Deformation prediction method of surrounding rock in mine tunnel based on neural network

Abstract

This application relates to the field of tunnel surrounding rock prediction technology, and more particularly to a neural network-based method for predicting deformation of mine tunnel surrounding rock. The method comprises: selecting at least one target tunnel section within the mined area of ​​a tunnel currently being mined; inputting tunnel width, tunnel height, mining progress map, material tensor, and the position encoding of the target tunnel section into a prediction network to predict the maximum stress, maximum strain, and plastic zone area of ​​each target tunnel section; and issuing an early warning of surrounding rock deformation based on the maximum stress, maximum strain, and plastic zone area of ​​each target tunnel section. Through the technical solution of this application, accurate prediction of mine tunnel surrounding rock deformation during mining can be achieved, thereby issuing a timely early warning.

Description

Technical Field

[0001] The present application relates to the field of tunnel surrounding rock prediction technology, and in particular to a method for predicting deformation of mine tunnel surrounding rock based on a neural network. Background Art

[0002] The roadway backfill method is a commonly used mining method for gold mining. After a section is mined, the goaf is backfilled with colloidal tailings, paste, or other filling materials. The backfill can withstand some rock pressure, reducing roof subsidence and sidewall displacement to maintain the stability of the surrounding rock. During the mining process, the mining progress affects the stress on the roadway, and thus its stability.

[0003] At present, the patent application document with publication number CN105260575A discloses a method for predicting tunnel surrounding rock deformation based on a neural network. The method includes: obtaining the key influencing factors of the surrounding rock through the hierarchical analysis method, establishing a BP neural network model, inputting training samples formed by a data group obtained by monitoring under different geological conditions in the early stage, and the data structure is a multidimensional matrix. The training sample data is used to train the neural network system through the trainlm function; using the trained neural network to predict the initial deformation of the tunnel excavation, and according to the input index parameters, the surrounding rock top plate sinking amount, bottom plate upward displacement, tunnel side strain and the maximum plastic zone failure depth generated are obtained. The neural network will predict the tunnel deformation in the early stage of excavation according to the prediction request, select appropriate support parameters to support and control the tunnel, and prevent safety accidents caused by rock instability.

[0004] The above method trains the BP neural network model based on the data sets obtained from monitoring under different geological conditions, thereby predicting the tunnel deformation in the early stage of excavation. However, the above method ignores the impact of the mining progress on the stress conditions of the tunnel during the mining process, and cannot achieve accurate prediction of the deformation of the surrounding rock of the mine tunnel during the mining process. Summary of the Invention

[0005] In order to solve the technical problem of being unable to accurately predict the deformation of mine tunnel surrounding rocks during mining, the present application provides a mine tunnel surrounding rock deformation prediction method based on a neural network, which can accurately predict the deformation of mine tunnel surrounding rocks during mining, thereby issuing early warnings in a timely manner.

[0006] In a first aspect, the present application provides a method for predicting deformation of surrounding rock in mine tunnels based on a neural network, the prediction method comprising: selecting at least one target tunnel section in a mined area in a tunnel being mined; inputting a mining progress map, a material tensor and a position code of the target tunnel section into a prediction network to predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, wherein the mining progress map includes the mining status of each position point in all tunnels, and the mining status includes filled, mined and unmined; the position code includes the tunnel ID of the tunnel and the horizontal distance between the target tunnel section and the mining site; the material tensor includes the mechanical parameters of the upper wall surrounding rock, the lower wall surrounding rock, the ore body and the filling material; judging whether the tunnel being mined is deformed based on the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, and issuing a warning signal if deformation occurs.

[0007] In the process of mining gold using the road-fill mining method, as the mining process progresses, the stress conditions of the tunnel being mined will also change. In order to timely determine whether the tunnel being mined has deformed, at least one target tunnel section is selected from the mined area of ​​the tunnel being mined, and the tunnel width, tunnel height, mining progress map, material tensor and position coding of the target tunnel section are input into the prediction network to predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section. The tunnel width and tunnel height can reflect the size information of the target tunnel section, the mining progress map and material tensor can reflect the mechanical parameters of the material around the target tunnel section, and the position coding of the target tunnel section is used to determine the position information of the target tunnel section. The size information, position information and mechanical parameters of the surrounding materials of the target tunnel section are comprehensively considered to accurately predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, thereby determining whether the tunnel being mined has deformed. If deformation occurs, an early warning signal is issued.

[0008] Preferably, the mechanical parameters include elastic modulus, Poisson's ratio, internal friction angle, density, tensile strength and cohesion.

[0009] Preferably, selecting at least one target roadway section includes: calculating the horizontal distance from each roadway section to the stope in the mined area of ​​the roadway being mined; and selecting at least one target roadway section based on the horizontal distance.

[0010] Preferably, selecting at least one target tunnel section based on the horizontal distance includes: the interval between any tunnel sections is the absolute value of the difference in horizontal distance, and selecting multiple target tunnel sections from each tunnel section at a preset interval.

[0011] Preferably, selecting at least one target tunnel section based on the horizontal distance includes: selecting a tunnel section with a horizontal distance of 0 as the target tunnel section, and calculating a selection coefficient, wherein the selection coefficient is positively correlated with the horizontal distance of the target tunnel section; taking the product of the selection coefficient and the minimum interval as the selection interval, and taking the tunnel section corresponding to the sum of the horizontal distance of the target tunnel section and the selection interval as the next target tunnel section; and selecting the target tunnel sections in sequence until the next target tunnel section cannot be obtained.

[0012] The stope is the active area of ​​mining and the main source of disturbance that affects the stress conditions of each tunnel section. Therefore, the closer to the stope, the faster the stress conditions change. Selecting dense sections in areas close to the stope and sparse sections in areas far from the stope can accurately capture the changing trend of the stress conditions to accurately obtain deformation prediction results, while reducing the amount of calculation.

[0013] Preferably, the prediction network includes a convolution subnetwork and a regression subnetwork; the convolution subnetwork is used to extract features from the mining progress map to obtain progress features; after splicing the progress features, material tensors and position codes of the target tunnel sections, the splicing results are input into the regression subnetwork to predict the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section.

[0014] An end-to-end neural network structure is provided to predict the maximum stress, maximum strain and plastic zone area of ​​any target tunnel section.

[0015] Preferably, the training method of the prediction network includes: numerically simulating the mining process based on the material tensor, obtaining the stress cloud map, strain cloud map and plastic area distribution map of the tunnel section coded at any position under any mining progress map, and then obtaining the stress label, strain label and plastic area label; inputting the material tensor, the mining progress map and the position code into the prediction network to predict the maximum stress, maximum strain and plastic area, performing weighted summation on the first deviation between the maximum stress and the stress label, the second deviation between the maximum strain and the strain label, and the third deviation between the plastic area and the plastic area label to obtain the loss function for iterative training of the prediction network; in response to the loss function being less than the preset loss or the number of iterations being greater than the preset number, completing the training of the prediction network.

[0016] The mining process was numerically simulated using FLAC3D software to obtain training data for training the prediction network, which was then constrained to learn the mapping relationship between maximum stress, maximum strain, plastic zone area, and input information.

[0017] Preferably, the loss function Satisfies the relationship:

[0018] ; 、 and are the maximum stress, maximum strain and plastic region area predicted by the prediction network, 、 and They are stress label, strain label and plastic area label, 、 and are the first weight, the second weight and the third weight respectively, and satisfy .

[0019] Preferably, the method for obtaining the first weight includes: during the numerical simulation process, counting the maximum stress of any tunnel section under different mining progress to obtain a stress curve, and taking the average variance of the stress curves of multiple tunnel sections as the stress variance; taking the ratio of the stress variance to the sum of the variances as the first weight, and the sum of the variances is the sum of the stress variance, the strain variance and the plastic zone area variance.

[0020] Since the maximum stress, maximum strain and plastic zone area are affected to different degrees by mining progress, the first weight is assigned to the maximum stress according to the change of the maximum stress with the mining progress during the numerical simulation process. The greater the change of the maximum stress with the mining progress, the greater the first weight is, to ensure the prediction accuracy of the maximum stress.

[0021] Preferably, determining whether a tunnel being mined has been deformed includes: taking the horizontal distance between the target tunnel section and the mining site as the horizontal coordinate and the maximum stress of the target tunnel section as the vertical coordinate to obtain a stress scatter curve, and interpolating the stress scatter curve using an interpolation algorithm to obtain the maximum stress of each tunnel section in the mined area of ​​the tunnel being mined; following the same steps to obtain the maximum strain and plastic zone area of ​​each tunnel section in the mined area of ​​the tunnel being mined; in response to the maximum stress of any tunnel section being greater than the maximum allowable stress, or the maximum strain of any tunnel section being greater than the maximum allowable strain, or the plastic zone area of ​​any tunnel section being greater than the maximum allowable plastic zone area, determining that the tunnel being mined has been deformed.

[0022] The technical solution of this application has the following beneficial technical effects:

[0023] In the process of mining gold using the road-fill mining method, as the mining process progresses, the stress conditions of the tunnel being mined will also change. In order to timely judge whether the tunnel being mined has deformed, at least one target tunnel section is selected in the mined area of ​​the tunnel being mined, and the tunnel width, tunnel height, mining progress map, material tensor and position coding of the target tunnel section are input into the prediction network to predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section. Among them, the tunnel width and tunnel height can reflect the size information of the target tunnel section, the mining progress map and material tensor can reflect the mechanical parameters of the material around the target tunnel section, and the position coding of the target tunnel section is used to determine the position information of the target tunnel section. The size information, position information and mechanical parameters of the surrounding materials of the target tunnel section are comprehensively combined to accurately predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, thereby judging whether the tunnel being mined has deformed. If deformation occurs, an early warning signal is issued.

[0024] Furthermore, when it is monitored that the mining progress increases by a preset length, at least one target tunnel section is selected again to determine whether the tunnel being mined is deformed after the mining progress changes, thereby realizing real-time early warning of mine tunnel surrounding rock deformation during the mining process. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flowchart of a method for predicting surrounding rock deformation in mine tunnels based on a neural network according to an embodiment of the present application.

[0026] Figure 2 Schematic diagram of a mine model according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] According to the first aspect of the present application, the present application provides a method for predicting deformation of mine tunnel surrounding rock based on a neural network. During the mining process of gold mines using the approach-and-fill mining method, it predicts whether deformation of the tunnel surrounding rock occurs and issues early warning information in a timely manner.

[0029] For ease of understanding, the operation process of the approach and filling mining method is introduced here. The approach and filling mining method divides the ore body into multiple layers according to different heights, and each layer corresponds to a layer tunnel. According to the mining order of multiple layer tunnels, the approach and filling mining method can be divided into upward approach mining and downward approach mining. Among them, the upward approach mining starts from the lowest layer tunnel and mines upward in sequence, while the downward approach mining starts from the highest layer tunnel and mines downward in sequence.

[0030] In the approach and backfill mining method, the ore body is divided into four layers according to different heights, and each layer corresponds to a layered roadway. The layered roadways are recorded as Roadway 1, Roadway 2, Roadway 3 and Roadway 4 from top to bottom; if upward approach mining is used to operate on the ore body with a fixed roadway width and roadway height, then Roadway 4 is mined first. At this time, the area around Roadway 4 is all ore body material; after completing the mining of Roadway 4, Roadway 4 is backfilled with filling material first, and then the mining process of Roadway 3 is started. At this time, the material below Roadway 3 is filling material, and the material in other positions is still ore body material; the mining of each layered roadway is completed in sequence according to the same method.

[0031] Among them, the tunnel width and tunnel height can determine the size information of the tunnel cross section.

[0032] Figure 1 This is a flow chart of a method for predicting surrounding rock deformation in mine tunnels based on a neural network according to an embodiment of the present application. Figure 1 As shown, the neural network-based mine tunnel surrounding rock deformation prediction method includes steps S101 to S103, which are described in detail below.

[0033] S101, selecting at least one target tunnel section in a mined area of ​​a tunnel being mined.

[0034] In one embodiment, the target tunnel section is any section within the mined area of ​​the tunnel being mined, and whether the tunnel being mined is deformed is determined by predicting the maximum stress, maximum strain and plastic zone area in the target tunnel section.

[0035] Specifically, selecting at least one target roadway section includes: calculating the horizontal distance from each roadway section to the stope in the mined area of ​​the roadway being mined; and selecting at least one target roadway section based on the horizontal distance.

[0036] It is understandable that the mined area in the tunnel being mined contains multiple tunnel sections. If the maximum stress, maximum strain and plastic zone area are predicted for all tunnel sections, although the accuracy of the surrounding rock deformation prediction can be ensured, it will undoubtedly consume a huge amount of calculation. Therefore, it is necessary to select multiple target tunnel sections from the mined area in the tunnel being mined. By predicting the maximum stress, maximum strain and plastic zone area for multiple target tunnel sections, the amount of calculation can be reduced while accurately obtaining the deformation prediction results.

[0037] In one embodiment, selecting at least one target tunnel section based on the horizontal distance includes: the interval between any tunnel sections is the absolute value of the difference in horizontal distance, and selecting multiple target tunnel sections from each tunnel section at a preset interval.

[0038] That is to say, the absolute value of the difference in horizontal distances between adjacent target tunnel sections is equal to the preset interval. In the embodiment of the present application, the preset interval is 5 meters.

[0039] In another embodiment, the stope is the active area of ​​mining and is the main source of disturbance that affects the stress conditions of each roadway section. Therefore, the closer to the stope, the faster the stress conditions change. In order to accurately capture the changing trend of the stress conditions, dense target roadway sections need to be arranged. Relatively speaking, as the horizontal distance between the section and the stope increases, the disturbance intensity decreases, and the trend of the stress conditions tends to be stable. The diluted target roadway sections are sufficient to reflect the changing trend of the stress conditions. At this time, sparse target roadway sections can be arranged to avoid redundant sampling. Therefore, dense sections are selected in the area close to the stope, and sparse sections are selected in the area far from the stope. This can accurately capture the changing trend of the stress conditions to accurately obtain deformation prediction results, and reduce the amount of calculation; wherein, the stress conditions include maximum stress, maximum strain, and plastic zone area.

[0040] Specifically, selecting at least one target tunnel section based on the horizontal distance includes: selecting a tunnel section with a horizontal distance of 0 as the target tunnel section, and calculating a selection coefficient, wherein the selection coefficient is positively correlated with the horizontal distance of the target tunnel section; taking the product of the selection coefficient and the minimum interval as the selection interval, and taking the tunnel section corresponding to the sum of the horizontal distance of the target tunnel section and the selection interval as the next target tunnel section; and selecting the target tunnel sections in sequence until the next target tunnel section cannot be obtained.

[0041] The horizontal distance is The selection coefficient of the target tunnel section for: , is an adjustment coefficient, which is used to control the speed at which the selected interval increases with the horizontal distance. In the embodiment of the present application, the adjustment coefficient is 0.1; the minimum interval is 2 meters.

[0042] For example, first select the tunnel section with a horizontal distance of 0 as the target tunnel section, then the value of the selection coefficient is 1, and the selection interval is the product of the minimum interval and the selection coefficient, that is, the value of the selection interval is 2 meters, then the next target tunnel section is the tunnel section with a horizontal distance of 2 meters; further calculate that the selection coefficient of the tunnel section with a horizontal distance of 2 meters is 1.2, and the value of the selection interval is 2.4 meters, then the next target tunnel section is the tunnel section with a horizontal distance of 4.4 meters; select multiple target tunnel sections in turn.

[0043] In this way, multiple target tunnel sections are selected in the mined area of ​​the tunnel being mined. By predicting the maximum stress, maximum strain and plastic zone area of ​​all target tunnel sections, the distribution of the maximum stress, maximum strain and plastic zone area in the mined area of ​​the tunnel being mined can be accurately obtained.

[0044] S102, the tunnel width, tunnel height, mining progress map, material tensor and position encoding of the target tunnel section are input into the prediction network to predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section.

[0045] In one embodiment, the tunnel width, tunnel height, mining progress map, material tensor, and the location of target tunnel sections are encoded and input into a prediction network to predict the maximum stress, maximum strain, and plastic zone area of ​​each target tunnel section. The prediction network is used to predict the maximum stress, maximum strain, and plastic zone area of ​​each target tunnel section. The tunnel width and tunnel height can determine the size of the tunnel section.

[0046] It should be noted that due to the dimensional differences between the tunnel width, tunnel height and material tensor in the input data, a standardization method should be used to eliminate the dimensions to avoid large numerical differences caused by different dimensions.

[0047] The mining progress map is a single-channel, two-dimensional image that includes the mining status of each point in all lanes, including filled, mined, and unmined. Specifically, the mining progress map is obtained by vertically projecting all lanes within the mine onto a horizontal plane to form a top-down view. The mining status of each point in all lanes is then mapped onto this top-down view to create the mining progress map. This map not only effectively reflects the distribution and relative positions of the lanes, but also intuitively reflects the mining status of each point in all lanes.

[0048] For example, the mine mining plan includes a total of 4 layered tunnels, and each layered tunnel corresponds to a tunnel ID, which are recorded as tunnel 1, tunnel 2, tunnel 3 and tunnel 4 in sequence; if the mining status of each position point in tunnel 1 is filled, it means that the mining and filling of tunnel 1 have been completed, and the mechanical parameters of each position point in tunnel 1 are the mechanical parameters of the filling material; if the mining status of each position point in tunnel 3 and tunnel 4 is unmined, it means that tunnel 3 and tunnel 4 have not yet started mining, and the mechanical parameters of each position point in tunnel 3 and tunnel 4 are the mechanical parameters of the ore body material; if the mining status of some position points in tunnel 2 is mined, and the mining status of other position points is unmined, it means that tunnel 2 is a tunnel being mined, and all position points with a mining status of mined constitute the mined area in the tunnel being mined, and all position points with a mining status of unmined constitute the unmined area in the tunnel being mined, there is no material in the mined area of ​​tunnel 2, and the mechanical parameters in the unmined area of ​​tunnel 2 are the mechanical parameters of the ore body material.

[0049] The material tensor includes mechanical parameters of the upper wall surrounding rock, the lower wall surrounding rock, the ore body and the filling material, and the mechanical parameters include elastic modulus, Poisson's ratio, internal friction angle, density, tensile strength and cohesion.

[0050] It can be understood that the combined mining progress map and material tensor can accurately reflect the material distribution of any cross section within the mined area of ​​the mining tunnel.

[0051] Among them, the position code of the target tunnel section includes the tunnel ID of the tunnel and the horizontal distance between the target tunnel section and the mining area. The target tunnel section can be uniquely and accurately located through the tunnel ID of the tunnel and the horizontal distance to the mining area; wherein, the mining area is the active area of ​​gold mining, and the location information of the mining area can be determined based on the mining progress map. The mining area position is the junction of the mined area and the unmined area in tunnel 2.

[0052] The tunnel width, tunnel height, mining progress map, material tensor and position encoding of the target tunnel section are taken as inputs of the prediction network, so that the prediction network can learn the material distribution of the target tunnel section under the mining progress map, and then predict the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section.

[0053] It can be understood that, while the mining progress map and material tensor remain unchanged, by changing the position coding of the target tunnel section, the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section under the mining progress map can be obtained.

[0054] In one embodiment, the prediction network includes a convolution subnetwork and a regression subnetwork. The convolution subnetwork is used to extract features from the mining progress map to obtain progress features. After splicing the tunnel width, tunnel height, the progress features, material tensors and position encoding of the target tunnel section, the splicing results are input into the regression subnetwork to predict the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section.

[0055] The convolutional subnetwork can employ existing convolutional neural networks such as ResNet or VGGNet, while the regression subnetwork is a BP neural network. The progress feature is a column vector with A rows and 1 column, where the value of A is related to the network structure of the convolutional subnetwork. The material tensor includes six mechanical parameters, namely, the upper wall surrounding rock, the lower wall surrounding rock, the ore body, and the fill material. Therefore, the material tensor can be considered a column vector with 24 rows and 1 column. The position encoding of the target roadway section can be considered a column vector with 2 rows and 1 column, with the roadway width and height both being specific values. Therefore, the input layer of the regression subnetwork consists of 24 + 2 + 2 + A neurons, and the output layer of the regression subnetwork consists of 3 neurons, corresponding to the maximum stress, maximum strain, and plastic zone area of ​​the target roadway section, respectively.

[0056] In one embodiment, in order to ensure that the prediction network can accurately predict the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section, the prediction network needs to be trained to constrain the prediction network to learn the mapping relationship between the maximum stress, maximum strain, plastic zone area and input information.

[0057] Specifically, the training method of the prediction network includes: establishing a mine model based on the material tensor, and numerically simulating the mining process with the tunnel width and tunnel height, obtaining the stress cloud map, strain cloud map and plastic area distribution map of the tunnel section encoded at any position under any mining progress map, and obtaining the stress label, strain label and plastic area label of multiple input samples; inputting the input samples into the prediction network to predict the maximum stress, maximum strain and plastic area, performing weighted summation on the first deviation between the maximum stress and the stress label, the second deviation between the maximum strain and the strain label, and the third deviation between the plastic area and the plastic area label to obtain the loss function, and iteratively training the prediction network; in response to the loss function being less than the preset loss or the number of iterations being greater than the preset number, completing the training of the prediction network.

[0058] Among them, the preset loss is 0.01; the preset number of times is 300.

[0059] Among them, the numerical simulation process is implemented using FLAC3D software (Fast Lagrangian Analysis of Continua). FLAC3D software is suitable for simulating material deformation and complex deformation states as well as deformation of highly nonlinear geological materials. It can obtain stress cloud maps, strain cloud maps and plastic area distribution maps of any tunnel section, and further standardize the maximum stress value in the stress cloud map to obtain a stress label, standardize the maximum strain in the strain cloud map to obtain a strain label, and standardize the area of ​​the plastic area in the plastic area distribution map to obtain a plastic area label; that is, the strain label, stress label and plastic area label are all dimensionless values.

[0060] See Figure 2 , is a schematic diagram of a mine model according to an embodiment of the present application. In the mine model, the mechanical parameters of the ore body, hanging wall surrounding rock, footwall surrounding rock, and the filling material used are shown in Table 1.

[0061] Table 1 Mechanical parameters of ore body, hanging wall surrounding rock, footwall surrounding rock and filling materials used

[0062]

[0063] Specifically, the loss function Satisfies the relationship:

[0064] ; 、 and are the maximum stress, maximum strain and plastic region area predicted by the prediction network, 、 and They are stress label, strain label and plastic area label, 、 and are the first weight, the second weight and the third weight respectively, and satisfy .

[0065] The first, second, and third weights are used to control the degree of attention paid to the maximum stress, maximum strain, and plastic zone area during the prediction network training process. For example, a larger value for the first weight indicates that greater attention is paid to the maximum stress during the prediction network training process. This can significantly change the value of the loss function when there is a prediction error in the maximum stress, thereby ensuring that the prediction network can output an accurate maximum stress.

[0066] In one embodiment, since the maximum stress, maximum strain, and plastic zone area are affected to different degrees by the mining progress, during the numerical simulation process, the change of the maximum stress with the mining progress is analyzed. If the change is greater, it means that the maximum stress is more greatly affected by the mining progress. In order to accurately predict whether the mine surrounding rock will deform, it is necessary to focus on the prediction accuracy of the maximum stress.

[0067] Specifically, the method for obtaining the first weight includes: during the numerical simulation process, statistically calculating the maximum stress of any tunnel section under different mining progress to obtain a stress curve, and taking the average variance of the stress curves of multiple tunnel sections as the stress variance; taking the ratio of the stress variance to the sum of the variances as the first weight, and the sum of the variances is the sum of the stress variance, the strain variance and the plastic zone area variance.

[0068] The strain variance and plastic zone area variance are obtained in the same way as the stress variance. The second weight is the ratio of the strain variance to the sum of the variances, and the third weight is the ratio of the plastic zone area variance to the sum of the variances.

[0069] For example, the training of the prediction network is completed, and the trained prediction network can output the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section based on the mining progress map, material tensor and position coding of the target tunnel section.

[0070] S103, judging whether the tunnel being mined has deformed based on the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, and issuing an early warning signal if deformation has occurred.

[0071] In one embodiment, determining whether a tunnel being mined has been deformed includes: using the horizontal distance between the target tunnel section and the mining area as the horizontal coordinate and the maximum stress of the target tunnel section as the vertical coordinate to obtain a stress scatter curve, and interpolating the stress scatter curve using an interpolation algorithm to obtain the maximum stress of each tunnel section in the mined area of ​​the tunnel being mined; following the same steps to obtain the maximum strain and plastic zone area of ​​each tunnel section in the mined area of ​​the tunnel being mined; and determining that the tunnel being mined has been deformed in response to the maximum stress of any tunnel section being greater than the maximum allowable stress, or the maximum strain of any tunnel section being greater than the maximum allowable strain, or the plastic zone area of ​​any tunnel section being greater than the maximum allowable plastic zone area.

[0072] The interpolation algorithm adopts radial basis interpolation method or spatial Kriging interpolation method.

[0073] The maximum allowable stress, maximum allowable strain, and maximum allowable plastic zone area are set by those skilled in the art based on the Hoek-Brown strength criterion. Taking maximum stress as an example, the maximum stress of each target roadway section can reflect the change in maximum stress within the mined area under the mining progress corresponding to the mining progress map. The maximum stress of each target roadway section is interpolated to obtain the maximum stress of each roadway section within the mined area. If the maximum stress of a roadway section exceeds the maximum allowable stress, it indicates that there is a risk of deformation in the mining roadway, thus issuing an early warning.

[0074] It should be noted that the mining progress is collected in real time. When it is monitored that the mining progress increases by a preset length, at least one target tunnel section is selected again in the mined area of ​​the tunnel being mined to determine whether the tunnel being mined is deformed after the mining progress changes.

[0075] The preset length is 5 meters. A laser ranging sensor can be used to monitor changes in the length of the mined area in the mining tunnel. When the mining progress increases by the preset length, it indicates that the stress conditions in the mining tunnel will change. At this time, the target tunnel section needs to be re-selected to achieve real-time early warning of surrounding rock deformation in the mine tunnel during the mining process, prompting staff to adjust support parameters in a timely manner to ensure the stability of the mining tunnel.

[0076] It should be noted that a person skilled in the art may make a number of modifications and improvements without departing from the concept of the present application, and these modifications and improvements are all within the scope of protection of the present application. Therefore, the scope of protection of the patent application shall be based on the appended claims.

Claims

1. A method for predicting surrounding rock deformation of mine tunnels based on neural networks, characterized in that: The prediction method includes: calculating the horizontal distance from each roadway section to the stope in the mined area of ​​the roadway being mined; selecting at least one target roadway section based on the horizontal distance; Selecting at least one target roadway section according to the horizontal distance includes: selecting a roadway section with a horizontal distance of 0 as the target roadway section, and calculating a selection coefficient, wherein the selection coefficient is positively correlated with the horizontal distance of the target roadway section; multiplying the selection coefficient and the minimum interval as the selection interval, and taking the roadway section corresponding to the sum of the horizontal distance of the target roadway section and the selection interval as the next target roadway section; selecting the target roadway sections in sequence until the next target roadway section cannot be obtained; the horizontal distance is The selection coefficient of the target tunnel section for: , is the adjustment coefficient, which is used to control the speed at which the selection interval increases with the horizontal distance; The tunnel width, tunnel height, mining progress map, material tensor and position codes of the target tunnel section are input into the prediction network to predict the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, wherein the mining progress map includes the mining status of each position point in all tunnels, and the mining status includes filled, mined and unmined; the position code includes the tunnel ID of the tunnel and the horizontal distance between the target tunnel section and the stope; the material tensor includes the mechanical parameters of the upper wall surrounding rock, lower wall surrounding rock, ore body and filling material; the prediction network includes a convolution subnetwork and a regression subnetwork; the convolution subnetwork is used to extract features from the mining progress map to obtain progress features; after splicing the tunnel width, tunnel height, progress features, material tensor and position codes of the target tunnel section, the splicing results are input into the regression subnetwork to predict the maximum stress, maximum strain and plastic zone area of ​​the target tunnel section; Based on the maximum stress, maximum strain and plastic zone area of ​​each target tunnel section, it is determined whether the tunnel being mined has deformed. If deformation occurs, an early warning signal is issued.

2. The method for predicting deformation of surrounding rock in mine tunnels based on neural network according to claim 1, characterized in that: The mechanical parameters include elastic modulus, Poisson's ratio, internal friction angle, density, tensile strength and cohesion.

3. The method for predicting deformation of surrounding rock in mine tunnels based on neural network according to claim 1, characterized in that: The training method of the prediction network includes: establishing a mine model based on the material tensor, and numerically simulating the mining process with the tunnel width and tunnel height, obtaining the stress cloud map, strain cloud map and plastic area distribution map of the tunnel section encoded at any position under any mining progress map, and obtaining the stress labels, strain labels and plastic area labels of multiple input samples; inputting the input samples into the prediction network to predict the maximum stress, maximum strain and plastic area, performing weighted summation on the first deviation between the maximum stress and the stress label, the second deviation between the maximum strain and the strain label, and the third deviation between the plastic area and the plastic area label to obtain a loss function, and iteratively training the prediction network; in response to the loss function being less than a preset loss or the number of iterations being greater than a preset number, completing the training of the prediction network.

4. The method for predicting deformation of surrounding rock in mine tunnels based on neural network according to claim 3, characterized in that: The loss function Satisfies the relationship: ; 、 and are the maximum stress, maximum strain and plastic region area predicted by the prediction network, 、 and They are stress label, strain label and plastic area label, 、 and are the first weight, the second weight and the third weight respectively, and satisfy .

5. The method for predicting deformation of surrounding rock in mine tunnels based on neural network according to claim 3, characterized in that: The method for obtaining the first weight includes: In the process of numerical simulation of mining process, the maximum stress of any tunnel section under different mining progress is counted to obtain stress curve, and the average variance of stress curves of multiple tunnel sections is taken as stress variance; The ratio of the stress variance to the sum of variances is used as the first weight, where the sum of variances is the sum of the stress variance, the strain variance, and the plastic region area variance.

6. The method for predicting deformation of surrounding rock in mine tunnels based on neural network according to claim 1, characterized in that: Determining whether a tunnel being mined has deformed includes: The horizontal distance between the target roadway section and the stope is used as the abscissa, and the maximum stress of the target roadway section is used as the ordinate to obtain a stress scatter curve. The stress scatter curve is interpolated using an interpolation algorithm to obtain the maximum stress of each roadway section in the mined area of ​​the roadway being mined. Follow the same steps to obtain the maximum strain and plastic area of ​​each roadway cross section in the mined area of ​​the roadway being mined; In response to the maximum stress of any roadway section being greater than the maximum allowable stress, or the maximum strain of any roadway section being greater than the maximum allowable strain, or the plastic zone area of ​​any roadway section being greater than the maximum allowable plastic zone area, it is determined that the roadway being mined is deformed.

Images