Tunnel underground water pressure-limiting discharge management system and method based on data driving

Abstract

The invention belongs to the technical field of tunnel engineering and hydrogeological engineering, and particularly relates to a tunnel underground water pressure-limiting discharge management system and method based on data driving. A tunnel underground water pressure-limiting discharge management system based on data driving comprises a monitoring module, a processing module, a control module and an execution module. The monitoring module is used for collecting contact pressure, water pressure, steel bar stress, steel arch internal force, lining displacement, steel bar strain, underground water level and water turbidity as multi-source monitoring data; the contact pressure comprises the contact pressure between the primary support and the secondary lining and the contact pressure between the primary support and the surrounding rock. And the processing module is used for dynamically adjusting the weight coefficient of each monitoring index through an adaptive particle swarm optimization algorithm according to the multi-source information, and calculating a comprehensive early warning index # imgabs0 #. According to the scheme, through multi-source data real-time monitoring and self-adaptive algorithm dynamic adjustment, accurate drainage early warning is achieved, the structural risk is reduced, and safety and reliability are improved.

Description

Technical Field

[0001] This solution belongs to the technical fields of tunnel engineering and hydrogeological engineering, and specifically relates to a data-driven tunnel groundwater pressure-limited discharge management system and method. Background Art

[0002] my country's operational transportation tunnels are unprecedented in scale. Water and mud inrush are a significant engineering hazard in tunnel construction in karst regions and a key factor in causing partial collapse of lining structures during operation. Pressure-limited groundwater discharge throughout the lifecycle of tunnel projects plays a crucial role in preventing severe economic losses and adverse social impacts.

[0003] Traditional tunnel groundwater drainage systems primarily rely on passive drainage, activating pumps in an emergency when a single control indicator, such as water pressure, reaches a threshold. This system has significant limitations: First, the activation of drainage measures relies heavily on manual experience, resulting in a delayed response and difficulty in real-time monitoring and rapid response to changes in groundwater pressure. Second, in actual projects, the imprecision of fixed thresholds prevents accurate response to drainage needs, making misjudgments and missed detections prone to failure, and failing to effectively prevent disasters caused by excessive groundwater pressure. The imprecise drainage methods of traditional tunnel groundwater drainage systems compromise tunnel safety and reliability, increasing operational risks. Summary of the Invention

[0004] The purpose of this solution is to provide a data-driven tunnel groundwater pressure-limited discharge management system and method to address the problem that the inaccurate drainage methods of traditional tunnel groundwater discharge systems affect the safety and reliability of tunnels and increase operational risks.

[0005] To achieve the above objectives, this solution provides a data-driven tunnel groundwater pressure-limited discharge management system, including:

[0006] A monitoring module is used to collect contact pressure, water pressure, steel stress, steel arch internal force, lining displacement, steel strain, groundwater level, and water turbidity as multi-source monitoring data; the contact pressure includes the contact pressure between the primary support and the secondary lining and the contact pressure between the primary support and the surrounding rock;

[0007] The processing module is used to dynamically adjust the weight coefficients of each monitoring indicator through the adaptive particle swarm optimization algorithm according to the multi-source information, and calculate the comprehensive early warning index TSI. The calculation formula of the TSI is shown in the following formula (1):

[0008]

[0009] Among them, λ is the correction coefficient of the actual engineering scenario, w iThe weight coefficients of each monitoring indicator are obtained based on the adaptive particle swarm optimization algorithm, and the sum is 1; S i is the evaluation coefficient of each multi-source monitoring data. If the multi-source monitoring data does not exceed the corresponding preset safety threshold, then S i ＝1 , if it exceeds the corresponding preset safety threshold, then D 监测 is multi-source monitoring data, D 安全 is the corresponding preset safety threshold;

[0010] The control module is used to output the four-level response of pressure-limited emission according to the comprehensive warning index TSI and generate a corresponding disposal plan. The relationship between TSI and the four-level response of pressure-limited emission is shown in the following formula (2):

[0011]

[0012] The execution module is used to execute the pressure-limited discharge operation according to the disposal plan.

[0013] And, a data-driven tunnel groundwater pressure-limited discharge management method using a data-driven tunnel groundwater pressure-limited discharge management system.

[0014] The principle and technical effectiveness of this solution are as follows: This solution uses a monitoring module to collect eight types of heterogeneous data (such as water pressure, stress, and displacement). The processing module then uses an adaptive particle swarm algorithm to dynamically calculate the weights of each indicator. When a particular indicator is abnormal, its contribution is amplified by the monitoring indicator weight coefficient, thereby triggering an early warning before complex risks emerge. Compared with traditional methods, this solution overcomes the traditional reliance on manual experience and delayed response. Through real-time multi-source data collection and dynamic analysis, it can rapidly respond to changes in groundwater pressure and other factors, accurately determine drainage needs, reduce misjudgments and missed detections, and effectively prevent disasters caused by excessive groundwater pressure. This precise drainage management reduces the adverse effects of groundwater on tunnel lining structures, reduces the risk of local lining collapse, ensures the structural safety of the tunnel during operation, and improves its overall reliability. Furthermore, this solution's rational drainage strategies and emergency measures avoid the severe economic losses and adverse social impacts caused by improper drainage, and reduces risks during tunnel operations.

[0015] Secondly, this solution incorporates turbidity data. When turbidity increases due to increased sediment content, the drainage threshold is automatically lowered to prevent the risk of pipe blockage. This solution uses a particle swarm algorithm to periodically update weights, automatically adjusting sensitive parameters as the tunnel ages, thereby extending its service life.

[0016] Furthermore, during the operation of this solution, continuously collected multi-source monitoring data will be accumulated into a database, providing data support for further optimization of the adaptive particle swarm optimization algorithm, adjustment of weight coefficients, and improvement of the early warning indicator system, thereby continuously improving the performance of this solution. Furthermore, multi-source data and comprehensive analysis results are not only used for immediate drainage decisions, but also provide a reference for long-term tunnel maintenance planning and reinforcement measures, ensuring the safety of the tunnel throughout its life cycle from a broader perspective.

[0017] In summary, this solution achieves accurate drainage warnings, reduces structural risks, and increases safety and reliability through real-time monitoring of multi-source data and dynamic adjustment of adaptive algorithms. It also introduces turbidity detection to prevent blockages and continuously optimizes algorithms to improve performance, ensuring safety throughout the tunnel's lifecycle and reducing economic losses.

[0018] Furthermore, the processing module is further used to establish three control indicators according to the water pressure, the contact pressure between the primary support and the secondary lining, and the contact pressure between the primary support and the surrounding rock, and use the control indicators according to the following formulas (3) and (4):

[0019] The calculation of comprehensive safety factor H is shown as follows:

[0020] H=min(H i ) (3),

[0021]

[0022] Among them, H i is the safety factor of each control index, i = 1, 2, 3; P1 is the measured water pressure, P2 is the measured contact pressure between the primary support and the surrounding rock, P3 is the measured contact pressure between the primary support and the secondary lining, P w is the allowable safety threshold of each monitoring parameter, and f is the calculation rule of the safety factor of each control indicator.

[0023] Furthermore, the comprehensive safety factor H quantitatively describes the safety status of the structure based on the following thresholds: if H ≥ 3, all monitoring parameters are small relative to the preset safety thresholds, the groundwater drainage system is in an inefficient operating state, and the drainage measures should be appropriately adjusted; if 3> H ≥ 1, all monitoring parameters do not exceed the preset safety thresholds, and the groundwater drainage measures meet the current actual situation; if H < 1, at least one monitoring parameter exceeds the preset safety threshold, the groundwater drainage measures cannot meet the current actual situation, and remedial measures should be taken immediately; activate the backup drainage channel and grouting reinforcement.

[0024] This solution not only achieves an accurate assessment of the safety status of the tunnel structure through dynamic calculation of the comprehensive safety factor and classification of graded thresholds, avoiding the misjudgment or omission problem of traditional fixed thresholds, but also significantly improves the scientific nature of drainage control through multi-parameter collaborative analysis (such as water pressure, initial support contact pressure, etc.). At the same time, the graded response mechanism based on the H value (such as optimized drainage when H ≥ 3, emergency grouting when H < 1) forms a full-process management from early warning to disposal, effectively reducing the risk of sudden water and mud, thereby transforming traditional passive drainage into active prevention and control. Furthermore, the comprehensive safety factor H not only serves drainage control, but can also be expanded to a structural health monitoring indicator, laying the foundation for the subsequent introduction of intelligent technologies (such as machine learning and digital twins), and ultimately promoting the upgrade of tunnel operation and maintenance from experience-based to data-driven standardized management.

[0025] Furthermore, the processing module also uses an adaptive particle swarm optimization algorithm coupled with an LSTM time series prediction model to dynamically adjust the weight coefficients through the following steps:

[0026] S10: Input historical multi-source monitoring data sequence X t ={x t-n ,x t-n+1 ,...,x t}, where x t Output the predicted value of the next k steps for the multi-source monitoring data at time t Establish an LSTM time series prediction model and establish the formula as shown in the following formula (5):

[0027]

[0028] Among them, f t ,i t ,o t They are forget gate, input gate and output gate respectively, C t is the cell state, h t is the hidden state, W * and b * is a trainable parameter, σ is the Sigmoid function, and ⊙ represents the Hadamard product;

[0029] S20: Prediction results based on the LSTM model Construct the objective function J(β i ) to minimize the volatility variance of the comprehensive early warning indicator I, J(β i ) function is shown in the following formula (6):

[0030]

[0031] Among them, m is the length of the sliding time window, λ is the regularization coefficient; the position of the sth particle in the particle swarm is and speed The update formula is shown in the following formula (7):

[0032]

[0033] Among them, the inertia weight ω(t) is adaptively adjusted with the number of iterations, and the adjustment formula is shown in the following formula (8):

[0034]

[0035] in, is the optimal solution for the particle, g best is the global optimal solution, c1, c2 are learning factors, r1, r2~U(0,1), T max is the maximum number of iterations;

[0036] S30: When a sudden change in groundwater level or surrounding rock fracture rate is detected, the LSTM model is retrained and W is updated. * and b * , and regenerate the initial distribution of the particle swarm.

[0037] By combining an adaptive particle swarm algorithm with an LSTM time series prediction model, this solution not only achieves dynamic optimization of weight coefficients, enabling drainage control to accurately match real-time geological and hydrological changes, but also significantly improves the solution's ability to respond to sudden risks through predictive modeling. While enhancing drainage accuracy, this solution effectively balances computational efficiency and optimization accuracy using an intelligent algorithm adaptive adjustment mechanism, avoiding the misjudgment problem caused by traditional fixed thresholds. In addition, the long-term stability of this solution has been significantly improved. False alarms have been reduced through regularization constraints and variance minimization design, while the LSTM time series prediction function further empowers this solution with advanced warning and preventive maintenance capabilities. In terms of scalability, the framework is compatible with multi-source monitoring data, providing technical support for tunnel life cycle management. Ultimately, while improving safety, it reduces operation and maintenance energy consumption and equipment loss, achieving a comprehensive upgrade from passive response to intelligent pre-control.

[0038] Furthermore, the monitoring module is also used to collect the seepage volume Q, pore water pressure P p and surrounding rock deformation rate V d When the processing module calculates the comprehensive early warning index, it also calculates the seepage flow Q, pore water pressure P p and surrounding rock deformation rate V d As a dynamic parameter, and expanding the calculation formula of the comprehensive early warning index, the expanded formula is shown in the following formula (9):

[0039]

[0040] Among them, Q0, P p0、V d0 are the benchmark thresholds of seepage rate, pore water pressure and surrounding rock deformation rate respectively; β n+1 , β n+2 , β n+3 The dynamic weight coefficient of the newly added parameter is optimized through the LSTM-APSO coupling model to meet When Q>Q0 or P p >P p0 or V d >V d0 When γ n+1 , γ n+2 , γ n+3 Take 1.5, otherwise take 1;

[0041] The dynamic parameter coupling analysis method identifies the seepage path and locates the potential water inflow point through the Darcy law and the mutation detection of the seepage flow rate Q. The calculation formula of Q is shown in the following formula (10):

[0042]

[0043] Where k is the permeability coefficient, A is the water flow area, is the hydraulic gradient;

[0044] Combined pore water pressure P p and deformation rate V d Calculate the radius R of the plastic zone of the surrounding rock p , R p The calculation formula is shown in the following formula (11):

[0045]

[0046] Where R0 is the tunnel radius, σ c is the compressive strength of surrounding rock, G is the shear modulus, δ c is the critical deformation;

[0047] Add a new loss function term to the LSTM-APSO model and update β synchronously through gradient descent n+1 , β n+2 , β n+3 , the loss function term is shown in the following formula (12):

[0048]

[0049] By introducing dynamic monitoring parameters such as seepage rate, pore water pressure, and surrounding rock deformation rate, this solution not only significantly improves the comprehensive early warning indicator system, enabling the solution to identify potential risks earlier, but also significantly improves the accuracy of disaster warnings by establishing a seepage path analysis model and surrounding rock stability assessment method. This solution uses a dynamic weight adjustment mechanism and an optimized loss function to effectively reduce the false alarm rate while significantly enhancing anti-interference capabilities, thereby comprehensively improving overall reliability. In addition, through intelligent threshold learning and monitoring network optimization, this solution can extend the service life of tunnel structures while reducing emergency maintenance costs. Its flexible architectural design enables it to adapt to various geological environments and construction conditions.

[0050] Furthermore, the processing module implements dynamic correction and early warning optimization of the prediction model through the following steps:

[0051] S40: Define the prediction error function E t To evaluate the model accuracy, E t The calculation formula is shown in the following formula (13):

[0052]

[0053] in is the predicted value of the i-th indicator at time t, D i (t) is the actual monitoring value; when E t When the error is greater than the preset threshold, the model retraining process is triggered. The retraining process is shown in the following formula (14):

[0054]

[0055] θ is the LSTM model parameter, η is the learning rate;

[0056] S50: Establish a sliding window confidence evaluation function, as shown in the following formula (15):

[0057]

[0058] Where T is the window length, I is the indicator function; when C T When ≥0.95, the model is judged to meet the prediction confidence standard;

[0059] S60: After the confidence condition is met, the predicted value is adopted Make early warning decisions for the period Δt ahead, and the early warning decision index I predict As shown in the following formula (16):

[0060]

[0061] According to the early warning decision index I predictTriggering a yellow warning, orange warning, or red warning is as shown in the following formula (17):

[0062]

[0063] S70: Establish a dual optimization objective, as shown in the following formula (18):

[0064]

[0065] The coordinated update of model parameters θ and weight coefficients β is achieved through alternating optimization.

[0066] This solution utilizes a dynamic learning mechanism to continuously optimize the prediction model, ensuring high-precision predictions over the long term. It also utilizes a multi-stage verification process to ensure scientifically sound early warning decisions. First, a comprehensive model self-assessment system is established, automatically triggering a retraining process when prediction deviations exceed the acceptable range, enabling the solution to self-improve. Second, a confidence management mechanism is introduced, ensuring that only rigorously verified predictions are used for early warning decisions, effectively ensuring early warning reliability. Finally, a collaborative optimization strategy is employed to simultaneously improve model accuracy and parameter sensitivity. Technically, this solution significantly extends early warning lead times, significantly reduces the likelihood of false alarms, and maintains scalability through a modular design. In terms of engineering applications, it supports flexible adaptation to diverse monitoring scenarios and provides a basis for preventive maintenance decisions. In terms of economic operation and maintenance, it effectively reduces the frequency of emergency response and the workload of manual review. Through deep integration with other system modules, this solution not only improves the closed-loop management process from data collection to early warning decisions, but also promotes the transformation and upgrade of tunnel safety management from post-event response to pre-event prevention.

[0067] Furthermore, the control module combines the early warning model with the four-level response of pressure-limited discharge, uses the prediction model to estimate the discharge effect during discharge, and dynamically adjusts the discharge response according to the prediction result. The combination of the early warning model and the four-level response of pressure-limited discharge is achieved through the discharge effect prediction model, and the discharge effect prediction model establishes the discharge volume-water pressure response function shown in the following formula (19):

[0068]

[0069] Among them, Q e is the planned emissions; is the predicted pore water pressure at time t; w j is the response level weight coefficient, j = 1, 2, 3, 4, corresponding to the four levels of response respectively; τ j is the time decay constant of each level;

[0070] The control module optimizes the response level selection by the dynamic response adjustment algorithm shown in the following formula (20) and formulates the constraint conditions shown in the following formula (21):

[0071]

[0072] Among them, η j is the target pressure adjustment coefficient for each level; ξ j is the threshold value of the allowable pressure change rate; The upper and lower limits of displacement for each level.

[0073] Furthermore, the control module updates the parameters of the emission effect prediction model in real time through a feedback correction mechanism; wherein, the adaptive learning rate is calculated as follows: the initial learning rate is used as the numerator, and 1 plus the sum of the squares of the difference between the actual pore water pressure and the predicted pore water pressure is used as the denominator; the control module constructs a comprehensive benefit function through a multi-objective optimization control module; the comprehensive benefit function includes three parts: the first part is 1 minus the ratio of the maximum safe pore water pressure excess value to the failure pore water pressure, and then multiplied by the corresponding coefficient; the second part is the sum of each planned emission amount multiplied by the corresponding cost coefficient, and then multiplied by the corresponding coefficient; the third part is the sum of the ratio of the actual operation times of each equipment to the maximum allowable operation times, and then multiplied by the corresponding coefficient; these three parts together constitute the comprehensive benefit function.

[0074] First, by establishing a response relationship model between drainage volume and water pressure changes, this solution can intelligently match the optimal response level, significantly improving drainage efficiency while ensuring structural safety. Second, the introduction of an adaptive learning mechanism enables this solution to continuously track environmental changes and continuously optimize control parameters. Finally, the multi-objective optimization framework balances key factors such as safety, economy, and equipment maintenance, forming a scientific basis for decision-making. In terms of operation and maintenance, the intelligent scheduling of drainage equipment workloads not only ensures the reliability of this solution but also extends the service life of key equipment. In terms of economic benefits, this solution significantly reduces operating costs by precisely controlling the balance between drainage volume and energy consumption. In terms of management decision-making, the visual analysis interface provides managers with multi-dimensional data support.

[0075] Furthermore, when the execution module performs the pressure-limited discharge operation, it obtains the water turbidity data collected by the monitoring module, associates the water turbidity data with the discharge time, analyzes the changing trend of the water turbidity during the current discharge process, and summarizes the clarification rules; at the same time, the volume of the newly added water source is calculated by obtaining the discharge volume and water storage volume during the discharge process, and the change in the volume of the newly added water source is combined with the changing trend of the water turbidity, to analyze the impact of the newly added water source on the water turbidity during the discharge process; based on the analysis results, a drainage power prediction model is established, and the drainage power prediction model takes the real-time water turbidity data and the discharge time as input, and combines the influencing factors of the newly added water source to predict the optimal power required for subsequent drainage; when the execution module performs the pressure-limited discharge operation, it continuously monitors the change in the water turbidity. If it is found that there is a deviation between the changing trend of the water turbidity and the output result of the drainage power prediction model, the model is corrected in time.

[0076] This solution utilizes the characteristic that the turbidity of water bodies clarifies over time, and analyzes its changing trends in combination with the discharge time. This can determine the water purification rules during the drainage process in real time, provide a basic basis for power adjustment, and avoid misjudgment of drainage strategies due to abnormal fluctuations in turbidity. Secondly, this solution calculates the volume of new water sources through drainage and water storage, and combines it with the turbidity change trend for analysis. It can accurately identify the impact of sudden water sources on the drainage process, avoid the increase in turbidity caused by the new water source from being misjudged as drainage failure, and improve the adaptability of this solution to complex hydrological environments. Furthermore, this solution constructs a power prediction model based on real-time turbidity data, discharge time, and the influencing factors of new water sources. It can predict the optimal drainage power under different working conditions in advance, avoid power waste while ensuring drainage efficiency, and achieve energy-saving operation. Finally, this solution continuously monitors turbidity changes and modifies the prediction model, forming a closed-loop control system called "monitor-predict-adjust." This allows the drainage equipment's power and startup time to be dynamically optimized according to actual operating conditions. This ensures efficient and stable operation despite sudden changes in water sources and turbidity fluctuations. This significantly improves the intelligence and reliability of tunnel groundwater pressure-limited drainage, reducing structural safety risks caused by untimely drainage or improper power matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a functional module diagram of a data-driven tunnel groundwater pressure-limited discharge management system in an embodiment of the present invention.

[0078] Figure 2 This is a flow chart of a processing module in an embodiment of the present invention by coupling an adaptive particle swarm optimization algorithm with a long short-term memory network time series prediction model.

[0079] Figure 3 This is a flow chart of the processing module in an embodiment of the present invention for implementing dynamic correction and early warning optimization of the prediction model. DETAILED DESCRIPTION

[0080] The following will clearly and completely describe the concept and technical effects of the present invention in conjunction with the embodiments to fully understand the purpose, features and effects of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention:

[0081] like Figure 1 As shown, a data-driven tunnel groundwater pressure-limited discharge management system includes a monitoring module, a processing module, a control module and an execution module.

[0082] The monitoring module is used to collect contact pressure, water pressure, steel stress, steel arch internal force, lining displacement, steel strain, groundwater level and water turbidity as multi-source monitoring data; the contact pressure includes the contact pressure between the primary support and the secondary lining and the contact pressure between the primary support and the surrounding rock.

[0083] The processing module is used to dynamically adjust the weight coefficients of each monitoring indicator through the adaptive particle swarm optimization algorithm according to the multi-source information, and calculate the comprehensive early warning index TSI. The calculation formula of the TSI is shown in the following formula (1):

[0084]

[0085] Among them, λ is the correction coefficient of the actual engineering scenario, w i The weight coefficients of each monitoring indicator are obtained based on the adaptive particle swarm optimization algorithm, and the sum is 1; S i is the evaluation coefficient of each multi-source monitoring data. If the multi-source monitoring data does not exceed the corresponding preset safety threshold, then S i =1, if it exceeds the corresponding preset safety threshold, then S i =

[0086] D 监测 is multi-source monitoring data, D 安全 is the corresponding preset safety threshold.

[0087] Specifically, S i is the evaluation coefficient of multi-source monitoring data, where S1 corresponds to water pressure, S2 corresponds to the contact pressure between primary support and surrounding rock, S3 corresponds to the contact pressure between primary support and secondary lining, S4 corresponds to steel bar stress, S5 corresponds to the internal force of the steel arch of primary support, S6 corresponds to lining displacement, S7 corresponds to steel bar strain, and S8 corresponds to the rate of groundwater level rise.

[0088] S9 corresponds to the turbidity level of the water.

[0089] The processing module is further used to establish three control indicators according to the water pressure, the contact pressure between the primary support and the secondary lining, and the contact pressure between the primary support and the surrounding rock, and use the control indicators to calculate the comprehensive safety factor H as shown in the following formulas (3) and (4):

[0090] H=min(H i ) (3),

[0091]

[0092] Among them, H i is the safety factor of each control index, i = 1, 2, 3; P1 is the measured water pressure, P2 is the measured contact pressure between the primary support and the surrounding rock, P3 is the measured contact pressure between the primary support and the secondary lining, P w is the allowable safety threshold of each monitoring parameter, and f is the calculation rule of the safety factor of each control indicator.

[0093] The control module is used to output the four-level response of pressure-limited emission according to the comprehensive warning index TSI and generate a corresponding disposal plan. The relationship between TSI and the four-level response of pressure-limited emission is shown in the following formula (2):

[0094]

[0095] The execution module is used to execute the pressure-limited discharge operation according to the disposal plan.

[0096] Specifically, the comprehensive safety factor H quantitatively describes the structural safety status according to the following thresholds:

[0097] If H ≥ 3, all monitoring parameters are small relative to the preset safety thresholds, the groundwater drainage system is in an inefficient operating state, and drainage measures should be appropriately adjusted;

[0098] If 3>H≥1, all monitoring parameters do not exceed the preset safety thresholds, and the groundwater discharge measures meet the current actual situation;

[0099] If H<1, then at least one monitoring parameter exceeds the preset safety threshold, and the groundwater discharge measures cannot meet the current actual situation. Remedial measures should be taken immediately; activate the backup drainage channel and grouting reinforcement.

[0100] Among them, Figure 2As shown in the figure, the processing module realizes dynamic optimization of the weight coefficients of monitoring indicators by coupling the adaptive particle swarm optimization algorithm (APSO) with the long short-term memory network (LSTM) time series prediction model. The specific process is as follows: first, the historical multi-source monitoring data sequence is input, and the forget gate, input gate, output gate and cell state calculation mechanism of the LSTM model are used to build a time series prediction model to output the parameter prediction values ​​for the next k steps; secondly, the objective function is constructed based on the LSTM prediction results, and the position and velocity of the weight coefficient particles are dynamically updated through the particle swarm algorithm. At the same time, the inertia weight is adaptively adjusted with the number of iterations to minimize the fluctuation variance of the comprehensive early warning indicator; finally, when a sudden change in groundwater level or a change in surrounding rock fracture rate is monitored, the LSTM model is triggered to retrain, the model trainable parameters are updated, and the initial distribution of the particle swarm is regenerated to realize dynamic optimization of the weight coefficient and adaptive update of the model.

[0101] Specifically, the processing module also uses an adaptive particle swarm optimization algorithm coupled with the LSTM time series prediction model to dynamically adjust the weight coefficient through the following steps:

[0102] S10: Input historical multi-source monitoring data sequence X t ={x t-n ,x t-n+1 ,...,x t}, where x t Output the predicted value of the next k steps for the multi-source monitoring data at time t Establish an LSTM time series prediction model and establish the formula as shown in the following formula (5):

[0103]

[0104] Among them, f t ,i t ,o t They are forget gate, input gate and output gate respectively, C t is the cell state, h t is the hidden state, W * and b * is a trainable parameter, σ is the Sigmoid function, and ⊙ represents the Hadamard product;

[0105] S20: Prediction results based on the LSTM model Construct the objective function J(β i ) to minimize the volatility variance of the comprehensive early warning indicator I, J(β i ) function is shown in the following formula (6):

[0106]

[0107] Among them, m is the length of the sliding time window, λ is the regularization coefficient; the position of the sth particle in the particle swarm is and speed The update formula is shown in the following formula (7):

[0108]

[0109] Among them, the inertia weight ω(t) is adaptively adjusted with the number of iterations, and the adjustment formula is shown in the following formula (8):

[0110]

[0111] in, is the optimal solution for the particle, g best is the global optimal solution, c1, c2 are learning factors, r1, r2~U(0,1), T max is the maximum number of iterations;

[0112] S30: When a sudden change in groundwater level is detected Or when the surrounding rock fracture rate changes, the LSTM model is triggered to retrain and update W * and b * , and regenerate the initial distribution of the particle swarm.

[0113] The monitoring module is also used to collect the seepage volume Q, pore water pressure P p and surrounding rock deformation rate V d When the processing module calculates the comprehensive early warning index, it also calculates the seepage flow Q, pore water pressure P p and surrounding rock deformation rate V d As a dynamic parameter, and expanding the calculation formula of the comprehensive early warning index, the expanded formula is shown in the following formula (9):

[0114]

[0115] Among them, Q0, P p0 、V d0 are the benchmark thresholds of seepage rate, pore water pressure and surrounding rock deformation rate respectively; β n+1 , β n+2 , β n+3 The dynamic weight coefficient of the newly added parameter is optimized through the LSTM-APSO coupling model to meet When Q>Q0 or P p >P p0 or V d >V d0 When γ n+1 , γ n+2 , γ n+3 Take 1.5, otherwise take 1;

[0116] The dynamic parameter coupling analysis method identifies the seepage path and locates the potential water inflow point through the Darcy law and the mutation detection of the seepage flow rate Q. The calculation formula of Q is shown in the following formula (10):

[0117]

[0118] Where k is the permeability coefficient, A is the water flow area, is the hydraulic gradient;

[0119] Combined pore water pressure P p and deformation rate V d Calculate the radius R of the plastic zone of the surrounding rock p , R p The calculation formula is shown in the following formula (11):

[0120]

[0121] Where R0 is the tunnel radius, σ c is the compressive strength of surrounding rock, G is the shear modulus, δ c is the critical deformation;

[0122] Add a new loss function term to the LSTM-APSO model and update β synchronously through gradient descent n+1 , β n+2 , β n+3 , the loss function term is shown in the following formula (12):

[0123]

[0124] Among them, Figure 3 As shown in the figure, the processing module realizes dynamic correction and early warning optimization of the prediction model through the following process: first, the prediction error function is defined, and the model accuracy is evaluated by comparing the predicted value of the indicator with the actual monitoring value. When the error exceeds the preset threshold, the model retraining is triggered, and the LSTM parameters are updated based on gradient descent; secondly, a sliding window confidence evaluation function is established, and the model confidence is determined by calculating the proportion of error exceeding the limit in the window. When the confidence reaches 0.95 or above, the model is judged to be reliable; then, after the confidence condition is met, the predicted value is used to calculate the early warning decision index of the advance period, and yellow, orange, and red level warnings are triggered according to the index range; finally, a dual optimization objective including the error term and the weight regularization term is constructed, and the coordinated update of the LSTM model parameters and the indicator weight coefficient is achieved through alternating optimization, forming a closed-loop optimization mechanism of "error evaluation-confidence verification-advance warning-parameter optimization".

[0125] Specifically, the processing module implements dynamic correction and early warning optimization of the prediction model through the following steps:

[0126] S40: Define the prediction error function Et To evaluate the model accuracy, E t The calculation formula is shown in the following formula (13):

[0127]

[0128] in is the predicted value of the i-th indicator at time t, D i (t) is the actual monitoring value; when E t When the error is greater than the preset threshold, the model retraining process is triggered. The retraining process is shown in the following formula (14):

[0129]

[0130] θ is the LSTM model parameter, η is the learning rate;

[0131] S50: Establish a sliding window confidence evaluation function, as shown in the following formula (15):

[0132]

[0133] Where T is the window length, I is the indicator function; when C T When ≥0.95, the model is judged to meet the prediction confidence standard;

[0134] S60: After the confidence condition is met, the predicted value is adopted Make early warning decisions for the period Δt ahead, and the early warning decision index I predict As shown in the following formula (16):

[0135]

[0136] According to the early warning decision index I predict Triggering a yellow warning, orange warning, or red warning is as shown in the following formula (17):

[0137]

[0138] S70: Establish a dual optimization objective, as shown in the following formula (18):

[0139]

[0140] The coordinated update of model parameters θ and weight coefficients β is achieved through alternating optimization.

[0141] Among them, the control module combines the early warning model with the four-level response of pressure-limited discharge. During discharge, the prediction model is used to estimate the discharge effect, and the discharge response is dynamically adjusted according to the prediction result. The combination of the early warning model and the four-level response of pressure-limited discharge is realized through the discharge effect prediction model. The discharge effect prediction model establishes the discharge volume-water pressure response function shown in the following formula (19):

[0142]

[0143] Among them, Q e is the planned emissions; is the predicted pore water pressure at time t; w j is the response level weight coefficient, j = 1, 2, 3, 4, corresponding to the four levels of response respectively; τ j is the time decay constant of each level;

[0144] The control module optimizes the response level selection through the dynamic response adjustment algorithm shown in the following formula (20) and formulates the constraints shown in the following formula (21):

[0145]

[0146] Among them, η j is the target pressure adjustment coefficient for each level; ξ j is the threshold value of the allowable pressure change rate; The upper and lower limits of displacement for each level.

[0147] The control module updates the parameters of the emission effect prediction model in real time through the feedback correction mechanism shown in the following formula (22):

[0148]

[0149] Where γ is the adaptive learning rate,

[0150] And through the multi-objective optimization control module, the comprehensive benefit function shown in the following formula (23) is constructed:

[0151]

[0152] Among them, when the execution module performs the pressure-limited discharge operation, it obtains the water turbidity data collected by the monitoring module, associates the water turbidity data with the discharge time, analyzes the changing trend of the water turbidity during the current discharge process, and summarizes the clarification rules; at the same time, by obtaining the drainage volume and water storage volume during the discharge process, the volume of the newly added water source is calculated, and the change in the volume of the newly added water source is combined with the changing trend of the water turbidity, and the impact of the newly added water source on the water turbidity during the discharge process is analyzed; based on the analysis results, a drainage power prediction model is established. The drainage power prediction model takes the real-time water turbidity data and the discharge time as input, and combines the influencing factors of the newly added water source to predict the optimal power required for subsequent drainage; when the execution module performs the pressure-limited discharge operation, it continuously monitors the change in the water turbidity. If it is found that there is a deviation between the changing trend of the water turbidity and the output result of the drainage power prediction model, the model is corrected in time.

[0153] This embodiment also includes a data-driven tunnel groundwater pressure-limited discharge management method that uses a data-driven tunnel groundwater pressure-limited discharge management system.

[0154] In specific implementation, an operating tunnel crossing a water-rich karst area entered a regular monitoring and maintenance phase five years after its completion and opening to traffic. The tunnel is 3.2 kilometers long and buried at depths of 150-280 meters. Its composite lining structure was affected by seasonal rainfall and groundwater level fluctuations during operation, resulting in repeated problems such as lining leakage and increased localized surrounding rock deformation. To ensure operational safety and prevent disasters such as water inrush and structural cracking, a data-driven groundwater pressure-limiting and discharge management system was fully implemented in operational monitoring, enabling dynamic regulation of groundwater pressure and intelligent assessment of structural safety.

[0155] The pore water pressure behind the lining is monitored by distributed optical fiber sensors, which are mainly deployed in the corresponding sections of geological faults. A monitoring point is set every 50 meters to obtain the water pressure value P in real time. p The safety threshold is set at 1.2 MPa. Ultrasonic flow meters are installed in drainage ditches to continuously monitor the seepage volume Q, with the benchmark value being 30m per day. 3 / h, when the seepage rate suddenly increases by more than 50m 3 / h triggers a preliminary warning.

[0156] The contact pressure between the primary support and the secondary lining, and between the primary support and the surrounding rock, is monitored using vibrating-wire stress gauges, with a set of sensors deployed every 100 meters to assess the stress state of the lining structure. The internal forces of the steel arch and the strain of the steel bars are collected in real time using resistance strain gauges, with a focus on monitoring weak points such as expansion joints and intersections.

[0157] The laser rangefinder scans the tunnel wall regularly to monitor the lining displacement rate V dThe threshold is set at 0.2 mm / d; the water turbidity sensor detects the sediment content of the drainage in real time. When the turbidity exceeds 50 NTU, it is automatically identified as a potential risk of sediment loss.

[0158] All data is connected to the cloud platform through the IoT gateway, forming a sliding monitoring window containing 72 hours of historical data, providing a basis for subsequent prediction and optimization.

[0159] The LSTM model uses historical data such as water pressure and seepage to predict parameter trends over the next 24 hours. For example, before the rainy season, the model, by studying historical data, predicts water pressure fluctuations caused by rising groundwater levels and issues a warning of peak pressure for the next three days. The particle swarm optimization algorithm (APSO) dynamically adjusts the weights of various monitoring indicators. When the water pressure in a certain section is continuously above the threshold and the seepage rate increases simultaneously, the algorithm automatically increases the weights of water pressure (40%) and seepage rate (30%), reducing the weights of secondary indicators to generate targeted warnings.

[0160] Based on three core indicators, including water pressure and initial support contact pressure, the comprehensive safety factor H is calculated. If H < 1 (a certain indicator exceeds the limit), the system immediately triggers an orange alert and simultaneously generates a work order containing drainage solutions and structural reinforcement recommendations. A typical example: During a period of continuous heavy rain in a certain month, the water pressure in the K2+300 section reached 1.5MPa (threshold 1.2MPa), and the comprehensive safety factor H = 0.8. The system identified this as a high-risk state, automatically initiated full-power drainage, and coordinated with the maintenance unit to carry out pre-grouting.

[0161] The control module triggers a graded response based on the comprehensive warning index (TSI):

[0162] Safe state (TSI < 1.5): Maintain normal drainage mode, with daily drainage volume controlled at 20-30m 3 , generate structural health reports weekly.

[0163] Early warning status (1.5 ≤ TSI < 1.75): Intensified monitoring (hourly data collection) is initiated, drainage pump operation frequency is optimized, and drainage volume is increased during off-peak hours at night to reduce water pressure buildup. For example, if monitoring reveals a lining displacement rate of 0.18 mm / day (close to the threshold of 0.2 mm / day) in a certain section in a certain month, the system adjusts drainage schedules, increasing drainage volume by 30% during the early morning hours when traffic is low. After three days, the displacement rate returns to a safe range.

[0164] Alarm status (1.75≤TSI<2): forced activation of the backup drainage channel, joint traffic management department to implement speed limit (40km / h), and simultaneous notification of the maintenance team to conduct on-site inspections. In a certain month, the seepage volume in the K1+800 section suddenly increased to 65m 3 / h, the system triggered an alarm response and reduced the water pressure from 1.4MPa to 1.1MPa within 2 hours, thus preventing the expansion of lining cracks.

[0165] Dangerous state (TSI ≥ 2): Fully automatic emergency response: full power drainage, lane closure, activation of emergency lighting in the tunnel, and notification of the competent authorities via the SMS platform.

[0166] Based on the prediction results and the optimized weight coefficients, the system automatically generates a drainage control plan: when the predicted pore water pressure is close to the safety upper limit, the drainage flow rate is increased to reduce the water pressure, while the potential impact of drainage on the stability of the surrounding rock is evaluated; if the deformation rate of the surrounding rock increases abnormally, the system synchronously adjusts the coordination between drainage and support measures, such as using intermittent drainage to reduce formation disturbance and coordinating with advance grouting to reinforce the fracture zone.

[0167] A prediction-monitoring comparison and correction mechanism has been established. When the error between measured data and model predictions exceeds a preset threshold, a model retraining process is automatically triggered. For example, if a sudden downpour during a certain period causes a surge in seepage volume exceeding the model's expectations, the system will collect abnormal data during the rainstorm in real time and update the LSTM model parameters to improve its adaptability to extreme conditions.

[0168] A multi-objective optimization system encompassing safety, economy, and equipment loss is constructed. Safety benefits focus on avoiding risky accidents, economic costs focus on energy and material consumption during drainage operations, and equipment loss focuses on the operating life of drainage pumps and monitoring instruments. Adaptive algorithms dynamically balance these three factors. For example, energy-saving drainage modes are used to reduce costs during low-risk periods, while equipment load is appropriately increased during high-risk periods, prioritizing safety.

[0169] The above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A data-driven tunnel groundwater pressure-limited discharge management system, characterized in that: include: A monitoring module is used to collect contact pressure, water pressure, steel stress, steel arch internal force, lining displacement, steel strain, groundwater level, and water turbidity as multi-source monitoring data; the contact pressure includes the contact pressure between the primary support and the secondary lining and the contact pressure between the primary support and the surrounding rock; The processing module is used to dynamically adjust the weight coefficients of each monitoring indicator through the adaptive particle swarm optimization algorithm according to the multi-source information, and calculate the comprehensive early warning index TSI. The calculation formula of the TSI is shown in the following formula (1): Among them, λ is the correction coefficient of the actual engineering scenario, w i The weight coefficients of each monitoring indicator are obtained based on the adaptive particle swarm optimization algorithm, and the sum is 1; S i is the evaluation coefficient of each multi-source monitoring data. If the multi-source monitoring data does not exceed the corresponding preset safety threshold, then S i =1, if it exceeds the corresponding preset safety threshold, then D 监测 is multi-source monitoring data, D 安全 is the corresponding preset safety threshold; The control module is used to output the four-level response of pressure-limited emission according to the comprehensive warning index TSI and generate a corresponding disposal plan. The relationship between TSI and the four-level response of pressure-limited emission is shown in the following formula (2): The execution module is used to execute the pressure-limited discharge operation according to the disposal plan.

2. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 1, characterized in that: The processing module is further used to establish three control indicators according to the water pressure, the contact pressure between the primary support and the secondary lining, and the contact pressure between the primary support and the surrounding rock, and use the control indicators to calculate the comprehensive safety factor H as shown in the following formulas (3) and (4): H=min(H i ) (3), Among them, H i is the safety factor of each control index, i = 1, 2, 3; P1 is the measured water pressure, P2 is the measured contact pressure between the primary support and the surrounding rock, P3 is the measured contact pressure between the primary support and the secondary lining, P w is the allowable safety threshold of each monitoring parameter, and f is the calculation rule of the safety factor of each control indicator.

3. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 2, characterized in that: The comprehensive safety factor H quantitatively describes the structural safety status based on the following thresholds: If H ≥ 3, all monitoring parameters are small relative to the preset safety thresholds, the groundwater drainage system is in an inefficient operating state, and drainage measures should be appropriately adjusted; If 3>H≥1, all monitoring parameters do not exceed the preset safety thresholds, and the groundwater discharge measures meet the current actual situation; If H<1, then at least one monitoring parameter exceeds the preset safety threshold, and the groundwater discharge measures cannot meet the current actual situation. Remedial measures should be taken immediately; activate the backup drainage channel and grouting reinforcement.

4. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 3, characterized in that: The processing module also uses an adaptive particle swarm optimization algorithm coupled with an LSTM time series prediction model to dynamically adjust the weight coefficients through the following steps: S10: Input historical multi-source monitoring data sequence X t ={x t-n ,x t-n+1 ,...,x t }, where x t Output the predicted value of the next k steps for the multi-source monitoring data at time t Establish an LSTM time series prediction model and establish the formula as shown in the following formula (5): Among them, f t ,i t ,o t They are forget gate, input gate and output gate respectively, C t is the cell state, h t is the hidden state, W * and b * is a trainable parameter, σ is the Sigmoid function, and ⊙ represents the Hadamard product; S20: Prediction results based on the LSTM model Construct the objective function J(β i ) to minimize the volatility variance of the comprehensive early warning indicator I, J(β i ) function is shown in the following formula (6): Among them, m is the length of the sliding time window, λ is the regularization coefficient; the position of the sth particle in the particle swarm is and speed The update formula is shown in the following formula (7): Among them, the inertia weight ω(t) is adaptively adjusted with the number of iterations, and the adjustment formula is shown in the following formula (8): in, is the optimal solution for the particle, g best is the global optimal solution, c1, c2 are learning factors, r1, r2~U(0,1), T max is the maximum number of iterations; S30: When a sudden change in groundwater level or surrounding rock fracture rate is detected, the LSTM model is retrained and W is updated. * and b * , and regenerate the initial distribution of the particle swarm.

5. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 4, characterized in that: The monitoring module is also used to collect the seepage volume Q, pore water pressure P p and surrounding rock deformation rate V d When the processing module calculates the comprehensive early warning index, it also calculates the seepage flow Q, pore water pressure P p and surrounding rock deformation rate V d As a dynamic parameter, and expanding the calculation formula of the comprehensive early warning index, the expanded formula is shown in the following formula (9): Among them, Q0, P p0 、V d0 are the benchmark thresholds of seepage rate, pore water pressure and surrounding rock deformation rate respectively; β n+1 , β n+2 , β n+3 The dynamic weight coefficient of the newly added parameter is optimized through the LSTM-APSO coupling model to meet When Q>Q0 or P p >P p0 or V d >V d0 When γ n+1 , γ n+2 , γ n+3 Take 1.5, otherwise take 1; The dynamic parameter coupling analysis method identifies the seepage path and locates the potential water inflow point through the Darcy law and the mutation detection of the seepage flow rate Q. The calculation formula of Q is shown in the following formula (10): Where k is the permeability coefficient, A is the water flow area, is the hydraulic gradient; Combined pore water pressure P p and deformation rate V d Calculate the radius R of the plastic zone of the surrounding rock p , R p The calculation formula is shown in the following formula (11): Where R0 is the tunnel radius, σ c is the compressive strength of surrounding rock, G is the shear modulus, δ c is the critical deformation; Add a new loss function term to the LSTM-APSO model and update β synchronously through gradient descent n+1 , β n+2 , β n+3 , the loss function term is shown in the following formula (12):

6. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 5, characterized in that: The processing module implements dynamic correction and early warning optimization of the prediction model through the following steps: S40: Define the prediction error function E t To evaluate the model accuracy, E t The calculation formula is shown in the following formula (13): in is the predicted value of the i-th indicator at time t, D i (t) is the actual monitoring value; when E t When the error is greater than the preset threshold, the model retraining process is triggered. The retraining process is shown in the following formula (14): θ is the LSTM model parameter, η is the learning rate; S50: Establish a sliding window confidence evaluation function, as shown in the following formula (15): Where T is the window length, I is the indicator function; when C T When ≥0.95, the model is judged to meet the prediction confidence standard; S60: After the confidence condition is met, the predicted value is adopted Make early warning decisions for the period Δt ahead, and the early warning decision index I predict As shown in the following formula (16): According to the early warning decision index I predict Triggering a yellow warning, orange warning, or red warning is as shown in the following formula (17): S70: Establish a dual optimization objective, as shown in the following formula (18): The coordinated update of model parameters θ and weight coefficients β is achieved through alternating optimization.

7. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 6, characterized in that: The control module combines the early warning model with the four-level response of pressure-limited discharge, uses the prediction model to estimate the discharge effect during discharge, and dynamically adjusts the discharge response according to the prediction result. The combination of the early warning model and the four-level response of pressure-limited discharge is achieved through the discharge effect prediction model, which establishes the discharge volume-water pressure response function shown in the following formula (19): Among them, Q e is the planned emissions; is the predicted pore water pressure at time t; w j is the response level weight coefficient, j = 1, 2, 3, 4, corresponding to the four levels of response respectively; τ j is the time decay constant of each level; The control module optimizes the response level selection by the dynamic response adjustment algorithm shown in the following formula (20) and formulates the constraint conditions shown in the following formula (21): Among them, η j is the target pressure adjustment coefficient for each level; ξ j is the threshold value of the allowable pressure change rate; The upper and lower limits of displacement for each level.

8. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 7, characterized in that: The control module updates the parameters of the discharge effect prediction model in real time through a feedback correction mechanism; wherein the adaptive learning rate is calculated by taking the initial learning rate as the numerator and 1 plus the sum of the squares of the difference between the actual pore water pressure and the predicted pore water pressure as the denominator; The control module constructs a comprehensive benefit function through a multi-objective optimization control module; the comprehensive benefit function consists of three parts: the first part is to subtract the ratio of the maximum safe pore water pressure excess value to the failure pore water pressure from 1, and then multiply it by the corresponding coefficient; the second part is to sum up the planned discharge amounts multiplied by the corresponding cost coefficients, and then multiply it by the corresponding coefficient; the third part is to sum up the ratios of the actual operation times of each equipment to the maximum allowable operation times, and then multiply it by the corresponding coefficient; the sum of these three parts constitutes the comprehensive benefit function.

9. The data-driven tunnel groundwater pressure-limited discharge management system according to claim 8, characterized in that: When the execution module performs the pressure-limited discharge operation, it obtains the water turbidity data collected by the monitoring module, associates the water turbidity data with the discharge time, analyzes the changing trend of the water turbidity during the current discharge process, and summarizes the clarification rules; at the same time, by obtaining the discharge volume and water storage capacity during the discharge process, it calculates the volume of the newly added water source, combines the change in the volume of the newly added water source with the changing trend of the water turbidity, and analyzes the impact of the newly added water source on the water turbidity during the discharge process; Based on the analysis results, a drainage power prediction model is established. The drainage power prediction model uses real-time data on water turbidity and discharge time as input, and combines the influencing factors of new water sources to predict the optimal power required for subsequent drainage. When the execution module performs the pressure-limited discharge operation, it continuously monitors changes in water turbidity. If a deviation is found between the changing trend of water turbidity and the output results of the drainage power prediction model, the model is corrected in a timely manner.

10. A data-driven tunnel groundwater pressure-limited discharge management method, characterized in that: A data-driven tunnel groundwater pressure-limited discharge management system as described in any one of claims 1 to 9 is used.

Images