An underground pipe gallery construction geological disaster risk dynamic management and control system

Abstract

The application discloses a kind of underground pipe gallery construction geological disaster risk dynamic management and control systems, it is related to the field of underground pipe gallery construction geological disaster risk control, in construction section layout multi-source heterogeneous perception terminal, real-time collection displacement, seepage pressure, stress, vibration and other multidimensional geological and construction monitoring data, edge node carries out adjacent point collaborative verification to collected data to exclude pseudo anomaly, according to risk state self-adaptive adjustment perception terminal sampling frequency, extract standardized risk characteristics and complete low-risk event on-site rapid disposal, cloud management and control platform receives the risk characteristic data uploaded on edge side, fusion construction section geological parameters, current process information, surrounding environmental data, based on time series risk deduction model calculation corresponding geological disaster occurrence probability.The application effectively solves the problem of high false alarm and missed alarm rate of existing control mode, and improves the accuracy and foresight of risk analysis significantly, realizes the whole-process closed-loop control from risk identification to disposal verification.

Description

Technical Field

[0001] This invention relates to the field of geological hazard risk management and control technology for underground utility tunnel construction, and in particular to a dynamic management and control system for geological hazard risks during underground utility tunnel construction. Background Technology

[0002] Urban underground utility tunnels are core infrastructure supporting the operation of urban public services. In recent years, the scale of utility tunnel construction in my country has continued to expand. Construction sections generally pass through complex geological conditions such as water-rich sand layers, weak surrounding rock, and shallow backfill soil. Some sections need to pass close to urban main roads, dense residential areas, and existing buildings and structures. During construction, geological disasters such as landslides, water inrushes, collapses, and ground subsidence are easily induced. Once the risks become unstable, they will cause casualties, property losses, and adverse social impacts. Therefore, accurate, dynamic, and closed-loop management of geological disaster risks throughout the entire construction cycle is a core requirement for the safety management of utility tunnel construction.

[0003] One of the mainstream control solutions in the industry is the fixed threshold monitoring and early warning system. This type of system deploys displacement, stress, and seepage pressure sensing terminals in the construction area, transmits the raw monitoring data to the back-end management terminal, and pre-sets fixed early warning thresholds for various monitoring parameters. When the collected parameter value exceeds the threshold, an audible and visual alarm is triggered. This type of solution has low deployment cost and simple implementation logic, and is widely used in small and medium-sized low-risk utility tunnel construction projects. However, it does not consider the differences in parameter response patterns under different geological sections and different construction procedures. The fixed sampling frequency has an inherent contradiction between redundant data and missed data due to abrupt changes. The static threshold judgment mode has a high false alarm and missed alarm rate. It can only achieve post-event alarm and cannot conduct risk prediction in advance. Moreover, after the early warning is triggered, there is no closed-loop process for issuing control instructions and tracking their execution, which cannot effectively curb the evolution of risks.

[0004] Another mainstream control solution is a semi-intelligent control model dominated by human experience. In this type of solution, dedicated safety engineers regularly review monitoring data, combine preliminary geological survey data with their own construction experience to determine the risk level, and then manually issue corresponding control measures and organize their implementation on-site. This type of solution can flexibly adapt to the personalized control needs of special working conditions and is widely used in high-risk key utility tunnel projects. However, its risk assessment results are highly dependent on the individual experience and professional ability of engineers. There is poor consistency in the assessments of different personnel, the response is delayed, it is impossible to quantify the probability of disaster occurrence, there is a lack of standardized effect verification mechanism after the control measures are issued, emergency resource scheduling relies on manual coordination, the response speed is slow, and the risk data throughout the process is stored in a scattered manner, making it impossible to reuse control experience across projects. Summary of the Invention

[0005] This invention proposes a dynamic management and control system for geological disaster risks during underground utility tunnel construction, in order to solve the problems mentioned in the prior art, such as high false alarm and missed alarm rates, insufficient early warning lead time, non-closed-loop management, and delayed response.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a dynamic management and control system for geological disaster risks in underground utility tunnel construction, comprising a multi-source heterogeneous sensing terminal cluster deployed at the tunnel construction face, temporary support structure, surrounding soil and rock layers and surface subsidence impact zone, an edge computing node connected to the sensing terminal cluster via a redundant dual-link communication network composed of a gigabit wired industrial Ethernet ring network and a 5G industry slice private network, a cloud-side management and control platform connected to the edge computing node via a national cryptographic SM2 encrypted transmission link, and multiple types of on-site construction management and control terminals that communicate with the cloud-side management and control platform; The multi-source heterogeneous sensing terminal cluster includes a fixed inclinometer, a steel bar stress gauge, a pore water piezometer, a construction vibration sensor, a stratified settlement gauge, and a rock detection sensor in front of the tunnel face that moves with the tunneling equipment. It collects parameters such as deep horizontal displacement of the soil and rock layers, stress and strain of the support structure, groundwater seepage pressure, construction disturbance vibration acceleration, surface stratified settlement, and integrity parameters of the surrounding rock in front of the tunnel face in real time according to the configured sampling frequency. It also simultaneously marks the construction process labels corresponding to the collection time to form a raw sensing dataset with process identification. The dual-link communication network automatically triggers millisecond-level seamless switching when the wired link transmission delay exceeds 200ms or the packet loss rate is higher than 3%, ensuring the continuity of sensing data transmission. Edge computing nodes are deployed in industrial explosion-proof cabinets within a safe distance behind the working face. After filtering out outliers, performing multi-source time-series alignment with a time synchronization accuracy of 10ms, and desensitizing sensitive fields on the original sensing dataset, the abnormal mutation features of each single-dimensional parameter are extracted to form a standardized initial risk vector, which is then uploaded to the cloud-based management and control platform through a low-latency transmission channel. The cloud-based management platform accesses the preliminary geological survey records of the utility tunnel construction section, the historical geological disaster case database of the same area, and real-time construction progress and tunneling parameter data. Combined with the initial risk assessment vector, it calculates the dynamic probability of occurrence of four core geological disasters—landslides, water inrushes, collapses, and ground subsidence—within a future rolling time window. The probability calculation uses the following formula: ;in For the first The real-time dynamic probability of occurrence of geological disasters, ranging from 0 to 1. The four types of disasters are landslide, water inrush, collapse, and ground subsidence, respectively. α, β, and γ are the weight coefficients of three factors: the credibility of the perceived data, the impact of process disturbance, and the impact of temporal evolution. The sum of the three is 1, and the calibration is based on historical disaster case data. For the i-th sensing and monitoring parameter, the i-th... The contribution weight of a disaster type is such that the sum of the contribution weights of all parameters under the same disaster type is 1. Let be the deviation of the real-time value of the i-th sensing parameter from the safety baseline value at that location. Let be the critical early warning threshold for the i-th sensing parameter corresponding to the j-th type of disaster. Both are based on the same physical dimension, and the ratio is the degree of parameter overstepping the dimensionless value. For the current first Class of construction procedures for the first The inherent disturbance coefficient of the disaster type, with a value range of 0 to 1, is calibrated through statistical analysis of similar engineering construction cases; This represents the percentage of the cumulative duration of similar abnormal parameters at this location within the past 72 hours, with a value ranging from 0 to 1. The risk evolution rate coefficient is calibrated based on the average time from the emergence of an anomaly to its occurrence for similar disasters; For the current system time, This refers to the moment when the corresponding disaster type's abnormal characteristics were first detected at this location; The cloud-based management and control platform calculates... The risk level is divided into four levels: red, orange, yellow, and blue. Dynamic control instructions with unique traceability codes are generated based on the characteristics of the current construction process. These instructions are accurately pushed to the on-site construction control terminals of the corresponding positions. The changes in the perceived data after the execution of the control instructions are continuously tracked to verify the effect, forming a closed-loop control system for the entire process of geological disaster risk perception, assessment, handling, and feedback.

[0007] Furthermore, the deployment spacing and basic sampling frequency of the multi-source heterogeneous sensing terminal cluster are adaptively configured according to the surrounding rock grade determined by the previous geological survey. The deployment spacing of monitoring points in Class I-II hard surrounding rock sections is no more than 50m, and the basic sampling frequency is set to 1 time / 10min. The deployment spacing of monitoring points in Class III-IV soft and hard alternating surrounding rock sections is no more than 30m, and the basic sampling frequency is set to 1 time / 5min. The deployment spacing of monitoring points in Class V and above soft surrounding rock sections is no more than 15m, and the basic sampling frequency is set to 1 time / 1min. When the monitoring parameters of the point meet When the trigger condition is met, the terminal automatically increases the sampling frequency to a continuous sampling mode of 1 time / 1 second, and simultaneously marks the collected data with a high-priority transmission mark, skipping the regular data queuing sequence of edge nodes for priority transmission, ensuring high-resolution data collection in the early stages of risk anomalies, and avoiding missing key features of risk mutations due to excessively large sampling intervals.

[0008] Furthermore, the edge computing node uses a sliding window interquartile range algorithm with a length of 15 consecutive sampling points to filter out outliers in the original sensing dataset. During the processing, the parameter coordination trend of three adjacent monitoring points of the same type is checked simultaneously. When a single point data deviates from the normal range of the window but there is no synchronous change trend of adjacent points, it is determined to be a pseudo outlier caused by device electromagnetic interference or sensing error, and is replaced by the distance weighted interpolation result of adjacent valid collected values ​​within the same window. If a single point of data deviates and is accompanied by synchronous changes in parameters of adjacent points, it is determined to be a real outlier and retained. After outlier filtering, the edge nodes use the Network Time Protocol to perform time-series alignment of all terminal-collected data. Sensitive fields involving the precise coordinates of monitoring points, the identity information of workers, and core construction process parameters are encrypted and desensitized at the field level using the national cryptographic SM4 algorithm. This reduces the interference of invalid data on risk assessment results from the source, while also reducing the data transmission bandwidth usage and computational load of the cloud platform.

[0009] Furthermore, the standardized initial risk assessment vector includes four types of normalized feature values: real-time collected values ​​of each single-dimensional monitoring parameter, linear change rate over the past hour, cumulative change over the past 24 hours, and the difference between the parameter and the corresponding warning threshold. All feature values ​​are mapped to the 0 to 1 range using the extreme value method. Edge computing nodes directly determine blue-level risks based on threshold out-of-bounds conditions of single-dimensional parameters. For scenarios where there is a slight increase in seepage pressure at a single point but no synchronous anomalies in related parameters such as displacement and stress, targeted on-site inspection prompts are directly generated and pushed to the handheld terminals of inspection personnel closest to the abnormal point. There is no need to upload to the cloud-based management platform for secondary analysis. This allows the vast majority of low-level risk warnings to be handled at the edge, significantly shortening the response chain for low-risk events and reducing unnecessary computational overhead on the cloud-based platform.

[0010] Furthermore, the time-series risk simulation model on the cloud-based management and control platform adopts a long short-term memory network structure that integrates feature attention mechanism and process embedding layer. In addition to receiving standardized initial risk vector, the model input layer also embeds real-time operation parameters such as current construction process type, tunneling speed, support strength, and grouting pressure. During model training, a focus loss function is used to address the sample imbalance problem caused by the low proportion of historical disaster samples. The training dataset covers the full-cycle monitoring data of historical disasters during pipeline construction under the same geological conditions, normal monitoring data of risk-free construction sections, and parameter response benchmark data under different construction procedures. The model can only be launched online after its accuracy is verified by the validation set. The output results cover the probability of occurrence of various geological disasters and their corresponding risk levels in three rolling time windows: 4 hours, 8 hours, and 24 hours. The 4-hour window results serve for immediate response decisions, the 8-hour window results serve for dynamic adjustment of construction plans, and the 24-hour window results serve for pre-preparation of emergency resources, comprehensively covering the risk management needs at different time scales.

[0011] Furthermore, when generating dynamic control instructions, the cloud-based control platform strictly matches the corresponding control intensity according to the risk level determined by the assessment. The red risk level corresponds to the control measures of immediately stopping construction at the work face, organizing all workers to evacuate to a safe area, and activating the special emergency plan. The orange risk level corresponds to the control measures of suspending tunneling operations within the risk impact area, carrying out temporary support and reinforcement, and assigning dedicated personnel to monitor the risk points 24 hours a day. The yellow risk level corresponds to the control measures of optimizing tunneling parameters to reduce construction disturbance, increasing the monitoring frequency to three times the normal level, and increasing the frequency of on-site inspections. The blue risk level corresponds to the control measures of maintaining normal construction and continuously monitoring the trend of parameter changes. Each control instruction clearly indicates the responsible position for execution, the completion deadline, and the acceptance and verification standards. After the instruction is pushed out, the confirmation receipt of the receiving terminal is monitored in real time. If no confirmation is received within 3 minutes, the instruction is automatically escalated and pushed to the superior management position. This avoids over-control due to mismatch between control measures and risk levels, which may affect the construction progress, or insufficient control intensity, which may cause safety accidents.

[0012] Furthermore, the cloud-based management and control platform tracks and verifies the execution effect of management and control instructions by using a dynamic time warping algorithm to calculate the actual change curve of the perceived parameters within the preset verification window after the instructions are issued, and the matching degree with the expected management and control effect curve generated by the model. When the matching degree is lower than the qualified threshold of 0.7, the risk probability calculation process is automatically retried to adjust the risk level judgment result and optimize the management and control intensity and specific measures of the management and control instructions. The platform automatically tags the entire process data of this deviation case with special labels, and conducts incremental training of the model during the idle period of system computing resources in the early morning every day to update the network weight parameters of the time series risk inference model. The incremental training process does not affect the platform's normal risk assessment function, which solves the problem that traditional static assessment models cannot adapt to the dynamic changes of on-site geological conditions and the assessment accuracy gradually decreases as construction progresses, thus realizing continuous iterative optimization of the model's assessment capabilities.

[0013] Furthermore, the cloud-based management platform is integrated with a geological disaster emergency material dispatch module. This module pre-inputs data on the location of emergency material storage points, material types, inventory quantities, and real-time access status of transportation channels around all construction sections. When the risk assessment result is red or orange (high-risk level), it uses an improved shortest path algorithm to calculate the shortest transportation time path from each storage point to the risk-affected area, taking into account the personnel distribution and terrain conditions within the risk impact area. This generates a dispatch plan that includes the quantity of materials to be allocated, transportation routes, and information on the personnel to receive the materials, and pushes it to the emergency management terminal. At the same time, it links with the on-site video monitoring system to check the progress of dispatch instructions in real time. This solves the problems of traditional emergency dispatch relying on human experience, slow response speed, and insufficient material matching, and improves the efficiency of emergency response in high-risk scenarios.

[0014] Furthermore, the cloud-based management and control platform adopts a distributed block storage architecture to encrypt and store all raw perception data, risk assessment results, management and control instructions, and execution feedback data. The data retention period is no less than 5 years after the completion of the project. The platform establishes a multi-condition combination index according to three dimensions: construction section mileage, construction procedure type, and risk event level. It supports second-level traceability and query of the entire chain of risk evolution process, assessment records, instruction issuance, and execution feedback data at any time point. The platform regularly conducts statistical analysis on risk data of completed construction sections, extracts safety benchmark values ​​and early warning thresholds for monitoring parameters under different geological conditions and construction procedures, and dynamically calibrates the initial risk assessment parameters for subsequent construction sections. This solves the problems of scattered storage, lack of traceability, and difficulty in reusing control experience in traditional construction risk data, and provides reusable data support for risk management of similar underground utility tunnel construction.

[0015] Furthermore, the on-site construction control terminals include intrinsically safe handheld terminals for mining equipped with UWB high-precision positioning modules, explosion-proof audible and visual alarm terminals fixedly installed at the entrances and risk points of each work face, and vehicle-mounted display terminals installed in the cabs of construction machinery. Based on the holder's job authority and real-time location, the handheld terminal pushes control instructions within the corresponding scope of responsibility. When personnel enter a high-risk area where control measures have not been lifted, the terminal automatically vibrates to remind them. The audible and visual alarm terminal emits different frequencies of audible and visual warning signals according to the risk level. Under the red risk level, the loudest alarm is triggered and the emergency broadcast is activated simultaneously to guide personnel to evacuate in an orderly manner. When the vehicle terminal receives a high-risk control instruction, it automatically locks the tunneling operation permission of the construction machinery. Operation can only be resumed after the risk is cleared and the platform sends an unlocking instruction. This ensures that the control instructions can accurately reach all relevant work positions, avoids the delay in risk handling caused by information transmission gaps, and builds a solid end-point defense line for on-site risk control.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention effectively solves the inherent contradictions of redundant sampling modes and missed anomaly features in existing fixed threshold early warning systems, as well as the problems of high false alarm and missed alarm rates and insufficient early warning lead time caused by the failure to combine geological conditions and construction process differences. It adopts adaptive sampling configuration of multi-source heterogeneous sensing terminals, collaborative anomaly verification of multiple measuring points, and dynamic risk probability inference technology that integrates multi-dimensional features. It can accurately capture weak anomaly features in the early stage of risk, quantify the probability of geological disasters at different time scales, and greatly improve the accuracy and foresight of risk assessment. It does not require manual real-time monitoring of massive amounts of monitoring data, effectively reducing the workload of front-line safety management personnel.

[0017] This invention effectively solves the problems of existing experience-driven management models, such as high dependence on personnel professional capabilities, delayed risk response, mismatch between management intensity and risk level, lack of tracking of instruction implementation, and continuous decline in the accuracy of static assessment models, through a closed-loop management architecture that includes a low-risk direct handling mechanism at the edge, targeted push of hierarchical management instructions, dynamic verification of management effectiveness, and incremental iteration of the assessment model. Low-level risks can be quickly handled on-site, while high-level risks are automatically matched with appropriate management measures. Instructions directly reach workers and construction equipment, avoiding information transmission gaps. At the same time, the model can autonomously optimize its assessment capabilities as construction progresses, balancing the dual needs of safety management and construction efficiency.

[0018] This invention incorporates an intelligent emergency material dispatch module and a full lifecycle risk data management system, effectively solving the problems of low efficiency in existing emergency dispatch relying on manual coordination and difficulty in reusing risk management experience across projects. It can adapt to underground utility tunnel construction scenarios with different geological conditions, different surrounding environmental sensitivities, and different construction scales. It is suitable for conventional low-risk suburban utility tunnel construction projects, and can also meet the high-standard management requirements of extremely high-risk sections such as those passing under urban main roads and densely populated residential areas. It has good industry adaptability and promotion and application value. Attached Figure Description

[0019] Figure 1 This is the overall flowchart of the dynamic management and control system for geological disaster risks during underground utility tunnel construction proposed in this invention; Figure 2 This is a flowchart of the multi-source heterogeneous sensing data acquisition and edge preprocessing process of the present invention; Figure 3 This is a flowchart of the cloud-side time-series risk simulation and control instruction generation process of the present invention; Figure 4 This is a flowchart of the closed-loop verification and model iteration of risk control instructions in this invention; Figure 5 This is a flowchart illustrating the high-risk emergency response and on-site terminal collaboration process of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Reference Figures 1 to 5 This invention discloses a dynamic management and control system for geological disaster risks during underground utility tunnel construction. The system includes a multi-source heterogeneous sensing terminal cluster deployed at the tunnel construction face, temporary support structure, surrounding soil and rock layers, and surface subsidence impact zone; edge computing nodes connected to the sensing terminal cluster via a redundant dual-link communication network composed of a gigabit wired industrial Ethernet ring network and a 5G industry slice private network; a cloud-side management and control platform connected to the edge computing nodes via a national cryptographic SM2 encrypted transmission link; and various types of on-site construction management and control terminals that communicate with the cloud-side management and control platform.

[0022] The multi-source heterogeneous sensing terminal cluster includes a guide wheel type fixed inclinometer with a range of ±30° and a measurement accuracy of 0.01°, a vibrating wire type steel bar stress gauge with a range of ±200MPa, a pore water piezometer with a range of 0-2MPa and an accuracy of 0.1% of the full scale, a triaxial construction vibration sensor with a frequency response range of 1-1000Hz, a magnetic ring type layered settlement gauge with a measurement accuracy of 0.1mm, and a geological radar surrounding rock detection sensor mounted on the cutterhead of the tunneling equipment with a detection depth up to 30m in front of the tunnel face.

[0023] Various terminals collect data in real time from the entire working face of the utility tunnel construction, including deep horizontal displacement of the soil and rock layers, stress and strain of the support structure, groundwater seepage pressure, construction disturbance vibration acceleration, surface stratified settlement, and integrity parameters of the surrounding rock in front of the tunnel face, according to the configured sampling frequency. They also simultaneously label the construction process corresponding to the collection time, forming a raw perception dataset with process identification.

[0024] In the dual-link communication network, the wired link uses armored flame-retardant optical cables laid along the side wall of the pipe gallery to form a gigabit industrial Ethernet ring network with a self-healing time of less than 50ms. The 5G link uses operator-dedicated uRLLC industry slicing with an end-to-end transmission latency of less than 20ms. The edge node sends a heartbeat detection packet to both links every 100ms. When the wired link times out three consecutive heartbeat detections or the packet loss rate is higher than 3%, a millisecond-level seamless switch is automatically triggered. After the wired link recovers stability for 10 minutes, it smoothly switches back to the original link to avoid frequent link jitter and ensure continuous and uninterrupted data transmission.

[0025] The edge computing nodes are equipped with explosion-proof industrial control computers with industrial-grade Core i7 processors, 16GB of memory, and 1TB solid-state drives, with an IP65 protection rating. They are deployed in explosion-proof cabinets in temporary refuge chambers 200m behind the working face and can adapt to high dust and strong vibration construction environments ranging from -20℃ to 60℃. After filtering out outliers, performing multi-source time-series alignment with a time synchronization accuracy of 10ms, and desensitizing sensitive fields in the original sensing dataset, the nodes extract the abnormal mutation features of each single-dimensional parameter to form a standardized initial risk vector, which is then uploaded to the cloud-based management and control platform through a low-latency transmission channel.

[0026] The transmission link between edge nodes and the cloud platform uses the national cryptographic SM2 asymmetric algorithm for identity authentication, and the transmitted content is encrypted using the national cryptographic SM4 symmetric algorithm to prevent core construction data from being tampered with or stolen.

[0027] The cloud-based management and control platform is deployed on a local government cloud dedicated server cluster, configured with 32 CPU cores, 128G memory and distributed block storage resources. It accesses the preliminary geological survey ledger of the pipeline tunnel construction section, the historical geological disaster case database of the same area, and real-time construction progress and tunneling parameter data. The geological survey ledger covers the surrounding rock grade, groundwater level, distribution of adverse geological conditions, and location information of surrounding pipelines and buildings for each mileage section along the pipeline tunnel. The historical disaster case database covers the full-process monitoring data, handling process and cause analysis results of 27 water inrush, collapse, subsidence and landslide disasters that occurred in the local area during underground space construction in the past 10 years. The real-time construction data comes from the construction management system and includes dynamic parameters such as current tunneling mileage, process type, tunneling speed, grouting pressure, and support strength.

[0028] The platform combines the initial risk vector assessment with rolling calculations of the dynamic probability of occurrence of four types of core geological disasters within a preset future time window. The probability calculation uses the following formula: ; In the formula For the first The real-time dynamic probability of occurrence of geological disasters, ranging from 0 to 1. The four types of disasters are landslide, water inrush, collapse, and ground subsidence, respectively. α, β, and γ are the weight coefficients of three factors: the credibility of perceived data, the impact of process disturbance, and the impact of temporal evolution. The sum of the three is 1. The calibration is performed by logistic regression training based on historical cases of 100 construction sections in the same region. For the first The sensing and monitoring parameters for the first The contribution weight of a disaster type is such that the sum of the contribution weights of all parameters under the same disaster type is 1. Let be the deviation of the real-time value of the i-th sensing parameter from the safety baseline value at that location. Let be the critical early warning threshold for the i-th sensing parameter corresponding to the j-th type of disaster. Both are based on the same physical dimension, and the ratio is the degree of parameter overstepping the dimensionless value. The inherent disturbance coefficient of the current type c construction process to the type j disaster is defined as the coefficient of disturbance, with a value ranging from 0 to 1, and is calibrated by statistical analysis of the frequency of disaster occurrence in similar engineering construction cases. This represents the percentage of the cumulative duration of similar abnormal parameters at this location within the past 72 hours, with a value ranging from 0 to 1. The risk evolution rate coefficient is calibrated based on the average time from the emergence of an anomaly to its occurrence for similar disasters; For the current system time, This is the moment when the corresponding disaster type's abnormal characteristics are first detected at this location.

[0029] The platform calculates The value is divided into four risk levels. A value less than 0.2 indicates low risk (blue), between 0.2 and 0.5 indicates moderate risk (yellow), between 0.5 and 0.8 indicates higher risk (orange), and greater than or equal to 0.8 indicates high risk (red). Dynamic control instructions with a unique 20-digit traceability code are generated based on the characteristics of the current construction process. The first 6 digits of the code represent the construction section mileage, the middle 8 digits represent the generation timestamp, and the last 6 digits represent the instruction type and level identifier. The instructions are accurately pushed to the on-site construction control terminals of the corresponding positions. The changes in perceived data after the execution of the control instructions are continuously tracked to verify the effectiveness. This forms a closed-loop control system for the entire process of geological disaster risk perception, assessment, handling, and feedback. This fundamentally solves the problems of poor reliability of the perception link, reliance on single-parameter static thresholds for risk assessment, failure to consider construction disturbances and risk evolution patterns, and numerous missed and false alarms and untimely responses caused by the open-loop control process without feedback in traditional utility tunnel risk control.

[0030] This invention also discloses the specific logic of adaptively configuring the deployment spacing and basic sampling frequency of the multi-source heterogeneous sensing terminal cluster based on the surrounding rock grade determined by the previous geological survey. According to the surrounding rock classification standard of the urban underground pipe gallery construction specification, the joints of Class I to II hard and intact surrounding rock sections are not developed and have strong self-stabilizing ability. The deployment spacing of monitoring points is no more than 50m, and the basic sampling frequency is set to once every 10 minutes, so as to reduce the generation of redundant data while ensuring monitoring coverage. In Class III to IV soft and hard alternating rock sections, joints are well-developed and self-stabilizing capacity is generally low. The spacing between monitoring points should not exceed 30m, and the basic sampling frequency should be set to once every 5 minutes to balance data volume and monitoring accuracy. In Class V and above weak surrounding rock, water-rich sand layer, and backfill soil sections, the rock mass is broken and self-stabilizing capacity is poor. The spacing between monitoring points should not exceed 15m, and the basic sampling frequency should be set to once per minute to increase the monitoring density in high-risk sections.

[0031] When the monitoring parameters of a location meet the trigger condition of exceeding the limit by 0.6 or more, it indicates that the risk has entered the nascent stage. The terminal automatically increases the sampling frequency to a continuous sampling mode of 1 time per second, and simultaneously marks the collected data with a high-priority transmission mark, skipping the regular data queuing sequence of edge nodes for priority transmission. The sensing terminal is locally configured with a 16G cache, which can cache more than 10 minutes of high-frequency data in continuous sampling mode. Even if the transmission is temporarily interrupted, it will be automatically retransmitted after the link is restored, avoiding the omission of key features of risk mutation.

[0032] This setting effectively resolves the inherent contradictions of traditional fixed sampling frequencies, namely, low sampling frequencies are prone to missing risk mutation characteristics, leading to false negatives, while high sampling frequencies generate a large amount of redundant data in daily conditions, occupying transmission and storage resources.

[0033] This invention also discloses the specific implementation logic of outlier filtering, temporal alignment, and desensitization processing of the original sensing dataset by the edge computing node. Outlier filtering employs a sliding window interquartile range algorithm covering 15 consecutive sampling points. It calculates the upper and lower quartiles of all collected values ​​within the window, marking values ​​exceeding 1.5 times the interquartile range as suspected outliers. Simultaneously, it verifies the coordinated change trend of parameters at three adjacent monitoring points of the same type. When a single point's data deviates from the normal range of the window but adjacent points show no synchronized change trend, it is determined to be a pseudo-outlier caused by electromagnetic interference or external force collision. The result is then replaced with a distance-weighted interpolation of adjacent valid collected values ​​within the same window, with higher weights for valid values ​​closer to the outlier during the interpolation process. If a single data point deviates while adjacent data points show synchronous changes in the same direction, it is considered a genuine geological anomaly and retained. After outlier filtering, the edge node acts as a local time server, sending a time synchronization packet to all sensing terminals every minute using the Network Time Protocol (NTP). This keeps the data acquisition time difference between different terminals within 10ms, preventing errors in parameter causality judgments caused by timing misalignments.

[0034] In the data anonymization process, sensitive fields involving precise coordinates of monitoring points, identification information of workers, and core construction process parameters are encrypted at the field level using the national cryptographic algorithm SM4. Only authorized personnel can decrypt and view the data using a dedicated key. All data transmitted outward from edge nodes has undergone anonymization processing, reducing the risk of sensitive information leakage from the source.

[0035] This processing logic effectively solves the problems of high false alarm rates and high data security risks caused by construction interference, timing misalignment, and insufficient protection of sensitive information in traditional systems, while reducing bandwidth consumption caused by invalid data transmission.

[0036] This invention also discloses the composition of a standardized initial risk assessment vector and a mechanism for handling low-risk situations on the edge. The standardized initial risk assessment vector includes four types of feature values: real-time collected values ​​of each single-dimensional monitoring parameter, the linear change rate over the past hour, the cumulative change over the past 24 hours, and the difference between the value and the corresponding parameter's warning threshold. All feature values ​​are mapped to the 0-1 interval using the extreme value method to eliminate the influence of differences in the dimensions of different parameters on the calculation results.

[0037] The rate of change over the past hour is obtained by performing a least-squares linear fit on the sampling points over the past hour to obtain the slope value, and the cumulative change over the past 24 hours is the difference between the current value and the baseline value 24 hours ago. Edge computing nodes can directly determine blue-level risks based on the threshold exceedance of single-dimensional parameters. For scenarios where a single-point parameter increases slightly but related parameters do not show synchronous anomalies, such as a slight increase in the reading of a single piezometer but no change in related parameters such as displacement and stress, and the degree of exceedance does not reach the yellow risk threshold, there is no need to upload the data to the cloud platform for secondary analysis. Instead, targeted on-site inspection prompts are directly generated and pushed to the handheld terminals of the inspection personnel closest to the abnormal point, clearly marking the location of the abnormal point, the abnormal parameter situation, and the key points for on-site verification.

[0038] Actual testing shows that this mechanism can handle more than 80% of low-level alert-type risks at the edge, significantly shortening the response chain for low-risk events. The response time is reduced from 5 to 10 minutes in traditional cloud-based centralized processing to less than 1 minute, while effectively reducing unnecessary computing overhead on the cloud platform.

[0039] This invention also discloses the specific structure and operating logic of the time-series risk projection model mounted on the cloud-based management and control platform. The model adopts a two-layer long short-term memory network structure that integrates a feature attention mechanism and a process embedding layer. The hidden layer dimension is set to 128. In addition to receiving the standardized initial risk judgment vector, the model input layer also encodes seven common pipe gallery construction processes—blasting, tunneling, slag removal, initial support, grouting, secondary lining, and inspection—into 7-dimensional uniquely heated vectors through the process embedding layer. These vectors are then concatenated with the perceived feature vectors and input into the network, allowing the model to automatically learn the normal response range of parameters under different processes, thus avoiding the judgment of normal parameter fluctuations caused by processes as abnormalities.

[0040] The feature attention layer can dynamically adjust the weights of different input features based on real-time parameter changes. For example, it automatically increases the weight of seepage pressure parameters when seepage pressure rises rapidly, and automatically increases the weight of settlement parameters when surface subsidence develops rapidly, replacing the traditional fixed-weight judgment logic. During model training, a focus loss function is used to address the sample imbalance problem caused by the low proportion of historical disaster samples. Higher loss weights are assigned to disaster-related data with fewer samples to prevent the model from biasing towards classifying all samples as risk-free due to an excessively high proportion of normal samples. The training dataset is divided into training, validation, and test sets in a 7:2:1 ratio. The model can only be deployed online when the disaster identification accuracy on the validation set reaches over 90% and the false alarm rate is below 10%. The model outputs results covering the probability of occurrence and corresponding risk levels of various geological disasters in three rolling time windows: 4 hours, 8 hours, and 24 hours. The 4-hour window results serve for immediate response decisions, the 8-hour window results serve for dynamic adjustment of construction team work plans, and the 24-hour window results serve for pre-emptive preparation of emergency resources. This comprehensively covers the risk management needs at different time scales, solving the problems of traditional judgment models that do not consider process disturbances, fixed feature weights, and sample imbalance leading to false alarms and missed alarms, as well as the problem that they can only achieve immediate alarms and cannot support early prevention and control.

[0041] This invention also discloses a hierarchical generation and push confirmation mechanism for dynamic control commands. The cloud-based control platform strictly matches the corresponding control intensity according to the assessed risk level. The red high-risk level corresponds to the control measures of immediately stopping construction at the work site, organizing all workers to evacuate to a safe area along the nearest escape route, and activating the special emergency plan. Orange, a higher risk level, corresponds to control measures such as suspending tunneling operations within the risk impact area, carrying out temporary support and reinforcement, and assigning dedicated personnel to monitor risk points 24 hours a day; yellow, a general risk level, corresponds to control measures such as optimizing tunneling parameters to reduce construction disturbance, increasing monitoring frequency to three times the normal level, and increasing on-site inspection frequency; blue, a low risk level, corresponds to control measures such as maintaining normal construction and continuously monitoring parameter change trends.

[0042] Each control instruction clearly indicates the responsible position, completion deadline, and acceptance verification standards. After the instruction is pushed out, the system monitors the confirmation receipt of the receiving terminal in real time. When the terminal receives the instruction, a full-screen reminder pops up, which the operator must manually click to confirm before the reminder can be closed. If no confirmation receipt is received within 3 minutes, the system automatically escalates the instruction to the corresponding person's direct supervisor and on-site safety officer until confirmation feedback is received.

[0043] This mechanism effectively addresses the problems of traditional control measures relying on human experience for decision-making, over-control leading to delays in construction progress due to mismatch between control intensity and risk level, safety accidents caused by insufficient control intensity, and lack of feedback and accountability after instructions are issued.

[0044] This invention also discloses the tracking and verification logic for the execution effect of control commands and the incremental iteration logic for the model. After the cloud-based control platform issues control commands, it uses a dynamic time warping algorithm to calculate the matching degree between the actual change curve of the perceived parameters within a preset verification window after the command is issued and the expected control effect curve generated by the model. This algorithm can automatically align the time axis deviation of the two curves, avoiding similarity calculation errors caused by fluctuations in parameter response time. The matching degree ranges from 0 to 1, with a higher value indicating that the actual handling effect is more consistent with the expectation.

[0045] When the matching degree falls below the acceptable threshold of 0.7, it indicates that the existing control measures have not achieved the expected results and the risk is still developing. The system automatically re-triggers the risk probability calculation process, upgrades the risk level, and optimizes the intensity and specific measures of control instructions. During the system's idle computing resources from 2:00 AM to 4:00 AM each day, the platform automatically adds special labels to the full-process data of all risk events from the previous day and the deviation data of the handling effect, conducts incremental training of the model, updates the network weight parameters of the time-series risk inference model, and performs accuracy verification on the test set after training to ensure that the recognition accuracy does not decrease before going online to replace the old model. The entire update process does not affect the platform's normal risk assessment function.

[0046] This mechanism addresses the problems of traditional static assessment models being unable to adapt to dynamic changes in on-site geological conditions, the gradual decline in assessment accuracy as construction progresses, the lack of effective tracking of control measures, and the inability to detect risks out of control in a timely manner, thereby enabling continuous iterative optimization of the model's assessment capabilities.

[0047] This invention also discloses the specific operational logic of a geological disaster emergency material dispatch module. The module pre-enters the locations of emergency material storage points, material types, inventory quantities, and contact information for all construction sections. It connects with on-site personnel positioning systems, video surveillance systems, and access status data, updating in real-time information on passage blockages caused by building material stacking and equipment occupation within the utility tunnel. When the risk assessment result is red or orange (high-risk level), the module first calculates the risk impact radius based on the disaster type. Then, combining the material reserves at each storage point and the real-time access routes to the risk-affected area, it calculates the optimal material dispatch plan with the shortest transportation time as the objective. Transportation time is automatically calculated based on path length and transportation method; pedestrian transportation is calculated at an average speed of 1 meter per second, and electric vehicle transportation at an average speed of 5 meters per second. The dispatch plan clearly indicates the material allocation type, quantity, transportation route, and receiving personnel information, and pushes it to emergency management positions and material management personnel. Simultaneously, it automatically switches video surveillance to the material storage points and risk points, facilitating real-time monitoring of dispatch progress by management personnel.

[0048] This module fundamentally solves the problems of traditional emergency dispatch, such as reliance on human experience, low material matching, and slow response speed and insufficient emergency preparedness caused by unreasonable route planning, and greatly improves the efficiency of emergency response in high-risk scenarios.

[0049] This invention also discloses the platform's data storage, indexing, and experience reuse mechanisms. The cloud-based management and control platform adopts a distributed block storage architecture to encrypt and store all raw perception data, risk assessment results, management and control instructions, and execution feedback data. All data is divided into 64MB data blocks, and each data block is stored as a copy on three independent storage servers. The failure of a single server will not result in data loss, and the data retention period is no less than five years after the completion of the project, meeting the requirements for construction project archive management.

[0050] The platform employs a distributed search engine to establish a multi-condition combined index based on three dimensions: construction section mileage, construction procedure type, and risk event level. This supports second-level traceability and querying of the entire data chain, including risk evolution, assessment records, instruction issuance, and execution feedback, at any given time point. After each 100-meter construction section is completed, the platform automatically compiles statistics on the geological conditions, monitoring parameter response patterns, and risk event handling experience for that section. It then extracts the safety benchmark values ​​and early warning thresholds for monitoring parameters under the corresponding geological conditions, dynamically calibrating the initial risk assessment parameters for subsequent adjacent construction sections without requiring manual parameter recalibration.

[0051] This mechanism addresses the problems of traditional construction risk data being scattered across various independent devices, making traceability and querying difficult, and making it difficult to reuse management experience across different sections. It provides accumulable and reusable data support for risk management of similar underground utility tunnel construction projects.

[0052] This invention also discloses multi-type adaptation and linkage logic for on-site construction control terminals. The on-site construction control terminals include intrinsically safe handheld terminals for mining equipped with UWB high-precision positioning modules, explosion-proof audible and visual alarm terminals fixedly installed at the entrances and risk points of each work face, and vehicle-mounted display terminals installed in the cabs of construction machinery.

[0053] The handheld terminal has an explosion-proof rating of Exib I Mb and a positioning accuracy of up to 30 centimeters. It pushes control instructions within the scope of the user's responsibilities based on the user's job authority and real-time location. When personnel enter a high-risk area that has not been released from control, the terminal automatically triggers a vibration reminder to prevent personnel from accidentally entering a dangerous area. The sound pressure level of the audible and visual alarm terminal is not less than 120 decibels. It emits different frequencies of audible and visual warning signals according to the risk level. Under blue risk, it emits a slow-paced alert tone once per second; under yellow risk, it emits a fast-paced alert tone once every 0.5 seconds; under orange risk, it emits a continuous alert tone combined with a high-frequency red flash; under red risk, it triggers the loudest alarm and simultaneously activates an emergency broadcast, which announces the risk location and evacuation route by voice. When the vehicle-mounted terminal receives a red high-risk control command, it automatically sends a high-level lock signal to the construction machinery control system, cutting off the power circuit for tunneling operations. Work can only resume after the risk is eliminated and an unlock command is received from the platform. This avoids situations where workers have to wait for the risk to be eliminated and an unlock command to be received from the platform before resuming work, and prevents workers from blindly carrying out construction work without knowing the risks. It ensures that control commands can accurately reach all relevant work positions, opens up the information transmission link from platform decision-making to on-site execution in risk control, and completely solves the problems of multiple information transmission levels, untimely reception by on-site personnel, and command suspension caused by construction machinery not being linked with the control system under the traditional control model. It builds a solid risk control defense line from the end of on-site operations.

[0054] Reference Figure 1 This flowchart illustrates the system-level overall operational architecture for dynamic management and control of geological hazard risks during underground utility tunnel construction. First, a cluster of multi-source heterogeneous sensing terminals deployed at the construction site collects environmental and structural parameters such as soil and rock displacement, water pressure, and stress in real time, simultaneously labeling them with construction process tags to form a raw dataset. Then, edge computing nodes receive the data, perform preprocessing operations such as outlier filtering and multi-source time-series alignment, extract abrupt change features to form a standardized initial risk assessment vector, and transmit it to the cloud-based management platform. Finally, the cloud-based management platform integrates preliminary survey records and real-time tunneling parameters, uses a time-series risk extrapolation model to continuously calculate the probability of geological hazard occurrence, generates dynamic management instructions, pushes them to the on-site construction management terminals, and continuously tracks data changes to build a closed-loop management system throughout the entire process.

[0055] Reference Figure 2 This flowchart details the specific execution steps of adaptive front-end data acquisition and edge-side data processing. The sensing terminal cluster adaptively configures the sampling frequency based on the basic risk level determined during the survey phase. Once the magnitude of a sudden change in monitored parameters exceeds a set threshold, it automatically increases to the highest frequency for continuous sampling. After receiving the raw sensing dataset, the edge computing nodes use a sliding window interquartile range algorithm to accurately filter out outliers and perform weighted interpolation replacement. Multi-source data is time-series aligned using a unified time reference, and precise positioning information and personnel privacy fields are anonymized. Four types of time-series evolution feature values ​​are extracted to construct an initial risk assessment vector. The edge nodes also possess the edge autonomy capability to directly generate on-site alerts for general-level risk alerts.

[0056] Reference Figure 3This flowchart illustrates the core mechanism of the cloud-based control platform for in-depth geological disaster risk simulation and command matching. The platform deeply integrates the received initial risk assessment vector with historical disaster case libraries, preliminary survey records, and real-time tunneling parameters, inputting them into a long short-term memory network structure equipped with an attention mechanism. This time-series risk simulation model continuously calculates the probability and corresponding risk level of disasters such as landslides and water inrushes within different future time windows. Subsequently, the platform strictly matches different levels of control intensity based on the assessed risk level, distinguishing between response measures such as suspending construction, optimizing parameters, and maintaining normal construction. All generated control commands are clearly labeled with the corresponding responsible entity, time limits, and effectiveness evaluation benchmarks.

[0057] Reference Figure 4 This flowchart illustrates the dynamic tracking and verification of the effectiveness of control command execution, as well as the model's self-learning and evolution mechanism. After the cloud-based control platform issues commands and they are executed by on-site terminals, the platform begins to extract the actual change trends of perceived parameters within a preset time window and compares them with the expected control effect curve generated by the model in advance. When the system finds that the matching degree between the actual curve and the expected curve is lower than the qualified threshold, it automatically triggers a feedback mechanism to promptly adjust the current risk level judgment result and optimize subsequent control command parameters. At the same time, the platform extracts the deviation data generated in this process to supplement the model training dataset, driving the time-series risk extrapolation model to complete incremental iterative updates and continuously improve the accuracy of subsequent risk warnings.

[0058] Reference Figure 5 This flowchart illustrates the system's automated emergency material dispatching and on-site multi-terminal collaborative alarm mechanism in response to high-risk geological disasters. When the risk assessment result from the cloud-based management platform reaches a high-risk level, the system automatically connects to the emergency material dispatching module to quickly retrieve the material storage locations, types, and storage quantities in the surrounding area. Based on the expected impact range of the disaster, the system automatically plans the optimal emergency material dispatching route and distributes it to the site. Regarding terminal collaboration, explosion-proof handheld terminals receive targeted push messages based on the job responsibility list and require mandatory confirmation, while fixedly deployed audible and visual alarm terminals simultaneously issue audible and visual alarms at corresponding frequencies according to the risk level, ensuring rapid response and evacuation of personnel from dangerous areas.

[0059] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A dynamic management and control system for geological hazard risks during underground utility tunnel construction, characterized in that, include: Multi-source heterogeneous sensing terminal cluster: Real-time collection of deep horizontal displacement of soil and rock layers, stress and strain of support structure, groundwater seepage pressure, construction disturbance vibration acceleration, and surface layer settlement parameters of the entire working face of the pipe gallery construction according to the preset sampling frequency, and synchronous labeling of the construction process corresponding to the collection time to form an original sensing dataset with process identification. Edge computing nodes: After preprocessing the original sensing dataset by filtering out outliers, aligning multiple sources in time series, and desensitizing sensitive fields, the abnormal mutation features of each single-dimensional parameter are extracted to form a standardized initial risk vector, which is then transmitted to the cloud-side management and control platform through a low-latency transmission channel. The cloud-based management and control platform integrates the preliminary geological survey records of the utility tunnel construction section, the historical geological disaster case database of the same area, and real-time construction progress and tunneling parameter data. It inputs the initial risk vector into the trained time-series risk extrapolation model, and continuously calculates the probability of occurrence and corresponding risk level of four core geological disasters—landslides, water inrushes, collapses, and ground subsidence—within a preset future time window. It generates dynamic management and control instructions adapted to the actual operation based on the characteristics of the current construction procedures, and pushes them to the on-site construction management and control terminals of the corresponding positions. At the same time, it continuously tracks and verifies the trend of changes in the perceived data after the execution of management and control instructions, forming a closed-loop management and control system for the entire process of geological disaster risk from perception, assessment to disposal and feedback.

2. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The basic sampling frequency of the multi-source heterogeneous sensing terminal cluster is adaptively configured according to the basic risk level determined by the previous geological survey of the monitoring point. When the sudden change of the monitoring parameters of the point exceeds the preset trigger threshold, the sampling frequency is automatically increased to the highest level and the continuous sampling mode is triggered, and a high-priority data transmission identifier is sent to the edge computing node at the same time.

3. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The edge computing node uses a sliding window interquartile range algorithm to filter out outliers in the original sensing dataset. Collected values ​​that exceed 1.5 times the interquartile range between the upper and lower quartiles within the window are identified as outliers. The outliers are replaced by the weighted interpolation result of adjacent valid collected values ​​within the same window. After outlier filtering, the parameters collected by different terminals are aligned according to a unified time reference. Sensitive fields containing accurate information of construction personnel and precise coordinates of monitoring points are desensitized.

4. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The standardized initial risk assessment vector includes four types of feature values: real-time collected value of single-dimensional monitoring parameter, rate of change in the past hour, cumulative change in the past 24 hours, and the difference between the parameter and the preset warning threshold. The edge computing node directly determines the general-level warning risk based on the threshold of the single-dimensional parameter and directly pushes the on-site warning instruction to the construction control terminal near the corresponding location.

5. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The cloud-based management and control platform is equipped with a time-series risk projection model that adopts a long short-term memory network structure that integrates feature attention mechanism. The model training dataset covers the historical disaster full-cycle monitoring data of pipeline construction under the same geological conditions, the normal monitoring data of risk-free construction sections, and the parameter response benchmark data under different construction procedures. The model output results cover the probability of occurrence of various geological disasters and the corresponding risk levels in three rolling time windows of the next 4 hours, 8 hours, and 24 hours.

6. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 5, characterized in that, When generating dynamic control instructions, the cloud-based control platform matches the corresponding control intensity according to the assessed risk level. Higher risk levels correspond to control measures such as suspending construction at the corresponding work face, initiating emergency support preparation, and increasing monitoring frequency. Moderate risk levels correspond to control measures such as optimizing construction and tunneling parameters and increasing on-site inspection frequency. Low risk levels correspond to control measures such as maintaining normal construction and monitoring at the regular frequency. All control instructions are simultaneously marked with the corresponding responsible entity, execution time limit, and effect verification standards.

7. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 6, characterized in that, The cloud-based management and control platform tracks and verifies the execution effect of management and control instructions by calculating the matching degree between the changing trend of perceived parameters within a preset time window after the instructions are issued and the expected management and control effect curve inferred by the model. When the matching degree is lower than the preset qualified threshold, the risk level judgment result is automatically adjusted and the parameters of subsequent management and control instructions are optimized. At the same time, the deviation data is added to the model training dataset to complete the incremental iterative update of the time-series risk inference model.

8. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The cloud-based management and control platform is connected to a geological disaster emergency material dispatch module. When the risk assessment result determines that the risk level is high, it automatically retrieves information on the location, type, and quantity of emergency material storage points around the corresponding construction section, and generates the optimal emergency material dispatch path based on the risk impact range, which is then pushed to the on-site emergency management and control terminal to shorten the emergency response preparation time.

9. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The cloud-based management and control platform performs distributed encrypted storage of all raw sensing data, risk assessment results, management and control instructions, and execution feedback data. It establishes a full-domain data index according to three dimensions: construction section, construction procedure, and risk level, supporting full-process traceability and query of risk evolution at any time point, and providing reliable data support for the calibration of risk benchmark parameters in subsequent adjacent construction sections.

10. The dynamic management and control system for geological disaster risks during underground utility tunnel construction according to claim 1, characterized in that, The on-site construction control terminal includes an explosion-proof handheld terminal equipped with a high-precision positioning module and an audible and visual alarm terminal fixedly installed on each construction work surface. When receiving risk control instructions of different levels, the handheld terminal pushes targeted messages according to the job responsibility list and confirms the receipt status, and the audible and visual alarm terminal emits audible and visual alarm signals of corresponding frequencies according to the risk level.

Images