CN-121999340-A - Ore contour prediction model training method, ore contour prediction method and device

CN121999340ACN 121999340 ACN121999340 ACN 121999340ACN-121999340-A

Abstract

The application discloses a training method of a ore body contour prediction model, a method and a device for ore body contour prediction, and relates to the technical field of intelligent mines and the technical field of artificial intelligence, wherein training sample data for training the ore body contour prediction model is obtained, the training sample data comprises continuous N layered three-dimensional voxel model data of a mine, the continuous N layered three-dimensional voxel model data are used as training sample input data, and the ore body contour actual measurement data of the next layer of the N layered ore body contour actual measurement data; the method comprises the steps of inputting a ore body contour prediction model to be trained, obtaining output ore body contour prediction data of the next layer of N layers, determining whether the model training meets a convergence condition or not based on the obtained ore body contour prediction data and ore body contour actual measurement data, determining that the training of the ore body contour prediction model is completed if the model training meets the convergence condition, and adjusting model parameters of the ore body contour prediction model if the model training does not meet the convergence condition. By adopting the scheme, the ore body contour prediction model capable of effectively and accurately predicting the ore body contour is obtained.

Inventors

JING BOWEN
LV ZHIFENG
DING ZHE
NIE YANKAI
ZHANG JUNYOU

Assignees

北京爱熵科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (11)

1. The ore body contour prediction model training method is characterized by comprising the following steps of: acquiring training sample data for training a mineral body contour prediction model, wherein the training sample data comprises continuous N layered three-dimensional voxel model data of a mine as training sample input data, the three-dimensional voxel model data can represent mineral body distribution conditions of the N layered mineral bodies, and the training sample data also comprises mineral body contour actual measurement data of a next layer of the N layered mineral bodies; Inputting the training samples into data, inputting the ore body contour prediction model to be trained, and obtaining the output ore body contour prediction data of the next layer of the N layers; determining whether the model training meets a convergence condition or not based on the obtained ore body contour prediction data and the ore body contour actual measurement data; And if the convergence condition is met, determining that the training of the ore body contour prediction model is completed, and if the convergence condition is not met, adjusting model parameters of the ore body contour prediction model.
2. The method of claim 1, wherein the N layered three-dimensional voxel model data comprises voxel attribute structured-coding data representing an actual distribution of the N layered ore bodies, the voxel attribute structured-coding data representing the actual distribution of the N layered ore bodies being generated based on the N layered measured drawing data and geological survey data of the mine, or The three-dimensional voxel model data of the N layers comprises voxel attribute structured coding data representing actual distribution conditions of the N layers of ore bodies and voxel attribute structured coding data representing design distribution conditions of the N layers of ore bodies, the voxel attribute structured coding data representing the actual distribution conditions of the N layers of ore bodies is generated based on actual measurement drawing data and geological exploration data of the N layers of ore bodies of the mine, and the voxel attribute structured coding data representing the design distribution conditions of the N layers of ore bodies of the mine is generated based on the design drawing data of the N layers of ore bodies of the mine.
3. The method of claim 2, wherein the training sample input data further comprises voxel attribute structured coding data representing a ore body design distribution of the N next layers, the voxel attribute structured coding data representing the ore body design distribution of the N next layers being generated based on design drawing data of the N next layers of the mine.
4. A method according to claim 2 or 3, wherein the voxel properties of each voxel structure encoded data, comprising the following properties: a ore body contour tag that indicates whether the voxel is within an ore body contour; A lane position code indicating whether the voxel is located within a lane; Lithologic classification codes indicating which lithologic region the voxel is located in.
5. The method of claim 1, wherein the ore body contour prediction model is a convolutional neural network model; the ore body contour prediction model comprises a 3D convolution layer, a 3D maximum pooling layer and a 3D convolution LSTM layer which are sequentially connected; Wherein the 3D convolution layer receives input training sample input data; the 3D convolution LSTM layer outputs ore body contour prediction data of the N layered next layers.
6. The method of claim 1, wherein the determining whether the model training satisfies a convergence condition based on the obtained ore body contour prediction data and the ore body contour measured data comprises: based on the obtained ore body contour prediction data and the ore body contour actual measurement data, calculating a joint loss function by adopting the following formula: ; ; ; Wherein, the As a result of the calculation of the joint loss function, As a result of the computation of the Dice loss function term, For the calculation result of the regularized term, For the weights of the Dice loss function term, For the weights of the regularized terms, The measured data of the ith three-dimensional voxel in the measured data of the ore body profile of the next layer of the N layers, Prediction data for an ith three-dimensional voxel in the ore body contour prediction data of the next layer of the N layers, M is a total number of three-dimensional voxels of the next layer of the N layers, Is a preset smoothing term for preventing zero removal, For the regularization coefficient(s), The weight of the jth model parameter of the ore body contour prediction model is given, and m is the total number of model parameters of the ore body contour prediction model; and determining whether the model training meets a convergence condition or not based on the calculation result of the joint loss function.
7. A method for ore body contour prediction, comprising: Acquiring three-dimensional voxel model data of continuous N layering above layering to be predicted of a mine; Based on the obtained three-dimensional voxel model data, the ore body contour prediction model obtained by training by the method of any one of claims 1-6 is adopted to predict the ore body contour of the to-be-predicted layering of the mine, so as to obtain the ore body contour prediction data of the to-be-predicted layering.
8. A ore body contour prediction model training device, characterized by comprising: The training data acquisition module is used for acquiring training sample data for training a ore body contour prediction model, the training sample data comprises continuous N layered three-dimensional voxel model data of a mine as training sample input data, the three-dimensional voxel model data can represent ore body distribution conditions of the N layered ore bodies, and the training sample data also comprises ore body contour actual measurement data of the next layer of the N layered ore bodies; The training data processing module is used for inputting the training samples into data, inputting the ore body contour prediction model to be trained, and obtaining the output ore body contour prediction data of the next layer of the N layers; the convergence condition judging module is used for determining whether the model training meets a convergence condition or not based on the obtained ore body contour prediction data and the ore body contour actual measurement data; And the model training module is used for determining that the training of the ore body contour prediction model is finished if the convergence condition is met, and adjusting the model parameters of the ore body contour prediction model if the convergence condition is not met.
9. A seam contour prediction apparatus, comprising: The characteristic data acquisition module is used for acquiring three-dimensional voxel model data of continuous N layers above the layer to be predicted of the mine; And the ore body contour prediction module is used for predicting the ore body contour of the to-be-predicted layering of the mine by training the obtained ore body contour prediction model by adopting the method of any one of claims 1-6 based on the obtained three-dimensional voxel model data to obtain ore body contour prediction data of the to-be-predicted layering.
10. An electronic device comprising a processor and a machine-readable storage medium storing machine-executable instructions executable by the processor to cause the processor to perform the method of any one of claims 1-6 or to perform the method of claim 7.
11. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program which, when executed by a processor, implements the method of any of claims 1-6 or implements the method of claim 7.

Description

Ore contour prediction model training method, ore contour prediction method and device Technical Field The application relates to the technical field of intelligent mines and artificial intelligence, in particular to a training method of a ore body contour prediction model, a method and a device for predicting an ore body contour. Background Downhole mining is a complex process that typically involves a number of critical steps. In this series of processes, geologic modeling is the basis for the entire underground mine production, which directly affects the subsequent layered design, production techniques, and production scheme formulation. In actual underground mining, the method generally comprises the following steps of firstly drilling holes to obtain preliminary information of a mine body, then carrying out geological modeling based on drilling hole data to determine the outline of the mine body, layering underground according to the outline of the mine body, designing mining facilities such as a roadway and a drop shaft, and finally carrying out actual construction. However, this process has significant limitations, especially in the actual construction process, due to the deviation between the actual distribution and trend of the ore body and the design, the design drawing needs to be continuously adjusted to obtain the ore body profile of the current drawing. Along with the construction, the upper-layered roadway design drawing and the current-layered roadway design drawing are continuously corrected, the contour error of the ore body of the lower-layered roadway design drawing is larger and larger, the adjustment of the ore body contour of the design drawing is more and more complex, and the workload is larger and larger. Conventional geologic modeling methods typically construct three-dimensional models of ore bodies based on borehole data and geologic profiles, either by contour stitching or by volumetric data isosurface. Conventional geologic modeling methods typically rely on manual experience and judgment, which not only increases the effort, but may also lead to inaccuracy in the modeling results. Hierarchical designs in mining are an important application based on geologic modeling. The accuracy of the layered design directly affects the subsequent roadway design and mining efficiency. However, conventional hierarchical design methods are typically based on static geologic models, and it is difficult to adapt to dynamic changes in the actual distribution of ore bodies. As the construction proceeds, the actual distribution and trend of the ore body may deviate greatly from the design, which requires continuous adjustment of the design drawing to obtain the ore body profile of the current drawing. However, such adjustment processes tend to be time consuming and labor intensive, and the complexity of the adjustment increases as the construction proceeds. Disclosure of Invention The embodiment of the application provides a training method of an ore body contour prediction model, an ore body contour prediction method and an ore body contour prediction device, which are used for solving the problem of how to predict an ore body contour of underground layering of a mine in the prior art. The embodiment of the application provides a training method for a ore body contour prediction model, which comprises the following steps: acquiring training sample data for training a mineral body contour prediction model, wherein the training sample data comprises continuous N layered three-dimensional voxel model data of a mine as training sample input data, the three-dimensional voxel model data can represent mineral body distribution conditions of the N layered mineral bodies, and the training sample data also comprises mineral body contour actual measurement data of a next layer of the N layered mineral bodies; Inputting the training samples into data, inputting the ore body contour prediction model to be trained, and obtaining the output ore body contour prediction data of the next layer of the N layers; determining whether the model training meets a convergence condition or not based on the obtained ore body contour prediction data and the ore body contour actual measurement data; And if the convergence condition is met, determining that the training of the ore body contour prediction model is completed, and if the convergence condition is not met, adjusting model parameters of the ore body contour prediction model. Further, the three-dimensional voxel model data of the N layers comprises voxel attribute structured coding data representing the actual distribution situation of the ore bodies of the N layers, which are generated based on the actual measurement drawing data and the geological exploration data of the N layers of the mine, or The three-dimensional voxel model data of the N layers comprises voxel attribute structured coding data representing actual distribution conditions of the N layers of ore b