Surrounding rock grouting stone body sensing system and method based on advanced drilling while drilling test

Abstract

The invention provides a surrounding rock grouting stone body sensing system and method based on advanced drilling and testing while drilling. The method comprises the steps that real-time tunneling parameters and geological parameter data in the drilling process are obtained; processing the data, and determining the weights of the tunneling parameters and the geological parameters in a final analysis model by using a function model so as to construct the final analysis model; performing hyper-parameter optimization by using the five sub-models with different viewpoints and adopting a particle swarm optimization algorithm, training the model based on known tunneling parameters and geological parameters, inputting the tunneling parameters into the trained model, predicting unknown geological parameters, and sensing geological and mechanical characteristics of a grouting stone body; and fusing feature results perceived by the five sub-models after optimization through a D-S evidence theory to obtain final surrounding rock grouting stone body features. According to the method, the accuracy of sensing the surrounding rock grouting stone body can be remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of directional drilling advanced pre-grouting, and particularly relates to a surrounding rock grouting stone body perception system and method based on advanced drilling while drilling testing. BACKGROUND

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

[0003] In recent years, with the continuous advancement of infrastructure construction, the horizontal directional drilling technology widely used in the fields of municipal and oil and gas pipeline laying has become one of the research hotspots.

[0004] From the existing research work, in the aspect of long-distance directional drilling while drilling lithology prediction, there are more studies on the inversion of rock uniaxial compressive strength based on drilling parameters, and there are fewer studies on the relationship between other rock mechanics parameters and drilling parameters. There are too many parameters to be determined, a certain subjectivity, and low accuracy. With the development of artificial intelligence, neural networks, support vector machines, decision trees and other machine learning algorithms are gradually applied, but as an identification method, a single model often exists, and there is a problem of low accuracy. SUMMARY

[0005] In order to solve the above problems, the present application provides a surrounding rock grouting stone body perception system and method based on advanced drilling while drilling testing. The present application adopts a function method to establish a stone body parameter perception model for geological parameters and tunneling parameters, maps and models the perception of grouting stone body characteristics based on deep learning, and fuses the identification results of multiple models through D-S evidence theory to improve the accuracy.

[0006] According to some embodiments, the present application adopts the following technical solutions: A surrounding rock grouting stone body perception system based on advanced drilling while drilling testing, comprising: A data acquisition module for acquiring real-time tunneling parameter and geological parameter data during drilling; A preprocessing module including a data correction module and a data conversion module, the data correction module being used for removing abnormal values in the acquired data through a clustering algorithm, and the data conversion module being used for standardizing the tunneling parameter data through a Z-Score method; A weight allocation module for determining the weights of the tunneling parameters and the geological parameters in the final analysis model by using a function model to construct a final analysis model; The feature preliminary screening module includes an extreme learning machine sub-model, a back propagation artificial neural network sub-model, a radial basis function artificial neural network sub-model, a random forest sub-model and a K-nearest neighbor algorithm sub-model, all of which are optimized by a particle swarm optimization algorithm, and are trained based on known tunneling parameters and geological parameters, and the trained model is used to input the tunneling parameters to predict unknown geological parameters and perceive the geological and mechanical characteristics of the grouting rock mass; The multi-modal fusion module fuses the feature results perceived by the five sub-models after optimization through D-S evidence theory to obtain the final characteristics of the grouting rock mass.

[0007] As an optional implementation, the geological parameters include uniaxial compressive strength, initial ground stress state, groundwater water content state and structure surface state; and the tunneling parameters include thrust, penetration, torque and tunneling speed.

[0008] As an optional implementation, the data correction module uses a DBSCAN clustering algorithm to eliminate outliers in the obtained data.

[0009] As an optional implementation, the data transformation module uses a Z-Score standardization method to uniformly scale the data in each dimension to between (-1, 1).

[0010] As an optional implementation, the weight distribution module fits the known tunneling parameters and geological parameters to obtain four functions respectively representing the relationship between the four geological parameters and the tunneling parameters, and combines the four functions to establish a final function model. The function model is embedded in the recognition sub-model to enhance the perception of the geological and mechanical characteristics of the grouting rock mass.

[0011] As an optional implementation, the five sub-models input thrust and penetration, torque, and tunneling speed to obtain output features including rock mass thickness, inclination angle, compressive strength and permeability coefficient.

[0012] As an optional implementation, when the five sub-models are optimized by the particle swarm optimization algorithm, historical data is used for training, and after verification, the trained model is used to identify and predict the surrounding rock grade.

[0013] As an optional implementation, when the five sub-models are optimized by the particle swarm optimization algorithm, the particles update their positions and velocities by tracking individual historical optimum and global optimum, and the core parameters include learning factor and inertia weight.

[0014] The working method of the surrounding rock grouting rock mass perception system comprises the following steps: The data acquisition module acquires real-time tunneling parameters and geological parameter data in the drilling process; The data correction module removes abnormal values in the acquired data through a clustering algorithm, and the data conversion module normalizes the tunneling parameter data through a Z-Score method; The weight distribution module determines the weights of the tunneling parameters and the geological parameters in the final analysis model by using a function model; The extreme learning machine sub-model, the back propagation artificial neural network sub-model, the radial basis function artificial neural network sub-model, the random forest sub-model and the K-nearest neighbor algorithm sub-model are all optimized by using a particle swarm optimization algorithm, and based on the tunneling parameters, the mechanical characteristics of the grouting stone body are perceived; The multi-modal fusion module fuses the feature results perceived by the five optimized sub-models through a D-S evidence theory to obtain the final surrounding rock grouting stone body feature.

[0015] Compared with the prior art, the present application has the following advantages: (1) The present application overcomes the shortcomings of too many required parameters and strong subjectivity in stone body perception, and does not have the black box problem in artificial intelligence methods, which is convenient for workers to understand and apply.

[0016] (2) The present application applies a DBSCAN clustering algorithm to abnormal value processing, and uses a relatively advanced PSO optimization algorithm to optimize the selection of the hyperparameters of the sub-model.

[0017] (3) The present application improves the traditional D-S evidence theory in surrounding rock grade recognition, and uses a multi-model fusion recognition prediction model, which has good robustness and overcomes the shortcomings of low accuracy and one-sidedness of a single model.

[0018] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are used for explanation. BRIEF DESCRIPTION OF DRAWINGS

[0019] The drawings accompanying the specification of the present application serve to provide a further understanding of the present application, and the schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application.

[0020] Figure 1 is a surrounding rock grouting stone body perception system framework based on advanced drilling drilling testing. DETAILED DESCRIPTION

[0021] The present application will be further described below in combination with the drawings and embodiments.

[0022] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0025] Example 1 A rock mass sensing system based on advanced drilling and testing, such as Figure 1 As shown, it includes a data acquisition module, a preprocessing module, a weight allocation module, a feature screening module, and a multimodal fusion module.

[0026] in: The data acquisition module is used to acquire tunneling parameters such as rotational speed, torque, and pressure during real-time monitoring of the drilling process.

[0027] Geological parameters include uniaxial compressive strength, groundwater aquifer, initial geostress, and structural surface condition; tunneling parameters include thrust, penetration depth, torque, and tunneling speed.

[0028] The preprocessing module includes a data correction module and a data transformation module. The data correction module uses the DBSCAN clustering algorithm to remove outliers from the acquired data. Its principle is to find core objects (data points whose neighborhood contains at least...). n The data points (referred to as core objects) are connected to form different dense regions, and isolated data points not belonging to any cluster are considered outliers. The data transformation module uses the Z-Score method to uniformly scale all data to the range (-1, 1) to eliminate the influence of different dimensions between data points. The formula is:

[0029] in, The mean, The standard deviation is denoted as .

[0030] The weight allocation module constructs a function model for perceiving stone body parameters during drilling testing, as follows: First, the fitness value Y is defined:

[0031] Wherein X1, X2, X3, X4 respectively refer to the thrust, penetration, torque and digging speed, the adaptability of the directional drilling during the while drilling test process can be well represented by the three parameters, wherein Respectively refer to the weight of the four parameters, the weight is determined by principal component analysis.

[0032] The principal component analysis method determines the weight, which can be assisted by SPSS software, and the variance explanation rate table and loading coefficient table are obtained through the principal component analysis function in the software, and the data in the table is used for weight calculation: First, calculate the linear combination coefficient matrix, the formula is the loading coefficient divided by the corresponding square root.

[0033] Second, calculate the comprehensive score coefficient, the formula is cumulative (linear combination coefficient * variance explanation rate) / cumulative variance explanation rate.

[0034] Third, determine the weight, sum the comprehensive score coefficient and normalize to obtain the weight value of each index.

[0035] The weight distribution module adopts uniaxial compressive strength, groundwater water content state, initial ground stress state, structure surface state Y1, Y2, Y3, Y4 four parameters respectively to adapt to the value Y Function fitting, four function forms are obtained, four function forms are combined to establish the final function model. The function model is embedded in the identification sub-model.

[0036] The feature preliminary screening module includes five basic models of ELM, BP, RBF, RF and KNN.

[0037] Extreme learning machine (ELM) is a single hidden layer feedforward neural network architecture, and its core feature lies in the random initialization mechanism of the hidden layer parameters. The network is composed of an input layer, a nonlinear hidden layer and an output layer, and its breakthrough design is that the weight vector and bias value of the hidden layer neurons are generated in a random way, without iterative optimization through the back propagation algorithm. The output layer weight is directly calculated by an analytical method (such as least squares method). This non-iterative training method significantly improves the learning speed, but the model performance is sensitive to the distribution of the initial random parameters, and unreasonable initialization may lead to poor prediction stability.

[0038] Backpropagation (BP) is the fundamental architecture in the field of deep learning, adopting a multi-layer perceptron structure consisting of an input layer, multiple hidden layers, and an output layer. Its core learning mechanism is based on the chain rule of differentiation, calculating the predicted output through forward propagation, backpropagating the error gradient, and adjusting the connection weights and bias values between neurons layer by layer using gradient descent. Despite the need to optimize a large number of parameters, BP networks and their variants (such as CNN and RNN) have become standard tools in image recognition, speech processing, and other fields due to their powerful non-linear fitting capabilities, driving the deep learning revolution.

[0039] Random Forest (RF) is a representative algorithm in the field of ensemble learning, which improves the model's generalization ability by constructing a combination of multiple decision trees. This algorithm introduces double randomness: at the sample level, multiple training subsets are generated through Bootstrap sampling; at the feature level, a subset of features is randomly selected for evaluation at each node split. The final prediction result is integrated through voting (classification) or averaging (regression). Key hyperparameters include the number of decision trees (n_estimators) and the number of node split features (max_features), which directly affect the model's complexity and anti-overfitting ability.

[0040] Radial Basis Function (RBF) is a three-layer feedforward network, whose core design is to use radial basis functions (such as Gaussian functions) as activation functions in the hidden layer. Input signals are non-linearly mapped from low-dimensional to high-dimensional space through hidden layer nodes, and the output layer performs linear weighted combination of high-dimensional features. This network has global approximation characteristics and can theoretically approximate continuous functions with arbitrary precision. Key parameters include radial basis function centers (usually determined by clustering), width parameters (σ values), and output layer weights, which together determine the network's function approximation ability and classification boundary complexity.

[0041] K-Nearest Neighbors (KNN) is an instance-based supervised learning algorithm suitable for classification and regression tasks. Its core idea is "birds of a feather flock together": for a new sample, calculate its distance (such as Euclidean distance) from all samples in the training set, select the nearest K neighbors, and make predictions through majority voting (classification) or averaging (regression). The value of K needs to be manually set, with smaller values being sensitive to noise and larger values potentially blurring class boundaries. The algorithm does not require explicit training, but has high computational overhead during prediction, making it suitable for small-scale data and sensitive to feature scaling, requiring normalization. Its advantages lie in simplicity and intuitiveness, while its disadvantages include distance invalidity in high-dimensional data and low efficiency.

[0042] The hyperparameter settings of the above five basic models have a great impact on performance, so an optimization algorithm is needed to find the optimal hyperparameters. PSO (Particle Swarm Optimization) is a swarm intelligence optimization method inspired by bird flocking. Particles update their positions and velocities by tracking individual historical best and global best. The core parameters include learning factor (c1 / c2) and inertia weight (w), which have global search ability and fast convergence. It is suitable for high-dimensional continuous problems such as function optimization and parameter tuning, but it may fall into local optimum and is sensitive to parameters in high-dimensional scenarios.

[0043] Therefore, PSO-SVM, PSO-BP, PSO-RBF, PSO-RF, and PSO-KNN are obtained. Five initial identification sub-models are obtained.

[0044] The feature preliminary screening module, based on known excavation parameters and geological parameters, inputs thrust, penetration, torque, and excavation speed into the five sub-models after training, and outputs the grouting rock mass characteristic parameters (rock mass thickness, inclination angle, compressive strength, and permeability coefficient) predicted by the model. Global optimization algorithm is used to find the optimal hyperparameters. 80% of the historical data is used for training, and 20% is used for testing. The trained model is used to identify and predict the surrounding rock grade.

[0045] The traditional D-S evidence theory method in the multi-modal fusion module has the following characteristics: D-S evidence theory mainly consists of identification framework, basic probability distribution function, and evidence combination rule.

[0046] First, identify the framework Θ={ θ 1, θ 2,..., θ n}, θ j is a subset of the identification framework Θ, 2 Θ ={A|A⊆Θ} is its power set, and if the mapping m :2 Θ → [0,1] satisfies the following conditions:

[0047] where Ø is the empty set, m is the basic probability function (BPA) of θ , m(A i ) is the support probability of a certain evidence in the identification framework for the subset A i , if m(A i ) >0, then AIt is a focal element of BPA.

[0048] The confidence function for BPA is:

[0049] Bel(A) This represents the sum of the probabilities of all subsets of an event and the probability that supports the event.

[0050] The likelihood function of BPA is:

[0051] Pl(A) This represents the sum of probabilities that the intersection with the event is not empty.

[0052] Pl(A)>Bel(A) , representing the upper and lower limits of support for the event, respectively.

[0053] For two independent pieces of evidence within the same forensic framework, their BPAs are respectively m 1 and m 2. The focus elements are respectively A 1, ... A k and B 1, ... B k The combination rules are as follows:

[0054] Preferably, the multimodal fusion module defines the thickness, tilt angle, compressive strength, and permeability coefficient of the grouting rock body as Θ = (I, II, III, and IV), and selects PSO-SVM, PSO-BP, PSO-RBF, PSO-RF, and PSO-KNN models as fusion evidence. In Θ, the support probability set for each piece of evidence for the rock characteristic parameters is a basic probability assignment function (BPA). m i (i = 1, 2, ..., 5). The prediction of each feature parameter sample yields the DS evidence theory probability matrix, as shown in Table 1.

[0055] Table 1. Probability Matrix of D-S Evidence Theory

[0056] Then, the weights between the evidence are calculated according to the formula, and the recognition results are fused according to the fusion rule formula.

[0057] Weight calculation formula: Evidence distance (Jousselme distance): First, construct the compatibility matrix 𝐷, whose elements are:

[0058] Jousselme distance of any two evidences:

[0059]

[0060]

[0061]

[0062] The working method of the above system comprises the following steps: A, drilling ahead to test the required parameters; B, the abnormal data is removed through the data correction module; C, the data in all aspects is uniformly compressed to (-1, 1) through the data conversion module; D, the geological parameters and the tunneling parameters are input into the weight distribution model to obtain the final function model; E, during the drilling ahead while drilling, the tunneling parameters are input into the feature preliminary screening module, and the grouting stone features are preliminarily obtained through the five sub-models; F, based on the output result of the feature preliminary screening module, the identification results of the five sub-models are fused through the multi-modal fusion module, and finally the prediction result with higher accuracy is obtained.

[0063] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk memory, optical storage etc.) having computer-usable program code embodied therein. CD ROM

[0064] ​​​​​​The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0065] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0066] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 one or more flow or blocks

[0067] The above description is only preferred embodiment of the present application but not for limiting the present application. The present application can have various modifications and changes for those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made by those skilled in the art within the spirit and principle of the present application without paying creative labor should be included in the protection scope of the present application.

Claims

1. A surrounding rock grouting stone mass perception system based on advanced drilling while drilling testing, characterized in that, The application relates to a drilling rock mass grouting stone feature recognition method and device. The data acquisition module is used for acquiring real-time tunneling parameters and geological parameter data in a drilling process. The preprocessing module comprises a data correction module and a data conversion module, the data correction module is used for eliminating abnormal values in the acquired data through a clustering algorithm, and the data conversion module is used for standardizing the tunneling parameter data through a Z-Score method. The weight distribution module determines the weights of the tunneling parameters and the geological parameters in a final analysis model by using a function model, so as to construct the final analysis model. The feature preliminary screening module comprises an extreme learning machine submodel, a back propagation artificial neural network submodel, a radial basis function artificial neural network submodel, a random forest submodel and a K-nearest neighbor algorithm submodel, the five submodels are all optimized by using a particle swarm optimization algorithm, the known tunneling parameters and the geological parameters are used for training the models, the tunneling parameters are input into the trained models, unknown geological parameters are predicted, and the geological and mechanical features of the grouting stone are perceived. The multi-modal fusion module fuses the feature results perceived by the five submodels after optimization through a D-S evidence theory, and finally obtains the features of the surrounding rock grouting stone.

2. A system for sensing the body of rock grouted by the surrounding rock while drilling based on advance drilling testing according to claim 1, characterized in that, The geological parameters comprise uniaxial compressive strength, initial ground stress state, underground water content state and structure surface state; and the tunneling parameters comprise thrust, penetration, torque and tunneling speed.

3. A system for sensing the body of rock grouted by the surrounding rock drilling and testing based on advanced drilling according to claim 1, characterized in that, The data correction module adopts a DBSCAN clustering algorithm to eliminate abnormal values in the acquired data.

4. The system for sensing the grouting body of the surrounding rock based on the advanced drilling and the while-drilling testing according to claim 1, characterized in that, The data conversion module adopts a Z-Score standardization method to uniformly scale the data in each dimension to (-1, 1).

5. A system for sensing the body of rock grouted by the immediate drilling and testing of the rock surrounding the borehole according to claim 1, characterized in that, The weight distribution module performs fitting processing on the known tunneling parameters and the geological parameters, respectively obtains function forms about the relationships between the four geological parameters and the tunneling parameters, combines the four function forms to establish a final function model, embeds the function model into the identification submodel, and strengthens the perception of the geological and mechanical features of the grouting stone.

6. A system for sensing the body of rock grouted by the immediate drilling while testing the rock based on the advanced drilling according to claim 1, characterized in that, The five submodels input the thrust and the penetration, the torque and the tunneling speed, and obtain output features containing stone thickness, inclination angle, compressive strength and permeability coefficient.

7. A system for sensing the body of rock grouted by the immediate drilling and testing of the rock surrounding the borehole according to claim 1, characterized in that, When the five submodels are optimized by using the particle swarm optimization algorithm, historical data are used for training, and after verification, the trained model is used for realizing the identification and prediction of the surrounding rock grade.

8. A system for sensing the body of rock grouted by the immediate drilling and testing of the rock surrounding the borehole according to claim 1, characterized in that, When the five submodels are optimized by using the particle swarm optimization algorithm, the particles update the position and speed by tracking the individual historical optimum and the global optimum, and the core parameters include a learning factor and an inertia weight.

9. A method of operating a surrounding rock grouting petromine system according to any of claims 1 - 8, characterized in that, The data acquisition module acquires real-time tunneling parameters and geological parameter data in a drilling process. The data correction module eliminates abnormal values in the acquired data through a clustering algorithm, and the data conversion module standardizes the tunneling parameter data through a Z-Score method. The weight distribution module determines the weights of the tunneling parameters and the geological parameters in a final analysis model by using a function model. ​ The limit learning machine sub-model, the back propagation artificial neural network sub-model, the radial basis function artificial neural network sub-model, the random forest sub-model and the K-nearest neighbor algorithm sub-model are all optimized in super parameters by using a particle swarm optimization algorithm, and based on the tunneling parameters, the mechanical characteristics of the grouting stone body are perceived; The multi-modal fusion module fuses the feature results perceived by the five sub-models after optimization through D-S evidence theory, and obtains the final surrounding rock grouting stone body characteristics.

Images