Numerical simulation and transfer learning combined mine surface deformation prediction method

Abstract

The invention belongs to the technical field of mining, and provides a numerical simulation and transfer learning combined mine surface deformation prediction method, which comprises the steps of carrying out numerical simulation on ore body mining and carrying out preprocessing to obtain source domain data, calculating an actual surface deformation value of a mining area to obtain target domain data, constructing a surface settlement prediction model comprising an LSTM layer, a respe layer and a DCNN layer, and carrying out migration learning on the surface settlement prediction model. Inputting the source domain data into the ground surface subsidence prediction model to pre-train the ground surface subsidence prediction model, adding a full connection layer to the pre-trained ground surface subsidence prediction model, setting the full connection layer to be in a trainable state, taking the target domain data as a target domain for transfer learning of the frozen ground surface subsidence prediction model, and training again to obtain a ground surface subsidence prediction model; parameters of the frozen ground surface settlement prediction model are finely adjusted, weights are updated, a prediction model is obtained, to-be-predicted mine field geological data are input into the prediction model for prediction, and a predicted ground surface settlement value result is obtained. The method can predict the deformation of the earth surface in different mining stages.

Description

Technical Field

[0001] The invention belongs to the technical field of mining, and specifically discloses a method for predicting mine surface deformation by combining numerical simulation with transfer learning. Background Art

[0002] The development of mineral resources has brought huge economic benefits to countries around the world and provided necessary materials for social development and construction. It has become an indispensable part of human social development and civilization progress. With the rapid development of modern society and economy, the demand in the industrial field has continued to grow, and the pressure on mining has also increased. Traditional mining capacity can no longer meet my country's huge demand for mineral resources, so mine mining has begun to expand to deeper strata and more complex geological conditions.

[0003] In the process of underground mining, the mining of ore bodies will form goafs. If these goafs collapse, they may destroy the stress balance of the overlying rock strata, causing stratum movement and ground collapse, which in turn will damage ground buildings and production facilities, and may even induce geological disasters such as landslides and ground fissures, seriously threatening human life and property safety.

[0004] Since the introduction of transfer learning in the late 20th century, it has received widespread attention and has been mainly used in fields such as computer vision and natural language understanding. At present, typical applications of transfer learning include classification, recognition detection, positioning and prediction. In the problem of surface settlement prediction in underground mining, due to the large differences in the engineering backgrounds of various mines, although there is a certain reference between them, the dynamic research on rock strata and surface under the influence of large goafs in metal mines, as well as the changes in time and space, is not in-depth enough. Few studies have used transfer learning to predict surface settlement in underground mining. Therefore, in response to these problems, it is very necessary to study and design a new method of numerical simulation and transfer learning combined with mine surface deformation prediction to solve the existing problems in the prediction of surface settlement in underground mining. Summary of the invention

[0005] In order to solve the problem that the existing underground mining surface subsidence prediction is difficult and there are few studies on using transfer learning to predict underground mining surface subsidence, the present invention proposes a mine surface deformation prediction method combining numerical simulation with transfer learning.

[0006] The present invention provides a method for predicting mine surface deformation by combining numerical simulation with transfer learning, comprising the following steps:

[0007] S1. Numerical simulation of ore body mining is performed based on the geological data of the mine site, and the surface deformation values ​​calculated by numerical simulation are preprocessed to obtain source domain data;

[0008] S2. Calculate the actual surface deformation value of the mining area based on the basic principle of probability integral method according to the geological data of the mining site to obtain the target domain data;

[0009] S3. Constructing a surface subsidence prediction model based on numerical simulation and transfer learning combined with a long short-term memory neural network LSTM and a deep convolutional neural network DCNN, wherein the surface subsidence prediction model includes an LSTM layer, a reshape layer and a DCNN layer;

[0010] S4. Using the source domain data obtained in step S1 as the source domain for transfer learning of the surface subsidence prediction model, inputting the source domain data into the surface subsidence prediction model constructed in step S3 to pre-train the surface subsidence prediction model, thereby obtaining a pre-trained surface subsidence prediction model;

[0011] S5. Add a fully connected layer to the pre-trained surface subsidence prediction model obtained in step S4, set each layer of the pre-trained surface subsidence prediction model to a non-trainable state, set the fully connected layer to a trainable state, obtain a frozen surface subsidence prediction model, use the target domain data obtained in step S2 as the target domain for transfer learning of the frozen surface subsidence prediction model and re-train, fine-tune the parameters of the frozen surface subsidence prediction model and update the weights to obtain a prediction model;

[0012] S6. Input the mine site geological data to be predicted into the prediction model obtained in step S5 to perform prediction and obtain the predicted surface settlement value result.

[0013] According to a method for predicting mine surface deformation by combining numerical simulation with transfer learning in some embodiments of the present application, step S1 comprises the following steps:

[0014] S101. Scanning the mine site by drone to obtain the real-scene 3D visualization point cloud data of the mine area, establishing the initial mine model based on the real-scene 3D visualization point cloud data of the mine area, determining the spatial posture and spatial position of the fault according to the mine exploration report and constructing a solid surface for cutting, cutting the initial mine model in turn by Boolean operation to obtain a geological database;

[0015] S102. Based on the core photos taken from the mine site, the Mask-RCNN deep learning neural network model is used to identify and obtain the RQD values ​​of each segment of different boreholes. According to the lithology description in the borehole column chart, the lithology and weathering conditions of each segment of different boreholes are determined. The RMR values ​​of each segment of different boreholes are calculated in combination with the RQD values ​​of each segment of different boreholes. The mechanical parameters of the spatial variability of the rock mass are calculated according to the Hoek-Brown strength criterion and imported into the geological database to obtain the rock mechanical parameters of the actual geological conditions of the mining area;

[0016] S103. According to the rock mechanical parameters under the same working conditions, 9 points are selected as numerical simulation monitoring points with the target point as the center and 1m intervals in front, back, left and right. The ore body is excavated and backfilled in 16 steps, with 4 steps as a segment. The surface settlement data of each excavation monitoring point is extracted to obtain the surface deformation curve. The surface deformation value is obtained according to the surface deformation curve, and the surface deformation value is preprocessed to obtain the source domain data.

[0017] According to a method for predicting mine surface deformation that combines numerical simulation with transfer learning in some embodiments of the present application, step S2 includes determining the surface subsidence coefficient, horizontal movement coefficient, main influence angle tangent, mining influence propagation angle and inflection point offset based on the mine site geological data, and calculating the surface inclination, curvature and horizontal deformation of the mining area through the basic principle of probability integral method to obtain the actual surface deformation value of the mining area.

[0018] According to some embodiments of the present application, a method for predicting mine surface deformation combining numerical simulation and transfer learning is provided, wherein the surface settlement prediction model in step S3 includes three LSTM layers, one reshape layer and one DCNN layer, wherein the three LSTM layers serve as encoders for extracting global temporal features and deeper abstract features of the source domain data, the reshape layer is used to change the array structure and convert the input of the DCNN layer into three-channel pixel matrix data, and the DCNN layer serves as a decoder for extracting features of the pixel matrix data spatially and locally through local connections and shared weights, and outputting predicted surface deformation value data.

[0019] According to a method for predicting mine surface deformation combining numerical simulation and transfer learning according to some embodiments of the present application, the DCNN layer includes a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a third convolutional layer, a third maximum pooling layer, a flattening layer, a first fully connected layer, a first dropout layer, a second fully connected layer, a second dropout layer, a third fully connected layer and an output layer stacked in sequence, the activation functions of the first convolutional layer, the second convolutional layer, the third convolutional layer, the first fully connected layer and the second fully connected layer are all ReLU activation functions, and the activation function of the third fully connected layer is the Softmax activation function.

[0020] According to a mine surface deformation prediction method combining numerical simulation and transfer learning in some embodiments of the present application, the discarding rates of the first discarding layer and the second discarding layer are both 0.5.

[0021] According to a method for predicting mine surface deformation by combining numerical simulation and transfer learning in some embodiments of the present application, the data size output by the third convolutional layer is 13×16×32.

[0022] According to a method for predicting mine surface deformation combining numerical simulation and transfer learning in some embodiments of the present application, in step S5, the loss functions used include mean square error, mean absolute percentage error, mean absolute error and root mean square error;

[0023] The mean square error is shown in formula (1):

[0024]

[0025] Among them, MSE represents mean square error, N represents the number of data, and y i Represents the predicted value, y i ′ represents the true value;

[0026] The mean absolute percentage error is shown in formula (2):

[0027]

[0028] Among them, MAPE means mean absolute percentage error;

[0029] The mean absolute error is shown in formula (3):

[0030]

[0031] Among them, MAE means mean absolute error;

[0032] The root mean square error is shown in formula (4):

[0033]

[0034] Here, RMSE stands for root mean square error.

[0035] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program; the processor is used to execute the computer program in the memory to implement the above method.

[0036] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0037] The present invention proposes a method for predicting mine surface deformation by combining numerical simulation with transfer learning. The LSTM layer is placed at the front end of the model, which can more intuitively explain the patterns or changing trends in the time series, making the entire model more transparent when analyzing the contribution or causal relationship of the time series. The method can be used to predict the deformation size of the surface at different mining stages, provide a basis for safe production, and avoid casualties and property losses caused by excessive deformation and collapse of the surface. The degree to which neighboring buildings and facilities are affected by mining can be judged according to the prediction results of the method, providing a basis for determining the relocation scope and relocation time, and providing a reference for the reasonable arrangement of planned ground facilities such as production, office and life. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The present invention is a flow chart of a method for predicting mine surface deformation by combining numerical simulation with transfer learning.

[0039] Figure 2 is the surface deformation prediction result of the first segment excavation in the embodiment of the present invention, (a) is the loss of the training set and the test set, (b) is the comparison between the predicted value and the true value;

[0040] Figure 3 is the surface deformation prediction result of the second segmented excavation in the embodiment of the present invention, (a) is the loss of the training set and the test set, (b) is the comparison between the predicted value and the true value;

[0041] Figure 4 1 is the surface deformation prediction result of the third segmented excavation in the embodiment of the present invention, (a) is the loss of the training set and the test set, (b) is the comparison between the predicted value and the true value;

[0042] Figure 5 It is the surface deformation prediction result of the fourth segmented excavation in the embodiment of the present invention, (a) is the loss of the training set and the test set, and (b) is the comparison between the predicted value and the true value. DETAILED DESCRIPTION

[0043] The following embodiments of the present invention are described in further detail in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0044] Example 1: This example provides a method for predicting mine surface deformation by combining numerical simulation with transfer learning. Figure 1 As shown, the following steps are included:

[0045] S1. Numerical simulation of ore body mining is performed based on the geological data of the mine site, and the surface deformation values ​​calculated by numerical simulation are preprocessed to obtain source domain data;

[0046] S2. Calculate the actual surface deformation value of the mining area based on the basic principle of probability integral method according to the geological data of the mining site to obtain the target domain data;

[0047] S3. Construct a surface subsidence prediction model based on numerical simulation and transfer learning, which combines the long short-term memory neural network LSTM and the deep convolutional neural network DCNN. The surface subsidence prediction model includes LSTM layer, reshape layer and DCNN layer.

[0048] S4. Using the source domain data obtained in step S1 as the source domain for transfer learning of the surface subsidence prediction model, inputting the source domain data into the surface subsidence prediction model constructed in step S3 to pre-train the surface subsidence prediction model, thereby obtaining a pre-trained surface subsidence prediction model;

[0049] S5. Add a fully connected layer to the pre-trained surface subsidence prediction model obtained in step S4, freeze the parameters of each layer of the pre-trained surface subsidence prediction model, obtain a frozen surface subsidence prediction model, use the target domain data obtained in step S2 as the target domain for transfer learning of the frozen surface subsidence prediction model and re-train, fine-tune the parameters of the frozen surface subsidence prediction model and update the weights to obtain a prediction model;

[0050] S6. Input the mine site geological data to be predicted into the prediction model obtained in step S5 to perform prediction and obtain a prediction result.

[0051] Embodiment 2: This embodiment provides a method for predicting mine surface deformation by combining numerical simulation with transfer learning, comprising the following steps:

[0052] S1. Numerical simulation of ore body mining is performed based on the geological data of the mine site, and the surface deformation values ​​calculated by numerical simulation are preprocessed to obtain source domain data;

[0053] As a preferred embodiment of this invention, specifically, step S1 includes the following steps:

[0054] S101. Scan the mine site by drone to obtain the real-scene 3D visualization point cloud data of the mine area, establish the initial mine model based on the real-scene 3D visualization point cloud data of the mine area, determine the spatial posture and spatial position of the fault according to the mine exploration report and construct a solid surface for cutting, cut the initial mine model in turn through Boolean operation to obtain a geological database;

[0055] S102. Based on the core photos taken from the mine site, the Mask-RCNN deep learning neural network model is used to identify and obtain the RQD values ​​of each segment of different boreholes. According to the lithology description in the borehole column chart, the lithology and weathering conditions of each segment of different boreholes are determined. The RMR values ​​of each segment of different boreholes are calculated in combination with the RQD values ​​of each segment of different boreholes. The mechanical parameters of the spatial variability of the rock mass are calculated according to the Hoek-Brown strength criterion and imported into the geological database to obtain the rock mechanical parameters of the actual geological conditions of the mining area;

[0056] S103. According to the rock mechanical parameters under the same working conditions, 9 points were selected as numerical simulation monitoring points with the target point as the center and 1m intervals in front, back, left and right. The ore body was excavated and backfilled in 16 steps, with 4 steps as a segment. The surface settlement data of each excavation monitoring point was extracted to obtain the surface deformation curve. The surface deformation value was obtained based on the surface deformation curve, and the surface deformation value was preprocessed to obtain the source domain data.

[0057] S2. Calculate the actual surface deformation value of the mining area based on the basic principle of probability integral method according to the geological data of the mining site to obtain the target domain data;

[0058] As a preferred embodiment of this embodiment, specifically, the surface subsidence coefficient, horizontal movement coefficient, main influence angle tangent, mining influence propagation angle and turning point offset are determined according to the geological data of the mine site, and the surface inclination, curvature and horizontal deformation of the mining area are calculated through the basic principle of probability integration method to obtain the actual surface deformation value of the mining area. Specifically, a MATLAB program suitable for underground mining of metal mines can be compiled to calculate the surface deformation value, and to carry out calculation and analysis of the surface inclination, curvature and horizontal deformation.

[0059] S3. Construct a surface subsidence prediction model based on numerical simulation and transfer learning, which combines a long short-term memory neural network LSTM and a deep convolutional neural network DCNN. The surface subsidence prediction model includes an LSTM layer, a reshape layer, and a DCNN layer. In this embodiment, the LSTM layer is placed in front to directly output the feature vector or hidden state of the time series. These features can more intuitively explain the pattern or change trend in the time series. Compared with CNN-LSTM, the surface subsidence prediction model of this embodiment is more transparent when analyzing the contribution or causal relationship of the time series.

[0060] As the preferred embodiment of the present invention, specifically, the surface subsidence prediction model includes three LSTM layers, one reshape layer and one DCNN layer. The three LSTM layers are used as encoders to accurately extract time series features using their memory characteristics in time series to extract global time series features and deeper abstract features of source domain data. The reshape layer is used to change the array structure and convert the input of the DCNN layer into three-channel pixel matrix data. The DCNN layer is used as a decoder to obtain local features of the extracted data through the use of convolutional layers and pooling layers by means of local connections and shared weights, and is used for underlying expression and extraction of spatial feature information of the reconstructed space, and is used to extract features of pixel matrix data in space and locally, and output predicted surface deformation value data. The network structure of the LSTM layer is specifically used to process time series data. It is good at capturing long-term and short-term dependencies in data. When the data has a strong time dependency, placing the LSTM layer at the front end of the model can effectively extract global time series features, and then these features will be passed to the CNN layer to extract spatial or local patterns, making the entire model more suitable for processing such tasks. The memory mechanism of the LSTM layer can effectively filter out the noise and non-stationarity in the time series data, smooth or normalize the extracted time series features, and prepare for the subsequent CNN layer input. Especially when there is a lot of noise in the original time series signal, this preprocessing can reduce the burden of the CNN layer in processing complex time series. Therefore, placing the LSTM layer at the front end of the model can compress high-dimensional time series data into low-dimensional features through time series modeling, so that the subsequent CNN layer only needs to process the features extracted by the LSTM layer without directly processing the complex original data, thereby reducing the computational complexity of the CNN layer.

[0061] Specifically, the DCNN layer includes the first convolutional layer, the first maximum pooling layer, the second convolutional layer, the second maximum pooling layer, the third convolutional layer, the third maximum pooling layer, the flattening layer, the first fully connected layer, the first drop layer, the second fully connected layer, the second drop layer, the third fully connected layer and the output layer, which are stacked in sequence. The activation functions of the first convolutional layer, the second convolutional layer, the third convolutional layer, the first fully connected layer and the second fully connected layer are all ReLU activation functions, and the activation function of the third fully connected layer is Softmax activation function. The drop rates of the first drop layer and the second drop layer are both 0.5. The data size of the output of the third convolutional layer is 13×16×32. The specific structure of the surface subsidence prediction model is shown in Table 1:

[0062] Table 1 Structure of surface subsidence prediction model

[0063]

[0064] S4. Using the source domain data obtained in step S1 as the source domain for transfer learning of the surface subsidence prediction model, inputting the source domain data into the surface subsidence prediction model constructed in step S3 to pre-train the surface subsidence prediction model, thereby obtaining a pre-trained surface subsidence prediction model;

[0065] Specifically, the source domain data, specifically 9 numerical simulation monitoring data, are first input into the surface subsidence prediction model. The model first contains three LSTM layers, which are responsible for processing time series data and finally output a data of size 100×128. Next, the data passes through a reshape layer, which transforms the data structure so that the input of the DCNN layer is formatted as a three-channel pixel matrix data. Subsequently, the data is processed by three convolutional layers and one maximum pooling layer. The convolutional layer uses the ReLU activation function to identify the extracted features, which are then input into the maximum pooling layer. After this series of operations, the output data size becomes 13×16×32. After that, the data passes through a flattening layer to convert the multidimensional data into one dimension so that it can be input into the fully connected layer. Finally, the data is processed by three fully connected layers. In the input of the fully connected layer, a small dropout layer is added, and its dropout rate is set to 0.5 to prevent overfitting. After processing by these layers, the surface subsidence prediction model finally outputs the surface deformation value data.

[0066] S5. Add a fully connected layer to the pre-trained surface settlement prediction model obtained in step S4. The fully connected layer is used as the core part of model fine-tuning to adapt to the specific task of surface deformation prediction, and further transform the general features extracted by the pre-trained surface settlement prediction model into task features that are highly related to surface deformation prediction. Freeze the parameters of each layer of the pre-trained surface settlement prediction model, set all layers of the pre-trained surface settlement prediction model to a non-trainable state, avoid updating these parameters during the fine-tuning process, protect the features learned by the pre-trained surface settlement prediction model through large-scale data set training, ensure the stability of the pre-trained surface settlement prediction model during the transfer learning process, set the fully connected layer to a trainable state, obtain a frozen surface settlement prediction model, use the target domain data obtained in step S2 as the target domain for transfer learning of the frozen surface settlement prediction model and retrain, fine-tune the parameters of the fully connected layer of the frozen surface settlement prediction model and update the weights, the learning ability of the entire model is mainly concentrated on the newly added fully connected layer to reduce the computational cost and effectively avoid overfitting, and obtain the prediction model;

[0067] As a preferred embodiment of this invention, specifically, the surface settlement value predicted by the model is evaluated against the true value for loss, so as to evaluate the accuracy of the model in predicting surface deformation during underground mining. The specific calculation of the loss is completed through the loss function. The loss function accepts the predicted value and the true value of the model as input, and outputs a scalar value, namely the loss value, which represents the overall prediction error of the model on the entire data set. By plotting the changes of the training loss and the verification loss with the number of iterations, the changing trend of the loss value of the model during the training process can be intuitively presented, as well as whether problems such as overfitting or underfitting are encountered, so as to adjust the model structure and training strategy. The loss functions used include mean square error, mean absolute percentage error, mean absolute error and root mean square error;

[0068] The mean square error is shown in formula (1):

[0069]

[0070] Among them, MSE represents mean square error, N represents the number of data, and y i Represents the predicted value, y i ′ represents the true value;

[0071] The mean absolute percentage error is shown in formula (2):

[0072]

[0073] Among them, MAPE means mean absolute percentage error;

[0074] The mean absolute error is shown in formula (3):

[0075]

[0076] Among them, MAE means mean absolute error;

[0077] The root mean square error is shown in formula (4):

[0078]

[0079] Here, RMSE stands for root mean square error.

[0080] S6. Input the mine site geological data to be predicted into the prediction model obtained in step S5 to perform prediction and obtain the predicted surface settlement value result.

[0081] This embodiment selects the center of surface deformation movement as the target domain for verifying the prediction accuracy of the model based on the geological data of the mine site. With the target point as the center, 9 points are selected as numerical simulation monitoring points with an interval of 1m in front, back, left and right. The ore body is excavated and backfilled in 16 steps, and each 4 steps is a segment, including the first segment, the second segment, the third segment and the fourth segment. The surface settlement data of each excavation monitoring point is extracted to obtain the surface deformation curve. The 9 numerical simulation monitoring data are input into the model, iterative training is performed and the pre-trained model is saved. A new fully connected layer is added on top of the pre-trained model as the main part of fine-tuning to adapt to the new task. Freeze all parameters of the pre-trained model, use the theoretical calculation results of surface deformation as a small amount of new data sets for fine-tuning, and only train the added new layers to avoid destroying the features learned in the pre-trained model during the fine-tuning process. Finally, test and verify to evaluate the performance of the fine-tuned model. The prediction results are as follows Figure 2-Figure 5 As shown in Table 2, the error values ​​increased with the increase of ore output, but the MSE was less than 0.04, indicating that the model has a high accuracy in predicting surface deformation in underground mining, which can provide ideas for the study of similar mine-related issues. The ore output and prediction error values ​​of each segment excavation are shown in Table 2:

[0082] Table 2 Ore output and prediction error values ​​of each segment excavation

[0083]

[0084] Based on the above embodiments, an embodiment of the present application also provides an electronic device, which includes: one or more processors, a memory, and one or more programs; wherein the one or more programs are stored in the memory, and the one or more programs include instructions, and when the instructions are executed by the electronic device, the electronic device executes the method provided in the above embodiments.

[0085] Based on the above embodiments, the embodiments of the present application further provide a computer storage medium, in which a computer program is stored. When the computer program is executed by a computer, the computer executes the method provided in the above embodiments.

[0086] The storage medium may be any available medium that can be accessed by a computer. For example, but not limited to, a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer.

[0087] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0088] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0089] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0090] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0091] The embodiments of the present invention are given for the purpose of illustration and description, and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for specific uses.

Claims

1. A mine surface deformation prediction method combining numerical simulation and transfer learning, characterized in that: The steps include: S1. Numerical simulation of ore body mining is performed based on the geological data of the mine site, and the surface deformation values ​​calculated by numerical simulation are preprocessed to obtain source domain data; S2. Calculate the actual surface deformation value of the mining area based on the basic principle of probability integral method according to the geological data of the mining site to obtain the target domain data; S3. Constructing a surface subsidence prediction model based on numerical simulation and transfer learning combined with a long short-term memory neural network LSTM and a deep convolutional neural network DCNN, wherein the surface subsidence prediction model includes an LSTM layer, a reshape layer and a DCNN layer; S4. Using the source domain data obtained in step S1 as the source domain for transfer learning of the surface subsidence prediction model, inputting the source domain data into the surface subsidence prediction model constructed in step S3 to pre-train the surface subsidence prediction model, thereby obtaining a pre-trained surface subsidence prediction model; S5. Add a fully connected layer to the pre-trained surface subsidence prediction model obtained in step S4, set each layer of the pre-trained surface subsidence prediction model to a non-trainable state, set the fully connected layer to a trainable state, obtain a frozen surface subsidence prediction model, use the target domain data obtained in step S2 as the target domain for transfer learning of the frozen surface subsidence prediction model and re-train, fine-tune the parameters of the frozen surface subsidence prediction model and update the weights to obtain a prediction model; S6. Input the mine site geological data to be predicted into the prediction model obtained in step S5 to perform prediction and obtain the predicted surface settlement value result.

2. The method for predicting mine surface deformation by combining numerical simulation and transfer learning according to claim 1, characterized in that: The step S1 comprises the following steps: S101. Scanning the mine site by drone to obtain the real-scene 3D visualization point cloud data of the mine area, establishing the initial mine model based on the real-scene 3D visualization point cloud data of the mine area, determining the spatial posture and spatial position of the fault according to the mine exploration report and constructing a solid surface for cutting, cutting the initial mine model in turn by Boolean operation to obtain a geological database; S102. Based on the core photos taken from the mine site, the Mask-RCNN deep learning neural network model is used to identify and obtain the RQD values ​​of each segment of different boreholes. According to the lithology description in the borehole column chart, the lithology and weathering conditions of each segment of different boreholes are determined. The RMR values ​​of each segment of different boreholes are calculated in combination with the RQD values ​​of each segment of different boreholes. The mechanical parameters of the spatial variability of the rock mass are calculated according to the Hoek-Brown strength criterion and imported into the geological database to obtain the rock mechanical parameters of the actual geological conditions of the mining area; S103. According to the rock mechanical parameters under the same working conditions, 9 points are selected as numerical simulation monitoring points with the target point as the center and 1m intervals in front, back, left and right. The ore body is excavated and backfilled in 16 steps, with 4 steps as a segment. The surface settlement data of each excavation monitoring point is extracted to obtain the surface deformation curve. The surface deformation value is obtained according to the surface deformation curve, and the surface deformation value is preprocessed to obtain the source domain data.

3. The method for predicting mine surface deformation by combining numerical simulation and transfer learning according to claim 1 is characterized in that: The step S2 includes determining the surface subsidence coefficient, horizontal movement coefficient, main influence angle tangent, mining influence propagation angle and turning point offset according to the geological data at the mine site, and calculating the surface inclination, curvature and horizontal deformation of the mining area through the basic principle of probability integral method to obtain the actual surface deformation value of the mining area.

4. The method for predicting mine surface deformation by combining numerical simulation and transfer learning according to claim 1 is characterized in that: The surface subsidence prediction model in step S3 includes three LSTM layers, one reshape layer and one DCNN layer. The three LSTM layers are used as encoders to extract the global temporal features and deeper abstract features of the source domain data. The reshape layer is used to change the array structure and convert the input of the DCNN layer into three-channel pixel matrix data. The DCNN layer is used as a decoder to extract the features of the pixel matrix data spatially and locally through local connections and shared weights, and output the predicted surface deformation value data.

5. The method for predicting mine surface deformation by combining numerical simulation with transfer learning according to claim 4 is characterized in that: The DCNN layer includes a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer, a third convolutional layer, a third maximum pooling layer, a flattening layer, a first fully connected layer, a first drop layer, a second fully connected layer, a second drop layer, a third fully connected layer and an output layer stacked in sequence, the activation functions of the first convolutional layer, the second convolutional layer, the third convolutional layer, the first fully connected layer and the second fully connected layer are all ReLU activation functions, and the activation function of the third fully connected layer is the Softmax activation function.

6. The method for predicting mine surface deformation by combining numerical simulation with transfer learning according to claim 5, characterized in that: The discarding rates of the first discarding layer and the second discarding layer are both 0.

5.

7. The method for predicting mine surface deformation by combining numerical simulation and transfer learning according to claim 5, characterized in that: The data size of the output of the third convolutional layer is 13×16×32.

8. The method for predicting mine surface deformation by combining numerical simulation and transfer learning according to claim 1, characterized in that: In step S5, the loss functions used include mean square error, mean absolute percentage error, mean absolute error and root mean square error; The mean square error is shown in formula (1): Among them, MSE represents mean square error, N represents the number of data, and y i Represents the predicted value, y i ′ represents the true value; The mean absolute percentage error is shown in formula (2): Among them, MAPE means mean absolute percentage error; The mean absolute error is shown in formula (3): Among them, MAE means mean absolute error; The root mean square error is shown in formula (4): Here, RMSE stands for root mean square error.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory stores a computer program; and the processor is used to execute the computer program in the memory to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images