A method for judging the quality of surrounding rock of shield tunnel based on monitoring video

Abstract

The present invention provides a method for judging the surrounding rock quality of a shield tunnel based on monitoring video. The method uses a shield thrust index FPI and a cutter head torque index TPI as evaluation indicators of geological quality parameters, constructs a neural network model based on image features and address parameters of monitoring videos, and optimizes the model according to the rating indicators using a genetic algorithm to finally obtain a surrounding rock quality judgment model. Then, the grayscale feature statistics of a collected slag image are input into the surrounding rock quality judgment model to finally obtain a judgment result of the surrounding rock quality. The method respectively corresponds the grayscale feature statistics of the slag image with the surrounding rock quality parameters, namely FPI and TPI, as the input and output of the surrounding rock quality judgment model, which not only solves the problem that the input parameters of the traditional rock-machine mapping model cannot be updated, but also avoids the problem that the tunneling parameters collected by the shield machine through a PLC data acquisition system cannot be reasonably used.

Description

Technical Field

[0001] The invention relates to a method for judging the quality of surrounding rock of a shield tunnel based on monitoring video. Background Art

[0002] With the rapid development of urbanization, the rational development and utilization of underground space has become an effective way to expand urban capacity and functions. Shield construction has been widely used in various underground space construction in my country due to its advantages such as small impact on the environment, high excavation efficiency and strong adaptability to the strata. Shield machines usually include earth pressure balance shield machines and slurry balance shield machines. Among them, slurry balance shield machines resist the water and soil pressure of the excavation face by generating appropriate pressure mud in the mud and water tank and forming a weakly permeable mud film. The geological conditions determine the difficulty of shield excavation and are also an important factor in the safe excavation of shields. Therefore, it is necessary to study the relationship between the data generated by shield construction and the geological conditions.

[0003] The geological information used in the construction of traditional rock-machine mapping models is mostly based on surrounding rock parameters, such as surrounding rock grade. However, the surrounding rock grade is mainly given based on the previous geological survey report. For slurry shield machines, it is generally impossible to obtain surrounding rock properties in real time through surrounding rock coring. Therefore, if the surrounding rock grade is used as the input of the rock-machine mapping model, on the one hand, the surrounding rock grade cannot be updated in real time. On the other hand, the surrounding rock grade is mainly determined by the integrity of the rock mass and the strength of the rock, and cannot directly reflect the rock-machine mapping relationship under different strata.

[0004] At present, shield machines can obtain a large number of historical excavation parameters generated by shield construction in real time through PLC data acquisition systems, but these excavation parameters have not been timely and reasonably analyzed and utilized through big data. The selection of shield excavation parameters still requires shield drivers to combine geological survey data and geological information revealed by slag discharge, and adjust according to driver experience and operating manuals. In theory, shield slag discharge has a corresponding relationship with specific stratum conditions, so a model of slag discharge image and geological quality can be established to judge the quality of the shield's surrounding rock. Summary of the invention

[0005] In order to solve the above problems, a method is provided to judge the quality of surrounding rock according to the video image of slag discharge. The present invention adopts the following technical solutions:

[0006] The invention provides a method for judging the quality of surrounding rock of a shield tunnel based on monitoring video, which is characterized by comprising the following steps: step S1, installing a video sensor at a slag outlet of a shield tunnel, and performing parameter calibration and image correction on the video sensor; step S2, collecting slag images within a predetermined time interval based on the video sensor; step S3, graying the slag images to obtain gray slag images, and extracting gray feature statistics of the gray slag images; step S4, determining the geological quality parameter evaluation index of the excavation position corresponding to the slag image; step S5, taking the gray feature statistics as an input parameter and the geological quality parameter evaluation index as an output parameter, and constructing a neural network model of image features and geological parameters; step S6, using a genetic algorithm to perform hyperparameter optimization on the neural network model of image features and geological parameters according to the evaluation index, generating an optimal neural network model of image features and geological parameters, and using it as a surrounding rock quality judgment model; step S7, inputting the slag monitoring image collected by the video sensor into the surrounding rock quality judgment model, and obtaining the judgment result of the surrounding rock quality of the excavation position corresponding to the slag monitoring image.

[0007] The present invention provides a method for judging the surrounding rock quality of a shield tunnel based on monitoring video, which may also have such technical features, wherein, in step S1, the video sensor is a solar camera, the parameter calibration includes calibrating the internal parameter matrix and the external parameter matrix of the solar camera, establishing a mapping relationship between the world coordinate system, the camera coordinate system and the image coordinate system, and the image correction is to correct the distorted image based on inverse transformation.

[0008] The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video provided by the present invention may also have such a technical feature, wherein, in step S3, the extraction of gray feature statistical values ​​includes statistical parameters characterizing a gray co-occurrence matrix, the matrix size of the slag image is set to M×N, f(x, y) is the gray value at the point (x, y) of the slag image matrix, and i=f(x, y), j=f(x+dx, y+dy), and the gray co-occurrence matrix is:

[0009] p(i,j|d,q)=#{[(x,y),(x+dx,y+dy)∈(M,N)×(M,N)|d,q,i,j]}

[0010] In the formula, (dx, dy) represents the offset of the point pair, the offset is determined by the distance and the offset direction, # represents the number of elements in the set, and the statistical parameters that characterize the gray-level co-occurrence matrix are energy, entropy, and autocorrelation. The energy is:

[0011] ASE=∑ i ∑ j p(i, j|d, θ) 2 ,

[0012] Entropy is:

[0013] ENT=-∑ i ∑ j p(i,j|d,θ)lgp(i,j|d,θ),

[0014] The autocorrelation is:

[0015]

[0016] The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video provided by the present invention may also have such a technical feature, wherein, in step S4, the geological quality parameter evaluation index is the shield thrust index FPI and the cutter head torque index TPI, and the calculation formula of the shield thrust index FPI is:

[0017]

[0018] Where F is the total thrust, p is the penetration,

[0019] The calculation formula of the cutter head torque index TPI is:

[0020]

[0021] Where T is the cutter head torque and p is the penetration rate.

[0022] The shield tunnel surrounding rock quality judgment method based on monitoring video provided by the present invention may also have such a technical feature, wherein step S6 includes the following sub-steps: step S6-1, within the test range, randomly selecting the hyperparameters to be optimized to form an initial population and using the initial population as the current population; step S6-2, assigning the hyperparameters of each individual in the current population to the neural network model of image features and geological parameters respectively, using the k-fold cross-validation method for verification, obtaining the evaluation index of the neural network model of image features and geological parameters corresponding to each individual, and evaluating the fitness of the individuals in the current population based on the evaluation index; step S6-3, from the current population In the group, individuals with fitness greater than 0 and in the top 25% of the fitness ranking are selected as high-quality parent individuals for crossover and mutation to generate new offspring individuals, and the new offspring individuals are combined with the high-quality parent individuals to form a new current population; step S6-4, repeating steps S6-2 to S6-3 until the number of iterations reaches a predetermined number, and obtaining an individual with the best evaluation index, and the individual is used as the optimal hyperparameter combination of the neural network model of image features and geological parameters; step S6-5, assigning the optimal hyperparameter combination to the neural network model of image features and geological parameters, and obtaining the optimal neural network model of image features and geological parameters, that is, the surrounding rock quality judgment model.

[0023] The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video provided by the present invention may also have such a technical feature, wherein the hyperparameter combination includes the number of hidden layers of the neural network model of image features and geological parameters, the number of neurons in each hidden layer, the learning rate, the L2 regularization penalty weight of the weight, and the L2 regularization penalty weight of the bias.

[0024] The shield tunnel surrounding rock quality judgment method based on monitoring video provided by the present invention can also have such technical characteristics, wherein the evaluation indicators are mean absolute percentage error MAPE, root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 One of them.

[0025] Function and Effect of the Invention

[0026] According to a method for judging the surrounding rock quality of a shield tunnel based on monitoring video of the present invention, since a video sensor is installed at the shield slag outlet, the slag image can be collected in real time. The present invention uses the shield thrust index FPI and the cutter head torque index TPI as evaluation indicators of geological quality parameters, constructs and optimizes the neural network model of slag image features and geological quality parameters. Since the evaluation index has a good correlation with the surrounding rock parameters, it can be used as a bridge to connect the formation conditions and the excavation parameters, thereby establishing a corresponding relationship between the shield slag and the specific formation conditions. Moreover, the optimized model, i.e., the surrounding rock quality judgment model, takes the grayscale feature statistics based on the collected slag image as input, and finally outputs the surrounding rock quality judgment result of the excavation position corresponding to the slag image, effectively improving the judgment accuracy of the surrounding rock quality judgment model for the surrounding rock quality, greatly reducing the judgment error, not only solving the problem that the input parameters of the traditional rock-machine mapping model cannot be updated, but also avoiding the problem that the excavation parameters collected by the shield machine through the PLC data acquisition system cannot be reasonably used. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of a method for judging the quality of surrounding rock of a shield tunnel based on monitoring video in an embodiment of the present invention;

[0028] Figure 2 A flow chart of using a genetic algorithm to optimize a model in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following is a detailed description of the method for judging the surrounding rock quality of a shield tunnel based on monitoring video in combination with embodiments and drawings.

[0030] <Example>

[0031] Figure 1 It is a flow chart of a method for judging the quality of surrounding rock of a shield tunnel based on monitoring video in an embodiment of the present invention.

[0032] like Figure 1 As shown in FIG. 1 , the method for judging the surrounding rock quality of a shield tunnel based on monitoring video includes the following steps:

[0033] Step S1, installing a video sensor at the slag outlet of the shield machine, and performing parameter calibration and image correction on the video sensor.

[0034] In this embodiment, the video sensor is installed on the slag discharge scaffolding of the slurry shield, and the installation position and distance are based on the principle of taking the most ideal picture and not interfering with the construction.

[0035] The video sensor is a solar camera or camera with wireless network card transmission function, and the battery capacity of the solar camera is sufficient to last for more than 25 consecutive rainy days. To ensure the video quality, the solar camera configuration should be at least higher than 2 million pixels.

[0036] The parameter calibration in this step includes calibrating the intrinsic parameter matrix and extrinsic parameter matrix of the solar camera, thereby establishing a mapping relationship between the world coordinate system, the camera coordinate system and the image coordinate system, and then correcting the distorted image based on the inverse transformation.

[0037] Step S2, collecting slag images within a predetermined time interval based on a video sensor.

[0038] Step S3, graying the slag discharge image to obtain a grayscale slag discharge image, and extracting grayscale feature statistics of the grayscale slag discharge image.

[0039] In this step, the grayscale feature statistics are the grayscale mean and texture features of the image, and the extraction of the grayscale feature statistics includes statistical parameters that can characterize the grayscale co-occurrence matrix. Specifically:

[0040] The matrix size of the slag image is set to M×N, f(x, y) is the gray value at the point (x, y) of the slag image matrix, and i=f(x, y), j=f(x+dx, y+dy), thus, the gray level co-occurrence matrix is ​​expressed as:

[0041] p(i,j|d,q)=#{[(x,y),(x+dx,y+dy)∈(M,N)×(M,N)|d,q,i,j]}

[0042] In the formula, (dx, dy) represents the offset of the point pair, the offset is determined by the distance and the offset direction, and # represents the number of elements in the set.

[0043] The statistical parameters that characterize the gray-level co-occurrence matrix are energy, entropy, and autocorrelation, where energy is:

[0044] ASE=∑ i ∑ j p(i, j|d, θ) 2 ,

[0045] Entropy is:

[0046] ENT=-∑ i ∑ j p(i,j|d,θ)lgp(i,j|d,θ),

[0047] And the autocorrelation is:

[0048]

[0049] Step S4, determining the geological quality parameter evaluation index of the excavation position corresponding to the slag discharge image.

[0050] Since the shield thrust index FPI and cutterhead torque index TPI have a good correlation with surrounding rock parameters such as uniaxial compressive strength USC and rock integrity coefficient Kv, they can serve as a bridge between the formation conditions and excavation parameters.

[0051] In this embodiment, the geological quality parameter evaluation indicators are the shield thrust index FPI and the cutter head torque index TPI. The calculation formula of the shield thrust index FPI is:

[0052]

[0053] Where F is the total thrust, p is the penetration rate; the calculation formula of the cutter head torque index TPI is:

[0054]

[0055] Where T is the cutter head torque and p is the penetration rate.

[0056] Step S5, taking the grayscale feature statistics as input parameters and the geological quality parameter evaluation index as output parameters, constructing a neural network model of image features and geological parameters.

[0057] Step S6, using a genetic algorithm to perform hyperparameter optimization on the neural network model of image features and geological parameters according to the evaluation index, generating an optimal neural network model of image features and geological parameters, and using it as a surrounding rock quality judgment model.

[0058] In this embodiment, the evaluation indicators include mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R). 2 .

[0059] Figure 2A flow chart of using a genetic algorithm to optimize a model in an embodiment of the present invention.

[0060] like Figure 2 As shown, the optimization process includes the following steps:

[0061] Step S6-1: within the test range, randomly select the hyperparameters to be optimized to form an initial population and use the initial population as the current population.

[0062] In this embodiment, the neural network model of image features and geological parameters includes an input layer, an output layer and a hidden layer, wherein the hyperparameters include the number of hidden layers, the number of neurons in each hidden layer, the learning rate, the L2 regularization penalty weight of the weight, and the L2 regularization penalty weight of the bias.

[0063] Step S6-2, assign the hyperparameters of each individual in the current population to the neural network model of image features and geological parameters respectively, and use the k-fold cross-validation method to verify, so as to obtain the evaluation index of the neural network model of image features and geological parameters corresponding to each individual, and evaluate the fitness of the individuals in the current population based on the evaluation index.

[0064] The k-fold cross-validation method in this embodiment refers to randomly dividing the original data set into k parts, selecting (k-1) parts as training sets each time, and the remaining 1 part as a validation set. The cross-validation is repeated k times, and the average of the k accuracy rates is taken as the evaluation indicator of the final model. It can effectively avoid the occurrence of overfitting and underfitting states.

[0065] Step S6-3, select individuals with fitness greater than 0 and in the top 25% of fitness ranking from the current population as high-quality parent individuals for crossover and mutation to generate new offspring individuals, and combine the new offspring individuals with the high-quality parent individuals to form a new current population.

[0066] As the population continues to evolve, high-quality individuals continue to accumulate, and the mutation probability is adaptively reduced from generation to generation to retain high-quality individuals.

[0067] The calculation formula for the mutation probability is:

[0068]

[0069] In the formula, p i is the mutation probability of the i-th generation, p 0 is the initial mutation probability, ω is a random number in [0,1], and n is the total number of population evolution generations.

[0070] Step S6-4, repeating steps S6-2 to S6-3 until the number of iterations reaches a predetermined number, obtaining an individual with the best evaluation index, and taking the individual as the optimal hyperparameter combination.

[0071] Step S6-5, assigning the optimal hyperparameter combination to the neural network model of image features and geological parameters to obtain the optimal neural network model of image features and geological parameters, that is, the surrounding rock quality judgment model.

[0072] The indicators for evaluating the prediction accuracy of the model include mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R 2 ) Four evaluation indicators. In the embodiment, any one of the above four evaluation indicators can be used to evaluate the model.

[0073] Step S7, inputting the slag monitoring image collected by the video sensor into the surrounding rock quality judgment model to obtain the judgment result of the surrounding rock quality at the excavation position corresponding to the slag monitoring image.

[0074] In this embodiment, the judgment result of the surrounding rock quality is the shield thrust index FPI and the cutter head torque index TPI at the excavation position corresponding to the slag discharge monitoring image.

[0075] Example Function and Effect

[0076] According to a method for judging the surrounding rock quality of a shield tunnel based on monitoring video provided in this embodiment, the shield thrust index FPI and the cutter head torque index TPI are used as evaluation indicators of geological quality parameters, and a surrounding rock quality judgment model based on monitoring video is constructed. The grayscale feature statistics of the collected slag image are input into the surrounding rock quality judgment model to finally obtain the judgment result of the surrounding rock quality. This method corresponds the grayscale feature statistics of the slag image with the surrounding rock quality parameters (FPI and TPI) as the input and output of the surrounding rock quality judgment model, which not only solves the problem that the input parameters of the traditional rock-machine mapping model cannot be updated, but also avoids the problem that the tunneling parameters collected by the shield machine through the PLC data acquisition system cannot be reasonably used.

[0077] In the embodiment, since a video sensor is installed at the shield slag outlet, the slag image can be collected in real time, so that the input of the surrounding rock quality judgment model, that is, the grayscale feature statistics of the slag image can be updated in real time.

[0078] In the embodiment, since the shield thrust index FPI and the cutterhead torque index TPI have a good correlation with the surrounding rock parameters, they can serve as a bridge connecting the formation conditions and the excavation parameters, and establish a corresponding relationship between the shield slag discharge and the specific formation conditions, thereby establishing a neural network model of the slag discharge image characteristics and geological quality parameters.

[0079] In the embodiment, since a genetic algorithm is used and the neural network model of the initially constructed image features and geological parameters is iteratively optimized according to the evaluation index until a surrounding rock quality judgment model is generated, the judgment accuracy of the surrounding rock quality judgment model is effectively improved and the judgment error is greatly reduced.

[0080] The above embodiments are only used to illustrate specific implementation modes of the present invention, and the present invention is not limited to the description scope of the above embodiments.

Claims

1. A method for judging the quality of surrounding rock of a shield tunnel based on monitoring video, which is used to judge the quality of surrounding rock at the excavation position, and is characterized in that: The following steps are involved: Step S1, installing a video sensor at a slag outlet of a shield machine, and performing parameter calibration and image correction on the video sensor; Step S2, collecting slag images within a predetermined time interval based on the video sensor; Step S3, graying the slag discharge image to obtain a grayscale slag discharge image, and extracting grayscale feature statistics of the grayscale slag discharge image; Step S4, determining the geological quality parameter evaluation index of the excavation position corresponding to the slag discharge image; Step S5, using the grayscale feature statistics as input parameters and the geological quality parameter evaluation index as output parameters to construct a neural network model of image features and geological parameters; Step S6, using a genetic algorithm to perform hyperparameter optimization on the neural network model of the image features and geological parameters according to the evaluation index, generating an optimal neural network model of the image features and geological parameters, and using it as a surrounding rock quality judgment model; Step S7: inputting the slag monitoring image collected by the video sensor into the surrounding rock quality judgment model to obtain the judgment result of the surrounding rock quality at the excavation position corresponding to the slag monitoring image. Wherein, the step S6 includes the following sub-steps: Step S6-1, within the test range, randomly selecting hyperparameters to be optimized to form an initial population and using the initial population as the current population; Step S6-2, assigning the hyperparameters of each individual in the current population to the neural network model of the image features and geological parameters respectively, using a k-fold cross-validation method for verification, obtaining evaluation indicators of the neural network model of the image features and geological parameters corresponding to each individual, and evaluating the fitness of the individuals in the current population based on the evaluation indicators; Step S6-3, selecting individuals whose fitness is greater than 0 and in the top 25% of the fitness ranking from the current population as high-quality parent individuals for crossover and mutation to generate new offspring individuals, and combining the new offspring individuals with the high-quality parent individuals to form a new current population; Step S6-4, repeating the steps S6-2 to S6-3 until the number of iterations reaches a predetermined number, and obtaining an individual with the best evaluation index, which serves as the optimal hyperparameter combination of the neural network model of the image features and geological parameters; Step S6-5, assigning the optimal hyperparameter combination to the neural network model of the image features and geological parameters to obtain the optimal neural network model of the image features and geological parameters, that is, the surrounding rock quality judgment model.

2. The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video according to claim 1, characterized in that: in, In step S1, the video sensor is a solar camera. The parameter calibration includes calibrating the intrinsic parameter matrix and the extrinsic parameter matrix of the solar camera, and establishing a mapping relationship between the world coordinate system, the camera coordinate system and the image coordinate system. The image correction is to correct the distorted image based on inverse transformation.

3. The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video according to claim 1, characterized in that: in, In step S3, the extraction of the gray-level feature statistical value includes the statistical parameters characterizing the gray-level co-occurrence matrix. The matrix size of the slag image is set to M×N, f(x, y) is the gray value at the point (x, y) of the slag image matrix, and i=f(x, y), j=f(x+dx, y+dy), and the gray level co-occurrence matrix is: p(i,j|d,q)=#{[(x,y),(x+dx,y+dy)∈(M,N)×(M,N)|d,q,i,j]} In the formula, (dx, dy) represents the offset of the point pair, the offset is determined by the distance and the offset direction, # represents the number of elements in the set, The statistical parameters characterizing the gray-level co-occurrence matrix are energy, entropy and autocorrelation. The energy is: ASE=∑ i ∑ j p(i,j|d,θ) 2 , The entropy is: ENT=-∑ i ∑ j p(i,j|d,θ)lgp(i,j|d,θ), The autocorrelation is:

4. The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video according to claim 1, characterized in that: in, In step S4, the geological quality parameter evaluation indicators are the shield thrust index FPI and the cutter head torque index TPI. The calculation formula of the shield thrust index FPI is: Where F is the total thrust, p is the penetration, The calculation formula of the cutter head torque index TPI is: Where T is the cutter head torque and p is the penetration rate.

5. The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video according to claim 1, characterized in that: in, The hyperparameter combination includes the number of hidden layers of the neural network model of the image features and geological parameters, the number of neurons in each hidden layer, the learning rate, the L2 regularization penalty weight of the weight, and the L2 regularization penalty weight of the bias.

6. The method for judging the quality of surrounding rock of a shield tunnel based on monitoring video according to claim 5 is characterized in that: in, The evaluation indicators are mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and determination coefficient (R). 2 Any one of .

Images