Fire source identification and spatial positioning device and method in a cable tunnel

Abstract

The application relates to the technical field of power facility safety monitoring, in particular to a fire source identification and space positioning device in a cable tunnel, which comprises a multi-modal fire source detection module, a space positioning module, a fire prediction module and a main control communication module; the multi-modal fire source detection module, the space positioning module and the fire prediction module are bidirectionally connected with the main control communication module, and the main control communication module is wirelessly connected with a remote monitoring terminal. The long-wave infrared thermal imaging unit, the Mie scattering smoke detection unit and the electrochemical gas sensing unit are arranged to form a multi-modal detection architecture, and the Jousselme distance is introduced based on the improved D-S evidence theory to quantitatively measure the conflict degree between evidences, so that the problem that the decision reliability is reduced when the evidences conflict in the multi-sensor information fusion is effectively solved, and the fire false alarm rate and the fire missing alarm rate in the complex environment of high humidity, dust and the like in the cable tunnel are significantly reduced.

Description

Technical Field

[0001] This invention relates to the field of power facility safety monitoring technology, specifically to a device and method for fire source identification and spatial positioning in cable tunnels. Background Technology

[0002] With the continuous expansion of urban power grids, cable tunnels, as a crucial infrastructure for power transmission, are seeing their construction mileage and laying density increase year by year. Cable tunnels are enclosed, narrow spaces with limited ventilation. The densely packed cables operate under high loads for extended periods. In the event of a fire, the flames spread rapidly along the cable lines, and heat and toxic fumes accumulate quickly within the confined space, making firefighting extremely difficult and often resulting in widespread power outages and severe economic and social losses. Therefore, achieving early and accurate identification and rapid spatial location of fire sources within cable tunnels is of paramount importance for ensuring the security of urban power supply.

[0003] Currently, cable tunnel fire monitoring technology has the following main shortcomings.

[0004] Single-parameter detection is insufficient to cover all characteristic stages of fire development. Existing technologies mostly employ either temperature sensors or smoke sensors for monitoring. Temperature sensors are susceptible to interference from fluctuations in normal cable load and changes in ambient temperature, making them difficult to trigger effectively in the early stages of a fire when temperature rise is not yet significant. Smoke sensors are prone to false alarms in high-humidity, dusty environments like tunnels, and pose a risk of missing early-stage fires such as smoldering fires. For example, Chinese patent application CN120544357A discloses a method for early warning and control of cable tunnel fires, which establishes a three-dimensional model of the cable tunnel using a fire dynamics simulator (FDS) and deploys temperature measuring points. However, this method still relies primarily on temperature monitoring, limiting its ability to identify early-stage fires. Detection methods using a single physical quantity cannot comprehensively cover the characteristics of each stage of fire incubation, development, and spread, thus restricting the reliability and timeliness of fire early warning.

[0005] Multi-sensor fusion decision-making lacks an effective conflict resolution mechanism. To overcome the limitations of single-sensor detection, some technologies have attempted to use multi-sensor information fusion for fire detection. For example, Chinese patent application CN121389039A discloses a fire detection method and device based on the fusion of multimodal perception and DS evidence theory. This method collects video image data and temperature sensor data, generates basic probability assignments for each, and then uses DS evidence theory for fusion decision-making. However, this scheme only fuses video image and temperature sensor data. When the outputs of different sensors contradict each other (e.g., temperature rises but visual features are not obvious, or smoke appears but the temperature has not yet reached the threshold), the existing DS fusion framework lacks a quantitative measurement and dynamic adjustment mechanism for the degree of conflict between evidence. Conflicting evidence may dominate the fusion result, leading to misjudgment. Furthermore, this scheme is designed for general fire detection scenarios and is not specifically designed for the unique characteristics of the enclosed and narrow environment of cable tunnels.

[0006] Existing cable tunnel fire monitoring systems mostly rely on segment-level positioning triggered by a single sensor, with positioning accuracy only on the order of tens of meters, failing to provide the three-dimensional spatial coordinates of the fire source. Some technologies attempt to achieve target positioning using the TDOA (Time Difference of Arrival) positioning algorithm, or to locate abnormal heat sources using digital twin models combined with thermal anomaly functions, but these mostly employ a single ranging method. In narrow, enclosed spaces like cable tunnels, the multipath reflection of sound waves and temperature gradient effects significantly impact ranging accuracy, making it difficult for a single ranging method to meet the requirements of precise positioning. Although some technologies have proposed using the TDOA algorithm to achieve three-dimensional positioning through multi-node data fusion, the impact of tunnel ambient temperature changes on the accuracy of sound velocity ranging has not yet been resolved.

[0007] Most existing cable tunnel fire early warning systems can only provide alarms after a fire has occurred, and cannot predict the direction and speed of fire spread in real time. Some studies have used fire dynamics simulators (FDS) to create 3D models of cable tunnels for fire simulation; however, this type of numerical simulation method is computationally intensive and time-consuming, making it difficult to meet the timeliness requirements of real-time early warning. Other technologies use 3D models of utility tunnels for fire prediction simulation, but these also face the challenge of balancing real-time performance with computational accuracy.

[0008] Currently, fire monitoring equipment in cable tunnels is mostly deployed at fixed points. Due to limitations in tunnel length and sensor spacing, blind spots are difficult to avoid. Although there are technical solutions using inspection robots for cable tunnel inspection, existing solutions mainly focus on cable status identification and early warning, and have not yet integrated complete intelligent fire protection functions from fire source identification and spatial positioning to spread prediction. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a device and method for fire source identification and spatial positioning in cable tunnels.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a fire source identification and spatial positioning device for cable tunnels, comprising a multimodal fire source detection module, a spatial positioning module, a fire prediction module, and a main control communication module; the multimodal fire source detection module, the spatial positioning module, and the fire prediction module are bidirectionally connected to the main control communication module, and the main control communication module is wirelessly connected to a remote monitoring terminal; The multimodal fire source detection module is equipped with a multi-source fusion decision algorithm based on the improved DS evidence theory. It performs weighted evidence fusion of long-wave infrared thermal imaging radiation signal, Mie scattering smoke particle signal and electrochemical gas concentration signal, and outputs fire source identification results and confidence level. The spatial positioning module is equipped with a TDOA positioning algorithm based on acoustic and thermal joint ranging. It utilizes a dual-mode collaboration of distributed ultrasonic transducer array and thermal imaging pixel ranging, combined with tunnel environment temperature compensation, to output the three-dimensional spatial coordinates of the fire source. The fire prediction module is equipped with a fire spread extrapolation algorithm based on a thermal field-airflow coupling model. It incorporates the tunnel ventilation heat diffusion equation and the empirical spread rate formula to output the predicted fire spread direction and speed. The main control communication module is used for data scheduling, timing synchronization, and wireless communication with the remote monitoring terminal among the modules.

[0011] By setting up a multimodal fire source detection module, a spatial positioning module, and a fire prediction module, and achieving collaborative operation among the three based on a main control communication module, a full-chain intelligent fire early warning capability is formed, encompassing fire source identification, spatial positioning, and spread prediction. The multimodal detection architecture covers multidimensional characteristic signals at each stage of fire incubation, development, and spread, overcoming the limitations of detecting single physical quantities. The linkage response mechanism between the spatial positioning module and the fire prediction module enables the system to immediately initiate positioning and simulation after confirming the fire source, comprehensively improving the automation level and emergency response efficiency of cable tunnel fire prevention and control.

[0012] Preferably, the multimodal fire source detection module includes a long-wave infrared thermal imaging unit, a Mie scattering smoke detection unit, and an electrochemical gas sensing unit; the long-wave infrared thermal imaging unit operates in the 8–14 μm band and is used to acquire images of thermal radiation distribution within cable tunnels; the Mie scattering smoke detection unit adopts an infrared emitter-receiver pair structure and detects smoke particle concentration by forward-scattering light intensity; the electrochemical gas sensing unit includes a CO electrochemical sensor and a CO2 NDIR sensor.

[0013] By integrating three heterogeneous sensing methods—long-wave infrared thermal imaging (8–14 μm band), Mie scattering smoke detection, and electrochemical gas sensing—into a single module, simultaneous acquisition of heat radiation from the fire source, smoke particles, and combustion product gases is achieved. Long-wave infrared thermal imaging can penetrate smoke to obtain the distribution of high-temperature points even in smoke-covered environments; Mie scattering forward detection has high sensitivity to submicron-sized smoke particles; and electrochemical gas sensing has a response advantage to characteristic gases produced in the early stages of smoldering. These three methods complement each other, enabling the system to cover different fire development stages, including open flame, smoldering, and overheating, significantly broadening the response bandwidth of fire detection.

[0014] Preferably, the multi-source fusion decision algorithm based on the improved DS evidence theory includes: Define the recognition framework Θ={F,NF}, where F indicates the presence of a fire source and NF indicates the absence of a fire source; the basic probability allocation functions for each sensor are m_i(F), m_i(NF), and m_i(Θ), i=1,2,3, where m_i(Θ) is an uncertain allocation; Calculating the Jousselme distance measures the degree of conflict between pieces of evidence: d_J(m_p,m_q)=√[0.5·(m_p-m_q)ᵀ·D·(m_p-m_q)] Where D is the Jaccard similarity matrix, D(A,B)=|A∩B| / |A∪B|; Calculate the dynamic credibility weights of each evidence source based on Jousselme distance: w_i=[1-(1 / (n-1))·Σ_{j≠i}d_J(m_i,m_j)] / Σ_k[1-(1 / (n-1))·Σ_{j≠k}d_J(m_k,m_j)] The original basic probability assignment is weighted and corrected as follows: m'_i(A)=w_i·m_i(A), m'i(Θ)=1-Σ{A≠Θ}m'_i(A); The Dempster combination rule was used to perform pairwise fusion of the weighted and modified evidence: m(A)=[Σ_{B∩C=A}m'_p(B)·m'_q(C)] / (1-K) Where K = Σ_{B∩C=∅}m'_p(B)·m'_q(C) is the conflict coefficient; When the fusion result m(F)≥T_fire, it is determined that a fire source exists, where T_fire is the adaptive fire source determination threshold.

[0015] By introducing Jousselme distance to quantify the degree of conflict between pieces of evidence and dynamically assigning credibility weights to each evidence source accordingly, the influence of conflicting evidence in the fusion process is adaptively suppressed, effectively solving the problem of decreased decision reliability when evidence conflicts occur in multi-sensor information fusion. Compared to the simple threshold superposition or fixed weights in traditional DS fusion, this dynamic weight allocation mechanism can automatically reduce the weight of abnormal evidence when sensors are interfered with (such as a smoke sensor being falsely triggered in a dusty environment), preventing it from dominating the fusion result. This significantly reduces the false alarm and missed alarm rates of fires in complex environments such as high humidity and dust in cable tunnels.

[0016] Preferably, the TDOA positioning algorithm based on combined acoustic and thermal ranging includes: N positioning nodes are deployed at equal intervals on the top of the cable tunnel, where N≥3, and each positioning node integrates an ultrasonic transducer and a thermal imaging lens. Ultrasonic ranging uses the temperature compensation formula v_sound=331.3+0.606·T_c, where T_c is the tunnel ambient temperature, and the ultrasonic propagation distance d_ultra=v_sound·Δt / 2, where Δt is the ultrasonic transmission and reception time difference. The formula for thermal image pixel ranging is d_thermal=(f·S_real) / S_pixel, where f is the focal length of the thermal imager, S_real is the estimated actual size of the fire source, and S_pixel is the pixel size occupied by the fire source in the thermal image. Establish the system of equations for the TDOA hyperbola: √[(x-x_i) 2 +(y-y_i) 2 +(z-z_i) 2 ] - √[(x-x_j) 2 +(y-y_j) 2 +(z-z_j) 2 ] = c·Δt_ij, where (x_i,y_i,z_i) are the coordinates of the i-th positioning node, c is the speed of sound, and Δt_ij is the time difference between the arrival of the signal at the i-th node and the j-th node. The three-dimensional coordinates (x,y,z) of the fire source are obtained by using the Chan algorithm. Combined acoustic and thermal ranging data fusion: d_fused=(σ_u 2 ·d_thermal+σ_t 2 ·d_ultra) / (σ_u 2 +σ_t 2 ), whereσ_u 2 and σ_t 2 These are the variances of ultrasonic ranging and thermal imaging ranging, respectively.

[0017] By employing a dual-mode collaborative approach combining a distributed ultrasonic transducer array and thermal imaging pixel ranging, and incorporating real-time compensation for sound velocity based on tunnel ambient temperature, this invention overcomes the shortcomings of single ranging methods in terms of accuracy under the influence of tunnel multipath reflections and temperature gradients. After temperature compensation, the ultrasonic ranging error is reduced from ±3% to ±0.5%, while the thermal imaging pixel ranging achieves an accuracy of ±0.3m within a 10m range. The combined positioning error, after variance-weighted fusion of the two methods, is ≤0.5m. Compared to existing segment-level positioning (accuracy of tens of meters) or single TDOA positioning schemes, this invention achieves precise output of the fire source's three-dimensional spatial coordinates at the meter level, providing accurate location information for fire emergency response.

[0018] Preferably, the fire spread prediction algorithm based on the thermal field-airflow coupling model includes: Establish the heat diffusion equation for tunnel fires: T / t=α·∇ 2 T+Q̇ / (ρ·c_p)-v_wind· T / x Where α is the thermal diffusivity, Q̇ is the heat release rate of the fire source, ρ is the air density, c_p is the specific heat of air at constant pressure, and v_wind is the longitudinal wind speed of the tunnel; Heat release rate of ignition source in t 2 Growth model calculation: Q̇(t) = α_f·t 2 , where α_f is the fire growth coefficient; Formula for fire spread rate: v_f=k·(Q̇ / Q̇_0)^n·(1+β·v_wind)·exp(-γ·d_cable) Where k is a material constant, Q̇_0 is a reference heat release rate, n is a power exponent, β is a wind speed influence coefficient, γ is a cable spacing attenuation coefficient, and d_cable is the spacing between adjacent cables; The direction of spread is determined by the gradient of the thermal field distribution: θ_spread = arctan( T / y / T / x), where T / x and T / y represents the longitudinal and transverse temperature gradients of the tunnel, respectively.

[0019] By establishing equations including thermal diffusion and t 2The thermal-airflow coupling model, which combines the growth model and empirical spread rate formula, comprehensively considers the combined effects of multiple factors, such as tunnel longitudinal ventilation, cable spacing, and material properties, on fire spread. Compared to numerical simulation using a fire dynamics simulator (FDS) (which is computationally intensive and time-consuming), the analytical model of this invention can complete a single simulation calculation in seconds, achieving real-time prediction of fire spread direction and speed with a simulation error of ≤7%, providing forward-looking decision support for emergency evacuation route planning and fire-fighting resource allocation.

[0020] Preferably, the long-wave infrared thermal imaging unit calculates the target temperature according to the Stefan-Boltzmann law: M = ε·σ·T 4 Where M is the radiative exitance, ε is the emissivity, and σ = 5.67 × 10⁻⁶. -8 W / (m 2 ·K 4 ) is the Stefan-Boltzmann constant, and T is the absolute temperature; when the temperature exceeds the preset threshold T_alarm and the temperature rise rate exceeds dT / dt, infrared fire evidence m1 is generated; The formula for calculating the scattering cross section of the Mie scattering smoke detection unit is: σ_sca=(2π / k 2 )·Σ(2n+1)·(|a_n| 2 +|b_n| 2 ), where k=2π / λ is the wavenumber, and a_n and b_n are the Mie scattering coefficients; when the scattered light intensity exceeds the threshold V_smoke, smoke fire evidence m2 is generated; In the electrochemical gas sensing unit, the CO sensor has a range of 0–1000 ppm, and the CO2 sensor has a range of 0–5000 ppm; when the CO concentration exceeds 50 ppm or the CO2 concentration exceeds 1000 ppm, gas fire evidence m3 is generated.

[0021] The three sensors each provide clear technical basis and triggering conditions for generating fire source evidence. Infrared thermal imaging accurately converts radiation signals into temperature values ​​based on the Stefan-Boltzmann law, and combined with dual judgment conditions of temperature rise rate (temperature threshold 80℃ and temperature rise rate threshold 5℃ / min), it effectively avoids false triggering caused by normal load fluctuations in the cable. Mie scattering detection uses the scattering cross-section formula to achieve highly sensitive quantitative detection of submicron-sized smoke particles. The CO / CO2 gas threshold (50ppm / 1000ppm) is set based on the gas production characteristics of cable material combustion, and can capture gas signals in the early stage of smoldering. Each of the three evidence generation mechanisms has a clear physical basis and engineering adaptability.

[0022] Preferably, the main control communication module uses an ARM processor, runs a real-time operating system, and communicates with the remote monitoring terminal via a 5G or WiFi 6 wireless network.

[0023] By employing an ARM processor in conjunction with a real-time operating system, the timing determinism of data scheduling between modules is ensured, meeting the microsecond-level clock synchronization requirements for synchronous acquisition and fusion processing of multi-source sensor data; the 5G / WiFi 6 dual-mode wireless communication solution features high bandwidth and low latency, supporting a 10Hz burst data upload mode in emergency alarm conditions, ensuring the reliability of communication and the real-time performance of data transmission within the tunnel.

[0024] Preferably, the multimodal fire source detection module, the spatial positioning module, the fire prediction module, and the main control communication module are integrated and installed on a quadruped robot dog inspection platform, which autonomously walks and inspects the cable tunnel with the quadruped robot dog.

[0025] By integrating the aforementioned functional modules into a quadruped robot dog inspection platform, a seamless combination of mobile autonomous inspection and fixed-node TDOA positioning is achieved. The robot dog continuously collects multimodal sensor data while autonomously walking along the tunnel, overcoming the limitations of coverage blind spots inherent in fixed-point monitoring equipment. Simultaneously, the positioning node is fixedly deployed at the top of the tunnel, and the robot dog only triggers positioning measurements after a fire source is confirmed, ensuring positioning accuracy while avoiding interference from the mobile platform's own noise on TDOA measurements.

[0026] Preferably, the positioning nodes are deployed every 15m to 25m along the top of the cable tunnel. Each positioning node is connected to the main control communication module via wired or wireless means to collaboratively complete the TDOA measurement. The fire prediction module updates the thermal field distribution with a 5s cycle, predicts the fire spread trend in the next 60s, and uploads the prediction results to the remote monitoring terminal in real time through the main control communication module.

[0027] By limiting the spacing between positioning nodes (15m~25m) and the thermal field update cycle (5s), a reasonable balance is achieved between system deployment cost and positioning accuracy. This spacing ensures the number of nodes and geometric configuration required to solve the TDOA hyperbolic equation system while controlling hardware deployment costs. The 5s thermal field update cycle and 60s extrapolation time can complete the prediction of fire trend within the next minute with a single calculation time of about 50ms, meeting the engineering requirements of timeliness for emergency response.

[0028] This invention also discloses a method for fire source identification and spatial positioning in cable tunnels, using the aforementioned device, comprising the following steps: S1: Collect long-wave infrared thermal radiation signals, Mie scattering smoke particle signals and electrochemical gas concentration signals through the multimodal fire source detection module, and generate infrared fire evidence m1, smoke fire evidence m2 and gas fire evidence m3 respectively; S2: Using the multi-source fusion decision algorithm based on the improved D-S evidence theory, calculate the Jousselme distance between each piece of evidence and assign dynamic credibility weights accordingly, perform weighted correction on the original basic probability assignment, then use the Dempster combination rule for pairwise fusion, and determine that a fire source exists and output the confidence level when the fusion result m(F)≥T_fire; S3: After the presence of the fire source is determined, activate the spatial positioning module, each positioning node measures the ultrasonic propagation distance and the thermal image pixel ranging respectively, performs variance weighted fusion on the two ranging results, establishes TDOA hyperbolic equations and uses the Chan algorithm for solution, and outputs the three-dimensional spatial coordinates of the fire source; S4: The fire situation prediction module receives the temperature field data and fire source coordinates, calculates the fire spread speed and direction through the thermal diffusion equation and the fire spread rate formula based on the thermal field-airflow coupling model, and continuously updates the deduction results at a set period; S5: The main control communication module synchronously uploads the fire source identification result, confidence level, three-dimensional spatial coordinates, fire spread direction and speed to the remote monitoring terminal through the wireless network.

[0029] Steps S1 to S5 define the complete workflow from multi-source signal collection, evidence fusion determination, three-dimensional coordinate positioning to spread prediction and reporting, and a strict data flow dependency relationship is formed between each step — identification and confirmation is the precondition for starting positioning, the positioning coordinates are the input parameters for spread deduction, and the final results are aggregated to the main control module for unified reporting. This process ensures that the system maintains a low-power patrol mode (only the multimodal detection module works) when there is no fire, and quickly switches to the full-module linkage state after the fire source is confirmed, taking into account both system energy consumption and emergency response speed.

[0030] Compared with existing technologies, this invention has the following advantages: First, by setting up a multimodal detection architecture consisting of a long-wave infrared thermal imaging unit, a Mie scattering smoke detection unit, and an electrochemical gas sensing unit, and by introducing Jousselme distance based on an improved DS evidence theory to quantify the degree of conflict between evidence, the credibility weight of each evidence source is dynamically allocated accordingly. This adaptively suppresses the influence of conflicting evidence during the fusion process, effectively solving the problem of decreased decision reliability when evidence conflicts occur in multi-sensor information fusion, and significantly reducing the false alarm and missed alarm rates of fires in complex environments such as high humidity and dust in cable tunnels. Second, by using a distributed ultrasonic transducer array and thermal imaging pixel ranging in a dual-mode collaboration, and by introducing tunnel ambient temperature to compensate for sound velocity in real time, the ultrasonic ranging and thermal imaging ranging are fused using variance weighting. This overcomes the shortcomings of single ranging methods in terms of insufficient accuracy under the influence of tunnel multipath reflection and temperature gradient, achieving precise output of the fire source's meter-level three-dimensional spatial coordinates, providing accurate location information for fire emergency response. Third, by establishing a system including the thermal diffusion equation, t 2 The growth model and empirical spread rate formula of the thermal field-airflow coupling model comprehensively consider the coupling effects of multiple factors such as tunnel longitudinal ventilation, cable spacing and material properties. It can predict the speed and direction of fire spread in real time at the second level, providing forward-looking decision support for emergency evacuation route planning and fire fighting resource allocation. Fourth, the above modules are integrated into the quadruped robot dog inspection platform, realizing the organic combination of mobile autonomous inspection and fixed node TDOA positioning, eliminating the coverage blind spots of fixed point monitoring, forming a full-chain intelligent fire early warning capability from fire source identification, spatial positioning to spread prediction, and comprehensively improving the fire prevention and control level of cable tunnels. Attached Figure Description

[0031] Figure 1 This is an overall architecture diagram of the fire source identification and spatial positioning device in cable tunnels according to the present invention; Figure 2 This is a cross-sectional view of the sensor integration structure of the multimodal fire source detection module of the present invention; Figure 3 This is a schematic diagram of the TDOA spatial positioning principle of the spatial positioning module of the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0033] like Figure 1As shown, the fire source identification and spatial positioning device in the cable tunnel consists of a multimodal fire source detection module, a spatial positioning module, a fire prediction module, and a main control communication module. The multimodal fire source detection module, spatial positioning module, and fire prediction module are bidirectionally connected to the main control communication module to achieve data exchange and the issuance of control commands. The main control communication module interacts with a remote monitoring terminal via a wireless network.

[0034] In this embodiment, the four modules are integrated and mounted on the back support plate of a quadruped robot dog inspection platform. The robot dog walks autonomously along the cable tunnel, and its walking speed can be adjusted according to the environmental conditions inside the tunnel. The main control communication module uses an ARM Cortex-A72 processor as the main control chip, runs a Linux real-time operating system, and is responsible for data scheduling, timing synchronization, and communication management of each module. The main control communication module communicates with the remote monitoring terminal via a 5G or WiFi 6 wireless network.

[0035] like Figure 2 As shown, the multimodal fire source detection module integrates a long-wave infrared thermal imaging unit, a Mie scattering smoke detection unit, and an electrochemical gas sensing unit within a cylindrical protective housing. The protective housing is made of stainless steel with an IP67 protection rating. A germanium glass observation window is located on the front side, and a waterproof aviation connector is located on the rear side for power supply and signal transmission.

[0036] The long-wave infrared thermal imaging unit employs an uncooled microbolometer focal plane array detector, operating in the 8–14 μm band, with a resolution of 640 × 512, NETD ≤ 50 mK, a frame rate of 30 fps, and a focal length of 25 mm. The detector is mounted at the front of the protective housing and acquires images of the thermal radiation distribution within the cable tunnel through a germanium glass window. Temperature calculations are based on the Stefan-Boltzmann law. M=ε·σ·T 4 Where M is the radiative exitance, ε is the surface emissivity of the cable (0.92 for PVC cables), and σ = 5.67 × 10⁻⁶. -8 W / (m 2 ·K 4 () represents the Stefan-Boltzmann constant, and T represents the absolute temperature. The system is calibrated in a blackbody furnace before leaving the factory to establish a mapping curve between the radiation signal and temperature; the calibration data is stored in the processor's Flash memory. The temperature threshold T_alarm is set to 80℃, and the temperature rise rate threshold dT / dt is set to 5℃ / min. When the thermal imaging unit detects that the temperature exceeds T_alarm and the temperature rise rate exceeds dT / dt, infrared fire evidence m1 is generated. m1(F)=min[1, (T-T_alarm) / (T_max-T_alarm)·(dT / dt) / (dT / dt)_ref] m1(NF) = 1 - m1(F) - m1(Θ) Where T_max is the upper limit reference value of temperature, (dT / dt)_ref is the reference value of temperature rise rate, and m1(Θ) is determined by the nominal accuracy of the detector, with a value of 0.05 to 0.10.

[0037] The Mie scattering smoke detection unit employs an 850nm infrared emitter and a photodetector arranged opposite each other, with a scattering chamber in between. The emitter is driven by a constant current circuit, and the receiver is followed by a transimpedance amplifier and a bandpass filter. The output signal is acquired by an ADC and then sent to a processor. When smoke particles enter the scattering chamber, the forward scattered light intensity is proportional to the particle concentration. The Mie scattering cross-section is calculated using the following formula: σ_sca=(2π / k 2 )·Σ(2n+1)·(|a_n| 2 +|b_n| 2 ) Where k = 2π / λ is the wavenumber, and a_n and b_n are the Mie scattering coefficients. The scattered signal, after smoothing and filtering, is compared with the background baseline value, which is automatically obtained during power-on calibration and updated every 30 minutes. The scattered light intensity threshold V_smoke is set to 1.5 times the baseline value. When the scattered light intensity exceeds V_smoke and lasts for more than 1 second, smoke fire evidence m2 is generated. m2(F)=min[1, (V_sca-V_smoke) / (V_sat-V_smoke)] m2(NF) = 1 - m2(F) - m2(Θ) V_sat is taken as 5 times the baseline value, and m2(Θ) is determined by the level of environmental interference, with a value of 0.10 to 0.15.

[0038] The electrochemical gas sensing unit includes a CO electrochemical sensor (range 0–1000 ppm, resolution 1 ppm, response time T90 ≤ 30 s) and a CO2 NDIR sensor (range 0–5000 ppm, resolution 10 ppm, response time T90 ≤ 60 s). Both sensors are mounted at the rear of the protective housing and communicate with the external environment through a waterproof and breathable membrane. The sensors sample every 2 seconds, and the sampled values ​​are filtered by median (window length 5) before being entered into the discrimination logic. When the CO concentration exceeds 50 ppm or the CO2 concentration exceeds 1000 ppm, gas fire evidence m3 is generated. m3(F)=min[1, max((C_CO-50) / 450, (C_CO2-1000) / 4000)] m3(NF) = 1 - m3(F) - m3(Θ) m3(Θ) takes values ​​from 0.08 to 0.12.

[0039] Define the recognition framework Θ={F,NF}, where F indicates the presence of a fire source and NF indicates the absence of a fire source. The basic probability assignments generated by the three sensors are m_i(F), m_i(NF), and m_i(Θ) (i=1,2,3), where m_i(Θ) represents the uncertainty assignment. After the three types of sensor evidence are generated, a fusion decision is performed according to the following steps: The first step is to calculate the Jousselme distance to measure the degree of conflict between pieces of evidence: d_J(m_p,m_q)=√[0.5·(m_p-m_q)ᵀ·D·(m_p-m_q)] Where D is the Jaccard similarity matrix, D(A,B)=|A∩B| / |A∪B|. The larger this distance, the more serious the conflict between the two pieces of evidence.

[0040] The second step is to calculate the dynamic credibility weights of each evidence source based on the Jousselme distance: w_i=[1-(1 / (n-1))·Σ_{j≠i}d_J(m_i,m_j)] / Σ_k[1-(1 / (n-1))·Σ_{j≠k}d_J(m_k,m_j)] When a sensor is interfered with, causing its evidence to conflict significantly with other evidence, its weight is automatically reduced to prevent abnormal evidence from dominating the fusion result.

[0041] The third step is to perform a weighted correction on the original basic probability allocation: m'_i(A) = w_i·m_i(A) m'i(Θ)=1-Σ{A≠Θ}m'_i(A) The uncertainty allocation of each piece of evidence was renormalized after the correction.

[0042] The fourth step involves using Dempster's combination rule to perform pairwise fusion of the weighted and modified evidence: m(A)=[Σ_{B∩C=A}m'_p(B)·m'_q(C)] / (1-K) K=Σ_{B∩C=∅}m'_p(B)·m'_q(C) Where K is the conflict coefficient. The three sets of corrected evidence are then fused pairwise in turn to obtain the final fusion result.

[0043] Fifth step, determination: When the fusion result m(F)≥T_fire, output the fire source existence signal. T_fire is the adaptive fire source determination threshold, which is adaptively selected in the range of 0.60 to 0.75 according to the tunnel environment conditions.

[0044] The technical specifications of each sensor in the multimodal fire source detection module are shown in Table 1.

[0045] Table 1 Technical Specifications of Multimodal Sensors

[0046] The DS evidence fusion decision rules are shown in Table 2.

[0047] Table 2 DS Evidence Fusion Decision Rules

[0048] like Figure 3 As shown, the spatial positioning module employs a TDOA positioning algorithm based on combined acoustic and thermal ranging. A positioning node is deployed every 20m at the top of the cable tunnel, for a total of N nodes (N≥3, N=5 in this embodiment). Each node integrates a 40kHz ultrasonic transducer and an infrared thermal imaging lens. Each positioning node is connected to the main control communication module via an RS-485 bus to collaboratively complete TDOA measurements.

[0049] Ultrasonic ranging uses temperature compensation, and the velocity of sound is calculated using the following formula: v_sound = 331.3 + 0.606·T_c Where T_c represents the tunnel ambient temperature, which is synchronously acquired by the thermal imaging unit. The ultrasonic wave propagation distance is: d_ultra=v_sound·Δt / 2 Where Δt is the ultrasonic transceiver time difference, measured by a timer. Since the ambient temperature in cable tunnels is typically between 15 and 40°C, temperature compensation can reduce the ranging error from ±3% to ±0.5%.

[0050] Thermal image pixel ranging is calculated using the focal length and pixel size of the infrared thermal imager: d_thermal=(f·S_real) / S_pixel Where f is the focal length of the thermal imager (25mm), S_real is the estimated actual size of the fire source (calculated based on the temperature distribution pattern of the thermal image), and S_pixel is the pixel size occupied by the fire source in the thermal image. For a thermal imager with a resolution of 640×512 and a focal length of 25mm, the positioning accuracy at a distance of 10m can reach ±0.3m.

[0051] The fusion of acoustic and thermal ranging data employs a variance-weighted method: d_fused=(σ_u 2 ·d_thermal+σ_t 2 ·d_ultra) / (σ_u 2 +σ_t 2 ) whereσ_u2 and σ_t 2 These represent the variances of ultrasonic ranging and thermal ranging, respectively. The ranging accuracy of the fused method is superior to that of either method alone.

[0052] TDOA positioning establishes a system of hyperbolic equations: √[(x-x_i) 2 +(y-y_i) 2 +(z-z_i) 2 ] - √[(x-x_j) 2 +(y-y_j) 2 +(z-z_j) 2 The equation is: ] = c·Δt_ij, where (x_i, y_i, z_i) are the coordinates of the i-th location node, c is the speed of sound, and Δt_ij is the time difference between the arrival of the fire sound / heat signal at the i-th node and the j-th node. The Chan algorithm is used to solve this nonlinear system of equations. First, the hyperbolic equation is linearized, and an initial solution is obtained using the weighted least squares method. Then, iterative optimization is performed to finally obtain the three-dimensional coordinates (x, y, z) of the fire source.

[0053] The positioning accuracy test results are shown in Table 3.

[0054] Table 3 Positioning accuracy test results

[0055] The fire prediction module is equipped with a fire spread prediction algorithm based on a thermal field-airflow coupling model. The module receives temperature field data from the multimodal fire source detection module and fire source coordinates from the spatial positioning module, and combines this with tunnel ventilation parameters (obtained by a wind speed sensor mounted on a quadruped robot dog) to predict the direction and speed of fire spread in real time.

[0056] Establish the heat diffusion equation for tunnel fires: T / t=α·∇ 2 T+Q̇ / (ρ·c_p)-v_wind· T / x Where α is the thermal diffusivity (for air, α ≈ 2.2 × 10⁻⁶). -5 m 2 / s), Q̇ is the heat release rate of the fire source, and ρ is the air density (≈1.2kg / m³). 3 c_p is the specific heat of air at constant pressure (≈1005J / (kg·K)), and v_wind is the longitudinal wind speed in the tunnel.

[0057] Heat release rate of ignition source in t 2 Growth model calculation: Q̇(t)=α_f·t2 Where α_f is the fire growth coefficient. For PVC insulated cables in cable tunnels, α_f is taken as 0.046~0.188kW / s 2 .

[0058] Empirical formula for fire spread rate: v_f=k·(Q̇ / Q̇_0)^n·(1+β·v_wind)·exp(-γ·d_cable) Where k is the material constant (0.012 m / s for PVC cables), Q̇_0 is the reference heat release rate (500 kW), n is the power exponent (0.6), β is the wind speed influence coefficient (0.08 s / m), and γ is the cable spacing attenuation coefficient (0.5 m). -1 ), d_cable is the spacing between adjacent cables (typical value 0.05~0.15m).

[0059] The direction of propagation is determined by the thermal field distribution gradient: θ_spread=arctan( T / y / T / x) in T / x and T / y represents the longitudinal and transverse temperature gradients of the tunnel, respectively, obtained through differential calculation of the thermal imaging temperature field. The system updates the thermal field distribution with a period of 5 seconds to predict the fire spread trend within the next 60 seconds.

[0060] The parameters for predicting the spread of fire are shown in Table 4.

[0061] Table 4 Fire spread prediction parameters

[0062] The main control communication module uses an ARM Cortex-A72 processor and runs a Linux real-time operating system. It communicates with the remote monitoring terminal via a 5G or WiFi 6 wireless network. This module is responsible for data scheduling and timing synchronization between modules, coordinating the sampling triggering of the multimodal fire source detection module, the TDOA measurement initiation of the spatial positioning module, and the temperature field data acquisition of the fire prediction module. All data uploaded by each module is in JSON format, including timestamps, module IDs, data types, and data content.

[0063] This device is mounted on a quadruped robot dog inspection platform and autonomously moves along the cable tunnel with the robot dog for inspection. All modules are integrated and installed on the robot dog's back support plate. The robot dog uses an autonomous navigation system to walk along a preset inspection route, completing multimodal sensor data acquisition and real-time analysis during movement. The composition of each module of the system is shown in Table 5.

[0064] Table 5 System Composition

[0065] When the quadruped robot dog, equipped with this device, inspects cable tunnels, the multimodal fire source detection module continuously collects infrared thermal imaging, smoke scattering, and gas concentration data. When any sensor detects an abnormal signal, the DS evidence fusion decision-making process is automatically triggered. If the fusion result m(F) ≥ T_fire, the system immediately activates the spatial positioning module, and each positioning node collaboratively measures the distance to the fire source, calculating the three-dimensional coordinates of the fire source using the TDOA algorithm. Simultaneously, the fire prediction module extrapolates the fire spread trend in real time based on the temperature field gradient and tunnel ventilation parameters. All results are synchronously uploaded to the remote monitoring terminal via the main control communication module for maintenance personnel to develop emergency response plans.

[0066] The device was validated in a simulated cable tunnel environment.

[0067] Verification Experiment 1: Verification of the Accuracy of Fire Source Identification. Three scenarios were set up in a simulated cable tunnel environment: open flame, smoldering, and interference source. In the open flame scenario, all three sensors were triggered, m(F)=0.92, with a correct recognition rate of 100%. In the smoldering scenario, the smoke and gas sensors were triggered first, followed by the infrared sensor, m(F)=0.78, with a correct recognition rate of 95%. In the interference source (welding fumes) scenario, the smoke sensor was falsely triggered, but the gas sensor was not triggered. After DS fusion, m(F)=0.35, effectively suppressing false alarms.

[0068] Verification Experiment 2: Spatial Positioning Accuracy Verification. Fire sources were placed at distances of 5m, 10m, and 20m in a 20m long simulated tunnel, and 3-node TDOA positioning was used. The positioning error was 0.15m at 5m, 0.32m at 10m, and 0.58m at 20m, all meeting the meter-level positioning accuracy requirements. After introducing temperature compensation, the positioning accuracy at each distance improved by an average of 40%.

[0069] Verification Experiment 3: Fire Spread Prediction Verification. A PVC cable section was ignited in a simulated tunnel. The measured fire spread velocity was 0.015 m / s, while the model prediction was 0.014 m / s, with a relative error of 6.7%. The predicted spread direction deviation was ≤15°. The prediction update cycle was 5 seconds, meeting real-time requirements.

[0070] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A device for identifying and spatially locating fire sources in cable tunnels, characterized in that, It includes a multimodal fire source detection module, a spatial positioning module, a fire prediction module, and a main control communication module; the multimodal fire source detection module, the spatial positioning module, and the fire prediction module are bidirectionally connected to the main control communication module, and the main control communication module is wirelessly connected to a remote monitoring terminal; The multimodal fire source detection module is equipped with a multi-source fusion decision algorithm based on the improved DS evidence theory. It performs weighted evidence fusion of long-wave infrared thermal imaging radiation signal, Mie scattering smoke particle signal and electrochemical gas concentration signal, and outputs fire source identification results and confidence level. The spatial positioning module is equipped with a TDOA positioning algorithm based on acoustic and thermal joint ranging. It utilizes a dual-mode collaboration of distributed ultrasonic transducer array and thermal imaging pixel ranging, combined with tunnel environment temperature compensation, to output the three-dimensional spatial coordinates of the fire source. The fire prediction module is equipped with a fire spread extrapolation algorithm based on a thermal field-airflow coupling model. It incorporates the tunnel ventilation heat diffusion equation and the empirical spread rate formula to output the predicted fire spread direction and speed. The main control communication module is used for data scheduling, timing synchronization, and wireless communication with the remote monitoring terminal among the modules.

2. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The multimodal fire source detection module includes a long-wave infrared thermal imaging unit, a Mie scattering smoke detection unit, and an electrochemical gas sensing unit. The long-wave infrared thermal imaging unit operates in the 8–14 μm band and is used to acquire images of thermal radiation distribution within cable tunnels. The Mie scattering smoke detection unit employs an infrared emitter-receiver pair structure and detects smoke particle concentration by measuring forward-scattered light intensity. The electrochemical gas sensing unit includes a CO electrochemical sensor and a CO2 NDIR sensor.

3. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The multi-source fusion decision algorithm based on the improved DS evidence theory includes: Define the recognition framework Θ={F,NF}, where F indicates the presence of a fire source and NF indicates the absence of a fire source; the basic probability allocation functions for each sensor are m_i(F), m_i(NF), and m_i(Θ), i=1,2,3, where m_i(Θ) is an uncertain allocation; Calculating the Jousselme distance measures the degree of conflict between pieces of evidence: d_J(m_p,m_q)=√[0.5·(m_p-m_q)ᵀ·D·(m_p-m_q)] Where D is the Jaccard similarity matrix, D(A,B)=|A∩B| / |A∪B|; Calculate the dynamic credibility weights of each evidence source based on Jousselme distance: w_i=[1-(1 / (n-1))·Σ_{j≠i}d_J(m_i,m_j)] / Σ_k[1-(1 / (n-1))·Σ_{j≠k}d_J(m_k,m_j)] The original basic probability assignment is weighted and corrected as follows: m'_i(A)=w_i·m_i(A), m'i(Θ)=1-Σ{A≠Θ}m'_i(A); The Dempster combination rule was used to perform pairwise fusion of the weighted and modified evidence: m(A)=[Σ_{B∩C=A}m'_p(B)·m'_q(C)] / (1-K) Where K = Σ_{B∩C=∅}m'_p(B)·m'_q(C) is the conflict coefficient; When the fusion result m(F)≥T_fire, it is determined that a fire source exists, where T_fire is the adaptive fire source determination threshold.

4. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The TDOA positioning algorithm based on combined acoustic and thermal ranging includes: N positioning nodes are deployed at equal intervals on the top of the cable tunnel, where N≥3, and each positioning node integrates an ultrasonic transducer and a thermal imaging lens. Ultrasonic ranging uses the temperature compensation formula v_sound=331.3+0.606·T_c, where T_c is the tunnel ambient temperature, and the ultrasonic propagation distance d_ultra=v_sound·Δt / 2, where Δt is the ultrasonic transmission and reception time difference. The formula for thermal image pixel ranging is d_thermal=(f·S_real) / S_pixel, where f is the focal length of the thermal imager, S_real is the estimated actual size of the fire source, and S_pixel is the pixel size occupied by the fire source in the thermal image. Establish the system of equations for the TDOA hyperbola: √[(x-x_i) 2 +(y-y_i) 2 +(z-z_i) 2 ] - √[(x-x_j) 2 +(y-y_j) 2 +(z-z_j) 2 ] = c·Δt_ij, where (x_i,y_i,z_i) are the coordinates of the i-th positioning node, c is the speed of sound, and Δt_ij is the time difference between the arrival of the signal at the i-th node and the j-th node. The three-dimensional coordinates (x,y,z) of the fire source are obtained by using the Chan algorithm. Combined acoustic and thermal ranging data fusion: d_fused=(σ_u 2 ·d_thermal+σ_t 2 ·d_ultra) / (σ_u 2 +σ_t 2 ), whereσ_u 2 and σ_t 2 These are the variances of ultrasonic ranging and thermal imaging ranging, respectively.

5. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The fire spread prediction algorithm based on the thermal field-airflow coupling model includes: Establish the heat diffusion equation for tunnel fires: T / t=α·∇ 2 T+Q̇ / (ρ·c_p)-v_wind· T / x Where α is the thermal diffusivity, Q̇ is the heat release rate of the fire source, ρ is the air density, c_p is the specific heat of air at constant pressure, and v_wind is the longitudinal wind speed of the tunnel; Heat release rate of ignition source in t 2 Growth model calculation: Q̇(t) = α_f·t 2 , where α_f is the fire growth coefficient; Formula for fire spread rate: v_f=k·(Q̇ / Q̇_0)^n·(1+β·v_wind)·exp(-γ·d_cable) Where k is a material constant, Q̇_0 is a reference heat release rate, n is a power exponent, β is a wind speed influence coefficient, γ is a cable spacing attenuation coefficient, and d_cable is the spacing between adjacent cables; The direction of spread is determined by the gradient of the thermal field distribution: θ_spread = arctan( T / y / T / x), where T / x and T / y represents the longitudinal and transverse temperature gradients of the tunnel, respectively.

6. The fire source identification and spatial positioning device in a cable tunnel according to claim 2, characterized in that, The long-wave infrared thermal imaging unit calculates the target temperature according to the Stefan-Boltzmann law: M = ε·σ·T 4 Where M is the radiative exitance, ε is the emissivity, and σ = 5.67 × 10⁻⁶. -8 W / (m 2 ·K 4 ) is the Stefan-Boltzmann constant, and T is the absolute temperature; when the temperature exceeds the preset threshold T_alarm and the temperature rise rate exceeds dT / dt, infrared fire evidence m1 is generated; The formula for calculating the scattering cross section of the Mie scattering smoke detection unit is: σ_sca=(2π / k 2 )·Σ(2n+1)·(|a_n| 2 +|b_n| 2 ), where k=2π / λ is the wavenumber, and a_n and b_n are the Mie scattering coefficients; when the scattered light intensity exceeds the threshold V_smoke, smoke fire evidence m2 is generated; In the electrochemical gas sensing unit, the CO sensor has a range of 0–1000 ppm, and the CO2 sensor has a range of 0–5000 ppm; when the CO concentration exceeds 50 ppm or the CO2 concentration exceeds 1000 ppm, gas fire evidence m3 is generated.

7. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The main control communication module uses an ARM processor, runs a real-time operating system, and communicates with the remote monitoring terminal via a 5G or WiFi 6 wireless network.

8. The fire source identification and spatial positioning device in a cable tunnel according to claim 1, characterized in that, The multimodal fire source detection module, the spatial positioning module, the fire prediction module, and the main control communication module are integrated and installed on the quadruped robot dog inspection platform, which autonomously walks and inspects the cable tunnel with the quadruped robot dog.

9. The fire source identification and spatial positioning device in a cable tunnel according to claim 4, characterized in that, The positioning nodes are deployed every 15m to 25m along the top of the cable tunnel. Each positioning node is connected to the main control communication module via wired or wireless means to collaboratively complete the TDOA measurement. The fire prediction module updates the thermal field distribution with a 5s cycle, predicts the fire spread trend in the next 60s, and uploads the prediction results to the remote monitoring terminal in real time through the main control communication module.

10. A method for fire source identification and spatial positioning in a cable tunnel, characterized in that, The apparatus according to any one of claims 1 to 9 comprises the following steps: S1: The multimodal fire source detection module collects long-wave infrared thermal radiation signals, Mie scattered smoke particle signals, and electrochemical gas concentration signals to generate infrared fire evidence m1, smoke fire evidence m2, and gas fire evidence m3, respectively. S2: Based on the improved DS evidence theory, the multi-source fusion decision algorithm calculates the Jousselme distance between each piece of evidence and assigns dynamic confidence weights accordingly. After weighted correction of the original basic probability allocation, Dempster combination rules are used to fuse them one by one. When the fusion result m(F)≥T_fire, the fire source is determined to exist and the confidence level is output. S3: After determining that the fire source exists, the spatial positioning module is activated. Each positioning node measures the ultrasonic propagation distance and thermal image pixel ranging. After the two ranging results are fused by variance weighting, the TDOA hyperbolic equation system is established and solved using the Chan algorithm to output the three-dimensional spatial coordinates of the fire source. S4: The fire prediction module receives temperature field data and fire source coordinates, calculates the fire spread speed and direction based on the thermal field-airflow coupling model, and continuously updates the simulation results at a set period. S5: The main control communication module synchronously uploads the fire source identification results, confidence level, three-dimensional spatial coordinates, fire spread direction and speed to the remote monitoring terminal via wireless network.

Images