CN-121997187-A - Control method for stability of pressure regulating well rock mass supported by supporting while digging

CN121997187ACN 121997187 ACN121997187 ACN 121997187ACN-121997187-A

Abstract

The invention discloses a control method for rock mass stability of a surge shaft supported by a support during excavation, and relates to the technical field of intersection of hydraulic engineering and geotechnical engineering. The method comprises the steps of presetting distributed optical fiber sensors around an excavation working face to collect rock mass data to form a rock mass data set, fusing a random forest algorithm training stability judging model at a central control terminal through a parameter grid optimization method based on the historical data set, training a support scheme judging model through a LightGBM gradient lifting regression algorithm, inputting the rock mass data set into the stability judging model to output a current early warning grade, generating support parameters by the support scheme judging model based on the early warning grade and the rock mass data in combination with the stability grade, after excavation circulation is completed, bringing the data into a sample library to perform dual-mode incremental training, realizing model self-adaptive learning, ensuring continuous adaptation of the model along with geological condition changes, maintaining high judgment accuracy in the whole process, and providing powerful support for safety, high efficiency and economy of excavation construction of a pressure regulating well.

Inventors

ZHAO JIANBIN
WANG ZHONGSHENG
CHEN YANYAN
LI BAOJIANG
WANG LIZHENG
LUO FENG
GOU WEN
CAO RUI
LIU DONG
GUO YUXIANG
CHEN JINLONG

Assignees

中国水电建设集团十五工程局有限公司

Dates

Publication Date: 20260508
Application Date: 20251211

Claims (9)

1. The method for controlling the stability of the pressure regulating well rock mass supported by the support during excavation is characterized by comprising the following steps of: presetting distributed optical fiber sensors around the excavation working surface of the pressure regulating well, and synchronously collecting stress values, displacement amounts and crack development rates of rock mass in an excavation region to form a rock mass data set; Based on the historical data set, a random forest algorithm is fused at a central control terminal through a parameter grid optimization method to train a rock mass stability judging model, and a support scheme judging model is trained through LightGBM gradient lifting regression algorithm; based on a rock mass data set, outputting a current rock mass stability grade comprising a stability grade, a critical grade and a destabilization early-warning grade through a rock mass stability judging model, and starting a dynamic regulation and control mechanism when the critical grade or the destabilization early-warning grade is judged; Based on the current rock mass stability grade and rock mass data, generating optimal supporting parameters through a supporting scheme judging model, determining the primary supporting thickness of quick-setting sprayed concrete, the spacing of hollow grouting anchor rods and the grouting pressure of an anchor rod channel, controlling the opening of a two-stage pressure regulating valve set preset at the bottom of a pressure regulating well, and regulating the pressure of fluid in the well; After the excavation circulation is completed, the central control terminal brings the circulation data of the round into a model training sample library, and performs incremental training on the random forest classification model and the gradient lifting regression model until the excavation operation of the pressure regulating well is completed.
2. The method according to claim 1, wherein the preset distributed optical fiber sensor is specifically a preset sensor layout depth multiple Distributing a plurality of rings of distributed optical fiber sensors at equal angular intervals along the excavation outline of the pressure regulating well, synchronously collecting stress value, displacement and crack development rate of rock mass in an excavation area, wherein the distribution depth of each ring of sensors is the excavation radius The distributed optical fiber sensor collects rock strain data through a BOTDR technology and calculates a stress value, wherein the formula is as follows: ; Wherein, the As the stress value of the rock mass, Is the elastic modulus of the rock mass, The displacement is calculated by integrating the strain distribution curve of the optical fiber sensor, and the crack development rate is determined by the ratio of the stress change gradient and the time difference in the adjacent acquisition period.
3. The method according to claim 1, wherein the rock mass stability judgment model is trained by fusing a random forest algorithm through a parameter grid optimization method, specifically comprising the following steps: carrying out standardized processing on characteristic parameters and rock mass stress anisotropy coefficients in the historical data set, and constructing a characteristic set to be trained, wherein the formula is as follows: ; Wherein, the For the j-th characteristic parameter of the normalized i-th sample, The jth characteristic parameter of the original ith sample, Is the mean value of the j-th characteristic parameter, Standard deviation of the jth characteristic parameter; Optimizing key parameters of a random forest algorithm by adopting a parameter grid optimization method, wherein the key parameters comprise the number of decision trees, the maximum depth of the decision trees and the minimum number of samples of node splitting, and the optimal parameter combination is screened by taking the accuracy of 5-fold cross verification as an evaluation index; Based on the optimal parameter combination, constructing a random forest classification model, wherein the output of the model is the stability grade of the current rock mass, and the core influence characteristics are screened through the characteristic importance in the model training process, and the formula is as follows: ; Wherein, the The importance of the jth feature is scored, For the Gini coefficient reduction at the time of the j-th feature split in the k-th decision tree, Is the first Total number of features.
4. The method according to claim 2, wherein the rock mass stress anisotropy coefficient is, in particular, the maximum principal stress through the rock mass Minimum principal stress of rock mass Mean principal stress of rock mass Calculating stress anisotropy coefficient of rock mass The formula is: ; In the model construction process, a bag method is adopted to sample samples, the sampling proportion is a preset proportion of the total sample quantity, and the feature selection quantity of each decision tree is set as the total feature quantity Wherein Is the total number of features, and is controlled by the leaf node purity threshold.
5. The method of claim 1, wherein the LightGBM gradient lifting regression algorithm trains a support scheme decision model, specifically by introducing a coupling coefficient of rock mass stress and displacement as a key input feature, the formula is: ; Wherein, the As a result of the coupling coefficient, For the rock mass stress variation of adjacent acquisition cycles, For the rock mass displacement variation of adjacent acquisition cycles, Is the rock mass density; in the training process of the support scheme judgment model, a gradient descent method is adopted to minimize a loss function, and the formula is as follows: ; Wherein, the As a function of the loss, In order to train the number of samples, As the actual support parameter for the ith sample, And the model is trained by adopting characteristic box division processing, dividing continuous characteristics into a plurality of preset intervals, and improving the fitting capacity of the model to nonlinear relations.
6. The method of claim 1, wherein the dynamic regulation mechanism comprises a three-stage response strategy, specifically, a one-stage valve opening of a two-stage pressure regulating valve group is preset based on a pressure regulating well working condition 、、 And rock mass stress growth rate threshold ; When the critical level is determined, a first-stage response is started, and the opening degree of a first-stage valve of the two-stage pressure regulating valve group is regulated to be The second-stage valve is kept closed, and meanwhile, the primary support thickness of the quick setting type sprayed concrete is increased by a certain value on the basis of a model predicted value; when the instability early warning level is judged, and the stress growth rate of the rock mass is smaller than At the time, a secondary response is initiated, the opening degree of the primary valve is adjusted to The opening degree of the secondary valve is adjusted to The space between the hollow grouting anchor rods is reduced, and the grouting pressure is improved; when the instability early warning level is judged and the rock mass stress growth rate is not less than At this time, a three-stage response is initiated: the opening degree of the primary valve is adjusted to The opening degree of the secondary valve is adjusted to Suspending the excavation operation and performing advanced support, wherein the advanced support adopts pipe shed support.
7. The method according to claim 1, wherein the optimal support parameters are generated by a support scheme determination model, specifically, the primary support thickness of the rapid hardening type sprayed concrete is determined based on the rock mass crack development rate, and the formula is as follows: ; Wherein, the In order to correct the thickness of the primary support after correction, The model predicted primary support thickness is determined for the support plan, In order to correct the coefficient of the coefficient, Is the rate of development of rock mass fissures; Based on the pulling resistance requirement of the rock mass, the space of the hollow grouting anchor rods and the grouting pressure of the anchor rod channels are determined, and the formula is as follows: ; Wherein, the For the anti-pulling force of the anchor rod, Is the diameter of the anchor rod, For the anchoring length of the anchor rod, The bonding strength of the rock mass and the mortar is obtained; The grouting amount is calculated according to the porosity of the rock mass, and the calculation formula is as follows: ; Wherein, the Is the grouting amount of a single anchor rod, For the rock volume corresponding to the anchor rod anchoring section, In order to achieve a porosity of the rock mass, Is the filling coefficient of grouting.
8. The method of claim 1, wherein the two-stage pressure regulating valve set opening adjustment is coordinated with the well fluid pressure, and the pressure control formula is: ; Wherein, the For a target fluid pressure within the well, For the initial fluid pressure in the well, As a result of the pressure adjustment factor, The maximum stress value of the current rock mass; in the process of mutual coordination, the actual pressure in the well is collected in real time through the pressure sensor When (when) Presetting a pressure difference threshold in the well And automatically adjusting the opening of the two-stage pressure regulating valve group.
9. The method according to claim 1, wherein the incremental training is performed by presetting a threshold for starting a excavation cycle Per completion After the round of excavation circulation, a rock mass data set of the round of circulation, corresponding supporting scheme parameters and stability grade judging results form a new training sample, and an incremental learning algorithm is adopted to update parameters of a random forest classification model and a LightGBM gradient lifting regression model; wherein, the increment updating of the random forest model is performed by combining a strategy of reserving an optimal decision tree through adding a new decision tree, and presetting a decision tree importance threshold value Before feature importance scoring in historical training is reserved Newly added decision tree number is the initial decision tree number ; Incremental updating of LightGBM models adopts a hot start mode, a new model is initialized based on historical model parameters, and only a small number of rounds of iterative training are performed on the newly added samples.

Description

Control method for stability of pressure regulating well rock mass supported by supporting while digging Technical Field The invention relates to the technical field of intersection of hydraulic engineering and geotechnical engineering, in particular to a control method for stability of a pressure regulating well rock mass along with excavation and support. Background Along with the expansion of the hydraulic engineering of China to high-head, large-flow and complex geological condition areas, the construction requirement of a pressure regulating well in a water delivery system of a large hydropower station is continuously increased, the excavation scale is continuously enlarged, the geological environment is increasingly complex, and higher requirements are put forward on the stability control of rock mass in the excavation process of the pressure regulating well. The pressure regulating well is used as a core structure of a water delivery system of a hydropower station, and the excavation process of the pressure regulating well needs to face complex geological environments, wherein surrounding rock often presents multi-crack development and ground stress concentration characteristics, excavation disturbance easily causes rock mass displacement and crack expansion, and on the other hand, pressure fluctuation in the well can further aggravate rock mass mechanical property degradation and cause safety accidents such as collapse and water burst, however, the existing pressure regulating well rock mass control method still has the defect to be solved urgently. Chinese patent (publication No. CN 119146871A) discloses a surrounding rock deformation monitoring method based on three-dimensional point cloud, which obtains rock mass surface deformation data through laser scanning and gives an early warning, but does not construct the mapping relation between monitoring data and supporting parameters, an engineer is required to determine a supporting scheme after the early warning, the time required from the early warning to the implementation of average is more than 8 hours, and the timeliness requirement of 'supporting along with digging' cannot be met. Chinese patent (publication No. CN120260230 a) discloses a method for controlling stability of surrounding rock of high earthquake area pressure regulating chamber, optimizes supporting structural form through numerical simulation, but adopts fixed supporting parameters, and cannot be dynamically adjusted according to rock stress and crack development rate collected in real time in the process of excavation, so that insufficient supporting strength is easy to occur in V-type broken surrounding rock, or material waste is caused in II-type complete surrounding rock. The Chinese patent (publication No. CN 118542365A) discloses a surrounding rock grade prediction method based on LightGBM, although a machine learning algorithm is introduced, model training only depends on static data of historical engineering, excavation cycle data of the current engineering is not brought into a sample library for incremental training, so that the prediction accuracy of the model is reduced from 92% to 68% when geological conditions change, and the control accuracy cannot be continuously ensured. The existing control technology for the stability of the pressure regulating well rock mass still cannot realize the integrated control of real-time monitoring, dynamic modeling, accurate support and multi-factor cooperation, and a technical scheme capable of solving the defects is needed to be provided. Disclosure of Invention Based on the technical problems, the application discloses a control method for the stability of a surge shaft rock mass along with excavation and support, which comprises the following steps: presetting distributed optical fiber sensors around the excavation working surface of the pressure regulating well, and synchronously collecting stress values, displacement amounts and crack development rates of rock mass in an excavation region to form a rock mass data set; Based on the historical data set, a random forest algorithm is fused at a central control terminal through a parameter grid optimization method to train a rock mass stability judging model, and a support scheme judging model is trained through LightGBM gradient lifting regression algorithm; based on a rock mass data set, outputting a current rock mass stability grade comprising a stability grade, a critical grade and a destabilization early-warning grade through a rock mass stability judging model, and starting a dynamic regulation and control mechanism when the critical grade or the destabilization early-warning grade is judged; Based on the current rock mass stability grade and rock mass data, generating optimal supporting parameters through a supporting scheme judging model, determining the primary supporting thickness of quick-setting sprayed concrete, the spacing of hollow grouting anchor rods and the grou