Evaluation and analysis method of excavation support quality of underground powerhouse

Abstract

The present invention provides an underground powerhouse excavation and support quality evaluation and analysis method, which relates to the field of underground powerhouses and includes the following steps: S1, data collection and preprocessing: establishing a basic database, and collecting geological exploration data and on-site monitoring data of underground powerhouses to obtain geological parameter data, and preprocessing the geological parameter data, setting standard characteristic vectors as evaluation indicators; S2, model selection and integration strategy: selecting physical models and machine learning models based on the preprocessed geological parameter data. Through the coordination between the above structures, the following beneficial effects are achieved: First, the geostress calculation model that can integrate multiple factors can accurately grasp the geostress distribution in various parts of the underground powerhouse; second, the real-time monitoring system can be combined with big data and artificial intelligence algorithms to timely discover support hazards and dynamically optimize, effectively preventing surrounding rock instability; third, the multi-source data integration platform provides comprehensive and accurate information to assist decision makers in comprehensive analysis and avoid one-sided decision-making.

Description

Technical Field

[0001] The present invention relates to the field of underground powerhouses, and in particular to a method for evaluating and analyzing the excavation and support quality of underground powerhouses. Background Art

[0002] In the construction of underground power plants, the technical problems faced are complex, diverse and interrelated. The ground stress environment is complex and changeable. The magnitude and direction of ground stress in different regions and depths vary significantly. The ground stress in plate boundaries or areas with frequent geological tectonic activities is large and complex in direction. Stress concentration will occur in mountainous and canyon areas due to the influence of topography and landforms. The history of geological tectonic movements also causes uneven distribution of ground stress fields. The lithology and strength of surrounding rocks vary greatly. Different lithology surrounding rocks such as igneous rocks, sedimentary rocks, and metamorphic rocks have their own characteristics. Their strength is affected by internal factors such as mineral composition, particle structure, and degree of cementation, as well as external factors such as groundwater erosion. The support strength is closely related to the stability of the surrounding rock. When the support strength is insufficient, the weak surrounding rock is prone to plastic deformation under high ground stress, and the surrounding rock with developed joints and fissures will collapse, threatening construction safety and progress.

[0003] Due to the different ground stress environments of underground powerhouses and the different lithology and strength of the surrounding rocks, the support strength required to maintain the stability of the surrounding rocks also changes accordingly. Insufficient support strength may lead to local instability, collapse, or excessive deformation of the surrounding rocks, or even overall destruction. Excessive support strength will cause unnecessary waste. Summary of the Invention

[0004] In view of the shortcomings of the above-mentioned existing technologies, the technical problem to be solved by the present invention is to provide a method for evaluating and analyzing the quality of excavation support of underground powerhouses, which can realize accurate excavation support design, ensure that the support structure can effectively maintain the stability of the surrounding rock, and prevent the surrounding rock from becoming unstable, collapsing or excessively deforming.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is as follows: the present invention provides an underground powerhouse excavation support quality evaluation and analysis method, comprising the following steps:

[0006] S1. Data collection and preprocessing: Establish a basic database and collect geological exploration data and field monitoring data of underground powerhouses to obtain geological parameter data , and geological parameter data Perform preprocessing and set the standard feature vector as the evaluation index;

[0007] S2. Model selection and integration strategy: Based on the pre-processed geological parameter data Select physical models and machine learning models, and determine the corresponding integration strategy based on the selected physical models and machine learning models;

[0008] S3. Model training and iteration: Train the selected physical model and machine learning model, and alternately train the deep learning model and the physical model. Use the prediction results of the deep learning model to optimize the parameters of the physical model, and use the output results of the physical model to guide the training of the deep learning model.

[0009] S4. Integrated prediction and performance verification: Integrate the prediction results of the trained deep learning model and the physical model to obtain an integrated model, and use actual engineering data to verify the prediction performance of the integrated model.

[0010] In the preferred solution, the following steps are also included:

[0011] S5. Application and Continuous Optimization: Deploy the trained deep learning model, physical model, and integrated model into the actual engineering environment, so that it can receive real-time geological parameter data from the engineering site. , and perform prediction and analysis. During the deployment process, the model is seamlessly integrated with the data acquisition system and monitoring system of the engineering site to obtain the required data and output the prediction results. Based on the prediction results and the set standard feature vectors, the prediction performance of the integrated model is evaluated to obtain the performance evaluation results. Based on the performance evaluation results, necessary adjustments and optimizations are made to the integrated model, deep learning model, and physical model.

[0012] The deep learning model is deployed on the server at the engineering site, and the geological parameter data collected by the sensor is received in real time through the network connection. ,The physical model can be adjusted according to the actual situation on site and then perform calculation and analysis. The two work together to provide decision-making for engineering construction;

[0013] Continue to collect new prediction results, organize them and store them in the basic database, repeat steps S1 to S5, and use the new data to update and optimize the deep learning model and physical model;

[0014] In a preferred solution, in step S1, the stratum lithology, geological structure and ground stress conditions are determined based on geological survey data, and on-site monitoring data are collected by sensors to obtain construction dynamics;

[0015] Among them, the on-site monitoring data includes surrounding rock deformation, stress changes and support structure stress data;

[0016] Preprocessing includes geological parameter data Cleaning and conversion, which is used to organize data to remove duplicate, erroneous or incomplete records, convert non-numeric data into a form suitable for analysis, and standardize data from different sources, including unifying physical units and normalizing numerical ranges;

[0017] Set the standard eigenvector as the evaluation index, including the anchor support strength index of the standard parameters The range is 1.17 to 1.24, the standard parameter anchor support strength index ranges from 1.55 to 1.7;

[0018] in, and It is a range value under ideal conditions determined based on a large amount of engineering experience and theoretical analysis, and is used to measure the difference between actual engineering and standard conditions.

[0019] In the preferred solution, in step S2, the physical model adopts a physical model based on finite element analysis to simulate the physical behavior of anchor rods and anchor cables under extremely high ground stress, and the geological parameter data of the underground powerhouse are used to simulate the physical behavior of anchor rods and anchor cables under extremely high ground stress. Set model parameters and define boundary conditions in the physical model based on finite element analysis to simulate the constraints in actual engineering;

[0020] Build a deep learning model and choose convolutional neural network as the deep learning architecture;

[0021] Among them, the convolution layer of the convolutional neural network can be used to automatically extract local features in the data from the physical model based on finite element analysis, the pooling layer is used to reduce the data dimension, and the fully connected layer is used to integrate the features and make the final prediction. The key features are input into the deep learning model, including stress distribution characteristic diagram stress and deformation trend data deformation, from the calculation results of the physical model based on finite element analysis. The key features are used as additional input of the deep learning model to achieve an organic combination of the physical model based on finite element analysis and the deep learning model, so that the deep learning model can learn the relevant features of the physical process simulated by the physical model, and enhance the deep learning model's prediction ability for complex geological and stress conditions.

[0022] In the preferred solution, in step S3, the specific training steps are as follows:

[0023] Using the collected and pre-processed geological parameter data Train a deep learning model to predict the support strength index of anchor rods and cables, and use the pre-processed geological parameter data As a dataset , and set the data set The total sample size is , divided into training sets according to the ratio of 7:2:1 , validation set and test set ,Right now , , , through the training set The initially constructed deep learning model is trained with multiple rounds of supervised training, and the validation set Monitor the training results and use the test set Evaluate model performance;

[0024] in, , , ;

[0025] Define the loss function: ;

[0026] in, is the actual support strength index in the training set, is the prediction function corresponding to the deep learning model, For the training set Middle Input geological parameter data A sample of The parameters of the deep learning model are obtained by inputting geological parameter data and parameters , output the corresponding predicted value , and calculate the predicted value With actual value The average of the sum of squared differences of quantized model training sets The prediction error on ;

[0027] Use the stochastic gradient descent algorithm to update the model parameters. For the convolution layer parameters and , and its update formula is:

[0028] ;

[0029] ;

[0030] in, is the convolution kernel weight, is the convolutional layer bias, is the learning rate, and They are loss functions Convolutional layer parameters and The gradient of the training set is calculated and multiplied by the learning rate to update the adjustment parameters so that the deep learning model The loss function value on gradually decreases;

[0031] Fully connected layer parameters and The update formula is as follows:

[0032] ;

[0033] ;

[0034] in, is the weight of the fully connected layer, is the bias of the fully connected layer;

[0035] During the training process, in the validation set To monitor model performance on the validation set, calculate the loss on the validation set:

[0036] ;

[0037] in, For the validation set Sample size, For the validation set The actual support strength index, For the validation set Middle Input geological parameter data sample;

[0038] When the validation set When the loss stops decreasing or starts to increase for several consecutive rounds, early stopping can be used to stop training to prevent overfitting and appropriately reduce the learning rate. Then continue training;

[0039] The physical model is based on finite element analysis, and its basic equations are based on the principles of continuum mechanics. The stress-strain relationship of the surrounding rock of the underground powerhouse follows the generalized Hooke's law.

[0040] Assume that the calculation result of the physical model based on finite element analysis is ;

[0041] in, Represents the input parameters of the physical model based on finite element analysis, including geometry, material properties, boundary conditions, Represents the parameters of the physical model based on finite element analysis, including material parameters and boundary condition parameters;

[0042] Use experimental data to train a physical model based on finite element analysis to obtain experimental results , the experimental results Comparison with the physical model calculation results based on finite element analysis In contrast, the error function defined by the physical model based on finite element analysis is The formula is as follows:

[0043] ;

[0044] in, To compare the number of data points, The first The strain value corresponding to the data point is Indicates the physical model calculated based on finite element analysis and The corresponding The strain value of each data point;

[0045] Adjust the material parameters and boundary condition parameters in the physical model based on finite element analysis according to the errors;

[0046] Alternately train the deep learning model and the physical model based on finite element analysis, and use the anchor and cable support strength index results predicted by the deep learning model 、 As a reference input in physical models based on finite element analysis;

[0047] Assume that the equivalent stiffness adjustment formula of the updated anchor rod and anchor cable in the physical model based on finite element analysis is:

[0048] ;

[0049] ;

[0050] in, and are the equivalent stiffness of anchor rods before and after updating, and are the equivalent stiffness of the anchor cable before and after the update, is the adjustment factor, used to control the amplitude of the adjustment. is the anchor support strength index of standard parameters, is the anchor support strength index of standard parameters, and The function is to adjust the equivalent stiffness of anchor rods and cables in the physical model based on finite element analysis according to the difference between the anchor rod and cable support strength index predicted by the deep learning model and the anchor rod and cable support strength index of standard parameters;

[0051] Recalculate the physical model based on finite element analysis to obtain new physical model results, and feed the relevant features in the physical model results back to the deep learning model as a basis for adjusting the training data;

[0052] Assume that the physical model result fed back to the deep learning model is , then the training data of the deep learning model can be updated as:

[0053] ;

[0054] The deep learning model continues to be trained based on new training data, and the process continues to cycle.

[0055] In the preferred solution, in step S4, the prediction result of the deep learning model is , the prediction results of the physical model based on finite element analysis are , using the weighted average integration strategy, the integrated prediction results The formula is as follows:

[0056] ;

[0057] Weight coefficient and According to the deep learning model and the physical model based on finite element analysis in the validation set Performance determination on

[0058] Calculate the mean squared error of the deep learning model on the validation set and the mean square error of the physical model based on finite element analysis on the validation set ;

[0059] The weight coefficient calculation formula is:

[0060] ;

[0061] ;

[0062] Through weight distribution, the model with smaller mean square error can be given a larger weight to reduce the error and uncertainty of a single model;

[0063] The prediction performance of the ensemble model is verified using historical data, and the mean square error between the predicted value and the actual value is calculated. The formula is:

[0064] ;

[0065] in, To verify the amount of data, is the actual support strength index, The predicted values ​​for the ensemble model and;

[0066] The formula for calculating the mean absolute error between the predicted value and the actual value is:

[0067] ;

[0068] Draw a scatter plot of predicted values ​​and actual values. If the data points are closely distributed on the diagonal If the error is close to , it means that the prediction accuracy of the integrated model is high. The error between the prediction results of the integrated model and the prediction results of the individual deep learning model and the physical model is compared to evaluate the effectiveness of the integrated strategy. and , it means that the integrated prediction improves the prediction performance.

[0069] In the preferred solution, in step S5, when the integrated model is put into practical application, the integrated prediction results are Determine the bolt support strength index predicted by the integrated model and anchor support strength index ;

[0070] Calculate the bolt support strength index predicted by the integrated model and anchor support strength index Bolt support strength index with standard parameters and standard parameter anchor support strength index The Euclidean distance of

[0071] For the anchor support strength index, the Euclidean distance calculation formula is:

[0072] ;

[0073] Euclidean distance Formula used to quantify the bolt support strength index predicted by the integrated model Bolt support strength index with standard parameters the degree of deviation from the range;

[0074] For the anchor support strength index, the Euclidean distance calculation formula is:

[0075] ;

[0076] Euclidean distance Formula used to quantify the bolt support strength index predicted by the integrated model Bolt support strength index with standard parameters the degree of deviation from the scope;

[0077] According to the calculated Euclidean distance and Convert to confidence ;

[0078] Select values ​​based on the basic database to distinguish between obvious deviations and relatively close to the standard , and set is the preset maximum Euclidean distance threshold;

[0079] Based on the anchor support strength index The corresponding confidence , the calculation formula is expressed as:

[0080] ;

[0081] Based on the anchor support strength index The corresponding confidence , the calculation formula is expressed as:

[0082] ;

[0083] Among them, the confidence The interval range is from 0 to 1;

[0084] Confidence The closer it is to 1, the stronger the anchor support strength index predicted by the current anchor. Bolt support strength index with standard parameters If the deviation from the range is small, Anchor support strength index under standard parameters Range, conversely, confidence The closer it is to 0, the stronger the anchor support strength index predicted by the current anchor. Bolt support strength index with standard parameters The greater the deviation from the range, the Bolt support strength index that deviates from standard parameters scope;

[0085] Confidence The closer it is to 1, the stronger the anchor support strength index predicted by the current anchor cable. Anchor support strength index with standard parameters If the deviation from the range is small, Anchor support strength index under standard parameters Range, conversely, confidence The closer it is to 0, the stronger the anchor support strength index predicted by the current anchor. Anchor support strength index with standard parameters The greater the deviation from the range, the Anchor support strength index that deviates from standard parameters scope.

[0086] The present invention provides a method for evaluating and analyzing the quality of underground powerhouse excavation support. Through the coordination of the above structures, the following beneficial effects are achieved:

[0087] First, the geostress calculation model, which can integrate multiple factors, can accurately grasp the geostress distribution at various locations in the underground powerhouse, and accordingly determine the appropriate support strength to avoid improper support. At the same time, it can accurately measure the mechanical parameters of the surrounding rock and construct a personalized support design model. It can customize the optimal support scheme for different lithology surrounding rocks and improve the synergy between support and surrounding rock.

[0088] Second, the real-time monitoring system, combined with big data and artificial intelligence algorithms, can promptly detect support hazards and dynamically optimize them, effectively preventing surrounding rock instability. In addition, by combining reasonable support structures with waterproofing and drainage measures, it can resist natural erosion, maintain surrounding rock stability, and extend the service life of the project.

[0089] Third, the multi-source data integration platform provides comprehensive and accurate information to assist decision makers in comprehensive analysis and avoid one-sided decision-making. At the same time, it helps to compare different plans through quantitative evaluation models, quickly determine the optimal plan, improve decision-making efficiency and rationality, and ensure the smooth progress of the project. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The present invention will be further described below with reference to the accompanying drawings and examples:

[0091] Figure 1 It is a main structural diagram of the process of the present invention. DETAILED DESCRIPTION

[0092] In order to better understand the purpose, structure and function of the present invention, the embodiments and features in the embodiments of the present invention can be combined with each other without conflict. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0093] Example 1

[0094] like Figure 1 As shown, this embodiment provides a method for evaluating and analyzing the quality of underground powerhouse excavation support, which includes the following steps:

[0095] S1. Data collection and preprocessing: Establish a basic database and collect geological exploration data and field monitoring data of underground powerhouses to obtain geological parameter data , and geological parameter data Perform preprocessing and set the standard feature vector as the evaluation index;

[0096] In a preferred solution, in step S1, the stratum lithology, geological structure and ground stress conditions are determined based on geological survey data, and on-site monitoring data are collected by sensors to obtain construction dynamics;

[0097] Among them, the on-site monitoring data includes surrounding rock deformation, stress changes and support structure stress data;

[0098] Preprocessing includes geological parameter data Cleaning and conversion, which is used to organize data to remove duplicate, erroneous or incomplete records, convert non-numeric data into a form suitable for analysis, and standardize data from different sources, including unifying physical units and normalizing numerical ranges;

[0099] Set the standard eigenvector as the evaluation index, including the anchor support strength index of the standard parameters The range is 1.17 to 1.24, the standard parameter anchor support strength index ranges from 1.55 to 1.7;

[0100] in, and It is a range value under ideal conditions determined based on a large amount of engineering experience and theoretical analysis, and is used to measure the difference between actual engineering and standard conditions.

[0101] S2. Model selection and integration strategy: Based on the pre-processed geological parameter data Select physical models and machine learning models, and determine the corresponding integration strategy based on the selected physical models and machine learning models;

[0102] In the preferred solution, in step S2, the physical model adopts a physical model based on finite element analysis to simulate the physical behavior of anchor rods and anchor cables under extremely high ground stress, and the geological parameter data of the underground powerhouse are used to simulate the physical behavior of anchor rods and anchor cables under extremely high ground stress. Set model parameters and define boundary conditions in the physical model based on finite element analysis to simulate the constraints in actual engineering;

[0103] Build a deep learning model and choose convolutional neural network as the deep learning architecture;

[0104] Among them, the convolution layer of the convolutional neural network can be used to automatically extract local features in the data from the physical model based on finite element analysis, the pooling layer is used to reduce the data dimension, and the fully connected layer is used to integrate the features and make the final prediction. The key features are input into the deep learning model, including stress distribution characteristic diagram stress and deformation trend data deformation, from the calculation results of the physical model based on finite element analysis. The key features are used as additional input of the deep learning model to achieve an organic combination of the physical model based on finite element analysis and the deep learning model, so that the deep learning model can learn the relevant features of the physical process simulated by the physical model, and enhance the deep learning model's prediction ability for complex geological and stress conditions.

[0105] S3. Model training and iteration: Train the selected physical model and machine learning model, and alternately train the deep learning model and the physical model. Use the prediction results of the deep learning model to optimize the parameters of the physical model, and use the output results of the physical model to guide the training of the deep learning model.

[0106] In the preferred solution, in step S3, the specific training steps are as follows:

[0107] Using the collected and pre-processed geological parameter data Train a deep learning model to predict the support strength index of anchor rods and cables, and use the pre-processed geological parameter data As a dataset , and set the data set The total sample size is , divided into training sets according to the ratio of 7:2:1 , validation set and test set ,Right now , , , through the training set The initially constructed deep learning model is trained with multiple rounds of supervised training, and the validation set Monitor the training results and use the test set Evaluate model performance;

[0108] in, , , ;

[0109] Define the loss function: ;

[0110] in, is the actual support strength index in the training set, is the prediction function corresponding to the deep learning model, For the training set Middle Input geological parameter data A sample of The parameters of the deep learning model are obtained by inputting geological parameter data and parameters , output the corresponding predicted value , and calculate the predicted value With actual value The average of the sum of squared differences of quantized model training sets The prediction error on ;

[0111] Use the stochastic gradient descent algorithm to update the model parameters. For the convolution layer parameters and , and its update formula is:

[0112] ;

[0113] ;

[0114] in, is the convolution kernel weight, is the convolutional layer bias, is the learning rate, and They are loss functions Convolutional layer parameters and The gradient of the training set is calculated and multiplied by the learning rate to update the adjustment parameters so that the deep learning model The loss function value on gradually decreases;

[0115] Fully connected layer parameters and The update formula is as follows:

[0116] ;

[0117] ;

[0118] in, is the weight of the fully connected layer, is the bias of the fully connected layer;

[0119] During the training process, in the validation set To monitor model performance on the validation set, calculate the loss on the validation set:

[0120] ;

[0121] in, For the validation set Sample size, For the validation set The actual support strength index, For the validation set Middle Input geological parameter data sample;

[0122] When the validation set When the loss stops decreasing or starts to increase for several consecutive rounds, early stopping can be used to stop training to prevent overfitting and appropriately reduce the learning rate. Then continue training;

[0123] The physical model is based on finite element analysis, and its basic equations are based on the principles of continuum mechanics. The stress-strain relationship of the surrounding rock of the underground powerhouse follows the generalized Hooke's law.

[0124] Assume that the calculation result of the physical model based on finite element analysis is ;

[0125] in, Represents the input parameters of the physical model based on finite element analysis, including geometry, material properties, boundary conditions, Represents the parameters of the physical model based on finite element analysis, including material parameters and boundary condition parameters;

[0126] Use experimental data to train a physical model based on finite element analysis to obtain experimental results , the experimental results Comparison with the physical model calculation results based on finite element analysis In contrast, the error function defined by the physical model based on finite element analysis is The formula is as follows:

[0127] ;

[0128] in, To compare the number of data points, The first The strain value corresponding to the data point is Indicates the physical model calculated based on finite element analysis and The corresponding The strain value of each data point;

[0129] Adjust the material parameters and boundary condition parameters in the physical model based on finite element analysis according to the errors;

[0130] Alternately train the deep learning model and the physical model based on finite element analysis, and use the anchor and cable support strength index results predicted by the deep learning model 、 As a reference input in physical models based on finite element analysis;

[0131] Assume that the equivalent stiffness adjustment formula of the updated anchor rod and anchor cable in the physical model based on finite element analysis is:

[0132] ;

[0133] ;

[0134] in, and are the equivalent stiffness of anchor rods before and after updating, and are the equivalent stiffness of the anchor cable before and after the update, is the adjustment factor, used to control the adjustment amplitude. is the anchor support strength index of standard parameters, is the anchor support strength index of standard parameters, and The function is to adjust the equivalent stiffness of anchor rods and cables in the physical model based on finite element analysis according to the difference between the anchor rod and cable support strength index predicted by the deep learning model and the anchor rod and cable support strength index of standard parameters;

[0135] Recalculate the physical model based on finite element analysis to obtain new physical model results, and feed the relevant features in the physical model results back to the deep learning model as a basis for adjusting the training data;

[0136] Assume that the physical model result fed back to the deep learning model is , then the training data of the deep learning model can be updated as:

[0137] ;

[0138] The deep learning model continues to be trained based on new training data, and the process continues to cycle.

[0139] S4. Integrated prediction and performance verification: Integrate the prediction results of the trained deep learning model and the physical model to obtain an integrated model, and use actual engineering data to verify the prediction performance of the integrated model.

[0140] In the preferred solution, in step S4, the prediction result of the deep learning model is , the prediction results of the physical model based on finite element analysis are , using the weighted average integration strategy, the integrated prediction results The formula is as follows:

[0141] ;

[0142] Weight coefficient and According to the deep learning model and the physical model based on finite element analysis in the validation set Performance determination on

[0143] Calculate the mean squared error of the deep learning model on the validation set and the mean square error of the physical model based on finite element analysis on the validation set ;

[0144] The weight coefficient calculation formula is:

[0145] ;

[0146] ;

[0147] Through weight distribution, the model with smaller mean square error can be given a larger weight to reduce the error and uncertainty of a single model;

[0148] The prediction performance of the ensemble model is verified using historical data, and the mean square error between the predicted value and the actual value is calculated. The formula is:

[0149] ;

[0150] in, To verify the amount of data, is the actual support strength index, The predicted values ​​for the ensemble model and;

[0151] The formula for calculating the mean absolute error between the predicted value and the actual value is:

[0152] ;

[0153] Draw a scatter plot of predicted values ​​and actual values. If the data points are closely distributed on the diagonal If the error is close to , it means that the prediction accuracy of the integrated model is high. The error between the prediction results of the integrated model and those of the individual deep learning model and physical model is compared to evaluate the effectiveness of the integrated strategy. and , it means that the integrated prediction improves the prediction performance.

[0154] S5. Application and Continuous Optimization: Deploy the trained deep learning model, physical model, and integrated model into the actual engineering environment, so that it can receive real-time geological parameter data from the engineering site. , and perform prediction and analysis. During the deployment process, the model is seamlessly integrated with the data acquisition system and monitoring system of the engineering site to obtain the required data and output the prediction results. Based on the prediction results and the set standard feature vectors, the prediction performance of the integrated model is evaluated to obtain the performance evaluation results. Based on the performance evaluation results, necessary adjustments and optimizations are made to the integrated model, deep learning model, and physical model.

[0155] The deep learning model is deployed on the server at the engineering site, and the geological parameter data collected by the sensor is received in real time through the network connection. ,The physical model can be adjusted according to the actual situation on site and then perform calculation and analysis. The two work together to provide decision-making for engineering construction;

[0156] Continue to collect new prediction results, organize them and store them in the basic database, repeat steps S1 to S5, and use the new data to update and optimize the deep learning model and physical model;

[0157] In the preferred solution, in step S5, when the integrated model is put into practical application, the integrated prediction results are Determine the bolt support strength index predicted by the integrated model and anchor support strength index ;

[0158] Calculate the bolt support strength index predicted by the integrated model and anchor support strength index Bolt support strength index with standard parameters and standard parameter anchor support strength index The Euclidean distance of

[0159] For the anchor support strength index, the Euclidean distance calculation formula is:

[0160] ;

[0161] Euclidean distance Formula used to quantify the bolt support strength index predicted by the integrated model Bolt support strength index with standard parameters the degree of deviation from the scope;

[0162] For the anchor support strength index, the Euclidean distance calculation formula is:

[0163] ;

[0164] Euclidean distance Formula used to quantify the bolt support strength index predicted by the integrated model Bolt support strength index with standard parameters the degree of deviation from the scope;

[0165] According to the calculated Euclidean distance and Convert to confidence ;

[0166] Select values ​​based on the basic database to distinguish between obvious deviations and relatively close to the standard , and set is the preset maximum Euclidean distance threshold;

[0167] Based on the anchor support strength index The corresponding confidence , the calculation formula is expressed as:

[0168] ;

[0169] Based on the anchor support strength index The corresponding confidence , the calculation formula is expressed as:

[0170] ;

[0171] Among them, the confidence The interval range is from 0 to 1;

[0172] Confidence The closer it is to 1, the stronger the anchor support strength index predicted by the current anchor. Bolt support strength index with standard parameters If the deviation from the range is small, Anchor support strength index under standard parameters Range, conversely, confidence The closer it is to 0, the stronger the anchor support strength index predicted by the current anchor. Bolt support strength index with standard parameters The greater the deviation from the range, the Bolt support strength index that deviates from standard parameters scope;

[0173] Confidence The closer it is to 1, the stronger the anchor support strength index predicted by the current anchor cable. Anchor support strength index with standard parameters If the deviation from the range is small, Anchor support strength index under standard parameters Range, conversely, confidence The closer it is to 0, the stronger the anchor support strength index predicted by the current anchor. Anchor support strength index with standard parameters The greater the deviation from the range, the Anchor support strength index that deviates from standard parameters scope.

[0174] Although the present application has been described with reference to specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present application. Accordingly, this specification and the drawings are merely illustrative of the present application as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, the present application is intended to include such modifications and variations if they fall within the scope of the claims of the present application and their equivalents.

[0175] It should also be noted that the terms "first," "second," and the like in the specification and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that the terms used in this manner are interchangeable where appropriate to facilitate the description of the embodiments of the present invention.

Claims

1. A method for evaluating and analyzing the quality of underground powerhouse excavation support, characterized in that: The following steps are involved: S1. Data collection and preprocessing: Establish a basic database and collect geological exploration data and field monitoring data of the underground powerhouse to obtain geological parameter data x, preprocess the geological parameter data x, and set the standard characteristic vector as the evaluation index; S2. Model selection and integration strategy: Select a physical model and a machine learning model based on the preprocessed geological parameter data x, and determine the corresponding integration strategy based on the selected physical model and machine learning model; S3. Model training and iteration: Train the selected physical model and machine learning model, and alternately train the deep learning model and the physical model. Use the prediction results of the deep learning model to optimize the parameters of the physical model, and use the output results of the physical model to guide the training of the deep learning model. The collected and pre-processed geological parameter data x is used to train the deep learning model to predict the support strength index of anchor rods and cables. The pre-processed geological parameter data x is used as the data set T, and the total number of samples in the data set T is p. It is divided into the training set T and the training set T according to the ratio of 7:2:

1. t , validation set T v and the test set T e ; Update the parameters of the deep learning model using the stochastic gradient descent algorithm; The physical model is based on finite element analysis, and its basic equations are based on the principles of continuum mechanics. The stress-strain relationship of the surrounding rock of the underground powerhouse follows the generalized Hooke's law. Assume that the calculation result of the physical model based on finite element analysis is y FEA (x FEA ,θ FEA ); Among them, x FEA represents the input parameters of the physical model based on finite element analysis, including geometry, material properties, boundary conditions, θ FEA Represents the parameters of the physical model based on finite element analysis, including material parameters and boundary condition parameters; Use experimental data to train a physical model based on finite element analysis to obtain experimental results The experimental results Comparison with the physical model calculation results based on finite element analysis In contrast, the error function E defined by the physical model based on finite element analysis is FEA The formula is as follows: Among them, S is the number of comparison data points, represents the strain value corresponding to the sth data point obtained from the experiment, Indicates the physical model calculated based on finite element analysis and The corresponding strain value of the sth data point; Adjust the material parameters and boundary condition parameters in the physical model based on finite element analysis according to the errors; Alternately train the deep learning model and the physical model based on finite element analysis, and use the anchor and cable support strength index results predicted by the deep learning model As a reference input in physical models based on finite element analysis; Assume that the equivalent stiffness adjustment formula of the updated anchor rod and anchor cable in the physical model based on finite element analysis is: in, and are the equivalent stiffness of anchor rods before and after updating, and are the equivalent stiffness of the anchor cable before and after the update, γ is the adjustment factor used to control the amplitude of the adjustment, I 标准锚杆 The anchor support strength index of the standard parameter, I 标准锚索 is the anchor support strength index of standard parameters, and The function is to adjust the equivalent stiffness of anchor rods and cables in the physical model based on finite element analysis according to the difference between the anchor rod and cable support strength index predicted by the deep learning model and the anchor rod and cable support strength index of standard parameters; Recalculate the physical model based on finite element analysis to obtain new physical model results, and feed the relevant features in the physical model results back to the deep learning model as a basis for adjusting the training data; Assume that the physical model result fed back to the deep learning model is x f , then the training data of the deep learning model can be updated as: Continue training the deep learning model based on new training data, and repeat this process; S4. Integrated prediction and performance verification: Integrate the prediction results of the trained deep learning model and the physical model to obtain an integrated model, and use actual engineering data to verify the prediction performance of the integrated model.

2. The underground powerhouse excavation support quality evaluation and analysis method according to claim 1 is characterized in that: The following steps are also included: S5. Application and Continuous Optimization: Deploy the trained deep learning model, physical model, and integrated model into the actual engineering environment, allowing them to receive real-time geological parameter data x from the engineering site and perform predictions and analysis. During deployment, seamlessly integrate the model with the engineering site's data acquisition and monitoring systems to obtain the required data and output prediction results. Based on the prediction results and the set standard feature vectors, evaluate the prediction performance of the integrated model to obtain performance evaluation results. Based on the performance evaluation results, make necessary adjustments and optimizations to the integrated model, deep learning model, and physical model. The deep learning model is deployed on a server at the construction site, which receives real-time geological parameter data x collected by sensors through a network connection. The physical model can then adjust parameters based on the actual situation on site and perform calculations and analysis. The two work together to provide decision-making for engineering construction. Continue to collect new prediction results, organize them and store them in the basic database, repeat steps S1 to S5, and use the new data to update and optimize the deep learning model and physical model.

3. The underground powerhouse excavation support quality evaluation and analysis method according to claim 2 is characterized in that: In step S1, the stratum lithology, geological structure and ground stress conditions are determined based on geological survey data, and on-site monitoring data is collected through sensors to obtain construction dynamics; Among them, the on-site monitoring data includes surrounding rock deformation, stress changes and support structure stress data; Preprocessing includes cleaning and converting geological parameter data x, which is used to organize data to remove duplicate, erroneous or incomplete records, convert non-numerical data into a form suitable for analysis, and standardize data from different sources, including unifying physical quantity units and normalizing numerical ranges; Set the standard characteristic vector as the evaluation index, including the anchor support strength index I of the standard parameters 标准锚杆 The range is 1.17 to 1.24, the standard parameter anchor support strength index I 标准锚索 ranges from 1.55 to 1.7; Among them, I 标准锚杆 and I 标准锚索 It is a range value under ideal conditions determined based on a large amount of engineering experience and theoretical analysis, and is used to measure the difference between actual engineering and standard conditions.

4. The underground powerhouse excavation support quality evaluation and analysis method according to claim 3 is characterized in that: In step S2, a physical model based on finite element analysis is used to simulate the physical behavior of anchor rods and cables under extremely high ground stress. Model parameters are set according to the geological parameter data x of the underground powerhouse, and boundary conditions are defined in the physical model based on finite element analysis to simulate the constraints in actual engineering. Build a deep learning model and choose convolutional neural network as the deep learning architecture; Among them, the convolutional layer of the convolutional neural network can be used to automatically extract local features in the data from the physical model based on finite element analysis, the pooling layer is used to reduce the data dimension, and the fully connected layer is used to integrate the features and make the final prediction. The preprocessed geological parameter data x is input into the deep learning model, and the key features are extracted from the calculation results of the physical model based on finite element analysis, including stress distribution characteristic map stress and deformation trend data deformation. The key features are used as additional inputs of the deep learning model to achieve an organic combination of the physical model based on finite element analysis and the deep learning model, so that the deep learning model can learn the relevant features of the physical process simulated by the physical model, and enhance the deep learning model's prediction ability for complex geological and stress conditions.

5. The underground powerhouse excavation support quality evaluation and analysis method according to claim 4 is characterized in that: In step S3, the specific training steps are as follows: T t ={x i1 ,x i2 ,…,x in }, T v ={x j1 ,x j2 ,…,x jm }), T e ={x k1 ,x k2 ,…,x kl }, through the training set T t The initially constructed deep learning model is trained with multiple rounds of supervised training, and the validation set T v Monitor the training effect and use the test set T e Evaluate model performance; Where n = 0.7p, m = 0.2p, l = 0.1p; Define the loss function: Among them, y i is the actual support strength index in the training set, f DL is the prediction function corresponding to the deep learning model, x i is the training set T t The sample of geological parameter data x of the i-th input, θ DL is the parameter of the deep learning model, which is obtained by inputting geological parameter data x and parameter θ DL , output the corresponding predicted value f DL (x i ,θ DL ), and calculate the predicted value f DL (x i ,θ DL ) and the actual value y i The average of the sum of squared differences of quantized model training set T t The prediction error on ; For the convolutional layer parameter w k and b k , and its update formula is: Among them, w k is the convolution kernel weight, b k is the convolution layer bias, α is the learning rate, and They are the loss functions L DL For the convolutional layer parameters w k and b k The gradient of the training set T is calculated and multiplied by the learning rate to update the adjustment parameters so that the deep learning model t The loss function value on gradually decreases; Fully connected layer parameter w f and b f The update formula is as follows: Among them, w f is the weight of the fully connected layer, b f is the bias of the fully connected layer; During the training process, on the validation set T v To monitor model performance on the validation set, calculate the loss on the validation set: Among them, m is the validation set T v Sample size, y j is the validation set T v Actual support strength index, x j is the validation set T v The j-th input geological parameter data x sample; When the validation set T v When the loss stops decreasing or starts to increase for several consecutive rounds, early stopping can be used to stop training to prevent overfitting, and training can be continued after appropriately reducing the learning rate α.

6. The underground powerhouse excavation support quality evaluation and analysis method according to claim 5 is characterized in that: In step S4, let the prediction result of the deep learning model be y DL , the physical model prediction result based on finite element analysis is y FEA , using the weighted average integration strategy, the integrated prediction result y 集成 The formula is as follows: and 集成 =w1y DL +w2y FEA ; The weight coefficients w1 and w2 are calculated based on the deep learning model and the physical model based on finite element analysis on the validation set T v Performance determination on Calculate the mean squared error of the deep learning model on the validation set and the mean square error of the physical model based on finite element analysis on the validation set The weight coefficient calculation formula is: Through weight distribution, the model with smaller mean square error can be given a larger weight to reduce the error and uncertainty of a single model; The prediction performance of the ensemble model is verified using historical data, and the mean square error between the predicted value and the actual value is calculated. The formula is: Among them, q is the number of verification data, is the actual support strength index, The predicted values ​​for the ensemble model and; The formula for calculating the mean absolute error between the predicted value and the actual value is: Draw a scatter plot of the predicted values ​​and the actual values. If the data points are closely distributed near the diagonal y=x, it means that the prediction accuracy of the integrated model is high. Compare the errors of the prediction results of the integrated model with those of the individual deep learning models and physical models to evaluate the effectiveness of the integrated strategy. and ), it means that ensemble prediction improves the prediction performance.

7. The underground powerhouse excavation support quality evaluation and analysis method according to claim 6 is characterized in that: In step S5, when the integrated model is put into practical application, the integrated prediction result y 集成 Determine the anchor support strength index I predicted by the integrated model 锚杆预测 and anchor support strength index I 锚索预测 ; Calculate the anchor support strength index I predicted by the integrated model 锚杆预测 and anchor support strength index I 锚索预测 Compared with the standard parameter anchor support strength index I 标准锚杆 and standard parameter anchor support strength index I 标准锚索 The Euclidean distance of For the anchor support strength index, the Euclidean distance calculation formula is: Euclidean distance d 锚杆 The formula is used to quantify the bolt support strength index I predicted by the integrated model 锚杆预测 Compared with the standard parameter anchor support strength index I 标准锚杆 the degree of deviation from the scope; For the anchor support strength index, the Euclidean distance calculation formula is: Euclidean distance d 锚索 The formula is used to quantify the bolt support strength index I predicted by the integrated model 锚杆预测 Compared with the standard parameter anchor support strength index I 标索锚杆 the degree of deviation from the scope; According to the calculated Euclidean distance d 锚杆 and d 锚索 Convert to confidence C; Based on the basic database, the value d is selected to distinguish between obvious deviations and relatively close to the standard max , and set d max is the preset maximum Euclidean distance threshold; Based on the anchor support strength index I 锚杆预测 The corresponding confidence C 锚杆 , the calculation formula is expressed as: Based on the anchor support strength index I 锚索预测 The corresponding confidence C 锚索 , the calculation formula is expressed as: Among them, the confidence C 锚杆 The interval range is from 0 to 1; Confidence C 锚杆 The closer it is to 1, the stronger the anchor support strength index I is. 锚杆预测 Compared with the standard parameter anchor support strength index I 标准锚杆 If the deviation from the range is small, I 锚杆预测 Anchor support strength index I under standard parameters 标准锚杆 Range, conversely, confidence C 锚杆 The closer it is to 0, the stronger the anchor support strength index I is. 锚杆预测 Compared with the standard parameter anchor support strength index I 标准锚杆 If the deviation from the range is large, I 锚杆预测 Anchor support strength index I that deviates from standard parameters 标准锚杆 scope; Confidence C 锚索 The closer it is to 1, the stronger the anchor support strength index I is predicted to be. 锚索预测 Compared with the standard parameter anchor support strength index I 标准锚索 If the deviation from the range is small, I 锚索预测 Anchor support strength index I under standard parameters 标准锚索 Range, conversely, confidence C 锚索 The closer it is to 0, the stronger the anchor support strength index I is. 锚索预测 Compared with the standard parameter anchor support strength index I 标准锚索 If the deviation from the range is large, I 锚索预测 Anchor support strength index I that deviates from standard parameters 标准锚索 scope.

Images