An explosion risk grading early warning system based on fuzzy neural network

Abstract

The application discloses an explosion risk grading early warning system based on a fuzzy neural network, belongs to the technical field of safety monitoring and neural networks, and comprises a data acquisition unit, a parameter preprocessing unit, a fuzzification processing unit, a fuzzy neural network inference unit, a risk grading determination unit and an early warning output unit. The system acquires multi-dimensional environmental parameters in a pipe gallery, performs abnormal value elimination, normalization and fuzzy conversion, inputs a trained fuzzy neural network model, outputs an explosion risk evaluation value, is mapped into a corresponding risk grade, triggers grading early warning and linkage control, and has the advantages that the fuzzy neural network is adopted to realize nonlinear evaluation of explosion risk coupling of multiple parameters, the accuracy and robustness of explosion early warning in a complex pipe gallery environment are effectively improved, dynamic grading early warning and accurate linkage of explosion risk can be realized, the working strength of operation and maintenance personnel is reduced, and the application is suitable for safety operation and maintenance scenes of various underground pipe galleries such as urban comprehensive pipe galleries and cross-river tunnel pipe galleries.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring and neural network technology, specifically to an explosion risk classification and early warning system based on fuzzy neural networks. Background Technology

[0002] Urban underground utility tunnels carry various pipelines such as water supply, power supply, and communications, and are important infrastructure to ensure the normal operation of cities. As the scale of utility tunnel construction continues to expand, special structures such as cross-river tunnels have been put into use. These types of tunnels have enclosed internal spaces, dense pipelines, and high humidity. The accumulation of flammable and toxic gases can easily lead to explosions, poisoning, and other safety accidents, posing a significant threat to the structural safety and daily operation and maintenance of the utility tunnels.

[0003] Currently, most conventional gas early warning methods for utility tunnels rely on fixed thresholds for single gas concentrations to trigger alarms. These methods can only determine if a parameter exceeds a limit, but cannot differentiate between different levels of risk severity. They also fail to comprehensively consider the coupled influence of ambient temperature, air velocity, and other factors on gas diffusion and explosion conditions. In the complex operating environment of utility tunnels, sensors are susceptible to interference from temperature, humidity, and airflow fluctuations. Single-threshold alarms are prone to frequent false alarms and struggle to identify hidden risks arising from the combined effects of multiple parameters, leading to potential missed alarms. Furthermore, most existing systems only provide alarm notification functions and cannot automatically trigger tiered responses based on risk levels. They also cannot automatically optimize early warning standards according to long-term environmental changes in the utility tunnel's operation, failing to meet the development needs of intelligent operation and maintenance.

[0004] The combination of fuzzy logic and neural networks has the ability to process uncertain information and learn autonomously, making it suitable for risk assessment scenarios with multiple coupled factors. However, current applications for early warning of explosion risks in underground utility tunnels still lack in terms of scenario adaptability, hierarchical control mechanisms, and joint response capabilities, and cannot fully meet the actual needs of safe operation and maintenance of utility tunnels.

[0005] To address this, a fuzzy neural network-based explosion risk classification and early warning system is proposed. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies, such as the single method of early warning for explosions in utility tunnels, high false alarm rate, and inability to conduct graded assessments. This invention provides an explosion risk graded early warning system based on fuzzy neural networks. Through multi-parameter acquisition, fuzzification processing, and neural network inference, it achieves accurate graded early warning and coordinated control of explosion risks, thereby improving the intelligent level of safe operation and maintenance of utility tunnels.

[0007] The specific technical solution is as follows: An explosion risk classification and early warning system based on fuzzy neural networks includes: The data acquisition unit is used to collect multi-dimensional environmental parameters within the target monitoring area of ​​the utility tunnel; The parameter preprocessing unit is communicatively connected to the data acquisition unit and is used to perform outlier removal and normalization on the acquired multi-dimensional environmental parameters to obtain effective monitoring parameters. The fuzzification processing unit is communicatively connected to the parameter preprocessing unit and is used to calculate the fuzzy membership degree of the effective monitoring parameters and generate a fuzzified feature vector of the corresponding dimension. A fuzzy neural network inference unit is communicatively connected to the fuzzification processing unit. The fuzzy neural network inference unit has a built-in trained fuzzy neural network model, which is used to input the fuzzified feature vector into the fuzzy neural network model and calculate and output the explosion risk assessment value. The risk classification and determination unit is communicatively connected to the fuzzy neural network inference unit and is used to map the explosion risk assessment value to the corresponding explosion risk level according to a preset level threshold range. The early warning output unit is communicatively connected to the risk classification and determination unit, and is used to output an early warning signal of the corresponding level according to the explosion risk level, and generate corresponding linkage control commands.

[0008] The above solution establishes a complete system architecture covering environmental data acquisition, data preprocessing, feature conversion, risk reasoning, risk level determination, and early warning output. Each unit communicates sequentially according to the data flow, forming a closed-loop automated monitoring and early warning chain. Raw environmental data undergoes layer-by-layer processing and analysis, ultimately transforming into early warning signals that directly guide operational actions. No manual intervention is required in the intermediate judgment stages, effectively shortening the response time from parameter anomalies to early warning issuance. Compared to traditional alarm methods that rely solely on comparing a single parameter to a threshold, the multi-unit collaborative processing model integrates more dimensions of information to complete risk assessment, avoiding false alarms caused by fluctuations in a single parameter and capturing potential risks under the combined effect of multiple parameters, reducing the possibility of missed alarms. Simultaneously, the standardized processing flow ensures the uniformity of early warning judgment standards across different time periods and regions, improving the overall stability of utility tunnel safety management.

[0009] As a preferred embodiment of the present invention, the multi-dimensional environmental parameters include five categories of parameters: methane lower explosive limit percentage, hydrogen sulfide volume fraction, carbon monoxide volume fraction, ambient temperature, and air velocity; wherein the sampling range of the methane lower explosive limit percentage is 0–100% LEL, the sampling range of the hydrogen sulfide volume fraction is 0–100 ppm, the sampling range of the carbon monoxide volume fraction is 0–500 ppm, the sampling range of the ambient temperature is -20–80°C, and the sampling range of the air velocity is 0–15 m / s.

[0010] The above-mentioned approach clearly defines the various environmental parameters to be collected, covering multiple influencing factors directly related to risk formation, such as combustible gas concentration, toxic and harmful gas concentration, ambient temperature, and airflow conditions. Simultaneous collection of multiple parameters comprehensively reconstructs the environmental state of the monitored area, providing a sufficient data foundation for subsequent risk analysis and avoiding biased risk assessments due to insufficient parameter dimensions. Different types of parameters correspond to different conditions for risk formation: gas concentration corresponds to the material basis of risk occurrence, temperature corresponds to the environmental conditions for risk initiation, and flow velocity corresponds to the influencing conditions of gas diffusion. Multi-dimensional analysis can more accurately reflect the true risk level, adapting to the enclosed and complex airflow environment of the utility tunnel, and improving the comprehensiveness and reliability of risk assessment.

[0011] As a preferred embodiment of the present invention, the sampling frequency range of the data acquisition unit for various parameters is 1 time / minute to 1 time / second, and the sampling frequency can be increased within this range for high-risk monitoring areas.

[0012] The above scheme limits the frequency range of data acquisition while allowing for increased acquisition frequency in high-risk areas. This adjustable acquisition mode flexibly allocates monitoring resources based on the risk level of the monitored area. In high-risk areas, it obtains denser, continuous data, promptly capturing rapid parameter fluctuations and sudden anomalies, avoiding missing the risk escalation process due to excessively long sampling intervals. In low-risk areas, a relatively gradual acquisition pace can be maintained, reducing data transmission and storage pressure while ensuring monitoring coverage, lowering long-term energy consumption and costs, and achieving a reasonable balance between monitoring accuracy and operating costs.

[0013] As a preferred embodiment of the present invention, the fuzzification processing unit uses a Gaussian membership function to divide each dimension of effective monitoring parameters into fuzzy levels, dividing a single parameter into 3 to 5 fuzzy levels, with each fuzzy level corresponding to a Gaussian membership function; the center parameter and width parameter of the Gaussian membership function are configured according to the safety threshold range of the corresponding parameter.

[0014] The above solution defines the method for parameter fuzzification, employing continuous membership functions to classify parameter levels, with function parameters set based on safety thresholds. Through fuzzy level conversion, continuously changing quantitative parameters can be transformed into qualitative states that conform to safety control logic. This adapts to the characteristic of industrial scenarios where risk boundaries are not absolutely clear, avoiding the problem of repeated alarm triggers due to small fluctuations in parameters near critical values. Continuous membership functions enable smooth transitions between different levels, making changes in risk levels more closely resemble the gradual process of actual working conditions, avoiding abrupt judgments, improving the system's tolerance to normal environmental fluctuations, and making subsequent risk reasoning processes more consistent with the logic of human experience-based judgment.

[0015] As a preferred embodiment of the present invention, the fuzzy neural network model is a five-layer feedforward neural network structure, which includes, in sequence, an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer and an output layer. The number of nodes in the input layer is consistent with the number of dimensions of the effective monitoring parameters, and is used to input the fuzzy feature vector; The number of nodes in the fuzzification layer is the product of the number of nodes in the input layer and the number of fuzziness levels of a single parameter, and is used to calculate the membership value of each parameter for each fuzziness level. The number of nodes in the fuzzy inference layer is the same as the number of preset fuzzy rules. The total number of fuzzy rules is the Cartesian product of the fuzzy levels of each input parameter. Each node corresponds to one fuzzy rule, which is used to perform a fuzzy AND operation on the input membership value and output the activation intensity of the corresponding rule. The normalization layer has the same number of nodes as the fuzzy inference layer and is used to normalize the output of the fuzzy inference layer. The output layer is a single-node structure used for weighted summation and output of explosion risk assessment values ​​in the range of 0 to 1.

[0016] The above scheme clarifies the multi-layered feedforward network structure of the inference model, defining rules for the number of nodes and computational logic in each layer. The layered structure design provides clear functional boundaries for each step of data processing, facilitating subsequent debugging and optimization for different stages and reducing model maintenance difficulty. The correspondence between the number of nodes and input parameters and fuzziness levels ensures that the fuzzy features of each input parameter are fully calculated and processed, preventing information omissions. The fuzzy rules are set according to the combination of parameter levels, covering all working scenarios formed by parameter combinations and avoiding omissions of potential risk combinations. The corresponding fuzzy computational logic accurately reflects the risk superposition effect when multiple parameters act simultaneously, ensuring that the inference process conforms to the objective law of multiple factors jointly influencing risk, and the final output assessment result is closer to the true risk level.

[0017] As a preferred embodiment of the present invention, the training process of the fuzzy neural network model includes the following steps: Collect historical monitoring data of the utility tunnel and the corresponding explosion risk labeling results to construct training and validation datasets; The gradient descent algorithm is used to iteratively optimize the membership function parameters and fuzzy rule weight parameters of the fuzzy neural network. When the loss function value on the validation set is lower than the preset convergence threshold, or when the number of iterations reaches the set upper limit, training is stopped and the model parameters are saved. The loss function adopts the mean squared error function, and the preset convergence threshold ranges from 0.001 to 0.01.

[0018] The above approach defines the model's training process and optimization methods. A dataset is constructed using historical data, and model parameters are adjusted iteratively with corresponding convergence conditions set. Training the model based on historical data from actual operation allows it to learn the correspondence between environmental parameters and risk levels in specific utility tunnel scenarios, adapting to the structural and environmental differences of different utility tunnels. The iterative optimization process gradually reduces the deviation between the model's output and the actual labeled results, continuously improving the accuracy of model evaluation. Setting convergence criteria stops training when the model's accuracy reaches the expected level, controlling training time while ensuring model performance, balancing training efficiency and evaluation accuracy, and ensuring that the final model can stably output reliable evaluation results.

[0019] As a preferred embodiment of the present invention, the risk classification and determination unit divides the explosion risk into 3 to 5 levels, and the threshold range of the explosion risk assessment value corresponding to each level can be dynamically adjusted within the range of 0 to 1 according to the safety protection level of the utility tunnel.

[0020] The above-mentioned solution limits the number of risk levels and the method of threshold adjustment, dividing continuous assessment results into multiple discrete risk levels, with thresholds supporting dynamic adjustment. This multi-level risk classification distinguishes different levels of risk, changing the traditional binary judgment model of "no alert, no safety," allowing maintenance personnel to intuitively grasp the severity of current risks and take appropriate measures, avoiding over- or under-response issues. The dynamically adjustable threshold design adapts to different security levels in utility tunnel scenarios, flexibly adjusting the strictness of control based on the importance of the tunnel, the surrounding environment, and maintenance requirements. This allows the system to adapt to various application scenarios and meet differentiated security management needs.

[0021] As a preferred embodiment of the present invention, the warning signal output by the warning output unit includes at least two of the following: on-site audible and visual warning signal, operation and maintenance terminal push signal, and utility tunnel broadcast warning signal; different risk levels correspond to different warning intensities and push frequencies, and the higher the risk level, the higher the warning intensities and push frequencies.

[0022] The above-mentioned solution limits the output formats of early warning signals, with different warning intensities and push frequencies corresponding to different risk levels. This combination of multiple warning signals simultaneously covers different receiving scenarios for both on-site inspection personnel and back-end maintenance personnel, ensuring that risk information is promptly delivered to all relevant personnel without information omissions due to scenario limitations. Matching warning intensity and push frequency according to risk level allows for gentle reminders to maintenance personnel when the risk level is low, avoiding interference with normal maintenance work; while stronger reminders are used when the risk level is high to ensure sufficient attention is paid to high-risk situations, achieving a tiered response control logic and improving the actual effectiveness of early warning information.

[0023] As a preferred embodiment of the present invention, it further includes a linkage control unit. The input end of the linkage control unit is communicatively connected to the early warning output unit, and the output end is electrically connected to the control end of the ventilation, power supply, fire door and other equipment in the pipe gallery. It is used to receive linkage control commands and drive the corresponding equipment in the pipe gallery to perform linkage operations. The linkage operation includes at least one of starting the ventilation equipment in the corresponding area, cutting off the non-explosion-proof power supply circuit, and closing the fire door of the pipe gallery section.

[0024] The above solution involves adding a linkage control unit to the existing system, clarifying the connection relationships and executable linkage operations of the units. With the added linkage control function, the system can automatically drive field equipment to perform protective actions while issuing warnings, eliminating the need for maintenance personnel to remotely operate each device individually or rush to the site. This effectively shortens the response time for risk management, allowing for intervention measures to be taken in the early stages of risk escalation and preventing further escalation. Multiple linkage operations function from various dimensions, including diluting harmful gases, cutting off ignition sources, and isolating risk areas, forming a multi-layered protection mechanism. This effectively reduces the probability of accidents while controlling the scope of their impact, enhancing the proactive safety protection capabilities of the utility tunnel.

[0025] As a preferred embodiment of the present invention, the data acquisition unit includes multiple sets of distributed sensing nodes, which are evenly arranged along the extension direction of the pipe gallery at a preset interval; the preset interval ranges from 50 to 200 m, and each set of sensing nodes integrates a gas detection module, a temperature detection module and a wind speed detection module.

[0026] The above-mentioned solution defines the deployment method and node composition of the data acquisition units. Distributed nodes are evenly distributed along the utility tunnel at regular intervals, with each node integrating multiple detection modules. This distributed deployment covers all sections of the entire utility tunnel, achieving blind-spot-free monitoring along the entire length. It also allows for precise location of potential risks, facilitating rapid on-site response by maintenance personnel. The integration of multiple detection modules into a single node enables simultaneous acquisition of various environmental parameters at the same monitoring location, ensuring temporal consistency of data from different parameters at the same location. This improves the accuracy of subsequent multi-parameter coupling analysis. Furthermore, the integrated design reduces the complexity of on-site installation and wiring, facilitating unified maintenance and repair in the future.

[0027] As a preferred embodiment of the present invention, it further includes a model incremental update unit. The output end of the model incremental update unit is communicatively connected to the model parameter input end of the fuzzy neural network inference unit. It is used to periodically import newly added pipeline monitoring data and accident handling records, and use the original model parameters as initial values ​​to incrementally train and optimize the fuzzy neural network model using a small-batch gradient descent algorithm. The incremental training period ranges from 1 to 12 months.

[0028] The above approach involves adding incremental update units to the model, clarifying the unit connections, update methods, and update cycles. Regularly updating model parameters allows the model to continuously adapt to environmental changes during the operation of the utility tunnel, such as parameter drift due to equipment aging and baseline shifts caused by seasonal changes. This maintains the long-term evaluation accuracy of the model and prevents performance degradation due to gradual environmental changes. Incremental training based on existing model parameters eliminates the need to retrain the model from scratch, saving computational resources and update time. It also preserves the basic patterns already learned by the model, making parameter updates smoother and preventing significant fluctuations in model performance, thus ensuring the long-term stability and reliability of the system.

[0029] The present invention has the following beneficial effects: First, it improves the accuracy of early warning assessments. The system conducts coupled assessments based on multiple environmental parameters, covering the main factors affecting explosion risk, which is more comprehensive than single-parameter threshold assessments. Combining fuzzy processing and network inference, it can adapt to real-world scenarios where risk boundaries are unclear, reducing false alarms caused by environmental fluctuations. It can also identify hidden risks formed by the superposition of multiple factors, reducing the possibility of missed alarms, and the early warning results are more in line with the actual risk situation.

[0030] Secondly, it enables tiered management and control, improving operational efficiency. The system classifies risks into multiple levels, with different levels corresponding to different intensities of early warning and response measures. This changes the traditional binary model of "no alarm, no safety," avoiding frequent alarms that disrupt normal operations in low-risk situations while ensuring sufficient alerts in high-risk situations. Simultaneously, it automatically links on-site equipment to perform protective actions, eliminating the need for manual operation and shortening response time for risk management. This also reduces the monitoring and operational burden on maintenance personnel.

[0031] Third, it possesses excellent scenario adaptability. The risk classification threshold can be flexibly adjusted according to the safety level of the utility tunnel, making it suitable for scenarios with different protection requirements, such as ordinary urban utility tunnels and cross-river tunnel utility tunnels. The model supports incremental updates and can be continuously optimized as the environment of the utility tunnel changes over a long period of time without the need for a complete model reconstruction, ensuring the stability and applicability of the system in the long term.

[0032] Fourth, enhance the proactive safety protection capabilities of the utility tunnel. The system can not only issue early warnings, but also automatically drive ventilation, power outage, fireproofing and other equipment to form multi-layer protection. It can proactively intervene in the early stages of risk escalation, suppress gas accumulation, eliminate ignition sources and limit the spread of risk. It can shift from passive alarm to active prevention and control, effectively reduce the probability of accidents, narrow the scope of accident impact and improve the overall safety level of utility tunnel operation.

[0033] Fifth, deployment and maintenance are convenient. The distributed sensing nodes adopt an integrated design, simplifying on-site installation and wiring. Evenly distributed along the line can achieve full coverage. The modular system architecture has clear functional boundaries for each unit, facilitating later debugging and maintenance of different components and reducing the complexity of system operation and maintenance. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the overall architecture of the explosion risk classification and early warning system based on fuzzy neural network described in this invention; Figure 2 This is a schematic diagram of the five-layer structure of the fuzzy neural network model described in this invention; Figure 3 This is a schematic diagram of the workflow for explosion risk classification and early warning as described in this invention. Detailed Implementation

[0035] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0036] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual images. They should not be construed as limiting the scope of this application. To better illustrate the embodiments of the present invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0037] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present application. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0038] In the description of this invention, unless otherwise explicitly specified and limited, the term "connection" or similar designation indicating a connection between components should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0039] like Figure 1 As shown, this embodiment provides an explosion risk classification and early warning system based on fuzzy neural network, including a data acquisition unit 1, a parameter preprocessing unit 2, a fuzzification processing unit 3, a fuzzy neural network inference unit 4, a risk classification judgment unit 5, an early warning output unit 6, a linkage control unit 7, and a model incremental update unit 8.

[0040] 1. Data Acquisition Unit Data acquisition unit 1 consists of multiple sets of distributed sensing nodes, evenly distributed along the extension direction of the utility tunnel at preset intervals. The preset intervals can be adjusted within the range of 50-200m according to the section length of the utility tunnel. Each set of sensing nodes integrates a gas detection module, a temperature detection module, and a wind speed detection module, and can simultaneously collect five types of environmental parameters: methane lower explosive limit percentage, hydrogen sulfide volume fraction, carbon monoxide volume fraction, ambient temperature, and air velocity.

[0041] The sampling range for methane lower explosive limit percentage is 0–100% LEL, with a detection accuracy of not less than 1% LEL; the sampling range for hydrogen sulfide volume fraction is 0–100 ppm, with a detection accuracy of not less than 0.5 ppm; the sampling range for carbon monoxide volume fraction is 0–500 ppm, with a detection accuracy of not less than 1 ppm; the sampling range for ambient temperature is -20–80℃, with a detection accuracy of not less than 0.5℃; and the sampling range for air velocity is 0–15 m / s, with a detection accuracy of not less than 0.1 m / s.

[0042] The sampling frequency of the sensing nodes can be configured within the range of 1 time / minute to 1 time / second, depending on the safety level of the utility tunnel. The sampling frequency can be increased in high-risk areas. The collected raw data is transmitted to the parameter preprocessing unit 2 via a wired or wireless communication network.

[0043] 2. Parameter preprocessing unit After receiving the raw monitoring data, parameter preprocessing unit 2 first performs outlier removal: it uses the 3σ criterion to identify abnormal data that deviates from the normal range, and replaces the identified outliers with data interpolation between adjacent time points to eliminate invalid data caused by sensor failures and instantaneous interference.

[0044] Then, normalization is performed: a linear normalization method is used to map the values ​​of each dimension parameter to the interval of 0 to 1. The normalization formula is as follows: ;in These are the original parameter values. and These are the lower and upper limits of the range for this parameter, respectively. These are the normalized effective monitoring parameters.

[0045] The effective monitoring parameters obtained after preprocessing are transmitted to the fuzzification processing unit 3.

[0046] 3. Fuzzification Processing Unit The fuzzification processing unit 3 divides the effective monitoring parameters of each dimension into fuzzy levels. In this embodiment, a single parameter is divided into three fuzzy levels: low, medium, and high. Each fuzzy level corresponds to a Gaussian membership function, and the expression of the Gaussian membership function is as follows: ;in The central value of the membership function. The width of the membership function is denoted by , and the two parameters are configured based on the corresponding safety thresholds. For example, for the lower explosive limit percentage of methane, the center value for the low level corresponds to the midpoint of the safe concentration range, the medium level corresponds to the midpoint of the warning concentration range, and the high level corresponds to the midpoint of the dangerous concentration range.

[0047] The membership function is used to calculate the membership value of each effective monitoring parameter, which is consistent with the number of fuzzy levels. The membership values ​​of all parameters are combined to form a fuzzy feature vector, which is then input into the fuzzy neural network inference unit 4.

[0048] 4. Fuzzy Neural Network Inference Unit The fuzzy neural network inference unit 4 contains a pre-trained fuzzy neural network model. The model employs a five-layer feedforward network structure, such as... Figure 2 As shown, the layers are, in order: input layer 41, fuzzification layer 42, fuzzy inference layer 43, normalization layer 44, and output layer 45.

[0049] Input layer 41: The number of nodes is consistent with the number of dimensions of the effective monitoring parameters. In this embodiment, there are 5 nodes, corresponding to five types of monitoring parameters. Its function is to input the fuzzy feature vector into the network.

[0050] Fuzzification layer 42: Each input node corresponds to 3 fuzzy level nodes, so the total number of nodes is 5×3=15. Each node calculates the membership value of the corresponding parameter corresponding to the fuzzy level, and the output is the membership value between 0 and 1.

[0051] Fuzzy inference layer 43: Each node corresponds to one fuzzy rule, and the number of nodes is equal to the total number of fuzzy rules. The total number of rules is the Cartesian product of the fuzziness levels of each input parameter; in this embodiment, the 5 input parameters each correspond to 3 fuzziness levels, and the total number of rules is 3. 5 =243 rules. This embodiment uses Mamdani fuzzy inference. The applicability of each rule is the product of the membership degrees of each parameter in the antecedent of the rule. That is, the activation strength of the rule is obtained by multiplying the input membership values. Fuzzy rules can be set based on experience and historical data of safe operation and maintenance of utility tunnels. Typical rule examples include: if the methane concentration is high, the temperature is high, and the wind speed is low, the explosion risk is high; if the methane concentration is medium, the hydrogen sulfide concentration is high, and the temperature is medium, the explosion risk is medium; if the methane concentration is low and the wind speed is high, the explosion risk is low.

[0052] Normalization layer 44: The number of nodes is the same as that of the fuzzy inference layer. The activation intensity of all fuzzy rules is normalized so that the sum of the outputs of all nodes is 1.

[0053] Output layer 45: This is a single-node structure. It performs a weighted summation of the activation intensities of each rule after normalization and outputs the final explosion risk assessment value, which ranges from 0 to 1. The larger the value, the higher the explosion risk.

[0054] The training process of the fuzzy neural network model is as follows: Dataset construction: Historical monitoring data of the target utility tunnel were collected, including multi-dimensional environmental parameters under different working conditions. At the same time, combined with expert annotation and accident simulation results, the corresponding explosion risk assessment value was labeled for each set of data, and the data was divided into training set and validation set in a 7:3 ratio.

[0055] Parameter initialization: Randomly initialize the center value and width value of each membership function, as well as the weight parameters of the fuzzy rules.

[0056] Iterative training: Gradient descent is used, with mean squared error as the loss function, and backpropagation is employed to adjust the network parameters. The loss function expression is: ;in The number of training samples, These are the actual labeled values. This is the network output value.

[0057] Convergence criterion: Training stops when the loss function value on the validation set falls below a preset convergence threshold, or when the number of iterations reaches a set upper limit. The convergence threshold can be selected in the range of 0.001 to 0.01, and the upper limit for the number of iterations can be set to 1000 to 10000.

[0058] The trained model is saved to the inference unit for real-time risk assessment.

[0059] 5. Risk Classification and Determination Unit After receiving the explosion risk assessment value, the risk classification determination unit 5 maps it to the corresponding explosion risk level according to the preset level threshold range. In this embodiment, the risk is divided into four levels: safe level, early warning level, alert level, and dangerous level. The assessment value range corresponding to each level can be dynamically adjusted within the range of 0 to 1 according to the safety protection level of the utility tunnel. A typical configuration is: safe level 0 to 0.25, early warning level 0.25 to 0.5, alert level 0.5 to 0.75, and dangerous level 0.75 to 1.

[0060] Different types of utility tunnels can be managed differently by adjusting the threshold range. For example, the threshold can be appropriately lowered and the protection level increased for cross-river tunnel utility tunnels.

[0061] 6. Early warning output unit and linkage control unit Based on the determined explosion risk level, the early warning output unit 6 outputs a corresponding early warning signal: Safety level: No warning signals, system operating normally; Warning level: Push notification information to the maintenance terminal, but do not activate audible and visual warnings on-site; Alert Level: Activate on-site low-frequency audible and visual warnings, push alarm information to maintenance personnel, and prompt them to conduct on-site verification; Danger Level: Activate high-frequency, strong light and sound warnings, trigger utility tunnel broadcast alarms, and simultaneously push emergency alarms to all operation and maintenance terminals and the control center. The higher the risk level, the higher the frequency and the wider the coverage of information pushes.

[0062] Simultaneously, the early warning output unit 6 generates a corresponding linkage control command and sends it to the linkage control unit 7. The linkage control unit 7 then drives the corresponding equipment in the pipe gallery to perform the linkage operation. Warning level: No coordinated action is taken; parameter changes are continuously monitored. Alert Level: Activate ventilation equipment in the corresponding area to accelerate gas diffusion; Hazard level: Activate ventilation equipment, cut off non-explosion-proof power supply circuits in the corresponding area, close fire doors in the zone, and limit the spread of risk.

[0063] 7. Model Incremental Update Unit The incremental update unit 8 periodically collects newly added monitoring data and accident handling records from the utility tunnel, expanding the original training dataset. Using the original model parameters as initial values, it iteratively optimizes the fuzzy neural network model using a mini-batch gradient descent algorithm, updating the membership function parameters and rule weights to ensure the model continuously adapts to environmental changes in the utility tunnel. The incremental training cycle can be configured within the range of 1 to 12 months; in the initial stage of operation and maintenance, the update cycle can be shortened to improve the model's adaptation speed.

[0064] like Figure 3 As shown, the workflow of this system includes the following steps: S1: The data acquisition unit collects multi-dimensional environmental parameters within the pipe gallery at a set frequency; S2: The parameter preprocessing unit removes outliers and normalizes the raw data to obtain effective monitoring parameters; S3: The fuzzification processing unit calculates the fuzzy membership degree of the effective parameters to generate fuzzy feature vectors; S4: The fuzzy neural network inference unit inputs the feature vectors into the trained model and calculates and outputs the explosion risk assessment value; S5: The risk grading determination unit determines the explosion risk level based on the threshold range; S6: The early warning output unit triggers corresponding early warning and linkage operations based on the risk level; S7: Return to step S1 and continue cyclic monitoring.

[0065] It is worth noting that in the fuzzy neural network-based explosion risk classification and early warning system of this embodiment, the risk classification determination unit has a built-in dynamic risk threshold adjustment model. The dynamic risk threshold adjustment model calculates the risk offset based on the explosion risk assessment values ​​and the changing trends of multi-dimensional environmental parameters at the current and historical times using the following comprehensive risk evolution equation, and dynamically corrects the preset level threshold range: ; in: Let be the risk offset at time t, with a dimension of 1; The explosion risk assessment value output by the fuzzy neural network inference unit at time t has a range of [0,1] and a dimension of 1. Let be the normalized value of the i-th effective monitoring parameter at time t, with a value range of [0,1] and a dimension of 1. , To effectively monitor the number of dimensions of parameters; , is a first-order forward difference with a dimension of 1; , is a second-order forward difference with a dimension of 1; , is a first-order forward difference with a dimension of 1; , is a second-order forward difference with a dimension of 1; This is the risk inertia coefficient, with a value range of 0.2 to 0.6 and a dimension of 1. These are the parameter-driven coefficients, ranging from 0.1 to 0.4, with a dimension of 1. The coefficient is the cooperative coupling coefficient, with a value range of 0.05 to 0.2 and a dimension of 1. The risk contribution weights for the i-th dimension parameter satisfy the following conditions: Its dimension is 1; Let the coupling sensitivity weights of the i-th dimension parameter satisfy the following condition: Its dimension is 1; It is a nonlinear activation function used to constrain coupling effects to occur only in the worsening trend; The dynamic risk threshold adjustment model is based on the risk offset. The upper limit threshold of the current risk level is adjusted using the following formula: ; in This is the original upper limit threshold for the current risk level. The correction coefficient has a value range of 0.1 to 0.3 and a dimension of 1. when hour, The warning threshold has been lowered, and the standards have been tightened; when hour, The warning threshold has been raised and the standards have been relaxed; the revised threshold range is used to remap the explosion risk assessment value to the revised explosion risk level.

[0066] Parameter description table:

[0067] Typical configuration values ​​(standard scenario for urban integrated utility tunnels): ; (corresponding to methane %LEL, hydrogen sulfide, carbon monoxide, temperature, and wind speed, respectively). .

[0068] Example: Scenario description: Within a certain utility tunnel monitoring area, the system takes three consecutive samples (time interval) The data is as follows:

[0069] Current risk level: Safe, original upper limit threshold .

[0070] Step 1: Calculate the first-order difference ; ; Step 2: Calculate the second-order difference ; ; ; Step 3: Substitute into the risk evolution equation (using typical configuration values, ignoring parameters other than methane and temperature) Step 4: Threshold Correction This indicates that the risk is accelerating, and the threshold has been lowered from 0.25 to 0.2475, with the early warning standard being moderately tightened. Although currently... If the threshold for safety level is far exceeded, the system will directly trigger a higher-level warning. However, in actual scenarios with a slow escalation, this early tightening mechanism can gain a valuable warning lead time of several minutes to tens of minutes.

[0071] Technical effects: First, it proactively senses the inertia of risk evolution, enabling early warning. Traditional methods only... An alarm is triggered only after a fixed threshold is exceeded. This equation captures risk acceleration through second-order forward difference, predicting trends before the risk value exceeds the original threshold and tightening the warning benchmark in advance, significantly advancing the warning trigger time and gaining valuable time for on-site response.

[0072] Second, it accurately captures the synergistic deterioration effect of multiple parameters, reducing the false alarm rate. Utility tunnel explosions are often not caused by a single parameter abrupt change, but rather by the simultaneous deterioration of multiple factors such as increased combustible gas concentration, poor ventilation, and excessively high temperatures. The synergistic coupling term in this equation... This ensures that risk offsets are amplified only in scenarios where multiple parameters deteriorate simultaneously, allowing the warning threshold for such high-risk combinations to be lowered earlier. Compared to traditional methods that rely solely on linear superposition, the coupling term can more accurately reflect the nonlinear escalation characteristics of risk when multiple factors act together, effectively reducing the false negative rate.

[0073] Third, adaptive threshold adjustment mitigates the dual contradiction of false positives and false negatives. Fixed thresholds are prone to false negatives when risk rises slowly, and prone to false positives when fluctuations around the threshold. This equation is based on... Positive and negative bidirectional adjustment threshold: when (When risk worsens rapidly) the threshold is lowered to improve sensitivity; within the same system When the risk is mitigated or improved, the threshold is raised to filter out interference. This mechanism, which dynamically adjusts according to the evolving risk situation, significantly improves the problem of missed alarms without significantly increasing the overall false alarm rate.

[0074] Fourth, the calculation is simple and efficient, meeting real-time requirements. The equation only involves historical data from the current time step and the two previous time steps, requiring no complex integration or iterative calculations, and the computational complexity is only a fraction of the original value. It is fully compatible with the real-time sampling requirements of utility tunnel monitoring systems, which are measured in seconds. Furthermore, all parameters are dimensionless pure quantities, and the coefficient configuration is intuitive, facilitating engineering deployment and on-site parameter adjustment.

[0075] Working principle and process: The dynamic risk threshold adjustment model is built into the risk classification and determination unit as a key preprocessing step before risk level determination. The specific steps are as follows: Step 1: Historical Data Caching and Differential Calculation The fuzzy neural network inference unit outputs the current time. Explosion risk assessment value Calculate the first-order difference respectively , and second-order difference , .

[0076] Step 2: Comprehensive Calculation of Risk Offset Substitute the difference results into the comprehensive risk evolution equation, and proceed according to the preset... The configuration parameters are used to calculate the risk offset at the current moment. This offset characterizes the degree and direction of deviation of the current risk evolution from a stable state.

[0077] Step 3: Dynamic Adjustment of Risk Level Thresholds Read the original upper limit threshold corresponding to the current preset risk level (safe level, warning level, alert level, or dangerous level). According to the corrected formula Calculate the corrected threshold. If The upper limit threshold has been lowered, and the warning boundaries have been tightened; if The upper limit threshold has been raised, and the warning boundary has been relaxed.

[0078] Step 4: Risk Level Remapping Explosion risk assessment value The current risk level is reassessed by comparing it with the revised threshold range. If a threshold reduction results in… If the risk level falls into a higher risk range, the warning will be upgraded; if the threshold is raised and If it falls into a lower risk level range, it will be downgraded.

[0079] Step 5: Tiered Early Warning and Coordinated Execution The early warning output unit outputs a corresponding level of early warning signal and linkage control command based on the corrected risk level. When the risk level increases, it triggers stronger audible and visual warnings and more comprehensive linkage operations (such as starting ventilation, cutting off power supply, and closing fire doors). When the risk level decreases, the warning intensity is weakened accordingly.

[0080] Step 6: Rolling Data Updates The system stores the current data in a cache, continuously overwriting the oldest data, and then proceeds to the next sampling period for cyclical calculation, achieving continuous dynamic adjustment of the threshold range. The entire correction process is seamlessly integrated with the original system workflow, without altering the original data acquisition, preprocessing, and fuzzy inference chains; only the risk classification and determination stage introduces an intelligent dynamic threshold adjustment mechanism.

[0081] In summary, the working principle of the explosion risk classification and early warning system based on fuzzy neural networks provided in this embodiment is as follows: The system operates according to a complete chain of data acquisition, preprocessing, feature transformation, intelligent reasoning, hierarchical judgment, and linked output, while also supporting continuous model optimization.

[0082] Various sensing elements deployed along the lines continuously acquire environmental data such as gas concentration, temperature, and wind speed within the utility tunnel, and upload this data to the processing unit via a communication network. After data upload, invalid data with abnormal fluctuations is first identified and removed using statistical methods to eliminate the impact of sensor malfunctions and transient interference. Then, parameters of different dimensions are normalized to unify various parameters into the same numerical range, providing a basis for subsequent joint calculations.

[0083] The preprocessed valid parameters enter the fuzzification process. The quantitative parameter values ​​are converted into multiple fuzzy levels of membership results through a continuous membership function. This simulates the characteristics of the fuzzy transition of risk boundaries in manual judgment and avoids repeated alarm state jumps when the parameters fluctuate slightly near the critical value.

[0084] The fuzzified feature vectors are fed into a multi-layer feedforward network model. Through layer-by-layer operations across multiple nodes, the model performs inference calculations on the multi-dimensional fuzzy features according to preset combination rules, ultimately outputting continuous risk assessment results that intuitively reflect the explosion risk level of the current monitoring area. The model's operation rules and parameters are trained using historical data, enabling it to fit the risk change patterns under the combined effects of multiple parameters, making it more consistent with actual working conditions than simple threshold judgments.

[0085] After obtaining the risk assessment results, the system divides the continuous assessment results into multiple discrete risk levels according to the pre-set level threshold range. Different levels correspond to different early warning methods and linkage strategies. In the low-risk state, only background recording and prompts are made. As the risk increases, on-site audio-visual prompts, equipment linkage, and other measures are gradually triggered. The higher the risk, the stronger the response.

[0086] In addition, the system can periodically summarize newly added operational data and accident handling records, perform incremental training on the basis of the original model, slowly adjust the model parameters, adapt to the slow changes in the environmental baseline during the long-term operation of the utility tunnel, and ensure that the early warning standards always fit the current actual working conditions.

[0087] How to use: The use of this system is mainly divided into three stages: deployment and debugging, daily operation and iterative optimization.

[0088] During the deployment and commissioning phase, based on the total length of the utility tunnel, fire compartment settings, and risk distribution characteristics, sensing nodes integrating multiple detection functions were first deployed along the tunnel at reasonable intervals, completing the installation of power supply and communication lines for these nodes. A backend management platform was then built, connecting all sensing nodes with on-site control equipment. Past operational monitoring data of the target utility tunnel was collected, and combined with the experience of maintenance personnel and industry safety standards, risk levels were labeled for data under different operating conditions. This served as the basis for training the initial inference model. Simultaneously, based on the safety protection level of the utility tunnel, risk classification threshold ranges were set, matching corresponding early warning methods and linkage equipment action logic, completing the system's joint debugging and testing.

[0089] During routine operation, the system continuously collects environmental parameters within the utility tunnel at a set sampling frequency, automatically completing the entire process of data cleaning, normalization, fuzzy conversion, and risk inference, and determining the current risk level in real time. When the risk is at a low level, the system only records data and displays it routinely; when the risk rises to the corresponding level, it automatically pushes alerts to the maintenance terminal or activates on-site early warning devices, simultaneously driving the ventilation, power outage, and fireproofing equipment in the corresponding area to perform linked actions. Maintenance personnel conduct on-site verification and handling based on the early warning level and location information, without requiring continuous manual monitoring of the data.

[0090] During the iterative optimization phase, monitoring data, early warning records, and handling results from a certain period are regularly compiled. Validated data is added to the training samples to incrementally update the model and correct parameter deviations. Simultaneously, based on problems exposed in actual operation, the risk classification thresholds and linkage strategies are adjusted to ensure the system's early warning standards continuously adapt to the operational status of the utility tunnel, maintaining a stable early warning effect in the long term.

[0091] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. An explosion risk classification and early warning system based on fuzzy neural networks, characterized in that, include: The data acquisition unit is used to collect multi-dimensional environmental parameters within the target monitoring area of ​​the utility tunnel; The parameter preprocessing unit is communicatively connected to the data acquisition unit and is used to perform outlier removal and normalization on the acquired multi-dimensional environmental parameters to obtain effective monitoring parameters. The fuzzification processing unit is communicatively connected to the parameter preprocessing unit and is used to calculate the fuzzy membership degree of the effective monitoring parameters and generate a fuzzified feature vector of the corresponding dimension. A fuzzy neural network inference unit is communicatively connected to the fuzzification processing unit. The fuzzy neural network inference unit has a built-in trained fuzzy neural network model, which is used to input the fuzzified feature vector into the fuzzy neural network model and calculate and output the explosion risk assessment value. The risk classification and determination unit is communicatively connected to the fuzzy neural network inference unit and is used to map the explosion risk assessment value to the corresponding explosion risk level according to a preset level threshold range. The early warning output unit is communicatively connected to the risk classification and determination unit, and is used to output an early warning signal of the corresponding level according to the explosion risk level, and generate corresponding linkage control commands.

2. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The multi-dimensional environmental parameters include five categories: methane lower explosive limit percentage, hydrogen sulfide volume fraction, carbon monoxide volume fraction, ambient temperature, and air velocity. The methane lower explosive limit percentage is collected in the range of 0–100% LEL, the hydrogen sulfide volume fraction is collected in the range of 0–100 ppm, the carbon monoxide volume fraction is collected in the range of 0–500 ppm, the ambient temperature is collected in the range of -20–80°C, and the air velocity is collected in the range of 0–15 m / s.

3. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The data acquisition unit samples various parameters at a frequency ranging from 1 time / minute to 1 time / second. The sampling frequency can be increased within this range for high-risk monitoring areas.

4. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The fuzzification processing unit uses a Gaussian membership function to classify the fuzziness levels of each effective monitoring parameter, dividing a single parameter into 3 to 5 fuzziness levels, with each fuzziness level corresponding to a Gaussian membership function; the center parameter and width parameter of the Gaussian membership function are configured according to the safety threshold range of the corresponding parameter.

5. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The fuzzy neural network model is a five-layer feedforward neural network structure, which includes an input layer, a fuzzification layer, a fuzzy inference layer, a normalization layer, and an output layer in sequence. The number of nodes in the input layer is consistent with the number of dimensions of the effective monitoring parameters, and is used to input the fuzzy feature vector; The number of nodes in the fuzzification layer is the product of the number of nodes in the input layer and the number of fuzziness levels of a single parameter, and is used to calculate the membership value of each parameter for each fuzziness level. The number of nodes in the fuzzy inference layer is the same as the number of preset fuzzy rules. The total number of fuzzy rules is the Cartesian product of the fuzzy levels of each input parameter. Each node corresponds to one fuzzy rule, which is used to perform a fuzzy AND operation on the input membership value and output the activation intensity of the corresponding rule. The normalization layer has the same number of nodes as the fuzzy inference layer and is used to normalize the output of the fuzzy inference layer. The output layer is a single-node structure used for weighted summation and output of explosion risk assessment values ​​in the range of 0 to 1.

6. The explosion risk classification and early warning system based on fuzzy neural network according to claim 5, characterized in that, The training process of the fuzzy neural network model includes the following steps: Collect historical monitoring data of the utility tunnel and the corresponding explosion risk labeling results to construct training and validation datasets; The gradient descent algorithm is used to iteratively optimize the membership function parameters and fuzzy rule weight parameters of the fuzzy neural network. When the loss function value on the validation set is lower than the preset convergence threshold, or when the number of iterations reaches the set upper limit, training is stopped and the model parameters are saved. The loss function adopts the mean squared error function, and the preset convergence threshold ranges from 0.001 to 0.

01.

7. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The risk classification and determination unit divides the explosion risk into 3 to 5 levels, and the threshold range of the explosion risk assessment value corresponding to each level can be dynamically adjusted within the range of 0 to 1 according to the safety protection level of the utility tunnel.

8. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The warning signals output by the warning output unit include at least two of the following: on-site audible and visual warning signals, operation and maintenance terminal push signals, and utility tunnel broadcast warning signals; different risk levels correspond to different warning intensities and push frequencies, and the higher the risk level, the higher the warning intensity and push frequency.

9. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, It also includes a linkage control unit, whose input terminal is communicatively connected to the early warning output unit, and whose output terminal is electrically connected to the control terminal of equipment such as ventilation, power supply, and fire doors in the pipe gallery, for receiving linkage control commands and driving the corresponding equipment in the pipe gallery to perform linkage operations; the linkage operation includes at least one of starting the ventilation equipment in the corresponding area, cutting off the non-explosion-proof power supply circuit, and closing the fire doors of the pipe gallery zones.

10. The explosion risk classification and early warning system based on fuzzy neural network according to claim 1, characterized in that, The data acquisition unit includes multiple sets of distributed sensing nodes, which are evenly distributed along the extension direction of the pipe gallery at preset intervals. The preset interval ranges from 50 to 200 m, and each set of sensing nodes integrates a gas detection module, a temperature detection module, and a wind speed detection module.

Images