hierarchical management of the fire alarm equipment based on the comprehensive health score. The hierarchical management includes: Set a first health score and a high health score; If the overall health score of a device is greater than or equal to the high health score, then the baseline weight of the device is maintained or increased for participation in the fusion calculation. When the overall health score of a device falls between the first health score and the high health score, the baseline weight of the device is reduced proportionally. If the overall health score of a device is lower than the first health score, a progressive elimination operation is performed. The progressive elimination operation includes linearly reducing the effective weight of the device in the fusion calculation to zero over multiple consecutive detection cycles, and marking the device as an invalid device after the overall health score of the device is lower than the set failure risk score.

4. The multi-source fusion detection method for highway tunnel fires according to claim 1, characterized in that, In step three, the division of dynamic parameter zones is based on tunnel structure zones, environmental risk zones, and fire safety zones. The tunnel structure is divided into three sections: entrance section, middle section, and exit section. Environmental risk zones are defined based on dust concentration distribution, background temperature distribution, and ventilation airflow intensity. Fire safety zoning is based on the fire protection unit division in the tunnel fire protection design code.

5. The multi-source fusion detection method for highway tunnel fires according to claim 1, characterized in that, In step five, the baseline weight is dynamically adjusted based on the real-time health score of the alarm device to obtain the real-time expected weight. In step six, the comprehensive probability value of a fire is calculated, specifically including: Determine the dynamic parameter partition to which the alarm event belongs, and obtain the baseline weight vector for that partition. Fire detection probability threshold ; Obtain the overall health score of all currently online alarm devices, and find the corresponding weight decay coefficient based on the overall health score; then, convert the benchmark weight vector... Each component in the vector is multiplied by the weight decay coefficient of the corresponding device to generate the real-time expected weight vector. ; Calculate the overall probability value of a fire occurring Overall probability value The ratio of the sum of the real-time expected weights of all devices in alarm state to the sum of the real-time expected weights of all online devices is calculated using the following formula: ； in, This indicates the real-time expected weight of the alarm status device. This represents the real-time expected weight of all online status devices.

6. A multi-source fusion detection system for highway tunnel fires, applicable to the multi-source fusion detection method for highway tunnel fires as described in any one of claims 1 to 5, characterized in that, The system consists of: The data access module is used to receive and cache monitoring data streams uploaded by various types of fire alarm devices in real time; The equipment health assessment engine is used to continuously calculate and update the comprehensive health score of each fire alarm device based on the monitoring data stream and historical event database, and generate the equipment weight attenuation coefficient based on the score. The scene partitioning management engine is used to store and manage the partitioning information of tunnel dynamic parameter partitions, as well as the fire detection basic parameter set corresponding to each partition; The dynamic fusion decision core is used to obtain the basic parameter set corresponding to the event from the scene partition management engine after receiving alarm data, obtain the weight attenuation coefficient of the relevant equipment from the equipment health assessment engine, synthesize the real-time expected weight and calculate the comprehensive probability value of fire occurrence, and make a fire judgment by comparing the probability value with the threshold. The results output and feedback module is used to output the fire judgment results and related information, and to feed back the finally verified event results to the historical event database for updating the equipment health assessment.

7. The multi-source fusion detection system for highway tunnel fires according to claim 6, characterized in that, The device health assessment engine includes a performance indicator tracking unit and a health calculation unit; The performance tracking unit maintains an independent performance profile for each device, recording detection accuracy, false alarm rate, response latency, and data quality indicators in real time. The health calculation unit calls the performance file at a preset period, calculates the health score based on the comprehensive health score formula, and outputs the corresponding device status label and weight decay coefficient according to the preset health interval mapping table.

8. The multi-source fusion detection system for highway tunnel fires according to claim 6, characterized in that, The scene partition management engine includes a partition model database and a parameter set storage; The partition model database stores dynamic parameter partition digital models constructed based on tunnel construction information models, historical environmental monitoring data, and fluid dynamics simulation results; The parameter set storage stores a base parameter set that can be independently configured and updated for each partition. The base parameter set includes at least an equipment type baseline weight table and a partition fire determination probability threshold.

9. A multi-source fusion detection system for highway tunnel fires according to claim 6, characterized in that, The dynamic fusion decision core includes an event locator, a parameter synthesizer, and a probability decision maker; The event locator queries the scene partition management engine based on the device physical location information of the alarm data to determine the dynamic parameter partition to which the event belongs; The parameter synthesizer obtains the basic parameter set according to the partition number and receives the weight decay coefficient sent by the device health assessment engine, and generates a real-time expected weight vector through weighted calculation. The probability decision-maker uses a synthesized real-time expected weight vector to calculate the comprehensive probability value according to the formula, and compares it with the fire determination probability threshold of the partition to generate the final fire determination instruction.

10. A multi-source fusion detection system for highway tunnel fires according to claim 6, characterized in that, The result output and feedback module includes an instruction output interface and a learning feedback loop; The command output interface encapsulates the fire judgment command, the list of triggering devices and the event location partition information into a standard message and sends it to the tunnel monitoring platform and the emergency linkage control subsystem. The learning feedback loop receives the manual confirmation results from the monitoring platform or the final verification conclusion of the fire event provided by other verification subsystems. After associating the conclusion with the original alarm data, it archives it into the historical event database as the data basis for the equipment health assessment engine to perform the next round of performance index calculation.

Images