A method for quickly identifying the state of a TBM tunnel excavation face
Patent Information
- Application Number
- CN202610927914.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-25
- Publication Date
- 2026-10-30
AI Technical Summary
[0006]鉴于以上技术问题，本公开提供了一种TBM隧道开挖面状态快速辨识方法，解决了现有技术中数据噪声、权重主观性和复杂地质下的泛化问题
方法的核心创新在于时空对齐的引入，可处理数据的时间延迟和空间异质性，提高辨识速度和准确性。相比现有方法，本发明可将辨识时间缩短至分钟级，并支持TBM参数的自适应优化。
Smart Images
Figure CN123012949A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tunnel engineering technology, and in particular to a rapid identification method for the excavation face condition of TBM tunnels based on tunneling parameters and rock debris identification, which is especially applicable to the surrounding rock quality assessment and tunneling optimization in hard rock tunnel construction. Background Technology
[0002] Tunnel boring machines (TBMs) are widely used in modern tunnel engineering, and their tunneling efficiency and safety depend on the accurate identification of the excavation face condition. Traditional methods mainly rely on geological exploration, drilling, or manual observation, which are time-consuming, costly, and unable to respond in real time to complex geological changes. In existing technologies, some methods utilize TBM tunneling parameters (such as thrust, torque, and rotational speed) or rock debris characteristics for surrounding rock assessment, but often neglect the spatiotemporal fusion of multi-source data, leading to insufficient identification accuracy. For example, in high-stress areas, unstable surrounding rock may cause the machine to stall, and single-parameter analysis is insufficient to capture dynamic interactions. Recent studies have shown that integrating tunneling parameters and rock debris identification can improve prediction accuracy, but there is a lack of efficient spatiotemporal alignment mechanisms to handle the heterogeneity of time series and spatial features. This invention aims to solve these problems by providing a fast, real-time identification method.
[0003] TBMs, as highly efficient mechanized tunneling equipment, have proven their value in numerous large-scale projects worldwide, such as the Qinling Tunnel in China and the Gotthard Tunnel in Switzerland. According to the International Tunneling Association (ITA), TBMs can increase tunneling speeds by 5-10 times compared to traditional methods; however, their performance is highly dependent on the assessment of the geological conditions of the tunnel face. Inaccurate identification of the surrounding rock grade can lead to equipment jamming, rock bursts, or collapses, causing economic losses and safety hazards. Traditional geological exploration relies on borehole sampling and manual assessment, typically taking several days and conducted during construction interruptions, making it unsuitable for the continuous operation of TBMs. In complex geological conditions such as fault zones or high-water-pressure areas, these methods carry even higher risks, and costs can account for 10%-20% of the total project cost.
[0004] Existing technologies attempt to assess surrounding rock using TBM tunneling parameters. For example, thrust (F) and torque (T) reflect rock mass strength, while penetration (P) and advance rate (R) indicate tunneling resistance. Some models, such as the CSM model, predict rock mass quality based on these parameters but ignore direct evidence from rock debris. Rock debris analysis systems extract particle size distribution (D), shape factor (S), and color (C) using machine vision, indirectly inferring rock discontinuities and strength. Studies have shown that by integrating tunneling parameters and rock debris characteristics, rock mass classification accuracy can be improved from 75% to over 90% using a single method. However, these methods often neglect spatiotemporal fusion: tunneling parameters are time-series data, while rock debris characteristics are spatially distributed, leading to heterogeneity and difficulties in data alignment. In high-stress areas, stress release in the surrounding rock can trigger dynamic instability, and single analysis cannot capture the interactive effects of vibration signals or stress changes, increasing the risk of TBM jamming. Literature indicates that in such areas, TBM jamming events account for 30% of accidents, often due to the lack of real-time fusion mechanisms.
[0005] In recent years, deep learning and data fusion technologies have offered opportunities for improvement. For example, spatiotemporal feature fusion models use convolutional neural networks (CNNs) to extract spatial features, long short-term memory networks (LSTMs) to process time series, and achieve alignment through dynamic time warping (DTW). This method has reduced prediction time to the minute level in real-world testing and supports adaptive adjustment of TBM parameters. However, current research still faces challenges: data noise, weight subjectivity, and generalization problems under complex geological conditions. The lack of efficient spatiotemporal alignment mechanisms and the inability to handle time delays and spatial heterogeneity lead to insufficient accuracy. Summary of the Invention
[0006] In view of the above technical problems, this disclosure provides a rapid identification method for the excavation face condition of TBM tunnels, solving the problems of data noise, subjective weighting, and generalization under complex geological conditions in existing technologies. The lack of an efficient spatiotemporal alignment mechanism and the inability to handle time delays and spatial heterogeneity lead to insufficient accuracy. This invention solves these problems by establishing a multi-source database, using the entropy weighting method to calculate parameter weights, and introducing a spatiotemporal alignment algorithm. This method not only improves identification accuracy but also predicts the surrounding rock grade, supporting optimization of tunneling efficiency. In practical applications, such as hard rock tunnels, it can reduce construction interruptions and improve overall efficiency.
[0007] According to one aspect of this disclosure, a method for rapid identification of the excavation face condition of a TBM tunnel is provided, comprising the following steps: (1) Data acquisition and database establishment: Establish equipment information sample database and rock mass information sample database, collect tunneling parameters and rock debris characteristics during the TBM tunneling process. Tunneling parameters include cutterhead thrust, torque, penetration depth and propulsion speed. Rock debris characteristics include particle size distribution, shape and color characteristics, which are extracted by image processing algorithms. (2) Analysis of rock mass parameter weights in the ascending section: The rock mass information of the ascending section of the tunneling parameters is analyzed, and the entropy weight method is used to calculate the rock mass parameter weights under different conditions. The rock mass parameter weights include uniaxial compressive strength, rock mass integrity, and hardness. (3) Optimization of tunneling parameters in the stable section: The convergence conditions are determined through the dynamic interaction mechanism of rock-machine parameters, and the optimal solution of tunneling parameters in the stable section is solved by the improved quantum particle swarm optimization algorithm; the convergence conditions include equipment structural strength design, safety design, and quality design; The quantum particle swarm optimization algorithm introduces the Logistic mapping formula: ; in, This represents the state value of the nth iteration, with a value ranging from [0,1]. It is an adjustable function, usually taking values in the range [0,4], used to control the dynamic behavior of the system.
[0008] (4) Multi-source data fusion and state identification prediction: Multi-source data is fused using a spatiotemporal alignment method, time series data is aligned using a dynamic time warping algorithm, and feature fusion is performed in combination with spatial features, including the geometric distribution of rock debris images, to achieve real-time identification of the excavation face state.
[0009] In some embodiments of this disclosure, the entropy weight method described in step (2) includes the following steps: (2.1) Data standardization, Calculate the standardized value using the following formula : ; in, This represents the standardized value of the i-th sample on the j-th indicator. This represents the original value of the i-th sample on the j-th indicator. and These are the maximum and minimum values for the j-th indicator, respectively; (2.2) Calculate the weight matrix. Calculate the weight of each sample under each indicator: ; (2.3) Calculate the index entropy value, ; like =0, defined =0; (2.4) Calculate the coefficient of difference. The coefficient of variation indicates the amount of information contained in the indicator: ; (2.5) Calculate the entropy weight of the j-th index. ; in, For the final weight, satisfying =1.
[0010] In some embodiments of this disclosure, the step (1) of collecting tunneling parameters and rock debris characteristics during the TBM tunneling process specifically includes: collecting tunneling parameters such as cutterhead thrust F, torque T, penetration depth P and advance speed R in real time through sensors; capturing rock debris images through camera equipment, and extracting average particle size D, shape factor S and color distribution C using image recognition algorithms; and storing the data according to the tunneling cycle into ascending and stable sections.
[0011] In some embodiments of this disclosure, the specific process of calculating the rock mass parameter weights under different equipment conditions using the entropy weight method in step (2) includes: standardizing the rock mass parameters of the ascending section, calculating the information entropy of each parameter, and then determining the uniaxial compressive strength σ. c Rock mass integrity coefficient K v Objective weighting of hardness; ; in, Indicates uniaxial compressive strength. is the maximum axial load when the rock mass fails, and A is the original cross-sectional area of the rock mass; ; in, Indicates the rock mass integrity coefficient. Indicates the longitudinal wave velocity of the rock mass. The longitudinal wave velocity of the same type of rock block.
[0012] In some embodiments of this disclosure, the spatiotemporal alignment and feature fusion in step (4) specifically include the following steps: the spatiotemporal alignment uses the DTW algorithm to align the time series of tunneling parameters and the spatial series of rock debris features: define a distance matrix, find the minimum path cumulative distance, and achieve time alignment; at the same time, the spatial alignment extracts the spatial features of the rock debris image through the convolutional neural network CNN and fuses them with the parameter sequence to form a spatiotemporal feature vector, which is then input into the long short-term memory network LSTM model for prediction.
[0013] In some embodiments of this disclosure, the state identification in step (4) specifically includes the following steps: based on the fusion features, the surrounding rock grade IV is identified using a BP neural network; at the same time, the improved QPSO algorithm is applied to optimize the stable section parameters: the particle swarm is initialized, chaotic search and neighborhood mutation are introduced to improve the global search capability; finally, the excavation face state is output, including rock mass quality indicators, and the tunneling efficiency is predicted.
[0014] The beneficial effects of this invention are as follows: The core innovation of this method lies in the introduction of spatiotemporal alignment, which can handle the temporal delay and spatial heterogeneity of data, improving identification speed and accuracy. Compared with existing methods, this invention can reduce identification time to the minute level and supports adaptive optimization of TBM parameters.
[0015] Multi-source data fusion for high identification accuracy: This invention integrates tunneling parameters and rock debris characteristics, and uses the DTW algorithm and CNN network to achieve spatiotemporal alignment, making full use of multi-source information during TBM tunneling and significantly improving the accuracy of excavation face condition identification. Actual tests show that this method achieves an identification accuracy of over 90% in long-distance hard rock tunnel engineering, outperforming traditional single-parameter identification methods.
[0016] The analysis is conducted in stages, revealing a clear mechanism: the tunneling cycle is divided into an ascending phase and a stable phase, which are handled separately. In the ascending phase, the entropy weight method is used to objectively determine the weights of rock mass parameters, providing a reliable foundation of rock mass information for subsequent optimization; in the stable phase, an improved quantum particle swarm optimization algorithm is used to optimize the process, making the optimization process more consistent with the actual physical process of TBM tunneling.
[0017] Algorithm innovation and strong optimization ability: Chaotic search and neighborhood mutation mechanism are introduced into the quantum particle swarm optimization algorithm, which effectively avoids the defect of traditional algorithms being prone to getting trapped in local optima, enhances global search ability and convergence speed, and can quickly solve for the optimal tunneling parameters of the stable section.
[0018] It combines real-time performance with practicality: Through spatiotemporal alignment and fusion, combined with an LSTM / BP neural network model, it achieves real-time identification of the excavation face condition and rapid classification of the surrounding rock grade, and can predict tunneling efficiency. The final optimal tunneling formula can be directly used for on-site construction guidance, and has significant engineering application value.
[0019] Good robustness: To address the data noise problem in actual engineering, filtering algorithms can be introduced for preprocessing, which enhances the robustness and adaptability of the method and makes it suitable for TBM construction under complex geological conditions. Attached Figure Description
[0020] Figure 1 A schematic diagram illustrating the technical path from scattered geological data to continuous three-dimensional geological models; The colored dots in the figure represent measured geological data points, and the layered surfaces are stratigraphic interfaces / geological property surfaces generated by interpolation algorithms. Figure 2 A schematic diagram of a one-dimensional convolutional neural network structure for tunnel excavation status recognition; Figure 3 A schematic diagram of a two-dimensional convolutional neural network for rock slag image recognition; Figure 4 A schematic diagram of a framework for generating excavation state feature sequences driven by multi-source heterogeneous data fusion; Detailed Implementation
[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Example 1
[0022] This example discloses a method for rapid identification of the excavation face condition of a TBM tunnel. (See also...) Figures 1-4 This includes the following steps: (1) Data acquisition and database establishment: Establish equipment information sample database and rock mass information sample database, collect tunneling parameters and rock debris characteristics during the TBM tunneling process. Tunneling parameters include cutterhead thrust, torque, penetration depth and propulsion speed. Rock debris characteristics include particle size distribution, shape and color characteristics, which are extracted by image processing algorithms. (2) Analysis of rock mass parameter weights in the ascending section: The rock mass information of the ascending section of the tunneling parameters is analyzed, and the entropy weight method is used to calculate the rock mass parameter weights under different conditions. The rock mass parameter weights include uniaxial compressive strength, rock mass integrity, and hardness. (3) Optimization of tunneling parameters in the stable section: The convergence conditions are determined through the dynamic interaction mechanism of rock-machine parameters, and the optimal solution of tunneling parameters in the stable section is solved by the improved quantum particle swarm optimization algorithm; the convergence conditions include equipment structural strength design, safety design, and quality design; The quantum particle swarm optimization algorithm introduces the Logistic mapping formula: ; in, This represents the state value of the nth iteration, with a value ranging from [0,1]. It is an adjustable function, usually taking values in the range [0,4], used to control the dynamic behavior of the system.
[0023] (4) Multi-source data fusion and state identification prediction: Multi-source data is fused using a spatiotemporal alignment method, time series data is aligned using a dynamic time warping algorithm, and feature fusion is performed in combination with spatial features, including the geometric distribution of rock debris images, to achieve real-time identification of the excavation face state.
[0024] The entropy weight method described in step (2) includes the following steps: (2.1) Data standardization, Calculate the standardized value using the following formula : ; in, This represents the standardized value of the i-th sample on the j-th indicator. This represents the original value of the i-th sample on the j-th indicator. and These are the maximum and minimum values for the j-th indicator, respectively.
[0025] (2.2) Calculate the weight matrix. Calculate the weight of each sample under each indicator: ; (2.3) Calculate the index entropy value, ; like =0, defined =0; (2.4) Calculate the coefficient of difference. The coefficient of variation indicates the amount of information contained in the indicator: ; (2.5) Calculate the entropy weight of the j-th index. ; in, For the final weight, satisfying =1.
[0026] The specific steps in step (1) for collecting tunneling parameters and rock debris characteristics during the TBM tunneling process include: collecting tunneling parameters such as cutterhead thrust F, torque T, penetration depth P, and propulsion speed R in real time through sensors; capturing rock debris images through camera equipment and extracting average particle size D, shape factor S, and color distribution C using image recognition algorithms; and storing the data according to the tunneling cycle into ascending and stable sections.
[0027] The specific process of calculating the rock mass parameter weights under different equipment conditions using the entropy weight method in step (2) includes: standardizing the rock mass parameters of the ascending section, calculating the information entropy of each parameter, and then determining the uniaxial compressive strength σ. c Rock mass integrity coefficient K v Objective weighting of hardness; ; in, Indicates uniaxial compressive strength. is the maximum axial load when the rock mass fails, and A is the original cross-sectional area of the rock mass; ; in, Indicates the rock mass integrity coefficient. Indicates the longitudinal wave velocity of the rock mass. The longitudinal wave velocity of the same type of rock block.
[0028] Step (4) includes the following steps for spatiotemporal alignment and feature fusion: Spatiotemporal alignment uses the DTW algorithm to align the time series of tunneling parameters and the spatial series of rock debris features: define the distance matrix, find the minimum path cumulative distance, and achieve time alignment; at the same time, spatial alignment extracts the spatial features of the rock debris image through the convolutional neural network CNN and fuses them with the parameter sequence to form a spatiotemporal feature vector, which is then input into the long short-term memory network LSTM model for prediction.
[0029] The state identification in step (4) specifically includes the following steps: Based on the fusion features, the BP neural network is used to identify the surrounding rock grade IV; at the same time, the improved QPSO algorithm is applied to optimize the stable section parameters: initialize the particle swarm, introduce chaotic search and neighborhood mutation to improve the global search capability; finally output the excavation face state, including rock mass quality indicators, and predict the tunneling efficiency.
[0030] See Figure 1 Text information such as geological exploration, advanced geological forecasting energy information, and construction logs. This fully demonstrates the technical path from "fragmented geological data" to "continuous three-dimensional geological models": By using multi-source data such as borehole data, advanced forecasting data, and construction logs, and employing spatial algorithms such as Kriging interpolation, the geological conditions of unknown underground areas can be reconstructed, providing a visual basis for tunnel construction and risk warning.
[0031] The colored dots in the figure represent known geological data points collected at different locations and depths (such as lithology and surrounding rock grade data at different depths in the borehole); the layered surfaces in the middle are stratigraphic interfaces / geological property surfaces fitted by interpolation algorithms (such as boundaries between different surrounding rock grades and rock strata interfaces). This process uses known discrete data points to reconstruct geological information from unknown locations, transforming scattered data into a continuous three-dimensional geological model. Geological modeling is performed using the Kriging interpolation formula shown in Equation 1.
[0032]
[0033] in, This represents the estimated geological properties at the unknown point X0 (such as the surrounding rock grade or rock layer depth at the unknown location). This represents the actual measured value of the i-th known data point. This represents the weight coefficient of the i-th known point. The Kriging method determines the weight based on spatial correlation (calculated through a semi-variogram).
[0034] See Figure 2 TBM tunneling timing information This paper presents a one-dimensional convolutional neural network (1D-CNN) model for tunnel excavation status recognition. It automatically extracts features from one-dimensional time-series data collected during tunnel construction (such as total thrust, cutterhead torque, cutterhead speed, and other tunneling parameters) and finally automatically determines the current excavation status.
[0035]
[0036] In the formula, X[] represents the i-th value of the output sequence, X[] represents the original one-dimensional time series data of the input, such as the total thrust, cutterhead torque, cutterhead speed and other tunneling parameters, w[k] represents the k-th parameter of the convolution kernel (weight), b is the bias term, K is the length of the convolution kernel, and s is the step size of the convolution, which controls the distance the convolution kernel slides each time.
[0037] See Figure 3 Rock slag image information The input raw rock debris image is processed through multiple convolution and pooling layers to automatically extract and compress features, which are then converted into class probabilities through a fully connected layer, and finally the predicted class of the image is output.
[0038] See Figure 4 Feature sequences oriented towards excavation status (1) Regarding the Encoder module, it integrates three types of data features: geological text information of geological spatial features after Kriging interpolation modeling, time-series information of TBM tunneling parameters extracted by 1D-CNN, and rock debris image information of image features extracted by 2D-CNN. Based on time-series location encoding, it uses attention mechanisms (α1, α2, α3) to determine which is more important in the current excavation state: geological data, TBM data, or image data. (2) Regarding the feature fusion module, it concatenates the feature ft from 1D time-series data and the feature fi from 2D image / geological data by channel dimension, merging them into a single dimension (d). t +d i The new feature vector is then used to perform a linear transformation on the spliced features using y=wf+b, making the fused features more suitable for subsequent tasks. At the same time, the attention weights α1, α2, and α3 are weighted and adjusted for different modal features. (3) Regarding the Decoder and output module, the Decoder receives the fused feature vector y and combines it with the temporal and attention information output by the Encoder to finally generate a feature sequence oriented towards the excavation state, which includes geological conditions, TBM tunneling data, and rock debris image information. This sequence can be directly used for subsequent excavation state classification, risk prediction, or trend analysis, realizing intelligent perception of tunnel construction driven by multi-source data.
[0039] The core innovation of this method lies in the introduction of spatiotemporal alignment, which can handle the temporal delay and spatial heterogeneity of data, improving identification speed and accuracy. Compared with existing methods, this invention can reduce identification time to the minute level and supports adaptive optimization of TBM parameters.
[0040] Based on multi-source data such as geological survey and construction log text information, TBM parameter time-series data, and rock debris image information, excavation state-related features are extracted. These features include text features, TBM time-series features, rock debris image features, and cutterhead vibration features, while also incorporating time-series embedding information. The excavation state-related features are first fed into an encoder to complete preliminary feature encoding and dimensional unification. The encoded modal features are then processed using an attention function, which calculates the importance of different features to the current excavation state, obtaining corresponding attention weights α1, α2, α3, and α4, with a weight sum of 1. The attention-weighted modal features are then concatenated and integrated: 1D time-series features and 2D spatial features are merged to obtain fused features. A linear transformation further unifies the feature space, making the fused features more compact. The fused features are then fed into a decoder to generate the final feature sequence oriented towards the excavation state. This feature sequence integrates multi-source information to reflect the core features of the TBM excavation state.
[0041] In practical tunnel engineering, this method has been applied to long-distance hard rock tunnels, and the database contains 159 sets of data. Tests show that the identification accuracy exceeds 90%, which is superior to traditional methods. Potential risks include data noise, which can be mitigated through filtering algorithms.
[0042] The classification and processing methods of multi-source parameters for excavation face condition identification are shown in the table below:
[0043] Although some preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.
[0044] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this application and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for rapid identification of the excavation face condition of a TBM tunnel, characterized in that: Includes the following steps: (1) Data acquisition and database establishment: Establish equipment information sample database and rock mass information sample database, collect tunneling parameters and rock debris characteristics during the TBM tunneling process. Tunneling parameters include cutterhead thrust, torque, penetration depth and propulsion speed. Rock debris characteristics include particle size distribution, shape and color characteristics, which are extracted by image processing algorithms. (2) Analysis of rock mass parameter weights in the ascending section: The rock mass information of the ascending section of the tunneling parameters is analyzed, and the entropy weight method is used to calculate the rock mass parameter weights under different conditions. The rock mass parameter weights include uniaxial compressive strength, rock mass integrity, and hardness. (3) Optimization of tunneling parameters in the stable section: The convergence conditions are determined through the dynamic interaction mechanism of rock-machine parameters, and the optimal solution of tunneling parameters in the stable section is solved by the improved quantum particle swarm optimization algorithm; the convergence conditions include equipment structural strength design, safety design, and quality design; The quantum particle swarm optimization algorithm introduces the Logistic mapping formula: ； in, This represents the state value of the nth iteration, with a value ranging from [0,1]. It is an adjustable function with a value range of [0,4], used to control the dynamic behavior of the system; (4) Multi-source data fusion and state identification prediction: Multi-source data is fused using a spatiotemporal alignment method, time series data is aligned using a dynamic time warping algorithm, and feature fusion is performed in combination with spatial features, including the geometric distribution of rock debris images, to achieve real-time identification of the excavation face state.
2. The method for rapid identification of the excavation face condition of a TBM tunnel as described in claim 1, characterized in that: The entropy weight method described in step (2) includes the following steps: (2.1) Data standardization, Calculate the standardized value using the following formula : ; in, This represents the standardized value of the i-th sample on the j-th indicator. This represents the original value of the i-th sample on the j-th indicator. and These are the maximum and minimum values for the j-th indicator, respectively; (2.2) Calculate the weight matrix. Calculate the weight of each sample under each indicator: ； (2.3) Calculate the index entropy value, ； like =0, defined =0; (2.4) Calculate the coefficient of difference. The coefficient of variation indicates the amount of information contained in the indicator: ； (2.5) Calculate the entropy weight of the j-th index. ； in, For the final weight, satisfying =1.
3. The method for rapid identification of the TBM tunnel excavation face condition as described in claim 1, characterized in that: The specific steps in step (1) for collecting tunneling parameters and rock debris characteristics during the TBM tunneling process include: collecting tunneling parameters such as cutterhead thrust F, torque T, penetration depth P, and propulsion speed R in real time through sensors; capturing rock debris images through camera equipment and extracting average particle size D, shape factor S, and color distribution C using image recognition algorithms; and storing the data according to the tunneling cycle into ascending and stable sections.
4. The method for rapid identification of the excavation face condition of a TBM tunnel as described in claim 1, characterized in that: The specific process of calculating the rock mass parameter weights under different equipment conditions using the entropy weight method in step (2) includes: standardizing the rock mass parameters of the ascending section, calculating the information entropy of each parameter, and then determining the uniaxial compressive strength σ. c Rock mass integrity coefficient K v Objective weighting of hardness; ； in, Indicates uniaxial compressive strength. is the maximum axial load when the rock mass fails, and A is the original cross-sectional area of the rock mass; ； in, Indicates the rock mass integrity coefficient. Indicates the longitudinal wave velocity of the rock mass. The longitudinal wave velocity of the same type of rock block.
5. The method for rapid identification of the excavation face condition of a TBM tunnel as described in claim 1, characterized in that: Step (4) includes the following steps for spatiotemporal alignment and feature fusion: Spatiotemporal alignment uses the DTW algorithm to align the time series of tunneling parameters and the spatial series of rock debris features: define the distance matrix, find the minimum path cumulative distance, and achieve time alignment; at the same time, spatial alignment extracts the spatial features of the rock debris image through the convolutional neural network CNN and fuses them with the parameter sequence to form a spatiotemporal feature vector, which is then input into the long short-term memory network LSTM model for prediction.
6. The method for rapid identification of the excavation face condition of a TBM tunnel as described in claim 1, characterized in that: The state identification in step (4) specifically includes the following steps: Based on the fusion features, the BP neural network is used to identify the surrounding rock grade IV; at the same time, the improved QPSO algorithm is applied to optimize the stable section parameters: initialize the particle swarm, introduce chaotic search and neighborhood mutation to improve the global search capability; finally output the excavation face state, including rock mass quality indicators, and predict the tunneling efficiency.