Tunnel surrounding rock quality evaluation index prediction method and system based on Attention-LSTM

Abstract

The invention relates to the field of tunnel engineering, and discloses an Attention-LSTM-based tunnel surrounding rock quality evaluation index prediction method and system, and the method comprises the steps: collecting training data, and constructing a sequence data set according to the collected training data; based on the LSTM network, a result output by the LSTM is weighted through an Attention layer, and an Attention-LSTM model is constructed; according to the constructed sequence data set, an Attention-LSTM model is trained, and a trained Attention-LSTM model is obtained; and collecting sample point data, and inputting the sample point data into the trained Attention-LSTM model to obtain the probability distribution of the hardness degree, the water burst state and the integrity degree corresponding to the sample points. According to the method, the hidden state output by the LSTM is weighted through the Attention layer, so that the model can pay more attention to characteristics and time steps which have important influences on surrounding rock quality evaluation, and the prediction precision of key indexes such as the hardness degree, the water burst state and the integrity degree is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of tunnel engineering, and in particular to a method and system for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM. Background Art

[0002] In tunnel engineering, accurately predicting the quality of tunnel surrounding rock is crucial for ensuring construction safety, optimizing design solutions, and reducing project costs. Traditional methods for evaluating surrounding rock quality rely primarily on geological surveys, empirical formulas, and field tests. While these methods can provide basic information on surrounding rock properties, they often suffer from limitations such as being time-consuming, costly, and susceptible to human influence. The rapid development of artificial intelligence (AI), particularly deep learning, offers new insights into tunnel surrounding rock quality due to its significant advantages in time series data processing and pattern recognition.

[0003] However, existing deep learning-based methods for predicting surrounding rock quality often use standard recurrent neural networks (RNNs) or their variants, such as long short-term memory (LSTM) networks. While these models can capture temporal dependencies when processing long sequences of data, they often struggle to effectively highlight key information, resulting in limited accuracy in assessing surrounding rock quality under complex geological conditions. Especially during tunnel excavation, surrounding rock physical properties and geological characteristics change dynamically with tunneling progress. Accurately extracting features critical to surrounding rock quality assessment from these sequence data has become an urgent challenge. Summary of the Invention

[0004] The present invention aims to overcome the shortcomings of the prior art and provide a method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM, which includes the following steps:

[0005] Step 1: Collect training data and construct a sequence data set based on the collected training data;

[0006] Step 2: Based on the LSTM network, the LSTM output results are weighted through the Attention layer to build an Attention-LSTM model;

[0007] Step 3: Train the Attention-LSTM model based on the constructed sequence dataset to obtain the trained Attention-LSTM model;

[0008] Step 4: Collect sample point data and input it into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample points.

[0009] Furthermore, the collecting of training data and constructing a sequence data set based on the collected training data include:

[0010] The elastic wave geophysical exploration equipment is used to detect the area in front of the tunnel working face to obtain the physical properties of the surrounding rock. A set of data is obtained for each set distance.

[0011] Based on the geological sketch data of the tunnel working face, the surrounding rock hardness, integrity, and groundwater status information data are extracted, and a set of data is obtained for each set extraction distance;

[0012] The sections with a serial relationship are divided according to the mileage serial number, and the samples with a set difference distance between the previous and the next are regarded as the same section; each training sample consists of geophysical data of some excavated sections and unexcavated sections, geological characteristics of the excavated sections and true labels.

[0013] Furthermore, the LSTM network is based on the Attention layer, and the LSTM output results are weighted to construct an Attention-LSTM model, including:

[0014] a. Encoding phase: The input sequence is encoded through one or more LSTM layers, and a hidden state htht is generated at each time step;

[0015] b. Attention calculation: At each decoding step, according to the current decoder state s t-1 and all encoder hidden states {h1,h2,...,h T}, and normalize it by the softmax function to get α i ; According to the attention weight α i The weighted sum of the encoder's hidden states gives the context vector C t ;

[0016] c. Decoding stage: the context vector C t Combined with the output of the previous decoder step, it is used as the input of the current decoding step and repeated until the entire output sequence is generated.

[0017] Furthermore, the collected sample point data is input into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample point, including:

[0018] Collect and store geophysical data and tunnel face data; eliminate outliers; perform feature engineering and construct indicators required for model learning; divide the data series and perform Z-Score normalization; input the trained Attention-LSTM model for prediction and output the probability distribution of hardness, water inflow status, and integrity corresponding to each sample point; conduct excavation verification and add cases, add the real data after excavation verification to the database, and repeat the process to improve accuracy.

[0019] Furthermore, the geophysical information data includes longitudinal wave velocity, shear wave velocity, density and Poisson's ratio.

[0020] Furthermore, the hardness in the geological sketch data is divided into extremely hard rock, hard rock, relatively soft rock, soft rock, and extremely soft rock; the integrity is divided into complete, relatively complete, relatively broken, broken, and extremely broken; the groundwater state is divided into dry, wet or dripping water, rain-like or linear flow water, and gushing water.

[0021] Furthermore, the Attention-LSTM model also includes a residual connection, which is added after the LSTM layer in the decoding stage.

[0022] A tunnel surrounding rock quality evaluation index prediction system based on Attention-LSTM, which applies the tunnel surrounding rock quality evaluation index prediction method based on Attention-LSTM, includes: a data collection module, a data preprocessing module, a model training module, a prediction module, a database module, and a data processing module; the data collection module, data preprocessing module, model training module, prediction module, and database module are respectively connected to the data processing module;

[0023] Wherein, the data collection module is used to collect geophysical information data and geological sketch data;

[0024] The data preprocessing module is used to construct sequence data sets, remove outliers, perform feature engineering and normalization;

[0025] The model training module is used to train the model using the Attention-LSTM algorithm;

[0026] The prediction module is used to make predictions based on input data and output the probability distribution of surrounding rock quality evaluation indicators;

[0027] The database module is used to store training data and real data after excavation verification to support continuous optimization and updating of the model.

[0028] The beneficial effects of the present invention are: improving prediction accuracy: weighting the hidden state of the LSTM output through the Attention layer enables the model to pay more attention to the features and time steps that have an important impact on the surrounding rock quality evaluation, effectively improving the prediction accuracy of key indicators such as hardness, water gushing state, and integrity.

[0029] Enhanced model robustness: This invention constructs a sequence data set containing geophysical information data and geological sketch data, and combines residual connections to enhance the learning ability of the model, so that the model exhibits stronger robustness and generalization ability when facing complex and changeable geological conditions.

[0030] Achieve dynamic prediction: The method supports real-time data updates as tunnel excavation progresses. Through continuous data input and model prediction, it achieves dynamic monitoring and evaluation of surrounding rock quality, providing timely and accurate information support for construction decision-making.

[0031] Optimized data management: Through integrated data collection, preprocessing, model training, prediction and database modules, a complete tunnel surrounding rock quality evaluation system has been built, achieving systematic management and efficient use of data, facilitating continuous optimization and updating of models. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flowchart of the tunnel surrounding rock quality evaluation index prediction method based on Attention-LSTM;

[0033] Figure 2 This is a schematic diagram of the principle of the tunnel surrounding rock quality evaluation index prediction system based on Attention-LSTM;

[0034] Figure 3 This is a structural diagram of Attention-LSTM. DETAILED DESCRIPTION

[0035] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0036] The features and performance of the present invention are further described in detail below with reference to the embodiments.

[0037] like Figure 1 As shown in the figure, the tunnel surrounding rock quality evaluation index prediction method based on Attention-LSTM includes the following steps:

[0038] Step 1: Collect training data and construct a sequence data set based on the collected training data;

[0039] Step 2: Based on the LSTM network, the LSTM output results are weighted through the Attention layer to build an Attention-LSTM model;

[0040] Step 3: Train the Attention-LSTM model based on the constructed sequence dataset to obtain the trained Attention-LSTM model;

[0041] Step 4: Collect sample point data and input it into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample points.

[0042] The collecting of training data and constructing a sequence data set based on the collected training data include:

[0043] The elastic wave geophysical exploration equipment is used to detect the area in front of the tunnel working face to obtain the physical properties of the surrounding rock. A set of data is obtained for each set distance.

[0044] Based on the geological sketch data of the tunnel working face, the surrounding rock hardness, integrity, and groundwater status information data are extracted, and a set of data is obtained for each set extraction distance;

[0045] The sections with a serial relationship are divided according to the mileage serial number, and the samples with a set difference distance between the previous and the next are regarded as the same section; each training sample consists of geophysical data of some excavated sections and unexcavated sections, geological characteristics of the excavated sections and true labels.

[0046] The LSTM network is based on the Attention layer, which weights the LSTM output results to construct the Attention-LSTM model, including:

[0047] a. Encoding phase: The input sequence is encoded through one or more LSTM layers, and a hidden state htht is generated at each time step;

[0048] b. Attention calculation: At each decoding step, according to the current decoder state s t-1 and all encoder hidden states {h1,h2,...,h T}, and normalize it by the softmax function to get α i ; According to the attention weight α i The weighted sum of the encoder's hidden states gives the context vector C t ;

[0049] c. Decoding stage: the context vector C t Combined with the output of the previous decoder step, it is used as the input of the current decoding step and repeated until the entire output sequence is generated.

[0050] The collected sample point data is input into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample point, including:

[0051] Collect and store geophysical data and tunnel face data; eliminate outliers; perform feature engineering and construct indicators required for model learning; divide the data series and perform Z-Score normalization; input the trained Attention-LSTM model for prediction and output the probability distribution of hardness, water inflow status, and integrity corresponding to each sample point; conduct excavation verification and add cases, add the real data after excavation verification to the database, and repeat the process to improve accuracy.

[0052] The geophysical exploration information data includes longitudinal wave velocity, shear wave velocity, density and Poisson's ratio.

[0053] The hardness of the geological sketch data is divided into extremely hard rock, hard rock, relatively soft rock, soft rock, and extremely soft rock; the integrity is divided into complete, relatively complete, relatively broken, broken, and extremely broken; the groundwater state is divided into dry, wet or dripping water, rain-like or linear flow water, and gushing water.

[0054] The Attention-LSTM model also includes residual connections, which are added after the LSTM layer in the decoding stage.

[0055] like Figure 2 As shown, the tunnel surrounding rock quality evaluation index prediction system based on Attention-LSTM applies the tunnel surrounding rock quality evaluation index prediction method based on Attention-LSTM, including: a data collection module, a data preprocessing module, a model training module, a prediction module, a database module and a data processing module; the data collection module, data preprocessing module, model training module, prediction module, and database module are respectively connected to the data processing module;

[0056] Wherein, the data collection module is used to collect geophysical information data and geological sketch data;

[0057] The data preprocessing module is used to construct sequence data sets, remove outliers, perform feature engineering and normalization;

[0058] The model training module is used to train the model using the Attention-LSTM algorithm;

[0059] The prediction module is used to make predictions based on input data and output the probability distribution of surrounding rock quality evaluation indicators;

[0060] The database module is used to store training data and real data after excavation verification to support continuous optimization and updating of the model.

[0061] Specifically, they include:

[0062] 1. Model Training

[0063] 1. Training data includes

[0064] (1) Geophysical information data: Using elastic wave geophysical equipment (including but not limited to TSP, HSP, SAP, etc.) to detect the area in front of the tunnel working face, the surrounding rock physical properties are obtained, including longitudinal wave velocity, shear wave velocity, density, and Poisson's ratio. A set of data is obtained every 120 m.

[0065] (2) Using the geological sketch data of the tunnel working face, extract information on the hardness, integrity, and groundwater status of the surrounding rock. A set of data can be obtained every 3 meters.

[0066] (3) Hardness is divided into: extremely hard rock, hard rock, relatively soft rock, soft rock, and extremely soft rock

[0067] The degree of integrity is divided into: complete, relatively complete, relatively broken, broken, and extremely broken

[0068] Groundwater conditions are divided into: dry, wet or dripping, rain-like or linear flow, and gushing flow.

[0069] 2. Use the Attention-LSTM machine learning algorithm for model training

[0070] ① Sequence dataset construction

[0071] a. Sequential segmentation. Geological features exhibit strong spatial continuity and autocorrelation, so a single sample is constructed from multiple adjacent geophysical data points. Since the data consists of multiple discontinuous tunnel sections, it is necessary to first segment the data into sequential segments. For each mileage number, samples within 100 meters of each other are considered to be the same segment.

[0072] b. Training dataset structure. Each training sample consists of three parts: geophysical data of some excavated and unexcavated sections, geological characteristics of the excavated sections, and true labels.

[0073] ②Attention-LSTM network

[0074] LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) architecture used to process and predict important events with very long intervals and delays in time series data. The Attention layer weights the LSTM output to improve the representation of local features. Finally, a fully connected layer with residual connections and a softmax function are used to determine the probability of each category. The overall structure is as follows: Figure 3 shown.

[0075] a. Encoding stage: The input sequence is encoded through one or more LSTM layers, and each time step produces a hidden state htht. These hidden states together constitute the encoded representation of the input sequence.

[0076] b. Attention calculation: At each decoding step, according to the current decoder state s t-1 and all encoder hidden states {h1,h2,...,h T}, calculate the attention weight α i The calculation method is usually to convert s t-1 and each h i It is fed into a feedforward neural network and then normalized by the softmax function to obtain α i . Use these weights to weight the hidden states of the encoder to get the context vector C t .

[0077] c. Decoding stage: context vector C t The decoder is combined with the output of the previous decoder step and used as the input for the current decoding step. The decoder is also an LSTM, which generates the output for the next time step based on the current input and the previous decoding state. This process is repeated until the entire output sequence is generated.

[0078] d. Residual Connections: Residual connections are a technique used in deep learning models to maintain or improve performance while increasing model depth. Applying residual connections in Attention-LSTM models typically involves appending the input of a layer directly to the output of the layer immediately following it. For LSTM models with attention mechanisms, residual connections can be added after the LSTM layer during the decoding phase. This involves appending the decoder input directly to the decoder output, or appending the output of one layer to the output of the next. This facilitates forward propagation of information and mitigates the problem of vanishing gradients during training.

[0079] 2. Elastic Wave Method Intelligent Prediction Steps

[0080] 1. Overall forecasting ideas:

[0081] (1) According to the time series, geophysical information and geological information can be input at different times. Each time the information is input, it is run once and a prediction result is output, namely the probability of each category of integrity, hardness and groundwater.

[0082] (2) For example, the first time the geological information of the tunnel face (3 meters) and the geophysical information (120 meters) are input, the first prediction result is obtained. After the second tunnel is excavated for 3 meters, the tunnel face information is input again, and the system is run again to output the surrounding rock results 117 meters ahead.

[0083] (3) And so on, infinitely approaching the true value

[0084] 2. Specific steps:

[0085] ① Geophysical data and tunnel face data obtained from a geophysical test

[0086] According to the proposed storage specifications, the stored data is a three-dimensional data body (.CSV format). The corresponding position of each sample point and the geophysical data indicators can be determined based on the xyz coordinates and excavation direction in the data.

[0087] ② Eliminate outliers

[0088] There may be samples with missing geophysical indicators or severely deviated values in the data. Samples with abnormal geophysical indicators are usually eliminated.

[0089] ③ Feature Engineering

[0090] There are only four columns of indicators in the original data: longitudinal wave velocity, ratio of longitudinal wave to shear wave velocity, density and Poisson's ratio. In order to facilitate model learning, indicators are constructed by taking the logarithm of skewed data and feature derivation.

[0091] ④Neural network prediction

[0092] Before training, the data sequence is first segmented into multiple, sequential segments based on the mileage number and mileage. Next, the data is normalized using Z-scores to prevent the data dimension from affecting the neural network's convergence. Finally, the trained neural network is fed into the prediction function, outputting the probability distribution of hardness, water inflow, and integrity for each sample point.

[0093] ⑤ Excavation verification and adding cases

[0094] Subsequently, the actual hardness, water inflow status, and integrity after excavation verification are continuously added to the database and repeatedly executed to improve the accuracy of the module.

[0095] The foregoing description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the form disclosed herein and should not be construed as excluding other embodiments. Rather, the present invention can be used in various other combinations, modifications, and environments and can be modified within the scope of the concept described herein through the above teachings or techniques or knowledge in the relevant field. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention are intended to be protected by the appended claims.

Claims

1. The tunnel surrounding rock quality evaluation index prediction method based on Attention-LSTM is characterized by: The steps include: Step 1: Collect training data and construct a sequence data set based on the collected training data; Step 2: Based on the LSTM network, the LSTM output results are weighted through the Attention layer to build an Attention-LSTM model; Step 3: Train the Attention-LSTM model based on the constructed sequence dataset to obtain the trained Attention-LSTM model; Step 4: Collect sample point data and input it into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample points.

2. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 1 is characterized in that: The collecting of training data and constructing a sequence data set based on the collected training data include: Using elastic wave geophysical equipment, the area in front of the tunnel working face is surveyed to obtain the physical properties of the surrounding rock. A set of geological sketch data for the tunnel working face is obtained after each set distance. Based on this geological sketch data, information on the surrounding rock hardness, integrity, and groundwater status is extracted. The sections with a serial relationship are divided according to the mileage serial number, and the samples with a set difference distance between the previous and the next are regarded as the same section; each training sample consists of geophysical data of some excavated sections and unexcavated sections, geological characteristics of the excavated sections and true labels.

3. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 2 is characterized in that: The LSTM network is based on the Attention layer, which weights the LSTM output results to construct the Attention-LSTM model, including: a. Encoding stage: The input sequence is encoded through one or more LSTM layers, producing a hidden state h at each time step. T ; b. Attention calculation: At each decoding step, according to the current decoder state s t-1 and all encoder hidden states {h1,h2,...,h T }, and get ɑ through softmax function normalization i ; According to the attention weight ɑ i The weighted sum of the encoder's hidden states gives the context vector C t ; c. Decoding stage: context vector C t Combined with the output of the previous decoder step, it is used as the input of the current decoding step and repeated until the entire output sequence is generated.

4. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 3 is characterized in that: The collected sample point data is input into the trained Attention-LSTM model to obtain the probability distribution of hardness, water gushing state, and integrity corresponding to the sample point, including: Collect geophysical data and tunnel face data and remove outliers; divide the data series and perform Z-Score normalization; input the trained Attention-LSTM model for prediction, and output the probability distribution of surrounding rock hardness, water gushing status, and integrity corresponding to each sample point; excavate and verify and add cases, add the real data after excavation verification to the database, and repeat the process to improve accuracy.

5. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 4 is characterized in that: The geophysical exploration information data includes longitudinal wave velocity, shear wave velocity, density and Poisson's ratio.

6. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 2 is characterized in that: The hardness of the geological sketch data is divided into extremely hard rock, hard rock, relatively soft rock, soft rock, and extremely soft rock; the integrity is divided into complete, relatively complete, relatively broken, broken, and extremely broken; the groundwater state is divided into dry, wet or dripping water, rain-like or linear flow water, and gushing water.

7. The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to claim 2 is characterized in that: The Attention-LSTM model also includes residual connections, which are added after the LSTM layer in the decoding stage.

8. The tunnel surrounding rock quality evaluation index prediction system based on Attention-LSTM is characterized by: The method for predicting tunnel surrounding rock quality evaluation indicators based on Attention-LSTM according to any one of claims 1 to 7 comprises: a data collection module, a data preprocessing module, a model training module, a prediction module, a database module, and a data processing module; the data collection module, the data preprocessing module, the model training module, the prediction module, and the database module are respectively connected to the data processing module; Wherein, the data collection module is used to collect geophysical information data and geological sketch data; The data preprocessing module is used to construct sequence data sets, remove outliers, perform feature engineering and normalization; The model training module is used to train the model using the Attention-LSTM algorithm; The prediction module is used to make predictions based on input data and output the probability distribution of surrounding rock quality evaluation indicators; The database module is used to store training data and real data after excavation verification to support continuous optimization and updating of the model.

Images