Engineering intact rock and drilling parameter determination method, system, medium, and device

Abstract

The application provides an engineering intact rock and drilling parameter determination method, system, medium and equipment, and belongs to the technical field of tunnels and underground engineering. The method comprises the following steps: acquiring multi-source time sequence data in a drilling process and preprocessing the multi-source time sequence data to obtain a multi-channel time sequence matrix; inputting the multi-channel time sequence matrix into a trained multi-parameter collaborative physical constraint time sequence neural network model, extracting a multi-scale fusion feature map through a multi-scale space feature extraction module; establishing a nonlinear collaborative mapping among drilling parameters through a parameter intersection attention module; mining forward and backward time sequence dependencies through a bidirectional time sequence modeling module; constructing a physical loss function based on a physical constraint output module in combination with a drilling process mechanics equation to output a probability distribution of each time window belonging to an intact rock paragraph; and finally determining the intact rock paragraph and the drilling parameters. The application significantly improves the accuracy of intact rock paragraph recognition and the reliability of drilling parameter interpretation.

Description

Technical Field

[0001] This invention belongs to the field of tunnel and underground engineering technology, and particularly relates to a method, system, medium and equipment for determining the integrity of rock and drilling parameters in engineering. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In tunnel and underground engineering construction, advanced geological drilling is a key technical means to reveal the geological conditions ahead of the tunnel face, determine the integrity of the surrounding rock, and ensure construction safety. During drilling, multi-source drilling parameters such as thrust, torque, rotational speed, and drilling rate can be collected in real time. These parameters directly reflect the difficulty of rock fracturing and the integrity of the rock mass, and are important bases for in-situ identification of intact rock sections and determination of representative drilling parameters. Intelligent lithological interpretation based on drilling parameters has significant engineering value for improving the accuracy of geological prediction and optimizing construction parameters.

[0004] Currently, the identification of intact rock segments and the interpretation of drilling parameters in engineering and scientific research still mainly rely on traditional numerical analysis methods, which generally have significant technical limitations. Most existing methods only select drilling rate as the core analytical indicator, using forward differencing and numerical differentiation to determine abrupt changes in the curve and thus delineate intact rock segments. These methods utilize only single-parameter information, ignoring the inherent mechanical coupling and temporal correlation between thrust, torque, rotational speed, and drilling rate. This makes them prone to misjudgment in fractured rock masses and jointed strata, resulting in poor robustness in identification.

[0005] Meanwhile, drilling parameters collected on-site are easily affected by drilling rig vibration, hydraulic fluctuations, and electromagnetic interference, containing a large amount of high-frequency noise. Traditional differential methods are highly sensitive to noise, significantly amplifying noise fluctuations and producing false peaks and abnormal jumps, making it difficult to distinguish between real lithological changes and noise interference, resulting in low accuracy and insufficient stability in lithological boundary identification. In addition, existing technologies rely on manual threshold setting based on borehole television images, which is highly subjective, cumbersome, and has low automation, making it difficult to adapt to complex and variable geological conditions and drilling conditions, and exhibiting weak generalization ability.

[0006] In recent years, deep learning has been applied in time series data analysis. However, conventional neural networks are mostly purely data-driven, failing to incorporate physical priors such as drilling mechanics and energy transfer relationships into the model. This can lead to outputs that contradict rock-breaking mechanisms, resulting in non-physical anomalies and insufficient engineering reliability and field applicability. Therefore, the industry urgently needs an intelligent interpretation method that can collaboratively utilize multi-source parameters and integrate deep learning feature extraction capabilities with physical mechanism constraints to achieve high-precision, robust, and automatic identification of complete rock sections and reliably determine representative drilling parameters. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, the present invention provides a method, system, medium, and equipment for determining complete rock sections and drilling parameters in engineering. It aims to solve the defects of the prior art, which only differentiates with respect to a single drilling rate, resulting in poor noise resistance and failure to utilize multi-parameter collaborative information, and significantly improves the accuracy of complete rock section identification and the reliability of drilling parameter interpretation.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for determining the integrity of rock in an engineering project and drilling parameters; Methods for determining the integrity of the rock and drilling parameters in engineering projects include: The multi-source drilling parameter time series data during the drilling process of the drilling rig are acquired and preprocessed to obtain a multi-channel time series matrix; The multi-channel time series matrix is ​​input into the trained multi-parameter collaborative physical constraint time series neural network model to determine the complete rock segment and the representative drilling parameters of the complete rock. The neural network model extracts features from the multi-channel time-series matrix using a multi-scale spatial feature extraction module to obtain a multi-scale fused feature map. This multi-scale fused feature map is then input into a parameter cross-attention module to establish a nonlinear collaborative mapping relationship between drilling parameters, outputting parameter fusion features. A bidirectional temporal modeling module performs forward and backward temporal dependency mining on the parameter fusion features to obtain depth feature representations. Based on a physical constraint output module, a physical loss function is constructed using the drilling process mechanics equations to output the probability distribution of each time window belonging to a complete rock segment. Complete rock segments are determined according to the probability distribution, and the multi-source drilling parameters within each segment are fused to obtain representative drilling parameters for the complete rock segment.

[0009] As a further technical solution, preprocessing is performed on the time-series data of multi-source drilling parameters, including: Remove unstable data segments during the initial hole opening and rod replacement phases of the drilling process; Wavelet thresholding was used to denoise the time series data of multi-source drilling parameters to eliminate high-frequency acquisition noise. The denoised data was then normalized using Z-score to eliminate dimensional differences. A sliding time window was used to slice the normalized time series data.

[0010] As a further technical solution, the multi-scale spatial feature extraction module specifically includes: three parallel one-dimensional convolutional branches, each branch followed by a batch normalization layer and a ReLU activation function; The outputs of the three branches are concatenated along the channel dimension, and then dimensionality is reduced and fused through one-dimensional convolution to obtain a multi-scale fused feature map.

[0011] As a further technical solution, the multi-scale fused feature map is input into the parameter cross-attention module to establish a nonlinear collaborative mapping relationship between various drilling parameters, and the parameter fusion features are output, including: The feature vectors corresponding to the four drilling parameters (thrust, torque, rotational speed, and drilling speed) in the multi-scale fused feature map are mapped to a query matrix Q, a key matrix K, and a value matrix V through independent linear transformations, respectively. The coupling attention weights between parameters are calculated using a scaled dot product attention mechanism:

[0012] in, is the dimension of the key matrix; T is the transpose; The value matrices of each parameter are weighted and aggregated according to the coupling weights to establish a nonlinear collaborative mapping of thrust-torque-rotation speed-drilling speed, and the parameter fusion features are output.

[0013] As a further technical solution, a bidirectional temporal modeling module is used to perform forward and backward temporal dependency mining on the parameter fusion features to obtain a deep feature representation, including: The parameter fusion features are input into a two-layer bidirectional long short-term memory network, wherein the forward long short-term memory network layer is used to mine the forward trend information of the time series data, and the backward long short-term memory network layer is used to mine the backward dependency information of the time series data. By concatenating the forward and backward hidden states, a deep feature representation containing global contextual information is obtained.

[0014] As a further technical solution, based on the physical constraint output module, a physical loss function is constructed by combining the mechanical equations of the drilling process, and the probability distribution of each time window belonging to a complete rock segment is output, including: Based on drilling mechanics theory, physical equations relating drilling speed, thrust, rotational speed, and rock integrity are established:

[0015] in, The drilling rate is predicted by the physical equations. For thrust, For rotational speed, Where σ is the drill bit diameter, and σ is the estimated uniaxial compressive strength of the rock. , , , , Physical coefficients related to drilling conditions; Constructing the physical consistency loss term :

[0016] in, This represents the batch sample size. For torque, This is the theoretical energy transfer efficiency benchmark for a complete rock segment. , These are the weighting coefficients; Constructing the time-series smoothing loss term :

[0017] in, Let be the probability that the i-th time window belongs to a complete rock segment; The total loss function is:

[0018] in, This is the total loss function; The binary cross-entropy classification loss is used. The neural network is trained by constraining the total loss function, and the probability distribution of each time window belonging to a complete rock segment is output.

[0019] As a further technical solution, complete rock segments are determined based on the probability distribution, and multi-source drilling parameters within the segments are fused to obtain representative drilling parameters for the complete rock segments, including: Set a probability threshold, extract time window sequences that are continuously higher than the probability threshold, perform morphological post-processing on the extracted sequences, remove isolated segments with a length less than the minimum number of sampling points, and merge adjacent segments with a gap less than a preset distance to obtain complete rock segments. Calculate the time-domain statistical characteristics of thrust, torque, rotational speed, and drilling speed within each complete rock segment. The time-domain statistical characteristics include mean, standard deviation, effective value, and coefficient of variation. The entropy weight method is used to determine the objective weights based on the information entropy of each parameter. The smaller the information entropy, the higher the weight is assigned. The time-domain statistical features are weighted and fused to obtain a multi-dimensional representative drilling parameter vector, which is used as the representative drilling parameter output corresponding to the complete rock segment.

[0020] A second aspect of the present invention provides a system for determining complete rock and drilling parameters in engineering.

[0021] A complete system for determining rock and drilling parameters for engineering projects, including: The data acquisition and preprocessing module is used to acquire and preprocess multi-source drilling parameter time-series data during the drilling process of the drilling rig to obtain a multi-channel time-series matrix; the multi-source drilling parameters include thrust, torque, rotational speed and drilling speed; A multi-parameter cooperative physical constraint temporal neural network model, the neural network model comprising: A multi-scale spatial feature extraction module is used to extract features from the multi-channel time series matrix to obtain a multi-scale fused feature map; The parameter cross-attention module is used to establish a nonlinear collaborative mapping relationship between various drilling parameters on the multi-scale fused feature map and output parameter fusion features. A bidirectional temporal modeling module is used to perform forward and backward temporal dependency mining on the parameter fusion features to obtain a deep feature representation; The physical constraint output module is used to construct a physical loss function by combining the mechanical equations of the drilling process, and output the probability distribution of each time window belonging to a complete rock segment; The segment determination and parameter fusion module is used to determine complete rock segments based on the probability distribution, and to fuse the multi-source drilling parameters within the segment to obtain representative drilling parameters for the complete rock.

[0022] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method for determining engineering intact rock and drilling parameters as described in the first aspect of the present invention.

[0023] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the method for determining engineering intact rock and drilling parameters as described in the first aspect of the present invention.

[0024] The above one or more technical solutions have the following beneficial effects: (1) This invention abandons the conventional method of relying solely on a single drilling speed parameter for numerical differentiation, and for the first time incorporates four drilling parameters—thrust, torque, rotational speed, and drilling speed—into a unified intelligent interpretation framework. By constructing a cross-attention module between parameters, this invention can adaptively learn the nonlinear coupling and synergistic relationships between parameters, and automatically discover complex synergistic discrimination rules such as "when thrust and torque are synchronously stable, even if there are slight fluctuations in drilling speed, the segment may still be intact rock." Compared with traditional methods, this invention fully utilizes the inherent correlation information of multiple source parameters, avoiding misjudgments caused by insufficient representation capabilities of a single parameter.

[0025] (2) This invention addresses the significant high-frequency noise in drilling parameters at the engineering site. In the preprocessing stage, wavelet thresholding is used to effectively suppress acquisition noise. Simultaneously, the model design employs multi-scale convolution and a bidirectional temporal modeling structure to directly extract robust deep features from noisy data, fundamentally avoiding the inherent defects of traditional differential methods that amplify noise signals. Furthermore, this invention eliminates the need for manually setting derivative thresholds or relying on manual comparison of borehole television images. Through end-to-end neural network learning and physical constraints, it automatically outputs the probability distribution of each time window belonging to intact rock and automatically determines segment boundaries based on threshold segmentation and morphological post-processing. The entire process achieves fully automated identification, effectively solving problems such as strong subjectivity and poor adaptability, and can be easily extended to different geological conditions and drilling scenarios.

[0026] (3) Based on drilling mechanics theory, this invention constructs a physical equation for drilling speed and an energy transfer efficiency equation, forming a physical consistency loss term. Simultaneously, it constructs a temporal smoothing loss term to constrain the continuity of segment identification. Through multi-task joint training, the neural network not only learns the statistical characteristics of data labels but is also forced to learn feature representations that conform to the rock-breaking physical mechanism, avoiding the non-physical abnormal results that a pure black-box model might output. The introduction of physical constraints significantly improves the model's generalization ability under unknown geological conditions, giving the output results clear mechanical interpretability and greatly enhancing the trust of engineering technicians in the intelligent interpretation results.

[0027] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0028] 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.

[0029] Figure 1 This is a flowchart of the method in the first embodiment.

[0030] Figure 2 The diagram below shows the results of identifying a complete rock segment in the first embodiment, where (a) is a line graph of drilling parameters obtained by the method of the present invention; and (b) is a diagram of the identified complete rock segment. Detailed Implementation

[0031] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0032] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0033] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0034] Example 1 This embodiment discloses a method for determining complete rock sections and drilling parameters in engineering. By constructing an improved neural network model, the method collaboratively analyzes the temporal characteristics of thrust, torque, rotational speed, and drilling speed, and introduces the mechanical equations of the drilling process as physical constraints to achieve high-precision automatic identification of complete rock sections and reliable determination of drilling parameters.

[0035] like Figure 1 As shown, the method for determining the complete rock mass and drilling parameters for the project includes: Step S1: Obtain multi-source drilling parameter time series data during the drilling process of the drilling rig and perform preprocessing to obtain a multi-channel time series matrix.

[0036] Time-series data of multi-source drilling parameters during the drilling process are acquired. These parameters include at least thrust F, torque T, rotational speed N, and drilling speed V. Due to the complexity of joint structures within the rock mass, drilling parameters fluctuate significantly when traversing fractures or faults. Furthermore, short-term instability in drill pressure and rotational speed occurs after each initial hole opening and rod change, leading to abnormal changes in drilling speed and torque. Therefore, the acquired time-series data is preprocessed to remove unstable data segments during the initial hole opening and rod change periods. Furthermore, the preprocessing includes: using wavelet thresholding to eliminate high-frequency acquisition noise; performing Z-score normalization on the denoised multi-source parameters to eliminate differences in units and numerical ranges between different parameters; and slicing the normalized time-series data using a sliding time window to construct a multi-channel time-series matrix X∈R^(W×4), where W is the window length and 4 corresponds to the four parameter channels. The step size of the sliding window is smaller than the window length to ensure temporal continuity between adjacent samples, facilitating subsequent bidirectional time-series modeling.

[0037] Step S2: Input the multi-channel time series matrix into the trained multi-parameter collaborative physical constraint time series neural network model, wherein the neural network model extracts features from the multi-channel time series matrix through a multi-scale spatial feature extraction module to obtain a multi-scale fused feature map.

[0038] The preprocessed multi-channel temporal matrix is ​​input into the multi-scale spatial feature extraction module of the multi-parameter co-physically constrained temporal neural network model. This module contains three parallel one-dimensional convolutional branches, employing causal convolutions with kernel sizes of 3, 5, and 7, respectively. Each branch is followed by a batch normalization layer and a ReLU activation function. The three branches capture the local response features of the drilling parameters at short-term (kernel size 3), medium-term (kernel size 5), and long-term (kernel size 7) time scales, respectively. After channel concatenation, dimensionality reduction and fusion are performed using 1×1 convolutions to obtain a multi-scale fused feature map Hs∈RW×Cs, where Cs is the number of channels after fusion.

[0039] The multi-scale spatial feature extraction module uses three parallel one-dimensional causal convolution branches with kernel sizes of 3, 5, and 7. Each branch outputs features after batch normalization and ReLU activation. The three features are concatenated along the channel dimension to obtain the concatenated features. Then, a 1×1 one-dimensional convolution is used to perform channel-weighted fusion and dimensionality reduction on the spliced ​​features, and the operation satisfies... In the formula , The weights and biases of the 1×1 convolution are respectively used to output multi-scale fused features. This approach can fully preserve the multi-scale drilling time-series features of short, medium, and long time periods, while achieving cross-scale feature coupling through convolutional adaptation and compressing channel redundancy dimensions.

[0040] The physical significance of this design lies in the fact that the differences in drilling parameters between intact rock sections and fractured sections are not only reflected in instantaneous fluctuations, but also in trend changes at different time scales. For example, when traversing fractures, the drilling rate may drop sharply in a very short time (short-scale characteristic), while after entering intact rock strata, the parameters will show a sustained and stable trend (long-scale characteristic). Multi-scale convolution can simultaneously capture these features at different levels.

[0041] Step S3: Input the multi-scale fused feature map into the parameter cross-attention module to establish a nonlinear collaborative mapping relationship between drilling parameters and output parameter fusion features.

[0042] A cross-attention module is implemented between the input parameters of the multi-scale fused feature map. This module treats the feature vectors corresponding to each drilling parameter as independent feature tokens, which are mapped to a query matrix Q, a key matrix K, and a value matrix V through independent linear transformations.

[0043]

[0044]

[0045] in, This indicates four parameter channels. It is a learnable linear projection matrix.

[0046] The coupling attention weights between parameters are calculated using a scaled dot product attention mechanism:

[0047] in, Let be the dimension of the key matrix, and be the scaling factor. To prevent the softmax gradient from vanishing due to excessively large dot product values.

[0048] The core innovation of this cross-attention mechanism lies in the fact that traditional methods treat parameters independently or simply concatenate them, failing to model the nonlinear coupling relationships between parameters. During drilling, increased thrust often leads to increased torque and changes in drilling speed, creating complex mechanical coupling among these three factors. The cross-attention module adaptively calculates the attention weights between parameters, automatically learning complex collaborative discrimination rules such as "when thrust and torque are synchronously stable, even with slight fluctuations in drilling speed, the segment may still be intact rock," significantly improving the model's ability to distinguish intact rock segments.

[0049] Step S4: The parameter fusion features are subjected to forward and backward temporal dependency mining through the bidirectional temporal modeling module to obtain a deep feature representation.

[0050] The fused parameters are input into the bidirectional temporal modeling module. This module employs a two-layer bidirectional long short-term memory (BiLSTM) network with 128 hidden layers and a dropout rate of 0.2 for regularization between layers. The internal computation of a single-step LSTM is as follows:

[0051] In the formula, , , These represent the states of the input gate, forget gate, and output gate, respectively. Candidate cell state, This represents the cell state after the update. Forward LSTM outputs the hidden state at time t; ⊙ represents the Hadamard product; , , , These correspond to the input weight matrices for the input gate, forget gate, output gate, and candidate cell state, respectively. , , , These are the cyclic weight matrices corresponding to the input gate, forget gate, output gate, and candidate cell state, respectively. , , , These correspond to the bias vectors of the input gate, forget gate, output gate, and candidate cell state, respectively. Input the feature vector for the current time step; This is the hidden state from the previous moment; This is the Sigmoid activation function.

[0052] The backward LSTM traverses the feature sequence in reverse order along the temporal path, extracting future contextual information from the back to the front. This corresponds to a single-step hidden state, and the calculation formula is the same as the forward LSTM structure, only the feature traversal order is reversed. This bidirectional structure can simultaneously mine forward trend information (historical context) and backward dependency information (future context) from the temporal data, effectively avoiding the problem of inaccurate identification at paragraph boundaries in unidirectional models. The output is a deep feature representation H containing global contextual information. t ∈R W×256 (Dimensions after bidirectional splicing).

[0053] Step S5: Based on the physical constraint output module, a physical loss function is constructed by combining the mechanical equations of the drilling process, and the probability distribution of each time window belonging to a complete rock segment is output.

[0054] The deep feature representation is input into the physical constraint output module. This module contains two parallel branches: a classification branch and a physical regression branch.

[0055] The classification branch outputs the probability p∈[0,1] that the current time window belongs to a complete rock segment through a fully connected layer (FC:256→128→1) and a sigmoid activation function.

[0056] The physical regression branch outputs estimated values ​​of each parameter within the current window through a fully connected layer (FC:256→128→4) to calculate the physical loss.

[0057] Furthermore, by constructing a physical constraint loss function, the training of the neural network is not only supervised by data labels but also constrained by the physical laws of drilling into the system. Specifically, this includes: (1) Physical consistency loss item : Based on drilling mechanics theory, a stable physical relationship exists between drilling rate, thrust, and rotational speed in intact rock sections. The drilling rate equation, corrected for lithology coefficients, is adopted:

[0058] Among them, C, , γ and δ are physical coefficients related to drilling conditions and drill bit type, which can be determined through field calibration or literature; D is the drill bit diameter; σ is the estimated value of the uniaxial compressive strength of the rock (as an intermediate variable for implicit learning of the network).

[0059] Construction of drilling rate reconstruction loss:

[0060] (2) Energy transfer efficiency loss item : During drilling, the input power is primarily used for rock breaking and friction consumption. In intact rock sections, the energy transfer efficiency should remain stable at the theoretical benchmark value. nearby:

[0061] Among them, molecules For the rotational input power, the denominator This represents the axial drilling power, and its ratio reflects the energy distribution efficiency.

[0062] (3) Time series smoothing loss term : The complete rock segment should have temporal continuity, and the probability output of adjacent windows should not change drastically.

[0063] (4) Total loss function:

[0064] in, The binary cross-entropy classification loss is used. To balance the weighting coefficients of each loss term.

[0065] By constraining the physical loss function, the neural network is forced to learn feature representations that conform to the physical laws of drilling during the training process, avoiding non-physical outputs that may be generated by a purely data-driven model, and significantly improving the model's generalization ability and engineering credibility under unknown geological conditions.

[0066] Step S6: Determine the complete rock segment based on the probability distribution, and perform fusion processing on the multi-source drilling parameters within the segment to obtain representative drilling parameters for the complete rock.

[0067] Based on the probability sequence {p1,p2,...,pL} output by the physical constraint output module, a probability threshold is set. (e.g., 0.7), extract all time window indices that are consecutively higher than the threshold. Perform morphological post-processing on the extracted sequences: remove sequences whose length is less than the minimum number of sampling points. Isolated segments (no fewer than 5 consecutive data points, determined based on actual drilling speed and sampling frequency); merging gaps are less than the preset distance. Adjacent paragraphs are separated to eliminate paragraph fragmentation caused by short-term noise.

[0068] The multi-source drilling parameters within the defined complete rock section are fused: the time-domain statistical characteristics of thrust, torque, rotational speed, and drilling speed within each section are calculated, including the mean. Standard deviation Effective value (RMS) and coefficient of variation (CV). The entropy weighting method is used to determine the objective weights of each parameter based on its information entropy. A lower information entropy indicates lower dispersion, higher stability, and greater information content for characterizing the complete rock within that segment; therefore, a higher weight is assigned. A multidimensional representative drilling parameter vector is obtained through weighted fusion. This serves as the final drilling parameter output corresponding to the complete rock section.

[0069] Specifically, firstly, for each identified complete rock segment, time-series data of four drilling parameters—thrust, torque, rotational speed, and drilling rate—are extracted from that segment. The mean, standard deviation, effective value, and coefficient of variation are calculated for each parameter to form the initial evaluation matrix. The rows of the matrix correspond to the sampling times or statistical samples within the segment, and the columns correspond to the four drilling parameters. Since the dimensions and numerical ranges of each parameter differ, each column needs to be normalized. The proportion of each sampled value in the sum of all sampled values ​​for that parameter is calculated to obtain the normalized probability matrix.

[0070] Secondly, based on the definition of information entropy, the information entropy value of each drilling parameter is calculated. First, normalize the sampled data of the j-th drilling parameter in the paragraph. Where n is the total number of paragraph samples; parameter information entropy .

[0071] For a given parameter, if its sampled values ​​fluctuate little and the data distribution is relatively concentrated within a segment, its information entropy is low, indicating that the parameter is highly stable within the intact rock segment and can more reliably reflect the integrity of the rock; therefore, it should be assigned a higher weight. Conversely, if the parameter fluctuates drastically and has a large degree of dispersion, its information entropy is high, indicating that it is greatly affected by noise or local fragmentation and has limited contribution to characterizing intact rocks; therefore, its weight should be reduced accordingly.

[0072] Finally, the time-domain statistical characteristics (such as the mean) of each parameter are weighted and summed with their corresponding weights to generate a multi-dimensional representative drill-in parameter vector. Specifically, based on... Calculate the objective weights of each parameter The obtained weights are used to sum the time-domain statistical characteristics of each parameter, and the resulting data are then fused to generate a representative drilling parameter vector. This vector integrates the stable characteristic information of multiple drilling parameters and can be used as a standardized output of the complete rock segment for subsequent lithology identification, surrounding rock classification, or construction parameter optimization.

[0073] This invention selects an advanced drilling section (depth 0-30m) from a tunnel project for verification. For example... Figure 2 As shown, the method of this invention automatically identified a total of 52 complete surrounding rock segments. Compared with borehole television images, the identification accuracy reached 94.3%, which is a significant improvement compared with the traditional single drilling rate difference method (accuracy 76.8%). Representative drilling parameters, compared with indoor core tests, showed that the calculated uniaxial compressive strength error was less than 12%, meeting engineering accuracy requirements.

[0074] Example 2 This embodiment discloses a system for determining complete rock formations and drilling parameters in engineering. A complete system for determining rock and drilling parameters for engineering projects, including: The data acquisition and preprocessing module is used to acquire and preprocess multi-source drilling parameter time-series data during the drilling process of the drilling rig to obtain a multi-channel time-series matrix; the multi-source drilling parameters include thrust, torque, rotational speed and drilling speed; A multi-parameter cooperative physical constraint temporal neural network model, the neural network model comprising: A multi-scale spatial feature extraction module is used to extract features from the multi-channel time series matrix to obtain a multi-scale fused feature map; The parameter cross-attention module is used to establish a nonlinear collaborative mapping relationship between various drilling parameters on the multi-scale fused feature map and output parameter fusion features. A bidirectional temporal modeling module is used to perform forward and backward temporal dependency mining on the parameter fusion features to obtain a deep feature representation; The physical constraint output module is used to construct a physical loss function by combining the mechanical equations of the drilling process, and output the probability distribution of each time window belonging to a complete rock segment; The segment determination and parameter fusion module is used to determine complete rock segments based on the probability distribution, and to fuse the multi-source drilling parameters within the segment to obtain representative drilling parameters for the complete rock.

[0075] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0076] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the method for determining engineering intact rock and drilling parameters as described in Example 1.

[0077] Example 4 The purpose of this embodiment is to provide an electronic device.

[0078] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for determining engineering intact rock and drilling parameters as described in Example 1.

[0079] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0080] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0081] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for determining the integrity of rock and drilling parameters in engineering projects, characterized in that, include: The multi-source drilling parameter time series data during the drilling process of the drilling rig are acquired and preprocessed to obtain a multi-channel time series matrix; The multi-channel time series matrix is ​​input into the trained multi-parameter collaborative physical constraint time series neural network model to determine the complete rock segment and the representative drilling parameters of the complete rock. The neural network model extracts features from the multi-channel temporal matrix through a multi-scale spatial feature extraction module to obtain a multi-scale fused feature map. The multi-scale fused feature map is then input into a parameter cross-attention module to establish a nonlinear collaborative mapping relationship between the drilling parameters and output parameter fusion features. The parameter fusion features are then subjected to forward and backward temporal dependency mining through a bidirectional temporal modeling module to obtain a deep feature representation. Based on the physical constraint output module, a physical loss function is constructed by combining the mechanical equations of the drilling process, and the probability distribution of each time window belonging to the complete rock segment is output. The complete rock segment is determined according to the probability distribution, and the multi-source drilling parameters within the segment are fused to obtain the representative drilling parameters of the complete rock.

2. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, Preprocessing of multi-source drilling parameter time-series data includes: Remove unstable data segments during the initial hole opening and rod replacement phases of the drilling process; Wavelet thresholding was used to denoise the time series data of multi-source drilling parameters to eliminate high-frequency acquisition noise. The denoised data was then normalized using Z-score to eliminate dimensional differences. A sliding time window was used to slice the normalized time series data.

3. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, The multi-scale spatial feature extraction module specifically includes: three parallel one-dimensional convolutional branches, each followed by a batch normalization layer and a ReLU activation function; The outputs of the three branches are concatenated along the channel dimension, and then dimensionality is reduced and fused through one-dimensional convolution to obtain a multi-scale fused feature map.

4. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, The multi-scale fused feature map is input into the parameter cross-attention module to establish a nonlinear collaborative mapping relationship between various drilling parameters, and the parameter fusion features are output, including: The feature vectors corresponding to the four drilling parameters (thrust, torque, rotational speed, and drilling speed) in the multi-scale fused feature map are mapped to a query matrix Q, a key matrix K, and a value matrix V through independent linear transformations, respectively. The coupling attention weights between parameters are calculated using a scaled dot product attention mechanism: in, is the dimension of the key matrix; T is the transpose; The value matrices of each parameter are weighted and aggregated according to the coupled attention weights to establish a nonlinear collaborative mapping of thrust-torque-rotation speed-drilling speed, and output parameter fusion features.

5. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, The parameter fusion features are subjected to forward and backward temporal dependency mining through a bidirectional temporal modeling module to obtain a deep feature representation, including: The parameter fusion features are input into a two-layer bidirectional long short-term memory network, wherein the forward long short-term memory network layer is used to mine the forward trend information of the time series data, and the backward long short-term memory network layer is used to mine the backward dependency information of the time series data. By concatenating the forward and backward hidden states, a deep feature representation containing global contextual information is obtained.

6. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, Based on the physical constraint output module, a physical loss function is constructed by combining the mechanical equations of the drilling process, and the probability distribution of each time window belonging to a complete rock segment is output, including: Based on drilling mechanics theory, physical equations relating drilling speed, thrust, rotational speed, and rock integrity are established: in, The drilling rate is predicted by the physical equations. For thrust, For rotational speed, Where σ is the drill bit diameter, and σ is the estimated uniaxial compressive strength of the rock. , , , , Physical coefficients related to drilling conditions; Constructing the physical consistency loss term : in, This represents the batch sample size. For torque, This is the theoretical energy transfer efficiency benchmark for a complete rock segment. , These are the weighting coefficients; Constructing the time-series smoothing loss term : in, Let be the probability that the i-th time window belongs to a complete rock segment; The total loss function is: in, This is the total loss function; The binary cross-entropy classification loss is used. The neural network is trained by constraining the total loss function, and the probability distribution of each time window belonging to a complete rock segment is output.

7. The method for determining the complete rock formation and drilling parameters as described in claim 1, characterized in that, Based on the probability distribution, complete rock segments are determined, and the multi-source drilling parameters within each segment are fused to obtain representative drilling parameters for the complete rock segment, including: Set a probability threshold, extract time window sequences that are continuously higher than the probability threshold, perform morphological post-processing on the extracted sequences, remove isolated segments with a length less than the minimum number of sampling points, and merge adjacent segments with a gap less than a preset distance to obtain complete rock segments. Calculate the time-domain statistical characteristics of thrust, torque, rotational speed, and drilling speed within each complete rock segment. The time-domain statistical characteristics include mean, standard deviation, effective value, and coefficient of variation. The entropy weight method is used to determine the objective weights based on the information entropy of each parameter. The smaller the information entropy, the higher the weight is assigned. The time-domain statistical features are weighted and fused to obtain a multi-dimensional representative drilling parameter vector, which is used as the representative drilling parameter output corresponding to the complete rock segment.

8. A system for determining complete rock and drilling parameters in engineering projects, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess multi-source drilling parameter time-series data during the drilling process of the drilling rig to obtain a multi-channel time-series matrix. The multi-source drilling parameters include thrust, torque, rotational speed, and drilling rate; A multi-parameter cooperative physical constraint temporal neural network model, the neural network model comprising: A multi-scale spatial feature extraction module is used to extract features from the multi-channel time series matrix to obtain a multi-scale fused feature map; The parameter cross-attention module is used to establish a nonlinear collaborative mapping relationship between various drilling parameters on the multi-scale fused feature map and output parameter fusion features. A bidirectional temporal modeling module is used to perform forward and backward temporal dependency mining on the parameter fusion features to obtain a deep feature representation; The physical constraint output module is used to construct a physical loss function by combining the mechanical equations of the drilling process, and output the probability distribution of each time window belonging to a complete rock segment; The segment determination and parameter fusion module is used to determine complete rock segments based on the probability distribution, and to fuse the multi-source drilling parameters within the segment to obtain representative drilling parameters for the complete rock.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the method for determining the complete rock and drilling parameters as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for determining the complete rock and drilling parameters of an engineering project as described in any one of claims 1-7.

Images