A topsis fusion theory solution of submarine tunnel water inflow prediction method
Patent Information
- Application Number
- CN202611114581.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-27
- Publication Date
- 2026-08-28
AI Technical Summary
理论解析法具有物理意义明确、参数解释性强的优点，但通常建立在均质、各向同性、边界规则等理想条件之上，难以准确描述海底隧道注浆圈、衬砌及施工扰动的动态影响
本发明的预测方法，涌水量预测模型训练时，利用综合权重进行加权后的输入特征作为输入，使得预测模型在训练初期即可获得影响因素重要性先验，根据隧道断面的相对贴近度确定注意力层约束，使预测模型能够根据断面风险等级自适应调整特征关注重点，隧道断面的相对贴近度根据加权标准化评价矩阵采用TOPSIS方法得到，从而实现了TOPSIS评价结果用于预测模型的训练，预测模型训练时，利用理论涌水量模型构造物理约束损失函数，实现了将渗流理论解作为物理损失约束引入模型训练，降低纯数据驱动模型偏离水文地质规律的风险，通过对初始涌水量预测模型的修正，实现了施工扰动与治理效果的动态修正，整个方法实现了TOPSIS评价结果、渗流理论解和深度学习模型的有效协同，兼顾了预测精度、物理一致性、可解释性和工程适应性。
Smart Images
Figure CN122654986A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tunnel engineering technology, specifically to a method for predicting water inflow in submarine tunnels using TOPSIS fusion theoretical solutions. Background Technology
[0002] The statements herein provide only background information in relation to this invention and do not necessarily constitute prior art.
[0003] Subsea tunnels typically traverse marine overburden, weathered rock masses, fault fracture zones, and jointed fracture zones. They are susceptible to water inrush, sudden water surges, or localized mudslides due to the combined effects of high seawater head, complex recharge boundaries, heterogeneous surrounding rock, and construction disturbances. Accurately predicting water inrush volumes at different cross-sections and construction stages is crucial for determining drainage system capacity, optimizing advanced geological forecasting and grouting schemes, and ensuring construction safety.
[0004] Existing methods for predicting water inflow mainly include theoretical analytical methods, numerical simulation methods, and data-driven methods. Theoretical analytical methods have the advantages of clear physical meaning and strong parameter interpretability, but they are usually based on ideal conditions such as homogeneity, isotropy, and regular boundaries, making it difficult to accurately describe the dynamic effects of grouting rings, linings, and construction disturbances in submarine tunnels. Numerical simulation methods can represent complex boundary conditions, but the modeling cost is high, and the results are sensitive to boundary conditions, mesh generation, and parameter values. Data-driven methods can learn the nonlinear relationships between multiple factors and water inflow, but pure data-driven models are prone to problems such as black-box architecture, insufficient physical constraints, and insufficient generalization ability for small-sample, high-risk conditions.
[0005] The TOPSIS method can be used for comprehensive evaluation of multiple indicators and can reflect the risk level of a cross-section relative to its ideal state. However, in traditional applications, the TOPSIS method is usually only used for risk classification and has not been further involved in feature learning and loss constraints of neural network models. Simply piecing together the TOPSIS evaluation results, seepage theory solutions, and deep learning models still makes it difficult to form an effective synergistic mechanism. Summary of the Invention
[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method for predicting the water inflow of submarine tunnels by fusing TOPSIS theoretical solutions, which overcomes the defects of the current methods for predicting the water inflow of submarine tunnels.
[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, embodiments of the present invention provide a method for predicting the inflow of water into a submarine tunnel, comprising the following steps: Evaluation indicators from the original engineering database of the submarine tunnel project were selected, and a weighted standardized evaluation matrix was constructed based on the comprehensive weights of the evaluation indicators determined by a combination of subjective and objective weighting methods. The relative proximity of the tunnel cross sections is obtained by combining the weighted standardized evaluation matrix with the TOPSIS method; The evaluation index is used as the input feature, and the input feature after weighting by comprehensive weight is used as the input. The feature group attention constraint is set according to the relative closeness. The physical constraint loss function is constructed by the pre-established theoretical water inflow model to train the model and obtain the initial water inflow prediction model. The initial water inflow prediction model was modified based on the blasting vibration damage coefficient and the grouting reinforcement coefficient to obtain the final water inflow prediction model. The water inflow of the tunnel section to be predicted is obtained by combining the input characteristics of the tunnel section to be predicted with the final water inflow prediction model.
[0008] Optionally, the method for obtaining the original engineering database for the undersea tunnel project is as follows: Geological and hydrological data, marine environmental data, construction parameter data, and measured water inflow data of multiple tunnel sections of the constructed subsea tunnel were obtained, with the geological and hydrological data, marine environmental data, and construction parameter data serving as evaluation indicators. The acquired data is processed for missing values, outlier identification, standardization, and time-series alignment to form the original engineering database for the submarine tunnel project.
[0009] Optionally, at least one of the following methods can be used to handle missing values: mean interpolation, median interpolation, adjacent section interpolation, or time-series interpolation. At least one of the following methods is used to identify outliers: interquartile range method, three-standard-deviation method, or engineering threshold method. Evaluation indicators with different dimensions are normalized to achieve standardization.
[0010] Optional geological and hydrological data in the evaluation indicators include seawater depth, tunnel burial depth, surrounding rock permeability coefficient, rock mass integrity coefficient, fault fracture zone width, and pore water pressure; Marine environmental data includes rainfall and tidal variations; Construction parameter data include the grouting ring permeability coefficient, lining permeability coefficient, excavation advance, and blasting vibration intensity.
[0011] Optionally, the subjective and objective weighting method adopts the AHP-entropy weighting method, and the comprehensive weight is obtained according to the AHP weight and the entropy weight.
[0012] Optionally, the method for obtaining the relative proximity to the tunnel cross-section using the TOPSIS method is as follows: The positive and negative ideal solutions are obtained based on the weighted standardized evaluation matrix; Obtain the first Euclidean distance from the tunnel cross-section to the ideal solution; Obtain the second Euclidean distance from the tunnel cross-section to the negative ideal solution; The relative proximity of the tunnel cross sections is obtained based on the first Euclidean distance and the second Euclidean distance.
[0013] Optionally, the method for setting the attention constraint of the feature group based on the relative proximity is as follows: obtain the water inrush risk level of the tunnel cross section based on the relative proximity of the tunnel cross section, and adjust the attention weight corresponding to each input feature according to the water inrush risk level.
[0014] Optionally, when establishing the theoretical inflow model, it is assumed that the seepage process conforms to Darcy's law; the surrounding rock, grouting ring, and lining can be equivalent to homogeneous isotropic media within a single calculation section; the seepage near the tunnel section is mainly radial confluence; and the interfaces between the grouting ring and the lining, and between the surrounding rock and the grouting ring, satisfy the conditions of continuous head and continuous flow.
[0015] Optionally, the objective function is to minimize the error between the measured water inflow at the tunnel cross section and the theoretical water inflow obtained from the theoretical water inflow model. The empirical parameters in the theoretical water inflow model are then inverted, and the inverted empirical parameters are updated to the theoretical water inflow model. A physical constraint loss function is constructed based on the updated theoretical inflow model.
[0016] Optionally, the inversion can be performed using the least squares method, genetic algorithm, particle swarm optimization algorithm, or Bayesian optimization algorithm.
[0017] The beneficial effects of this invention are as follows: The prediction method of this invention uses weighted input features with comprehensive weights as input during the training of the water inflow prediction model. This allows the prediction model to obtain prior knowledge of the importance of influencing factors in the early stages of training. Attention layer constraints are determined based on the relative proximity of the tunnel cross-section, enabling the prediction model to adaptively adjust the focus of features according to the risk level of the cross-section. The relative proximity of the tunnel cross-section is obtained using the TOPSIS method based on the weighted standardized evaluation matrix, thus realizing the use of TOPSIS evaluation results for prediction model training. During prediction model training, a physical constraint loss function is constructed using the theoretical water inflow model, realizing the introduction of the seepage theory solution as a physical loss constraint into the model training, reducing the risk of the purely data-driven model deviating from hydrogeological laws. By correcting the initial water inflow prediction model, dynamic correction of construction disturbance and treatment effect is achieved. The entire method realizes the effective synergy of TOPSIS evaluation results, seepage theory solution, and deep learning model, taking into account prediction accuracy, physical consistency, interpretability, and engineering adaptability. Attached Figure Description
[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0019] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention; Figure 2 This is a comprehensive weighting diagram of the evaluation indicators in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the equivalent three-dimensional radial seepage model established in Embodiment 1 of the present invention; Detailed Implementation It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0020] Example 1 This embodiment provides a method for predicting the inflow of water into a submarine tunnel by fusing theoretical solutions with TOPSIS, such as... Figure 1 As shown, it includes the following steps: Step S1: Select evaluation indicators from the original engineering database of the undersea tunnel project, and construct a weighted standardized evaluation matrix based on the comprehensive weights of the evaluation indicators determined by a combination of subjective and objective weighting methods. This includes the following specific steps: Step S1.1: Data acquisition and preprocessing, constructing the original engineering database for the submarine tunnel project, and selecting the set indicators in the original engineering database as evaluation indicators.
[0021] The method for constructing the original project database includes the following steps: Step S1.1.1: During the construction of the undersea tunnel, geological and hydrological data, marine environmental data, construction parameter data, and measured water inflow data of multiple tunnel sections of the constructed undersea tunnel project are collected. Among them, geological and hydrological data, marine environmental data, and construction parameter data are used as evaluation indicators.
[0022] The geological and hydrological data in the evaluation indicators include seawater depth, tunnel burial depth, surrounding rock permeability coefficient, rock mass integrity coefficient, fault fracture zone width, and pore water pressure.
[0023] The marine environmental data include rainfall and tidal level changes, with the tidal level changes being the difference between high and low tide levels.
[0024] The construction parameters include the grouting ring permeability coefficient, lining permeability coefficient, excavation advance, and blasting vibration intensity.
[0025] In a practical application, the original database of a certain section of an undersea tunnel project is shown in Table 1 below: Table 1. Original database of a certain section of the undersea tunnel project
[0026] In this embodiment, data such as geological survey reports, borehole exposure data, seawater depth data, tunnel burial depth, surrounding rock permeability coefficient, rock mass integrity coefficient, fault fracture zone width, pore water pressure, grouting ring permeability coefficient, lining permeability coefficient, excavation advance, blasting vibration velocity, and water inflow volume measured during construction or estimated according to standard experience are collected to construct the subsequent original engineering database.
[0027] Step S1.1.2: Perform missing value processing, outlier identification, standardization processing, and time-series alignment on the acquired data to form the original engineering database for the submarine tunnel project.
[0028] Among them, at least one of the following methods is used to handle missing values: mean interpolation, median interpolation, adjacent section interpolation, or temporal interpolation. At least one of the following methods is used to identify outliers: interquartile range method, three-standard-deviation method, or engineering threshold method. Standardization is achieved by performing positive, negative, and normalization processes on evaluation indicators with different dimensions.
[0029] The above methods can be implemented using existing technologies, and their specific steps will not be described in detail here. The timing alignment method can also be implemented using existing technologies, and will not be described in detail here.
[0030] Step S1.2: Construct a weighted standardized evaluation matrix based on the comprehensive weights of the evaluation indicators determined using a combination of subjective and objective weighting methods, including the following steps: Step S1.2.1: Determine the comprehensive weight of the evaluation index using a combination of subjective and objective weighting methods.
[0031] In this embodiment, as Figure 2 As shown, the subjective and objective weighting method is the AHP-entropy weighting method, where the AHP weights... wj AHP Entropy weights are obtained by combining initial values from the Analytic Hierarchy Process (AHP) with a judgment matrix and consistency checks. wj E The information entropy of standardized samples is used to obtain the AHP weights and entropy weights. Existing technologies can be used to obtain these weights, so they will not be described in detail here. The comprehensive weights... wj Calculate using the following formula:
[0032] in, mTo evaluate the number of indicators, j This refers to the evaluation indicator number.
[0033] The comprehensive weights of the various evaluation indicators obtained in the above practical applications are shown in Table 2 below: Table 2 Comprehensive Weights of Evaluation Indicators
[0034] Step S1.2.2: Based on the comprehensive weights corresponding to each evaluation index obtained in step S1.2.1, construct a weighted standardized evaluation matrix for all tunnel sections.
[0035] Step S2: Obtain the relative proximity of the tunnel cross-section based on the weighted standardized evaluation matrix and the TOPSIS method. This includes the following steps: Step S2.1: Obtain the positive and negative ideal solutions of the weighted standardized evaluation matrix; The methods for obtaining the positive and negative ideal solutions can be found using existing technologies, and will not be described in detail here.
[0036] Step S2.2: Obtain the first i The first Euclidean distance from each tunnel cross-section to the ideal solution.
[0037] Step S2.3: Obtain the first i The second Euclidean distance from the tunnel cross-section to the negative ideal solution
[0038] Step S2.4: Based on the first Euclidean distance Second Euclidean distance The relative proximity of the i-th tunnel section is obtained. C i .
[0039] Specifically:
[0040] Based on the relative proximity of the i-th tunnel cross section C i The risk level of water inrush at the tunnel cross-section is classified as follows: When the relative proximity is greater than the first set threshold, the corresponding tunnel section is determined to be a high-risk section. When the relative proximity is less than the second set threshold, the corresponding tunnel section is determined to be a low-risk section. When the relative proximity is neither greater than the first set threshold nor less than the second set threshold, the corresponding tunnel section is determined to be a medium-risk section.
[0041] In this embodiment, the first set threshold is 0.7, and the second set threshold is 0.3, that is, when C iA cross section with a value >0.7 is classified as high-risk; a cross section with a value ≤0.3 is classified as high-risk. C i When the value is ≤0.7, it is determined to be a medium-risk section; when C i A section with a value less than 0.3 is considered a low-risk section.
[0042] It is understandable that the first and second set thresholds can be adaptively adjusted based on the quantile thresholds of the engineering historical samples, which will not be described in detail here.
[0043] Step S3: Use the evaluation index as the input feature, the input feature weighted by the comprehensive weight as the input, set the feature group attention constraint according to the relative closeness, construct the physical constraint loss function with the pre-established theoretical water inflow model, and train the LSTM model to obtain the initial water inflow prediction model.
[0044] The method for constructing the theoretical inflow model is as follows: like Figure 3 As shown, the equivalent external head is determined based on seawater depth, tunnel burial depth, and pore water pressure. H w Based on the tunnel excavation radius r With 1 as the inner boundary and R0 as the outer boundary, the surrounding rock, grouting ring, and lining are respectively regarded as layered seepage media, and an equivalent three-dimensional radial seepage model of the interaction between seawater, surrounding rock, grouting ring, and lining is established.
[0045] In this embodiment, for a circular or near-circular cross-section of a submarine tunnel, the hydraulic recharge formed by seawater and the overlying rock mass is considered as the outer boundary head, the tunnel excavation outline or the inner edge of the lining is considered as the inner boundary, and the surrounding rock, grouting ring, and lining are respectively regarded as continuous media layers with different equivalent permeability coefficients. The equivalent three-dimensional radial seepage model is used to describe the seepage process of water flowing through the surrounding rock, grouting ring, and lining into the tunnel interior under high seabed head conditions.
[0046] In this embodiment, the equivalent three-dimensional radial seepage model satisfies the following assumptions: the seepage process conforms to Darcy's law; the surrounding rock, grouting ring, and lining can be equivalent to a homogeneous isotropic medium within a single calculation section; the seepage near the tunnel section is mainly radial confluence; and the interfaces between the grouting ring and the lining, and between the surrounding rock and the grouting ring, satisfy the conditions of continuous head and continuous flow.
[0047] Based on the equivalent three-dimensional radial seepage model, combined with the outer boundary conditions, inner boundary drainage conditions, and the continuous head and flow conditions at the interfaces of each zone, the theoretical inflow model in the form of seepage resistance in the surrounding rock-grouting ring-lining series is obtained as follows:
[0048] In the formula, QtheoryTheoretical flow rate H w The equivalent external head is calculated from seawater depth, tunnel burial depth, and pore water pressure. R 0 represents the radius of influence. r 1 represents the tunnel excavation radius or equivalent drainage radius. r 2 represents the outer radius of the lining. r 3 represents the outer radius of the grouting ring. k w The equivalent permeability coefficient of the surrounding rock is given by [the value of the value]. k g The equivalent permeability coefficient of the grouting ring is... k l The equivalent permeability coefficient of the lining.
[0049] Furthermore, to reduce the subjectivity in determining the radius of influence, equivalent permeability coefficient, and external head, empirical parameters of the theoretical inflow model were used based on the measured inflow during the construction period. H w , R 0、 k w , k g , k l Perform inversion calibration and convert the inverted empirical parameters H w , R 0、 k w , k g , k l The theoretical inflow model is updated to obtain the final updated theoretical inflow model.
[0050] Inversion objective function min J(θ) It can be set to:
[0051] in, Qobs,i For the first i The measured water inflow at each cross-section Qtheory,i(θ) For parameter vectors i The corresponding theoretical flow rate, i include R 0. At least one of the following: head correction factor, equivalent permeability factor correction factor for surrounding rock, equivalent permeability factor correction factor for grouting ring, and equivalent permeability factor correction factor for lining. i 0 represents the initial empirical parameter. m This is the regularization coefficient.
[0052] Parameter inversion can be performed using the least squares method, genetic algorithm, particle swarm optimization algorithm, or Bayesian optimization algorithm.
[0053] The inverted empirical parameters are fed back to the theoretical inflow model and updated simultaneously in the subsequent theoretical inflow model used to construct the physical constraint loss function, thus forming a closed loop of theoretical derivation—experimental inversion—model constraint.
[0054] The model training uses the LSTM model training method, which includes the following steps: In this embodiment, the evaluation index is used as the input feature, and the j-th standardized input feature is multiplied by the AHP-entropy weight before entering the LSTM. wj We obtain the weighted input features:
[0055] in, x i,t,j For the i-th cross section, the... t The time step, the first j Standardized input features; wj Let be the comprehensive weight of the j-th indicator; The input features are pre-weighted by the comprehensive weights.
[0056] During model training, all weighted input features are combined into an LSTM input sequence according to time windows. The LSTM outputs a hidden state at each time step. i Hidden state sequence of each tunnel section H i It can be represented as:
[0057] in, i For tunnel cross-sections or sample numbers, t For the current construction period, T The time window length, h i,t Indicates that LSTM is in the first... t The hidden state vector is output at each time step. The hidden state vector contains the current input and historical information from previous construction moments, such as the temporal characteristics of evaluation indicators like surrounding rock permeability, pore water pressure, cumulative blasting disturbance, grouting effect, tidal variation, and excavation progress.
[0058] For the i Section 1, the first t The time step and the first j The original attention score can be calculated using the following method based on the weighted input features:
[0059] in, e i,t,j For the firsti The tunnel cross section, the first t The time step, the first j The original attention scores for each input feature; h i,t For LSTM in the first i The tunnel cross section, the first t The hidden state vector output at each time step; Features are those that have been pre-weighted according to the overall weights; W h 、W x 、Toilet、 and b a All of these are trainable parameters of the attention layer. tanh is the hyperbolic tangent activation function.
[0060] The original attention scores of each weighted input feature are Softmax normalized to obtain the normalized i-th... i The tunnel cross section, the first t The time step, the first j Initial attention weights for each input feature :
[0061] Where m is the total number of weighted input features, and k = 1, 2, ..., m.
[0062] For high-risk sections: Calculate the sum of initial attention weights for the geological and hydrological data set. S h Specifically:
[0063] if S h ≥r h， Then keep the initial attention weights unchanged:
[0064] That is, the final attention weight equal to the initial attention weight .
[0065] if S h <p h， The following adjustments will be made:
[0066] rh The lower limit for attention weighting of high-risk sections is preferably 0.6. G h It is a set of evaluation indicators for geological and hydrological data groups.
[0067] The total attention weight for the geological and hydrological data group is increased to the specified lower limit while maintaining the original relative proportions of each indicator within the group; the non-geological and hydrological data group is compressed proportionally. After the correction, all attention weights are still non-negative and their sum is 1.
[0068] For low-risk sections: Calculate the sum of initial attention weights for the construction parameter data set. S c .
[0069]
[0070] like S c ≥ r c If so, the initial attention weights remain unchanged.
[0071]
[0072] That is, the final attention weight equal to the initial attention weight .
[0073] if S c <p c， The following adjustments will be made:
[0074] r c The lower limit of the attention weight for low-risk sections is preferably 0.5. G c It is a set of evaluation indicators for construction parameter data groups.
[0075] The total attention weight of the construction parameter data group is increased to the specified lower limit, while maintaining the original relative proportions of each indicator within the group; the non-construction parameter data group is compressed proportionally. After the correction, all attention weights are still non-negative and their sum is 1.
[0076] For medium-risk sections: Calculate the sum of the initial attention weights for each geological and hydrological data set. S h Sum of initial attention weights with the construction parameter data set S c :
[0077]
[0078] If | S h S c |≤ e If the initial attention weight remains unchanged, then the initial attention weight becomes the final attention weight.
[0079] If | S h S c |> e Then calculate the weight Δ that needs to be transferred.
[0080] In this embodiment e= 0.1, the calculation method for the transition weight Δ is as follows:
[0081] if ,but
[0082] if ,but
[0083] The method for adjusting attention weights is as follows:
[0084]
[0085]
[0086] G m For the marine environmental data set, this adjustment only shifts weights between the geological / hydrological group and the construction group; the attention weight of the marine environmental group remains unchanged. Therefore, the adjusted result satisfies... Furthermore, all attention weights are non-negative and their sum is 1.
[0087] Through the above steps, feature group attention constraints are set based on relative proximity during model training, and the evaluation results of the TOPSIS method can be directly used for feature learning of the neural network model.
[0088] We directly weight the hidden state vectors of the LSTM, and the context vector is:
[0089] αi,τ Indicates the first i Attention weights for each tunnel cross section and at the τth time step; t For the current predicted time, h i,τ For LSTM in the first i The hidden state vector output at the τth time step of each tunnel section, where Zi is the context vector.
[0090] The context vector is input into the regression layer to obtain the initial inflow prediction value. The initial inflow prediction value is evaluated using the physical loss constraint function. After multiple training sessions, the initial inflow prediction model is obtained.
[0091] Specifically, the physical loss constraint function is constructed using the updated theoretical inflow model, and the details are as follows: The physical loss constraint function L is:
[0092] in, Q true，i Let i be the measured inflow rate of the i-th sample. Q pred，i For the first i The model predicts the inflow rate for each sample. n Let λ be the sample size. C i (for relative closeness) C i Relevant theoretical constraint coefficients, Q theory,i For the first i The theoretical inflow rate for each sample is obtained based on the theoretical inflow rate model.
[0093] λ(C i ） The theoretical constraint coefficient is adaptively selected according to the risk level of water inrush. A smaller theoretical constraint coefficient is used for high-risk sections to enhance the model's ability to learn from complex geological anomalies, while a larger theoretical constraint coefficient is used for low-risk sections to enhance the physical consistency of the prediction results. A theoretical constraint coefficient between the two is used for medium-risk sections.
[0094] Specifically: The relative proximity of each tunnel section is calculated based on the weighted standardized evaluation matrix. C i Based on this, the risk level of water inrush is classified: when C i A cross-section with a value >0.7 is considered high-risk; a cross-section with a value ≤0.3 is considered high-risk. C i A cross-section with a value ≤ 0.7 is considered a medium-risk section; when Ci A cross-section with a value less than 0.3 is considered a low-risk section. Subsequently,C i Mapped to theoretical constraint coefficients λ ( C i ).
[0095] in: λ( C i ) =λ min + (λ max λ min (1) C i ) in, C i ∈ [0,1], and 0<λ min <λ max .thus, C i The larger the value, the higher the risk of the cross-section, λ( C i The closer to λ min ; C i The smaller the value, the lower the cross-sectional risk, λ( C i The closer to λ max .
[0096] λ min and λ max Existing methods can be used to predetermine the value, such as grid search, K-fold cross-validation, or Bayesian optimization, with the goal of minimizing the MAE, RMSE, or total loss of the validation set. These methods will not be described in detail here.
[0097] Step S4: Correct the initial water inflow prediction model based on the blasting vibration damage coefficient and the grouting reinforcement coefficient to obtain the final water inflow prediction model. Qfinal .
[0098] Specifically:
[0099] in, Qfinal This is the final predicted inflow rate. Qpred These are the preliminary predictions from the initial inflow prediction model. v The blasting vibration velocity, k before The permeability coefficient of the surrounding rock before grouting. k after The permeability coefficient of the surrounding rock after grouting. a , b , cThese are empirical coefficients obtained based on regression or parameter inversion of measured samples.
[0100] In this embodiment, after obtaining the final inflow prediction model, the performance of the inflow prediction model is evaluated using K-fold cross-validation, hold-out validation, or rolling time series validation. Evaluation indicators may include mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R²).
[0101] Step S5: Input the evaluation index of the tunnel cross section to be predicted into the final water inflow prediction model to obtain the predicted water inflow of the tunnel cross section and the confidence interval.
[0102] Furthermore, confidence intervals are determined using Monte Carlo Dropout, Bootstrap resampling, quantile regression, or residual distribution estimation.
[0103] In this embodiment, the prediction method uses weighted input features with comprehensive weights as input during the training of the water inflow prediction model. This allows the prediction model to obtain prior knowledge of the importance of influencing factors in the early stages of training. Attention layer constraints are determined based on the relative proximity of the tunnel cross-section, enabling the prediction model to adaptively adjust the focus of features according to the risk level of the cross-section. The relative proximity of the tunnel cross-section is obtained using the TOPSIS method based on the weighted standardized evaluation matrix, thus realizing the use of TOPSIS evaluation results for prediction model training. During prediction model training, a physical constraint loss function is constructed using the theoretical water inflow model, realizing the introduction of the seepage theory solution as a physical loss constraint into the model training, reducing the risk of the purely data-driven model deviating from hydrogeological laws. By correcting the initial water inflow prediction model, dynamic correction of construction disturbance and treatment effect is achieved. The entire method achieves effective synergy between TOPSIS evaluation results, seepage theory solution, and deep learning model, taking into account prediction accuracy, physical consistency, interpretability, and engineering adaptability. It can be used for construction water inflow early warning, drainage system design, and grouting parameter optimization.
[0104] Meanwhile, by continuously calibrating the empirical parameters of the theoretical inflow model through empirical parameter inversion, the applicability of the prediction method under different engineering conditions is improved.
[0105] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for predicting the inflow of water into a submarine tunnel by fusing theoretical solutions with TOPSIS, characterized in that, Includes the following steps: Evaluation indicators from the original engineering database of the submarine tunnel project were selected, and a weighted standardized evaluation matrix was constructed based on the comprehensive weights of the evaluation indicators determined by a combination of subjective and objective weighting methods. The relative proximity of the tunnel cross sections is obtained by combining the weighted standardized evaluation matrix with the TOPSIS method; The evaluation index is used as the input feature, and the input feature after weighting by comprehensive weight is used as the input. The feature group attention constraint is set according to the relative closeness. The physical constraint loss function is constructed by the pre-established theoretical water inflow model to train the model and obtain the initial water inflow prediction model. The initial water inflow prediction model was modified based on the blasting vibration damage coefficient and the grouting reinforcement coefficient to obtain the final water inflow prediction model. The water inflow of the tunnel section to be predicted is obtained by combining the input characteristics of the tunnel section to be predicted with the final water inflow prediction model.
2. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, The method for obtaining the original engineering database for the undersea tunnel project is as follows: Geological and hydrological data, marine environmental data, construction parameter data, and measured water inflow data of multiple tunnel sections of the constructed subsea tunnel were obtained, with the geological and hydrological data, marine environmental data, and construction parameter data serving as evaluation indicators. The acquired data is processed for missing values, outlier identification, standardization, and time-series alignment to form the original engineering database for the submarine tunnel project.
3. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 2, characterized in that, At least one of the following methods, namely mean interpolation, median interpolation, nearest-neighbor section interpolation, or temporal interpolation, is used to handle missing values. At least one of the following methods is used to identify outliers: interquartile range method, three-standard-deviation method, or engineering threshold method. Evaluation indicators with different dimensions are normalized to achieve standardization.
4. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 2, characterized in that, The geological and hydrological data in the evaluation indicators include seawater depth, tunnel burial depth, surrounding rock permeability coefficient, rock mass integrity coefficient, fault fracture zone width, and pore water pressure. Marine environmental data includes rainfall and tidal variations; Construction parameter data include the grouting ring permeability coefficient, lining permeability coefficient, excavation advance, and blasting vibration intensity.
5. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, The subjective and objective weighting method adopts the AHP-entropy weighting method, and the comprehensive weight is obtained according to the AHP weight and the entropy weight.
6. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, The method for obtaining the relative proximity to the tunnel cross-section using the TOPSIS method is as follows: The positive and negative ideal solutions are obtained based on the weighted standardized evaluation matrix; Obtain the first Euclidean distance from the tunnel cross-section to the ideal solution; Obtain the second Euclidean distance from the tunnel cross-section to the negative ideal solution; The relative proximity of the tunnel cross sections is obtained based on the first Euclidean distance and the second Euclidean distance.
7. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, The method for setting attention constraints for feature groups based on relative proximity is as follows: obtain the water inrush risk level of the tunnel cross section based on the relative proximity of the tunnel cross section, and adjust the attention weights corresponding to each input feature according to the water inrush risk level.
8. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, When establishing the theoretical inflow model, it is assumed that the seepage process conforms to Darcy's law; the surrounding rock, grouting ring, and lining can be equivalent to a homogeneous isotropic medium within a single calculation section; the seepage near the tunnel section is mainly radial confluence; and the interfaces between the grouting ring and the lining, and between the surrounding rock and the grouting ring, satisfy the conditions of continuous head and continuous flow.
9. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 1, characterized in that, The objective function is to minimize the error between the measured water inflow at the tunnel cross section and the theoretical water inflow obtained from the theoretical water inflow model. The empirical parameters in the theoretical water inflow model are then inverted, and the inverted empirical parameters are updated to the theoretical water inflow model. A physical constraint loss function is constructed based on the updated theoretical inflow model.
10. The method for predicting the inflow of water into a submarine tunnel using TOPSIS fusion theoretical solutions as described in claim 9, characterized in that, Inversion can be performed using the least squares method, genetic algorithm, particle swarm optimization algorithm, or Bayesian optimization algorithm.