Method, system and equipment for detecting combustible gas in high-temperature fire scene environment and medium

Abstract

The invention belongs to the technical field of fire-fighting monitoring and gas detection, and discloses a high-temperature fire scene environment combustible gas detection method, system, equipment and medium, and the method comprises the following steps: carrying out spectrum environment data collection and pretreatment operation on a high-temperature fire scene environment mixed gas sample cooled by an immersed liquid cooling double-convection structure; estimating the concentration of the target gas based on a deep regression neural network model, and classifying and identifying the target gas by using a support vector machine algorithm; the target gas concentration and the type identification result are subjected to weighted fusion processing; and performing risk grade judgment on the final detection value of the target gas and a preset threshold value to obtain a risk grade judgment result. According to the invention, rapid cooling sampling, spectrum identification and intelligent fusion analysis of multi-component combustible gas in an extreme fire scene environment can be realized, so that high-precision detection and risk dynamic early warning of fire scene gas components are completed.

Description

Technical Field

[0001] This invention relates to the field of fire monitoring and gas detection technology, and in particular to a method, system, equipment and medium for detecting combustible gases in high-temperature fire environments. Background Technology

[0002] In recent years, with the rapid development of the new energy and chemical industries, gas leaks and secondary explosions in high-temperature environments have become increasingly frequent, seriously threatening fire rescue safety and the stable operation of industrial plants. Existing gas detection equipment is mostly designed for ambient temperature conditions; its sensors are easily damaged by heat or experience signal drift in high-temperature fire environments, leading to distorted detection results and slow response, failing to meet the real-time monitoring needs under extreme conditions.

[0003] Currently, some high-temperature area monitoring systems on the market are trying to extend the lifespan of sensors by using heat-insulating housings and air cooling, but these measures still have problems such as low heat dissipation efficiency, uneven cooling, and long response time; although some infrared spectroscopy detection devices have high sensitivity, they are difficult to maintain stable identification under strong radiation, smoke and dust interference, and high temperature background.

[0004] Among them, Chinese patent application CN119959298A discloses a combustible gas detection system based on calculated diffusion rate, combustion heat index, and thermal conductivity. Although it can perform high-precision gas analysis, it is only suitable for early warning and cannot work in high-temperature environments. Chinese patent application CN115112593A proposes a multi-channel redundant combustible gas concentration detection method. By analyzing the calculation results of each group, it determines whether the sensor is faulty, improving the reliability, robustness, and anti-false alarm capability of the combustible gas concentration sensor. However, it can only detect gases at room temperature. Achieving high-precision and rapid identification and concentration inversion of combustible gases in extreme high-temperature fire environments has become a key technical challenge for early fire risk warning and rescue safety control.

[0005] Therefore, how to provide a method, system, equipment and medium for detecting combustible gases in high-temperature fire environments is an urgent problem to be solved. Summary of the Invention

[0006] This invention provides a method, system, device, and medium for detecting combustible gases in high-temperature fire environments to solve the problems mentioned above in the prior art.

[0007] According to a first aspect of the present invention, a method for detecting combustible gases in a high-temperature fire environment is provided.

[0008] In one embodiment, the method for detecting combustible gases in a high-temperature fire environment includes the following steps:

[0009] Based on a pre-configured sensor set, spectral environmental data acquisition and preprocessing operations are performed on a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure to form a spectral environmental parameter set.

[0010] Based on the spectral environmental parameter set, the target gas concentration is estimated using a deep regression-based neural network model to obtain the target gas concentration. Then, the target gas is classified and identified using a support vector machine algorithm to obtain the type identification result.

[0011] By employing a layered fusion strategy, the target gas concentration and type identification results are weighted and fused to obtain the final detection value of the target gas.

[0012] The final detected value of the target gas is compared with a preset threshold to determine the risk level, thus obtaining the risk level determination result.

[0013] In one embodiment, the process of acquiring and preprocessing spectral environmental data from a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure, based on a pre-configured sensor group, to form a spectral environmental parameter set, includes the following steps:

[0014] Based on the immersion liquid-cooled dual convection structure, the high-temperature fire environment mixed gas after sampling is cooled to obtain a cooled gas sample.

[0015] Using a pre-configured sensor array, environmental and spectral parameters are collected from the cooled gas sample to obtain spectral data and environmental data. The spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data.

[0016] Based on the temperature compensation model and combined with temperature data, the spectral data is numerically corrected to obtain corrected spectral data.

[0017] By integrating the calibrated spectral data with environmental data, a set of spectral environmental parameters is obtained.

[0018] In one embodiment, the submerged liquid-cooled dual-convective structure includes: a submerged dual-convective heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer, wherein the multi-stage cooling structure and the pretreatment structure are both connected to the submerged dual-convective heat exchanger, and the insulation layer is disposed on the outside of the submerged dual-convective heat exchanger.

[0019] The submerged double convection heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged double convection heat exchanger is a liquid-cooled medium to form strong convection heat transfer. The inner layer of the submerged double convection heat exchanger is a mixed gas of high-temperature fire environment.

[0020] The multi-stage cooling structure includes a primary cooling layer region and a secondary temperature homogenization region;

[0021] The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber.

[0022] The insulation layer includes a vacuum cavity and high-temperature insulation material.

[0023] In one embodiment, the step of performing numerical correction processing on the spectral data based on the temperature compensation model and incorporating temperature data to obtain corrected spectral data includes the following steps:

[0024] Zero-point calibration is performed on the spectral data to obtain calibrated spectral data;

[0025] The calibrated spectral data were subjected to background subtraction and low-pass filtering to obtain standardized spectral data.

[0026] Standardized spectral and temperature data are input into a temperature compensation model, and the standardized spectral data are numerically corrected using the temperature compensation model to obtain corrected spectral data.

[0027] In one embodiment, the expression for the temperature compensation model is:

[0028] A i,adj =A i,net -α i (T-T0)-β i (T-T0) 2 ;

[0029] In the formula, A i,adj Represents the temperature compensation model, A i,net This represents standardized spectral data, where T represents temperature data, and α... i and β i T0 represents the calibration coefficient, and T0 represents the calibration temperature.

[0030] In one embodiment, the deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer.

[0031] The activation function of the deep regression neural network model is the modified linear unit function;

[0032] The loss function of the deep regression neural network model is the mean squared error loss function;

[0033] The optimization algorithm for the deep regression neural network model is the adaptive moment estimation algorithm.

[0034] In one embodiment, the training of the deep regression neural network model includes: an offline calibration phase and an online adjustment phase;

[0035] The offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory.

[0036] The online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

[0037] In one embodiment, the expression for the mean squared error loss function is:

[0038]

[0039] In the formula, Y represents the mean squared error loss function, N represents the number of training samples, and Y represents the mean squared error loss function. (k) This represents the true concentration vector of the k-th sample. X represents the model's predicted value, θ represents the set of neural network parameters, and X represents the model's predicted value. (k) This represents the input feature vector of the k-th sample.

[0040] According to a second aspect of the present invention, a combustible gas detection system for high-temperature fire environments is provided.

[0041] In one embodiment, a combustible gas detection system for high-temperature fire environments includes:

[0042] The data acquisition module is used to collect and preprocess spectral environmental data from a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure, based on a pre-configured sensor group, to form a spectral environmental parameter set.

[0043] The intelligent analysis module is used to estimate the target gas concentration based on a deep regression neural network model according to the spectral environmental parameter set, obtain the target gas concentration, and use the support vector machine algorithm to classify and identify the target gas to obtain the type identification result.

[0044] The result fusion module is used to perform weighted fusion processing on the target gas concentration and type identification results using a layered fusion strategy to obtain the final detection value of the target gas.

[0045] The results display module is used to compare the final detection value of the target gas with a preset threshold to determine the risk level and obtain the risk level determination result.

[0046] In one embodiment, the data acquisition module is further configured to:

[0047] Based on the immersion liquid-cooled dual convection structure, the high-temperature fire environment mixed gas after sampling is cooled to obtain a cooled gas sample.

[0048] Using a pre-configured sensor array, environmental and spectral parameters are collected from the cooled gas sample to obtain spectral data and environmental data. The spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data.

[0049] Based on the temperature compensation model and combined with temperature data, the spectral data is numerically corrected to obtain corrected spectral data.

[0050] By integrating the calibrated spectral data with environmental data, a set of spectral environmental parameters is obtained.

[0051] In one embodiment, the submerged liquid-cooled dual-convective structure includes: a submerged dual-convective heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer, wherein the multi-stage cooling structure and the pretreatment structure are both connected to the submerged dual-convective heat exchanger, and the insulation layer is disposed on the outside of the submerged dual-convective heat exchanger.

[0052] The submerged double convection heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged double convection heat exchanger is a liquid-cooled medium to form strong convection heat transfer. The inner layer of the submerged double convection heat exchanger is a mixed gas of high-temperature fire environment.

[0053] The multi-stage cooling structure includes a primary cooling layer region and a secondary temperature homogenization region;

[0054] The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber.

[0055] The insulation layer includes a vacuum cavity and high-temperature insulation material.

[0056] In one embodiment, the step of performing numerical correction processing on the spectral data based on the temperature compensation model and incorporating temperature data to obtain corrected spectral data includes:

[0057] Zero-point calibration is performed on the spectral data to obtain calibrated spectral data;

[0058] The calibrated spectral data were subjected to background subtraction and low-pass filtering to obtain standardized spectral data.

[0059] Standardized spectral and temperature data are input into a temperature compensation model, and the standardized spectral data are numerically corrected using the temperature compensation model to obtain corrected spectral data.

[0060] In one embodiment, the deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer.

[0061] The activation function of the deep regression neural network model is the modified linear unit function;

[0062] The loss function of the deep regression neural network model is the mean squared error loss function;

[0063] The optimization algorithm for the deep regression neural network model is the adaptive moment estimation algorithm.

[0064] In one embodiment, the training of the deep regression neural network model includes: an offline calibration phase and an online adjustment phase;

[0065] The offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory.

[0066] The online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

[0067] According to a third aspect of the present invention, a computer device is provided.

[0068] In some embodiments, the computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method described above.

[0069] According to a fourth aspect of the present invention, a computer-readable storage medium is provided.

[0070] In one embodiment, a computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the above method.

[0071] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0072] 1. This invention enables rapid cooling sampling, spectral identification, and intelligent fusion analysis of multi-component combustible gases in extreme fire environments, thereby achieving high-precision detection of fire gas components and dynamic risk warning. It also features dual innovations in structural protection and algorithm collaboration, and can operate stably for a long time under high temperature conditions of 650-800℃. It is suitable for rapid detection and risk assessment at fire rescue and explosion accident sites.

[0073] 2. This invention significantly improves the high-temperature adaptability and operational stability of the sensing unit through the combination of dual convection liquid cooling and heat insulation cavity design, enabling stable detection in extreme environments.

[0074] 3. This invention achieves high-precision quantitative identification of multi-component combustible gases by introducing a fusion model of infrared multi-band spectral detection and intelligent algorithm.

[0075] 4. This invention utilizes the collaborative analysis of neural networks and support vector machines to improve the detection accuracy and anti-interference capability under complex atmospheric conditions.

[0076] 5. This invention has the advantages of high system integration, short response time and strong portability, and can be used for rapid risk assessment and decision support at fire rescue sites.

[0077] 6. The self-calibration and online fine-tuning capabilities of the algorithm in this invention can maintain detection accuracy over a long period of time and reduce maintenance costs.

[0078] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0079] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0080] Figure 1 This is a flowchart illustrating a method for detecting combustible gases in a high-temperature fire environment, according to an exemplary embodiment.

[0081] Figure 2 This is a schematic diagram illustrating the structural principle of gas collection, cooling, and detection in a combustible gas detection method for a high-temperature fire environment, according to an exemplary embodiment.

[0082] Figure 3 This is a flowchart illustrating the intelligent algorithm and data fusion process in a combustible gas detection method for a high-temperature fire environment, according to an exemplary embodiment.

[0083] Figure 4 This is a schematic diagram of a combustible gas detection system in a high-temperature fire environment, according to an exemplary embodiment.

[0084] Figure 5 This is a flowchart illustrating a method for detecting combustible gases in a high-temperature fire environment, according to an exemplary embodiment.

[0085] Figure 6 This is a schematic diagram of the structure of a computer device according to an exemplary embodiment. Detailed Implementation

[0086] The following description and accompanying drawings fully illustrate specific embodiments described herein to enable those skilled in the art to practice them. Some portions and features of certain embodiments may be included in or replace portions and features of other embodiments. The scope of the embodiments herein includes the entire scope of the claims and all available equivalents thereof. The various embodiments described herein are presented in a progressive manner, with each embodiment focusing on its differences from other embodiments; similar or identical parts between embodiments can be referred to interchangeably.

[0087] The modules in the apparatus or system of this application can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0088] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0089] Figure 1 An embodiment of a method for detecting combustible gases in a high-temperature fire environment according to the present invention is shown.

[0090] In this optional embodiment, the method for detecting combustible gases in a high-temperature fire environment includes the following steps:

[0091] Step S101: Based on the pre-configured sensor group, perform spectral environmental data acquisition and preprocessing operations on the high-temperature fire environment mixed gas sample after being cooled by the immersion liquid-cooled dual convection structure to form a spectral environmental parameter set.

[0092] Step S102: Based on the spectral environmental parameter set, the target gas concentration is estimated using a deep regression neural network model to obtain the target gas concentration, and the target gas is classified and identified using a support vector machine algorithm to obtain the type identification result.

[0093] Step S103: Using a layered fusion strategy, the target gas concentration and type identification results are weighted and fused to obtain the final detection value of the target gas.

[0094] Step S104: The final detection value of the target gas is compared with the preset threshold to determine the risk level, and the risk level determination result is obtained.

[0095] Specifically, the concentration result output by the neural network is weighted and fused with the classification result of the support vector machine. When the two results are consistent, the final detection value is output. When there is a difference, the one with higher confidence is used, and the result is transmitted to the display and alarm module to trigger the corresponding level of alarm.

[0096] It should be noted that after obtaining the risk level assessment result, the final detection value of the target gas and the risk level assessment result can be visualized. The visualization display is mainly responsible for displaying the detection results and triggering an alarm for high concentrations of combustible gas. It mainly includes a local display and an audible and visual alarm system, which can display the detected gas concentration and alarm level in real time, and transmit the detection results to the host computer control center via a wireless communication module to realize remote monitoring and data storage.

[0097] In this optional embodiment, when performing spectral environmental data acquisition and preprocessing operations on a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure based on a pre-configured sensor group to form a spectral environmental parameter set, the sampled high-temperature fire environment mixed gas can be cooled using the immersion liquid-cooled dual convection structure to obtain a cooled gas sample. Using the pre-configured sensor group, environmental parameters and spectral parameters are acquired from the cooled gas sample to obtain spectral data and environmental data. The spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data. Based on a temperature compensation model and combined with the temperature data, numerical correction processing is performed on the spectral data to obtain corrected spectral data. The corrected spectral data and environmental data are integrated to obtain the spectral environmental parameter set.

[0098] It should be noted that, as Figure 2 The diagram shows the structure of an intelligent combustible gas detection device for high-temperature fire environments. The outer shell is encapsulated with high-temperature resistant composite insulation material. Key components are protected by an insulated cavity and liquid-cooled channel, allowing the entire unit to operate continuously for over 30 minutes in an external environment of 800℃. The lightweight design of the entire unit facilitates transport by firefighters to the scene for emergency detection tasks. Sampling of mixed gases in high-temperature fire environments requires a sampling and delivery device, responsible for delivering the high-temperature mixed gases from the fire scene to the device's detection interior. This device includes an air inlet and a miniature high-temperature pump. The air inlet is made of a high-temperature resistant alloy, ensuring long-term operation in fire environments up to 1000℃. The miniature high-temperature pump can be replaced with a high-temperature resistant blower, achieving pump-suction sampling with an adjustable flow rate of 0.1–1.0 L / min. The device also includes an evaporator exhaust port / water injection port for discharging evaporative exhaust gases or injecting coolant into the system. An exhaust outlet is used to discharge treated exhaust gases from the system. Power and communication cables use high-temperature resistant silicone wire harnesses. Temperature sensors include: a gas temperature sensor before cooling and a gas temperature sensor after cooling. The pre-treatment unit is used to perform preliminary treatment on the incoming gas, the post-treatment unit is used to further treat the cooled gas, the flow sensor is used to monitor the gas flow rate in real time, and the gas pump is used to provide power for the gas flow.

[0099] The submerged liquid-cooled dual-convective structure is used to cool high-temperature mixed gas to within the temperature limit of gas detection. It mainly includes a submerged dual-convective heat exchanger, a multi-stage cooling structure, a pretreatment structure (unit), and an insulation layer. The submersible dual-convective heat exchanger uses an inner spiral copper tube or copper coil as the gas flow channel, with an outer layer of liquid-cooled medium (deionized water or antifreeze coolant) to form strong convective heat transfer. The copper tube length must be controlled to ≤0.5m to ensure the timeliness of the test results. A liquid-cooled circulation chamber is also included, connecting a temperature sensor and a temperature control unit. This chamber automatically adjusts the coolant flow rate based on real-time temperature, ensuring the outlet gas temperature remains stable below 40℃. The coolant medium can be deionized water or a glycol mixture. The coolant temperature is controlled via a closed-loop PID algorithm to maintain the gas outlet temperature below 40℃. The multi-stage cooling structure consists of a primary rapid cooling zone (large temperature difference heat exchange) and a secondary temperature homogenization zone (reducing pulsation / temperature gradient). The pretreatment unit includes a high-temperature filter (to trap particulate impurities from the fire scene), a moisture-absorbing layer (to ensure gas dryness), and a low-pressure buffer chamber (to smooth the flow rate). The insulation layer, through a vacuum chamber and high-temperature insulation materials, minimizes the impact of the high-temperature environment of the fire scene on the test results.

[0100] In addition, when collecting environmental and spectral parameters from the cooled gas samples, infrared gas detection technology was used for spectral parameter acquisition. The relationship between gas concentration and absorption intensity (Lambert-Beer Law) was utilized to identify gas components and determine their concentrations. The device is equipped with a 2-12μm multi-band infrared light source, capable of detecting various combustible gases. Environmental parameters included temperature T(t), humidity RH(t), and flow rate Q(t). The infrared spectral detection equipment (i.e....) Figure 2 The gas composition analysis device includes an infrared light source, a reflective detection cavity, an optical path shaping mirror, and a pyroelectric detector. The infrared light source emits wavelengths in the range of 2–12 μm, covering the absorption ranges of typical gases such as methane, hydrogen, and carbon monoxide. The detection cavity employs a multi-reflection optical path to extend the optical path length and improve detection sensitivity. The pyroelectric detector outputs multi-band absorption signals via a voltage stabilization circuit for subsequent algorithm processing.

[0101] In this optional embodiment, the submerged liquid-cooled dual-convective structure includes: a submerged dual-convective heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer. The multi-stage cooling structure and the pretreatment structure are both connected to the submerged dual-convective heat exchanger. The insulation layer is disposed on the outside of the submerged dual-convective heat exchanger. The submerged dual-convective heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged dual-convective heat exchanger is a liquid-cooled medium to form strong convective heat transfer, and the inner layer is a high-temperature fire environment mixed gas. The multi-stage cooling structure includes a primary cooling layer zone and a secondary temperature homogenization zone. The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber. The insulation layer includes a vacuum chamber and high-temperature insulation material.

[0102] In this optional embodiment, when the spectral data is numerically corrected based on the temperature compensation model and combined with temperature data to obtain corrected spectral data, zero-point calibration can be performed on the spectral data to obtain calibrated spectral data; background subtraction and low-pass filtering are then performed on the calibrated spectral data to obtain standardized spectral data; the standardized spectral data and temperature data are input into the temperature compensation model, and the standardized spectral data is numerically corrected using the temperature compensation model to obtain corrected spectral data.

[0103] It should be noted that correcting the detection data ensures real-time accuracy of the results. This is primarily achieved through an embedded controller and a data storage and wireless communication module. The embedded controller, consisting of an ARM or microcontroller with an edge GPU / NPU, is capable of controlling the sampling pump, coolant pump, temperature / flow rate acquisition, primary signal processing, and communication with algorithms (neural network / SVM execution). The data storage and wireless communication module is mainly used for uploading historical data and remote monitoring.

[0104] In this optional embodiment, the expression for the temperature compensation model is:

[0105] A i,adj =A i,net -α i (T-T0)-β i (T-T0) 2 ;

[0106] In the formula, A i,adj Represents the temperature compensation model, T i,net This represents standardized spectral data, where T represents temperature data, and α... i and β i T0 represents the calibration coefficient, and T0 represents the calibration temperature.

[0107] In this optional embodiment, the deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer; the activation function of the deep regression neural network model is the modified linear unit function (ReLU); the loss function of the deep regression neural network model is the mean squared error loss function (MSE); and the optimization algorithm of the deep regression neural network model is the adaptive moment estimation (Adam) algorithm.

[0108] In this optional embodiment, the training of the deep regression neural network model includes: an offline calibration stage and an online adjustment stage; the offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory; the online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

[0109] In this optional embodiment, the expression for the mean squared error loss function is:

[0110]

[0111] In the formula, Y represents the mean squared error loss function, N represents the number of training samples, and Y represents the mean squared error loss function. (k) This represents the true concentration vector of the k-th sample. X represents the model's predicted value, θ represents the set of neural network parameters, and X represents the model's predicted value. (k) This represents the input feature vector of the k-th sample.

[0112] When necessary, such as Figure 3 As shown, the high temperature at the fire site can cause infrared sensor response drift, changes in absorption coefficient with temperature, and alterations in optical scattering due to water / particle content in the sample. Directly calculating concentration using the original absorbance will lead to errors. Therefore, during the detection process, noise and background processing (filtering, background subtraction) is required first, followed by temperature-based numerical correction (using polynomials or lookup tables), and finally, the corrected signal is fed into the machine learning module.

[0113] For temperature compensation, the absorbance A is measured for each band i. i,meas First, the background is subtracted to obtain the net absorbance A. i,net Then perform temperature correction:

[0114] A i,adj =A i,net -α i (T-T0)-β i (T-T0) 2 ;

[0115] Where, αi and β i T0 represents the coefficient obtained through calibration experiments (different for each band), and T0 represents the calibration temperature.

[0116] This invention, based on infrared absorbance and environmental parameters, employs a deep regression-based neural network model to estimate the volume fraction (or concentration) of a target gas in real time. The feature vector input to the neural network consists of multi-band absorbance data that has undergone temperature compensation and background correction, along with environmental parameters, denoted as... in, X represents the absorbance value of the i-th band after preprocessing and temperature correction, T represents the temperature of the detection chamber, RH represents the relative humidity, and Q represents the real-time flow rate or velocity. The neural network takes this vector X as input and outputs the target gas concentration vector. in, This represents the estimated volume fraction or concentration value of the j-th gas (e.g., CH4, H2, CO, etc.).

[0117] To ensure the stability and generalization ability of the regression model, this invention employs a multi-layer fully connected neural network as the regression entity. The network consists of an input layer, two to three hidden layers, and an output layer. The hidden layers use the ReLU (Rectified Linear Unit) activation function, and the output layer uses linear activation to directly regress continuous concentration values. The model parameters are trained by minimizing the mean squared error (MSE) loss function, which is expressed as follows: Where N represents the number of training samples, Y (k) This represents the true concentration vector of the k-th sample. Let θ represent the model's predicted values, and let θ represent the set of neural network parameters. To prevent overfitting and improve field robustness, regularization techniques (such as L2 regularization) and early stopping are used during training, and dropout can be applied to the hidden layers. The optimization algorithm uses Adam or RMSprop to accelerate convergence and adapt to different feature scales.

[0118] To meet the computing power constraints of portable devices, the network structure is designed to balance accuracy and computational complexity: the recommended hidden layer structure is 32→16 (i.e., two layers, containing 32 and 16 neurons respectively) or 64→32→16 (i.e., three layers), and the input and output dimensions are adjusted according to the actual number of wavebands and the number of target gases.

[0119] Furthermore, model training is divided into two stages: offline calibration and online fine-tuning. In the offline calibration stage, training data is collected using a standard mixed gas in a laboratory or controlled environment. The data should cover typical operating conditions and noise distributions that the equipment may encounter during operation, including different temperatures, humidity levels, baseline backgrounds, and typical interfering gas combinations. For each band and each gas, a sufficient density of calibration points needs to be collected to establish a robust regression mapping. In the online fine-tuning stage, during field operation, the system performs small-amplitude incremental training (or parameter fine-tuning) based on real-time collected long-term data and a small number of reference samples to adapt to long-term drift and special operating conditions. To ensure the safety of online fine-tuning, a validation set and drift detection strategy are introduced. Model parameters are only updated when validation performance improves and statistical indicators stabilize (e.g., validation MSE decreases and confidence intervals converge). If online fine-tuning fails or an anomaly occurs, the system rolls back to the last stable model version and triggers a maintenance prompt.

[0120] Regarding the confidence assessment of regression prediction results, this invention quantifies the confidence of the output through residual statistics and uncertainty estimation. Specifically, it uses the residual distribution on the training set to estimate the confidence interval, or introduces prediction variance estimation into the regression model (e.g., using MC-dropout or a method based on Bayesian approximation), and marks low-confidence results as "requiring verification," which are then used in the final decision fusion along with the SVM judgment results.

[0121] Furthermore, to accurately identify gas types and address peak overlap issues, this invention employs a Support Vector Machine (SVM) as a classification module to determine the presence of the target gas, the dominant gas species, and the type of mixed gases. The features received by the SVM can be either the original absorbance vector after temperature compensation or... It can also be an intermediate feature (i.e., a deep feature) of a hidden layer or output layer of a neural network to improve the robustness of discrimination. For classification tasks, multi-class SVM or one-to-one or one-to-many strategies can be used to achieve multi-gas discrimination. Radial basis functions (RBFs) are preferred for the kernel function, and the parameter penalty term C and kernel parameter γ are determined through cross-validation to achieve good classification boundaries and generalization ability. The output of the SVM can be a discrete label (presence / absence, dominant gas category), or it can output a decision value (distance from the hyperplane) or a probability estimate (e.g., a probability mapping based on Platt scaling), which is used for subsequent fusion with the regression results of the neural network. This invention employs a stacked fusion strategy between concentration estimation and type determination: the neural network provides continuous concentration estimation, and the SVM provides gas presence and category probabilities. The outputs of both are weighted in the fusion unit and undergo consistency checks to obtain the final detection result. The fusion logic can be a weighted average or a rule engine. If the concentration of a gas output by the neural network is significantly higher than a threshold and the SVM simultaneously determines that the gas is present (with a probability exceeding the set threshold), then the presence of the gas is confirmed, and the concentration is reported as the regression value of the neural network. If the results of the neural network and the SVM are inconsistent, the fusion unit automatically adjusts the output based on the confidence levels of both: for example, when the SVM determines a high confidence level while the neural network determines a low confidence level, the protection mechanism is triggered based on the SVM's category determination, and the regression value is marked as "potentially subject to interference"; when both have low confidence levels, the system enters a maintenance prompt or requires manual review. The fusion unit also provides the fuzzy decision module with data such as concentration, category, and environmental change rate to implement risk classification and alarm threshold decisions.

[0122] To ensure the real-time performance and resource availability of portable devices, this invention recommends lightweighting of the neural network model during deployment. This includes techniques such as model pruning, quantization (e.g., 8-bit quantization), and knowledge distillation to enable low-latency operation on ARM-based embedded processors or edge computing units with NPUs. SVM models can also utilize kernel approximation or be run after feature dimensionality reduction to reduce computational load. During operation, the device maintains model version control and logging, storing all key inference results (concentration estimation, class determination, confidence scores) along with the original time-series input for future traceability and model retraining.

[0123] Figure 4 An embodiment of a combustible gas detection system for high-temperature fire environments according to the present invention is shown.

[0124] In this optional embodiment, the combustible gas detection system for high-temperature fire environments includes:

[0125] The data acquisition module 201 is used to perform spectral environmental data acquisition and preprocessing operations on a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure based on a pre-configured sensor group, to form a spectral environmental parameter set.

[0126] The intelligent analysis module 202 is used to estimate the target gas concentration based on a deep regression neural network model according to the spectral environmental parameter set, obtain the target gas concentration, and use the support vector machine algorithm to classify and identify the target gas to obtain the type identification result.

[0127] The result fusion module 203 is used to perform weighted fusion processing on the target gas concentration and type identification results using a layered fusion strategy to obtain the final detection value of the target gas.

[0128] The results display module 204 is used to compare the final detection value of the target gas with a preset threshold to determine the risk level and obtain the risk level determination result.

[0129] In this optional embodiment, the data acquisition module is further configured to: cool the sampled high-temperature fire environment mixed gas based on an immersion liquid-cooled dual convection structure to obtain a cooled gas sample; collect environmental parameters and spectral parameters from the cooled gas sample using a pre-configured sensor group to obtain spectral data and environmental data; the spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data; perform numerical correction processing on the spectral data based on a temperature compensation model and in conjunction with the temperature data to obtain corrected spectral data; and integrate the corrected spectral data with the environmental data to obtain a spectral environmental parameter set.

[0130] In this optional embodiment, the submerged liquid-cooled dual-convective structure includes: a submerged dual-convective heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer. The multi-stage cooling structure and the pretreatment structure are both connected to the submerged dual-convective heat exchanger. The insulation layer is disposed on the outside of the submerged dual-convective heat exchanger. The submerged dual-convective heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged dual-convective heat exchanger is a liquid-cooled medium to form strong convective heat transfer, and the inner layer is a high-temperature fire environment mixed gas. The multi-stage cooling structure includes a primary cooling layer zone and a secondary temperature homogenization zone. The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber. The insulation layer includes a vacuum chamber and high-temperature insulation material.

[0131] In this optional embodiment, the step of performing numerical correction processing on spectral data based on the temperature compensation model and in conjunction with temperature data to obtain corrected spectral data includes: performing zero-point calibration processing on the spectral data to obtain calibrated spectral data; performing background subtraction and low-pass filtering processing on the calibrated spectral data to obtain standardized spectral data; inputting the standardized spectral data and temperature data into the temperature compensation model, and using the temperature compensation model to perform numerical correction processing on the standardized spectral data to obtain corrected spectral data.

[0132] In this optional embodiment, the deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer; the activation function of the deep regression neural network model is a modified linear unit function; the loss function of the deep regression neural network model is a mean squared error loss function; and the optimization algorithm of the deep regression neural network model is an adaptive moment estimation algorithm.

[0133] In this optional embodiment, the training of the deep regression neural network model includes: an offline calibration stage and an online adjustment stage; the offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory; the online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

[0134] To facilitate understanding of the above technical solutions of the present invention, the following further explains the above technical solutions of the present invention from the perspectives of architecture and principle, such as... Figure 5 As shown, the specific steps for intelligent detection of combustible gases in high-temperature fire environments are as follows:

[0135] 1. Initiate self-test;

[0136] After powering on, the controller (main controller) performs a self-test, starts coolant circulation, and checks whether each temperature sensor is functioning properly; if the temperature is cooled to the preset range, it enters sampling standby mode.

[0137] 2. Pump suction sampling (sampling parameters);

[0138] The suction pump draws gas from the sampling port at a set flow rate, which then enters the first heat exchange section. The sampling path length is minimized to reduce hysteresis.

[0139] 3. Multi-stage cooling and pretreatment;

[0140] The gas is rapidly cooled (e.g., from 700℃ to ≤50℃) through a double convection formed by the inner spiral copper tube and the outer liquid. The cooling outlet is connected to a high-temperature filter, a moisture-absorbing layer, and a buffer chamber, where particles and moisture are removed before the gas enters the testing chamber.

[0141] 4. Spectral acquisition (infrared);

[0142] After entering the detection chamber, the light source emits light at a preset wavelength, and the detector collects multi-channel absorbance signals Ai(t). At the same time, environmental parameters such as temperature T(t), humidity RH(t), and flow rate Q(t) are also collected.

[0143] 5. Signal preprocessing and temperature compensation (online);

[0144] The original absorbance was zero-point calibrated, background subtracted, and low-pass filtered (to remove high-frequency noise). A temperature compensation model was then applied to obtain the corrected absorbance value.

[0145] 6. Data fusion and recognition (intelligent algorithms);

[0146] The preprocessed signal vector is input into a neural network to obtain a preliminary concentration estimate. Then, an SVM is used to determine the gas type and identify cross-interference. Finally, a risk level and alarm decision are generated through fuzzy rules.

[0147] 7. Output and storage;

[0148] The display screen shows the concentration and alarm level of each target gas in real time; key data is stored locally and can be uploaded to a remote server.

[0149] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores static and dynamic information data. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the above method embodiments.

[0150] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0151] In addition, the present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0152] In addition, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0154] This invention is not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this invention is limited only by the appended claims.

Claims

1. A method for detecting combustible gases in a high-temperature fire environment, characterized in that, Includes the following steps: Based on a pre-configured sensor set, spectral environmental data acquisition and preprocessing operations are performed on a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure to form a spectral environmental parameter set. Based on the spectral environmental parameter set, the target gas concentration is estimated using a deep regression-based neural network model to obtain the target gas concentration. Then, the target gas is classified and identified using a support vector machine algorithm to obtain the type identification result. By employing a layered fusion strategy, the target gas concentration and type identification results are weighted and fused to obtain the final detection value of the target gas. The final detected value of the target gas is compared with a preset threshold to determine the risk level, thus obtaining the risk level determination result.

2. The method for detecting combustible gases in a high-temperature fire environment according to claim 1, characterized in that, The process of acquiring and preprocessing spectral environmental data from a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual-convection structure, based on a pre-configured sensor group, to form a spectral environmental parameter set, includes the following steps: Based on the immersion liquid-cooled dual convection structure, the high-temperature fire environment mixed gas after sampling is cooled to obtain a cooled gas sample. Using a pre-configured sensor array, environmental and spectral parameters are collected from the cooled gas sample to obtain spectral data and environmental data. The spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data. Based on the temperature compensation model and combined with temperature data, the spectral data is numerically corrected to obtain corrected spectral data. By integrating the calibrated spectral data with environmental data, a set of spectral environmental parameters is obtained.

3. The method for detecting combustible gases in a high-temperature fire environment according to claim 2, characterized in that, The submerged liquid-cooled double convection structure includes: a submerged double convection heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer. The multi-stage cooling structure and the pretreatment structure are both connected to the submerged double convection heat exchanger, and the insulation layer is disposed on the outside of the submerged double convection heat exchanger. The submerged double convection heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged double convection heat exchanger is a liquid-cooled medium to form strong convection heat transfer. The inner layer of the submerged double convection heat exchanger is a mixed gas of high-temperature fire environment. The multi-stage cooling structure includes a primary cooling layer region and a secondary temperature homogenization region; The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber. The insulation layer includes a vacuum cavity and high-temperature insulation material.

4. The method for detecting combustible gases in a high-temperature fire environment according to claim 2, characterized in that, The process of numerically correcting spectral data based on a temperature compensation model and temperature data to obtain corrected spectral data includes the following steps: Zero-point calibration is performed on the spectral data to obtain calibrated spectral data; The calibrated spectral data were subjected to background subtraction and low-pass filtering to obtain standardized spectral data. Standardized spectral and temperature data are input into a temperature compensation model, and the standardized spectral data are numerically corrected using the temperature compensation model to obtain corrected spectral data.

5. The method for detecting combustible gases in a high-temperature fire environment according to claim 4, characterized in that, The expression for the temperature compensation model is: A i,adj =A i,net -α i (T-T0)-β i (T-T0) 2 ; In the formula, A i,adj Represents the temperature compensation model, A i,net This represents standardized spectral data, where T represents temperature data, and α... i and β i T0 represents the calibration coefficient, and T0 represents the calibration temperature.

6. The method for detecting combustible gases in a high-temperature fire environment according to claim 1, characterized in that, The deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer. The activation function of the deep regression neural network model is the modified linear unit function; The loss function of the deep regression neural network model is the mean squared error loss function; The optimization algorithm for the deep regression neural network model is the adaptive moment estimation algorithm.

7. The method for detecting combustible gases in a high-temperature fire environment according to claim 1, characterized in that, The training of the deep regression neural network model includes: an offline calibration phase and an online adjustment phase; The offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory. The online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

8. The method for detecting combustible gases in a high-temperature fire environment according to claim 6, characterized in that, The expression for the mean squared error loss function is: In the formula, Y represents the mean squared error loss function, N represents the number of training samples, and Y represents the mean squared error loss function. (k) This represents the true concentration vector of the k-th sample. X represents the model's predicted value, θ represents the set of neural network parameters, and X represents the model's predicted value. (k) This represents the input feature vector of the k-th sample.

9. A combustible gas detection system for high-temperature fire environments, characterized in that, include: The data acquisition module is used to collect and preprocess spectral environmental data from a high-temperature fire environment mixed gas sample cooled by an immersion liquid-cooled dual convection structure, based on a pre-configured sensor group, to form a spectral environmental parameter set. The intelligent analysis module is used to estimate the target gas concentration based on a deep regression neural network model according to the spectral environmental parameter set, obtain the target gas concentration, and use the support vector machine algorithm to classify and identify the target gas to obtain the type identification result. The result fusion module is used to perform weighted fusion processing on the target gas concentration and type identification results using a layered fusion strategy to obtain the final detection value of the target gas. The results display module is used to compare the final detection value of the target gas with a preset threshold to determine the risk level and obtain the risk level determination result.

10. The combustible gas detection system for high-temperature fire environments according to claim 9, characterized in that, The data acquisition module is also used for: Based on the immersion liquid-cooled dual convection structure, the high-temperature fire environment mixed gas after sampling is cooled to obtain a cooled gas sample. Using a pre-configured sensor array, environmental and spectral parameters are collected from the cooled gas sample to obtain spectral data and environmental data. The spectral data is a multi-channel absorbance signal, and the environmental data includes temperature data, humidity data, and flow rate data. Based on the temperature compensation model and combined with temperature data, the spectral data is numerically corrected to obtain corrected spectral data. By integrating the calibrated spectral data with environmental data, a set of spectral environmental parameters is obtained.

11. The combustible gas detection system for high-temperature fire environments according to claim 10, characterized in that, The submerged liquid-cooled double convection structure includes: a submerged double convection heat exchanger, a multi-stage cooling structure, a pretreatment structure, and an insulation layer. The multi-stage cooling structure and the pretreatment structure are both connected to the submerged double convection heat exchanger, and the insulation layer is disposed on the outside of the submerged double convection heat exchanger. The submerged double convection heat exchanger uses an internal spiral copper tube or copper coil as the gas flow channel. The outer layer of the submerged double convection heat exchanger is a liquid-cooled medium to form strong convection heat transfer. The inner layer of the submerged double convection heat exchanger is a mixed gas of high-temperature fire environment. The multi-stage cooling structure includes a primary cooling layer region and a secondary temperature homogenization region; The pretreatment structure includes a high-temperature filter element, a moisture-absorbing layer, and a low-pressure buffer chamber. The insulation layer includes a vacuum cavity and high-temperature insulation material.

12. The combustible gas detection system for high-temperature fire environments according to claim 10, characterized in that, The method of using a temperature compensation model and combining temperature data to perform numerical correction processing on spectral data to obtain corrected spectral data includes: Zero-point calibration is performed on the spectral data to obtain calibrated spectral data; The calibrated spectral data were subjected to background subtraction and low-pass filtering to obtain standardized spectral data. Standardized spectral and temperature data are input into a temperature compensation model, and the standardized spectral data are numerically corrected using the temperature compensation model to obtain corrected spectral data.

13. The combustible gas detection system for high-temperature fire environments according to claim 12, characterized in that, The deep regression neural network model adopts a multi-layer fully connected structure, which includes an input layer, several hidden layers, and an output layer. The activation function of the deep regression neural network model is the modified linear unit function; The loss function of the deep regression neural network model is the mean squared error loss function; The optimization algorithm for the deep regression neural network model is the adaptive moment estimation algorithm.

14. The combustible gas detection system for high-temperature fire environments according to claim 13, characterized in that, The training of the deep regression neural network model includes: an offline calibration phase and an online adjustment phase; The offline calibration stage is used to collect training data using a standard mixed gas in a preset controlled environment or laboratory. The online fine-tuning stage is used to adjust parameters based on real-time collected data samples during on-site operation, and to update the model parameters of the deep regression neural network model based on the validation set and drift detection strategy.

15. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images