Three-dimensional geological modeling method, system and equipment based on Markov chain and improved Monte Carlo

Abstract

The invention discloses a three-dimensional geological modeling method, system and equipment based on a Markov chain and improved Monte Carlo, and belongs to the technical field of engineering rock geological modeling. On the basis that three-dimensional point cloud data of a rock mass is obtained through three-dimensional laser scanning, Markov chain transition probability matrix calculation is carried out through existing actual drilling and virtual drilling technologies, and then the stratum state of virtual drilling is confirmed based on a superimposed Markov chain. And finally, performing correlation simulation on the stratum thickness corresponding to the stratum state based on an improved Monte Carlo method. Compared with a traditional artificial three-dimensional geologic modeling method, the method can be more comprehensively suitable for engineering rock three-dimensional geologic modeling scenes in various complex scenes; meanwhile, based on the automatically constructed three-dimensional geologic model, the later engineering rock mass stability can be efficiently evaluated, and important data support is provided for rock mass engineering stability evaluation; the method is easy to implement, flexible to operate and remarkable in effect, and the application requirement is met.

Description

Technical Field

[0001] The invention relates to the technical field of engineering rock mass geological modeling, and in particular to a three-dimensional geological modeling method, system and equipment based on Markov chain and improved Monte Carlo. Background Art

[0002] Geological modeling is a significant engineering task for geological survey and other work. It provides data basis for a series of work such as construction design of engineering rock mass, geological body estimation and safety and stability assessment. Traditional geological modeling is artificially constructed based on on-site drilling core data and geological profile data. However, in the face of complex geological conditions and less geological data, the geological model constructed based on the existing geological data often cannot truly reflect the geological information of the project. Since the existing geological modeling methods are overly dependent on geological drilling data, how to use limited drilling data to obtain a refined geological model has become a technical key in current geological modeling work.

[0003] At present, traditional geological modeling uses three-dimensional interpolation technology to deduce the state and thickness of the formation. The principle of three-dimensional space interpolation technology is to construct a continuous plane area based on discrete points and a certain characteristic attribute of the point. On-site technicians often use the lithology of the borehole as a field to construct a geological model of the entire area. With the development of random modeling theory, because it can characterize the uncertainty of the formation through means similar to random simulation, the use of random modeling theory can automatically and efficiently complete complex three-dimensional geological modeling work. However, due to the complexity and variability of geological conditions, traditional random modeling theory is only applicable to simple and uniformly thick engineering geological bodies, and is not applicable to engineering geological bodies with complex formation states and uneven formation thickness. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention aims to provide a three-dimensional geological modeling method, system and equipment based on Markov chain and improved Monte Carlo, which obtains point cloud data of engineering rock mass through unmanned aerial vehicle and three-dimensional laser scanning technology, and downsamples the point cloud data, selects virtual boreholes based on the point cloud data, predicts the stratum state based on the superimposed Markov chain, and then simulates the stratum thickness corresponding to the stratum state of each virtual borehole based on Monte Carlo, and finally determines the final three-dimensional geological model based on the maximum probability model selection principle.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0006] The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo includes the following steps:

[0007] S1: Acquire the three-dimensional point cloud data of the construction engineering geological body area;

[0008] S2: Based on the formation state and formation thickness data of the actual drilling in the construction area, the Markov chain is used to construct the formation state transition probability matrix, and the Monte Carlo method is used to determine the probability distribution model of the formation thickness;

[0009] S3: Determine the initial ground state of the virtual borehole through photogrammetry and lithology identification;

[0010] S4: Simulation of the formation state of the virtual borehole based on superimposed Markov chain;

[0011] S5: Determine the maximum probability attribute of the virtual borehole stratum, calculate the stratum boundary line based on the cumulative probability distribution model curve of the stratum state, and finally correct the stratum boundary line by setting verification boreholes to obtain the final geological model.

[0012] Furthermore, the specific operation of step S2 includes the following steps:

[0013] S201: Using Markov chain to construct the formation state transition probability matrix of real borehole;

[0014] S202: constructing a probability distribution model of the formation thickness of the actual borehole;

[0015] S203: Generate virtual borehole data using the point cloud data, and construct a formation state transition probability matrix of the virtual borehole.

[0016] Furthermore, the formation state transition probability matrix of the actual borehole in step S201 is expressed as

[0017]

[0018] Where P ij represents the probability of stratum i mutating into stratum j;

[0019] The calculation formula of the transition probability matrix of the virtual drilling in step S203 is:

[0020]

[0021] In the formula, represents the transition probability matrix of virtual drilling, β i represents the distance weight, Transition probability matrix representing the actual drilling.

[0022] Furthermore, the specific operation of the superimposed Markov chain described in step S401 is to adjust the order of the transfer probability matrix according to the formation state. The order of the transfer probability matrix is ​​calculated based on the number of formation state segments of the virtual borehole obtained by statistics. The order calculation formula of the transfer probability matrix is:

[0023]

[0024] In the formula, represents the probability of transitioning from formation state i to formation state j at s steps, represents the number of transitions from formation state i to formation state j at step s, M i is the total number of samples in stratum i.

[0025] Furthermore, the specific operation of step S5 includes the following steps:

[0026] S501: Calculating the formation state probability determined by superimposing the Markov chain in step S4;

[0027] S502: Select the simulation result corresponding to the maximum probability of the formation state, and determine the formation state of the virtual borehole elevation h;

[0028] S503: confirming the boundary corresponding to each stratum state of the virtual borehole;

[0029] S504: Correcting the boundary corresponding to each stratum state by verifying the drilling hole;

[0030] S505: Establish a three-dimensional geological model according to the stratum states and stratum state boundaries of the real borehole and the virtual borehole.

[0031] Further, in step S501, the probability of formation state i at elevation h is expressed as

[0032]

[0033] Where y is the number of superimposed Markov chain simulations, x i is the number of formation state i at level h.

[0034] Furthermore, in step S3, RandMixAugment, a neural network framework for identifying lithology based on deep learning, is used to identify the stratigraphic state of the surface. RandMixAugment divides the original data into original training data and test data. The original training data is enhanced by regional enhancement and image enhancement to obtain enhanced training data. The original training data and the enhanced training data are merged for deep learning training. The training process includes WideResNet, SupCon, Shake-Shake and Pyramid-Shake. Finally, the data prediction and evaluation results of the stratigraphic state are obtained, and the network framework is tested using the test data.

[0035] Furthermore, image-level enhancement is achieved through automatic enhancement, and region-level enhancement is achieved through masking and hybrid strategies;

[0036] The masking operation first randomly generates a mask box of size S1×Ts, where Ts is the size of the original training image. Next, the generated mask box is applied to the training image at a random position. The parameter S1 ranges from [0,1]. The mask box is smaller than the original image, and the color of the mask box is determined to be the average value of the RGB color of the image to which the randomization is applied.

[0037] The blending operation first randomly generates a patch of size S2×OS from the original training image, where OS is the original training image size and S2 is the maximum size of the patch; the generated patch is added at a random position where the masking box is applied to the original training image.

[0038] Furthermore, the present invention also includes a three-dimensional geological modeling system based on Markov chain and improved Monte Carlo, which specifically includes a data acquisition module, a real borehole data modeling module, a virtual borehole modeling module and a geological model building module;

[0039] The data acquisition module, the real borehole data modeling module, the virtual borehole modeling module and the geological model building module realize three-dimensional geological modeling based on the three-dimensional geological modeling method as described above.

[0040] Furthermore, the present invention also includes an electronic device, specifically including at least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the three-dimensional geological modeling method as described above.

[0041] The beneficial effects of the present invention are:

[0042] 1. The present invention proposes a three-dimensional geological modeling method based on Markov chain and improved Monte Carlo. On the basis of obtaining three-dimensional point cloud data of rock mass by three-dimensional laser scanning, the Markov chain transfer probability matrix is ​​calculated by existing actual drilling and virtual drilling technology, and then the formation state of virtual drilling is confirmed based on the superposition of Markov chain. Finally, the formation thickness corresponding to the formation state is simulated based on the improved Monte Carlo method. Compared with the traditional artificial three-dimensional geological modeling method, this method is more comprehensively applicable to the three-dimensional geological modeling scenes of engineering rock mass in various complex scenes. At the same time, the three-dimensional geological model based on this automatic construction can efficiently evaluate the stability of engineering rock mass in the later stage, and provide important data support for the evaluation of rock mass engineering stability. The present invention is simple to implement, flexible to operate, and has significant effect, meeting the application requirements.

[0043] 2. The present invention also proposes a new neural network framework RandMixAugment based on deep learning to identify lithology to identify the stratigraphic state of the surface. This method effectively combines the inherent characteristics of regional-level and image-level enhancement techniques, implements a random expansion process in the initial stage, eliminates the problem of additional search space, and improves the training speed. In addition, the masking and mixing processes in turn improve the generalization performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flow chart of the three-dimensional geological modeling method based on Markov chain and improved Monte Carlo in the present invention;

[0045] Figure 2 This is a schematic diagram of the principle of obtaining the probability frequency of real borehole formation transfer in the present invention.

[0046] Figure 3 This is a schematic diagram of the Markov chain superposition in the present invention.

[0047] Figure 4 This is a schematic diagram of the principle of confirming the stratum boundary in the present invention.

[0048] Figure 5 This is a principle block diagram of the neural network framework RandMixAugment for identifying lithology based on deep learning in the present invention.

[0049] Figure 6 This is a diagram of surface lithology identification results in an application example of the present invention.

[0050] Figure 7 It is the cumulative probability distribution curve of formation state 4 in the application example of the present invention.

[0051] Figure 8 It is the maximum probability model diagram in the application example of the present invention.

[0052] Fig. 9 This is the final model for modifying the ZK3-ZK7 geological profile in the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0054] Embodiment 1:

[0055] Embodiment 1 provides a three-dimensional geological modeling method based on Markov chain and improved Monte Carlo, as shown in the attached Figure 1 As shown, the specific steps include:

[0056] S1: Use UAV 3D laser scanning technology to obtain 3D point cloud data of the construction engineering geological body area;

[0057] Specifically, S101: clarify the measurement purpose and requirements, including the size of the measurement area, the required spatial resolution, the accuracy requirements of the data, etc.; at this stage, factors such as laws and regulations, flight permits, and safety risk assessments must also be considered;

[0058] S102: Select a suitable drone and payload; for 3D point cloud data collection, commonly used payloads are LiDAR or high-resolution cameras;

[0059] S103: Design a flight plan based on the characteristics and requirements of the survey area, including setting parameters such as flight altitude, speed, path, and overlap, to ensure that data coverage is comprehensive and accuracy meets requirements;

[0060] S104: After the flight, the collected raw data needs to be processed and analyzed to form a usable 3D point cloud. For LiDAR data, the point cloud needs to be pre-processed by voxel downsampling;

[0061] S105: Perform quality control and verification on the generated 3D point cloud data to ensure that it meets the project requirements, and finally store the data.

[0062] Further, step S2: based on the formation state and formation thickness data of the actual boreholes in the construction area, a formation state transition probability matrix is ​​constructed using a Markov chain, and a probability distribution model of formation thickness is determined using a Monte Carlo method;

[0063] Specifically, S201: constructing a formation state transition probability matrix of a real borehole using a Markov chain;

[0064] Using Markov chain to construct the formation state transition probability matrix requires obtaining the frequency of occurrence of the formation state of each real borehole. Therefore, the lowest and highest elevation values ​​of the formation in all real drilling processes are collected, and based on the principle of linear interpolation, the highest and lowest elevation values ​​are interpolated to obtain the frequency of the formation state.

[0065] The number of interpolation segments should take into account the efficiency and accuracy of stratigraphic modeling, and should also avoid different stratigraphic layers in the interpolation segments, such as the Figure 2 As shown in Figure 2. Z is obtained based on the statistics of m+1 real drilling holes. max and Z min , and divide the geological borehole into n segments Q1, Q2, Q3, ..., Q according to the maximum and minimum difference formation state frequencies. n Each drilling segment will correspond to the stratigraphic information S1, S2, S3, etc. of the drilling area. n The stratum status corresponding to each segment of stratum information is stratum 1, stratum 2, ... or empty stratum.

[0066] The final formation state transition probability matrix of the real borehole is expressed as

[0067]

[0068] Where P ij Represents the probability that stratum i mutates into stratum j.

[0069] S202: constructing a probability distribution model of the formation thickness of the actual borehole, the normal distribution data of the formation thickness of the actual borehole is shown in Table 1 below;

[0070] Table 1 Normal distribution data of formation thickness of real cobalt holes

[0071] Formation status Expected lognormal distribution Lognormal distribution variance 1 <![CDATA[E1]]> <![CDATA[σ1]]> 2 <![CDATA[E2]]> <![CDATA[σ2]]> 3 <![CDATA[E3]]> <![CDATA

[03] ]> ... ... ... n <![CDATA[E n ]]> <![CDATA[σ n ]]>

[0072] The normal distribution curve of the formation thickness is determined according to the expectation and variance. The Monte Carlo method is used to randomly sample on the normal distribution curve. The value corresponding to the sampling point is the thickness of the corresponding formation.

[0073] S203: Generate virtual borehole data using the point cloud data, and construct a formation state transition probability matrix of the virtual borehole;

[0074] Specifically, the virtual drilling is based on the construction area, and the three-dimensional point cloud of the regional surface is obtained based on the UAV oblique photogrammetry, and then the point cloud body generated after preprocessing is used as the virtual drilling of the area.

[0075] The Chebyshev distance weight is calculated according to the location information of the virtual borehole, and then the transition probability matrix corresponding to each real borehole is weighted and superimposed, so that the transition probability matrix of all virtual boreholes is obtained.

[0076] The Chebyshev distance weight reasonably avoids the problem of excessive distance weight calculation. It effectively reduces the probability matrix error generated by calculating the distance weight by calculating the maximum value of XC, YC and ZC. In order to accurately express the transfer probability matrix at the three-dimensional virtual drilling position, this method chooses the Chebyshev distance weighted form (Chebyshev distance). The Chebyshev distance is given at two points P = (p1, p2, ... p n ) and Q=(q1,q2,……q n ) In n-dimensional space, its Chebyshev distance can be expressed as Dchebyshev(P,Q)=max|p i -q i |, which is the distance weight β iThis formula represents the maximum value of two points in any dimension, which can solve the problem that the distance formula at high latitudes ignores the influence of a certain dimension on the weight coefficient. Finally, based on the linear superposition of the weight coefficient of each actual borehole and the corresponding transfer probability matrix, the final transfer probability matrix of the virtual borehole can be obtained. The calculation formula of the transfer probability matrix of the virtual borehole is shown in formula (2):

[0077]

[0078] In the formula, represents the transition probability matrix of virtual drilling, β i represents the distance weight, Transition probability matrix representing the actual drilling.

[0079] Further, step S3: determining the initial ground state of the virtual borehole through photogrammetry and lithology identification;

[0080] Specifically, this step can be determined by existing surface lithology identification methods, which will not be described in detail in this application.

[0081] Further, step S4: simulating the formation state of the virtual borehole based on the superimposed Markov chain;

[0082] Specifically, based on the superimposed Markov chain, the formation state of the virtual borehole is simulated starting from the surface;

[0083] Since the tth layer of the Markov chain is only related to the previous layer t-1, the Markov chain has no memory, which also shows that the previous and next states are independent of each other. However, there is a possibility of sudden change in the formation state. For example, 1, 1, 1, 1, ..., 2, 1, 1, ... (0, 1, 2 represent the formation state) appears in the formation state. If the Monte Carlo simulation is used in this case, it will cause the system to ignore 2 or ignore all 1s after 2 based on the principle of high probability in the previous step, so it is difficult to get the correct result in the end. Therefore, in the present invention, the superposition Markov chain principle is proposed to solve the problem of instability of prediction accuracy in traditional Markov chain speculation geological modeling.

[0084] The principle of superimposed Markov chain is to deal with complex geological environments by adjusting the order (step size) of the transfer probability matrix. The order of the transfer probability matrix is ​​calculated based on the number of formation state segments of the virtual borehole obtained by statistics, as shown in the following formula (4).

[0085]

[0086] In the formula, represents the probability of transitioning from formation state i to formation state j at s steps, represents the number of transitions from formation state i to formation state j at step s, M i is the total number of samples in stratum i.

[0087] For example, when the formation state of a certain borehole core is complex and changeable, the order of the probability transfer matrix should be increased, as shown in formula (5), which are the 1st, 2nd, and mth order state transfer probability matrices respectively; by increasing the order of the probability transfer matrix to cope with the complex and changeable formation state, is the probability that the formation state n is converted to itself through m steps; thus, the accuracy of Markov chain simulation of formation sequence can be improved by adjusting the step size factor to cope with different formation state complexities;

[0088]

[0089] The principle of superposition Markov chain theory is as follows Figure 3 As shown, when the core formation state of a real borehole or a virtual borehole is complex and changeable, a probability transfer matrix of multiple orders (step lengths) needs to be taken into consideration when calculating the formation state of the kth layer.

[0090] Further, step S5: determine the maximum probability attribute of the virtual borehole stratum, calculate the stratum boundary line based on the cumulative probability distribution model curve of the stratum state, and finally correct the stratum boundary line by setting a verification borehole to obtain the final geological model.

[0091] Specifically, S501: calculating the formation state probability determined by superimposing the Markov chain in step S4;

[0092] In the present invention, the probability of the formation state is calculated according to the geological probability attribute formula, and the probability of the formation state i at the elevation h is expressed as

[0093]

[0094] Where y is the number of superimposed Markov chain simulations, x i is the number of formation state i at level h;

[0095] S502: Select the simulation result corresponding to the maximum probability of the formation state, and determine the formation state of the virtual borehole elevation h;

[0096] S503: confirming the boundary corresponding to each stratum state of the virtual borehole;

[0097] The stratigraphic boundary line is obtained by solving the cumulative normal distribution probability model of each stratum thickness, and the corresponding probability is calculated by the cumulative curve of the model through the stratigraphic boundary line of the actual borehole. Figure 4As shown, taking ZK1 as an example, assuming that the stratigraphic boundary line of stratum 1 is calculated, the thickness of stratum 1 is substituted into the cumulative probability distribution curve of stratum 1, and the probability Pd1 of stratum 1 of the real borehole ZK1 can be obtained. Other virtual boreholes and their corresponding stratum thicknesses are calculated according to this method.

[0098] S504: Correcting the boundary corresponding to each stratum state by verifying the drilling hole;

[0099] S505: Establish a three-dimensional geological model according to the stratum states and stratum state boundaries of the real borehole and the virtual borehole.

[0100] Embodiment 2:

[0101] Based on the first embodiment, the second embodiment provides a method for quickly identifying the initial stratum state of the virtual borehole surface through photogrammetric images. Specifically, the stratum state of the surface is identified through a new neural network framework RandMixAugment based on deep learning to identify lithology, as shown in the attached figure. Figure 5 As shown in Figure 1, RandMixAugment uses automatic enhancement with masking and mixing operations, which can effectively combine the inherent characteristics of region-level enhancement and image-level enhancement.

[0102] Specifically, RandMixAugment divides the original data into original training data and test data. The original training data is enhanced by region-level enhancement and image-level enhancement to obtain enhanced training data. The original training data and the enhanced training data are merged for deep learning training. The training process includes wide residual module (increasing the number of feature channels by adding more convolution kernels and accelerating the convergence speed of the model to a certain extent), SupCon (loss function, which can improve the prediction accuracy by simply replacing the cross entropy), Shake-Shake (data enhancement data, interpolating two data in the feature space to synthesize a lot of meaningful augmented data) and Pyramid-Shake (same meaning as Shake-Shake). Finally, the data prediction and evaluation results of the formation state are obtained, and the network framework is tested using test data.

[0103] Among them, the automatic enhancement step corresponds to image-level enhancement, and the masking and blending strategy represents region-level enhancement. Two key parameters S1 and S2 are introduced in the masking and blending strategy, where the S1 parameter represents the maximum size of the masking box in the masking operation and the S2 parameter represents the maximum size of the patch created in the blending operation.

[0104] The specific process of automatic enhancement is: apply RandAugment(N,M) to the training image, and use grid search to customize the enhancement strategy of the model and dataset to remove the additional search space. The RandAugment method eliminates the separate search space using two grid search parameters (N,M), which specify the number of enhancements to be applied and their size. After the automatic enhancement operation, the image-level enhanced data is obtained.

[0105] The masking operation first randomly generates a mask box of size S1×Ts, where Ts is the size of the original training image. Next, the generated mask box is applied to the training image at a random position. The value range of parameter S1 is [0,1]. The generated mask box is smaller than the original image. At the same time, the color of the mask box is determined as the average value of the RGB color of the image to which the randomization is applied.

[0106] The blending operation first randomly generates a patch of size equivalent to S2×OS from the original training image, where OS is the original training image size and S2 is the maximum size of the patch; created from the original image; similar to parameter S1, the value of parameter S2 is restricted to the interval [0,1]. Therefore, the generated synthetic patch will be relatively smaller than the original image. Finally, the generated patch is added at a random position where the masking box is applied to the original training image. The region-level enhanced image is obtained after masking and blending operations.

[0107] The proposed method effectively combines the inherent properties of region-level and image-level enhancement techniques, implements a random augmentation process in the initial stage, eliminates the problem of additional search space, and improves the training speed. Moreover, the masking and mixing process in turn improves the generalization performance.

[0108] Embodiment three:

[0109] Based on the second embodiment, the present invention also provides a three-dimensional geological modeling system based on Markov chain and improved Monte Carlo.

[0110] Embodiment 4:

[0111] On the basis of the second embodiment, the present invention also provides a three-dimensional geological modeling device based on Markov chain and improved Monte Carlo.

[0112] Application examples:

[0113] The three-dimensional geological modeling method of the present invention is applied to an open-pit iron mine in Liaoning Province, China. According to the drill core and exploration line profile, it can be analyzed that there are 6 types of stratigraphic states in the area. The basic information of the actual boreholes is shown in Table 2 below.

[0114] Table 2 Basic information of real drilling

[0115]

[0116] As can be seen from Table 2, due to the complex surface undulation, engineering geological boreholes all have a certain inclination angle. Therefore, it is necessary to firstly process the real boreholes with the median inclination theorem, and perform the median theorem equivalent processing on the X and Y of all real boreholes, thus avoiding the influence of the borehole inclination angle on the formation modeling error.

[0117] Constructing the formation state transition probability matrix specifically includes the following steps:

[0118] (1) There are 6 stratigraphic states (named 1 to 6) in the geological model constructed this time. The stratigraphic thickness and stratigraphic state of all real boreholes in Table 1 are counted, and the maximum elevation of the real boreholes in the modeling area is 124.525 and the minimum elevation is -687.175. At the same time, considering the modeling efficiency and accuracy, the segment thickness of the linear interpolation is set to 2m, and the real boreholes are divided into 407 segments. Taking ZK3 data as an example, the transition probability matrix of the Markov chain constructed according to the segment thickness statistics of the stratigraphic state is shown in Table 3 below. The table is a 7×7 transition probability matrix, the first row represents the stratigraphic space, and the others are stratigraphic states 1-6.

[0119] Table 3 Transition probability matrix of Markov chain

[0120]

[0121] (2) Based on mine survey, point cloud extraction is performed by tilting the drone to shoot the model data, and point cloud data preprocessing functions such as voxel downsampling, point cloud data denoising, and edge extraction are performed on the high-precision point cloud data.

[0122] (3) The Chebyshev distance weight is calculated based on the location information of the virtual borehole, and then the transition probability matrix corresponding to each real borehole is weighted and superimposed, so that the transition probability matrix of all virtual boreholes is obtained. The transition probability matrix at a virtual borehole is shown in formula (7).

[0123]

[0124] The formation thickness is statistically analyzed based on the actual drilling in the modeling area, as shown in Table 4. As can be seen from Table 4, the variance and mean of formations 2 and 4 have obvious numerical differences compared with other formations. In order to prevent large errors in simulating formation thickness, the present invention uses a log-normal distribution model to simulate the thickness corresponding to its formation state. The log-normal distribution and its original variance and mean are shown in Table 4. It can be seen that the log-normal distribution model using logarithm 10 will not produce large numerical differences in expectation and variance, effectively reducing the formation thickness simulation error.

[0125] Table 4 Comparison of formation thickness distribution models

[0126]

[0127] Furthermore, the formation state and formation thickness are determined based on superimposed Markov and improved Monte Carlo.

[0128] The surface images of the open-pit mine are obtained through drone photogrammetry technology, and the lithology in the surface images is identified based on the deep learning framework. Figure 6 This is the result of deep learning and the final surface rock type identification result map.

[0129] In the present invention, the step size is set to 3 (Markov transition probability matrix third order) to superimpose and predict the virtual borehole formation state, and the simulation is repeated according to the number of Monte Carlo simulations, and the number of Monte Carlo simulations is selected to be 1000 times.

[0130] After determining the formation state, the system simulates the formation thickness according to the formation probability distribution model of log-normal distribution based on the Monte Carlo random sampling theory, and restores the thickness according to the logarithmic restoration formula with base 10 when the simulation thickness is completed, thus completing the simulation of the formation thickness. The simulation results are shown in Table 5 below, which shows the formation state and formation thickness of the virtual borehole after 1000 simulations, thus constructing the formation state and corresponding formation thickness of 1000 virtual boreholes simulated by Monte Carlo.

[0131] Table 5 Monte Carlo simulation results of formation state and formation thickness

[0132]

[0133] Furthermore, the maximum probability model of virtual drilling is determined.

[0134] The spatial position of the virtual borehole is recorded, and the borehole database is cataloged based on the stratigraphic state and stratigraphic thickness completed in each Monte Carlo simulation, ultimately forming 1,000 geological models.

[0135] (1) Confirmation of the most probable stratigraphic sequence

[0136] The present invention first determines the maximum probability formation state at the elevation based on the Monte Carlo simulation formation state, and then calculates the maximum formation state probability under all virtual boreholes based on the maximum formation state combined with the Chebyshev distance weight coefficient, and finally substitutes the probability into the cumulative probability distribution model to obtain the corresponding formation thickness. Taking the virtual borehole VZK-[0] between ZK-3 and ZK-7 as an example, the probability distribution under each formation state can be obtained by calculating the first maximum probability, as shown in Table 6 below. It can be seen from Table 4 that the initial maximum formation state of the virtual borehole is formation state 2.

[0137] Table 6 Virtual drilling VZK-[0] initial maximum state confirmation table

[0138]

[0139] (2) Determination of the maximum probability stratigraphic boundary line

[0140] The stratigraphic probability of the actual borehole elevation is calculated by the position of the initial elevation. The probability of stratigraphic state 2 at the initial elevation of ZK-3 is 1, and the probability of ZK-7 is 0.68553. The distance weight of VZK-543 is converted by the Chebyshev distance weight, and the stratigraphic state of the initial elevation and the corresponding maximum probability are selected by the maximum probability ratio. Finally, the probability is substituted into the cumulative normal distribution curve of state 2 to obtain the stratigraphic thickness of 56.59638853206067; the second layer first determines the maximum stratigraphic state at the elevation, and the stratigraphic thickness is taken according to the corresponding cumulative probability distribution curve of the stratigraphic state, and the corresponding thickness is calculated by analogy. When the superimposed thickness is greater than or equal to the threshold of the borehole stratigraphic thickness, the system determines that the simulation ends.

[0141] By determining the maximum formation state of the borehole and calculating the corresponding thickness, the maximum probability model of the borehole is constructed, as shown in the attached figure. Figure 7 As shown, the present invention selects the profile consisting of two real boreholes ZK-3 to ZK-7 as the display of the maximum probability model.

[0142] (3) Maximum Probability Verification Drilling Correction Adjustment

[0143] In order to obtain the final modified model of the engineering geological body, a verification borehole must be selected from the geological section. The present invention adds a verification borehole VZK[0] in the section between the actual boreholes ZK3 and ZK7. The basic geological profile and location information of the verification borehole VZK[0] are shown in Table 7 below.

[0144] Table 7 Geological profile of verification borehole VZK[0]

[0145]

[0146] The formation probability distribution of the verification borehole and its corresponding cumulative probability are incorporated into the real drilling sequence. Similarly, according to the maximum probability model calculation process, the verification correction model is calculated by the maximum probability formation state and its corresponding formation thickness, as shown in the attached figure. Figure 8 As shown, through Figure 7 and Figure 8By comparison, it can be seen that the geological profiles before and after the verification borehole construction have a large error. This further illustrates the necessity of setting up verification boreholes. Thus, the final work of 3D geological modeling is completed by selecting the maximum probability model, as shown in the attached figure. Fig. 9 shown.

[0147] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A three-dimensional geological modeling method based on Markov chain and improved Monte Carlo, characterized in that: The following steps are involved: S1: Acquire the three-dimensional point cloud data of the construction engineering geological body area; S2: Based on the formation state and formation thickness data of the actual drilling in the construction area, the Markov chain is used to construct the formation state transition probability matrix, and the Monte Carlo method is used to determine the probability distribution model of the formation thickness; S3: Determine the initial ground state of the virtual borehole through photogrammetry and lithology identification; S4: Simulation of the formation state of the virtual borehole based on superimposed Markov chain; S5: Determine the maximum probability attribute of the virtual borehole stratum, calculate the stratum boundary line based on the cumulative probability distribution model curve of the stratum state, and finally correct the stratum boundary line by setting verification boreholes to obtain the final geological model.

2. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 1, characterized in that: The specific operation of step S2 includes the following steps: S201: Using Markov chain to construct the formation state transition probability matrix of real borehole; S202: constructing a probability distribution model of the formation thickness of the actual borehole; S203: Generate virtual borehole data using the point cloud data, and construct a formation state transition probability matrix of the virtual borehole.

3. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 2, characterized in that: The formation state transition probability matrix of the actual borehole in step S201 is expressed as Where P ij represents the probability of stratum i mutating into stratum j; The calculation formula of the transition probability matrix of the virtual drilling in step S203 is: In the formula, represents the transition probability matrix of virtual drilling, β i represents the distance weight, Transition probability matrix representing the actual drilling.

4. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 3, characterized in that: The specific operation of the superimposed Markov chain described in step S401 is to adjust the order of the transfer probability matrix according to the formation state. The order of the transfer probability matrix is ​​calculated based on the number of formation state segments of the virtual borehole obtained by statistics. The order calculation formula of the transfer probability matrix is: In the formula, represents the probability of transitioning from formation state i to formation state j at s steps, represents the number of transitions from formation state i to formation state j at step s, M i is the total number of samples in stratum i.

5. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 4, characterized in that: The specific operation of step S5 includes the following steps: S501: Calculating the formation state probability determined by superimposing the Markov chain in step S4; S502: Select the simulation result corresponding to the maximum probability of the formation state, and determine the formation state of the virtual borehole elevation h; S503: confirming the boundary corresponding to each stratum state of the virtual borehole; S504: Correcting the boundary corresponding to each stratum state by verifying the drilling hole; S505: Establish a three-dimensional geological model according to the stratum states and stratum state boundaries of the real borehole and the virtual borehole.

6. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 5, characterized in that: In step S501, the probability of formation state i at elevation h is expressed as Where y is the number of superimposed Markov chain simulations, x i is the number of formation state i at level h.

7. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 1, characterized in that: In step S3, RandMixAugment, a neural network framework for identifying lithology based on deep learning, is used to identify the stratigraphic state of the surface. RandMixAugment divides the original data into original training data and test data. The original training data is enhanced by regional enhancement and image enhancement to obtain enhanced training data. The original training data and the enhanced training data are merged for deep learning training. The training process includes WideResNet, SupCon, Shake-Shake and Pyramid-Shake. Finally, the data prediction and evaluation results of the stratigraphic state are obtained, and the network framework is tested using the test data.

8. The three-dimensional geological modeling method based on Markov chain and improved Monte Carlo according to claim 7, characterized in that: Image-level enhancement is achieved through automatic enhancement, and region-level enhancement is achieved through masking and hybrid strategies; The masking operation first randomly generates a mask box of size S1×Ts, where Ts is the size of the original training image. Next, the generated mask box is applied to the training image at a random position. The parameter S1 ranges from [0,1]. The mask box is smaller than the original image, and the color of the mask box is determined to be the average value of the RGB color of the image to which the randomization is applied. The blending operation first randomly generates a patch of size S2×OS from the original training image, where OS is the original training image size and S2 is the maximum size of the patch; the generated patch is added at a random position where the masking box is applied to the original training image.

9. A three-dimensional geological modeling system based on Markov chain and improved Monte Carlo, characterized by: It includes a data acquisition module, a real borehole data modeling module, a virtual borehole modeling module and a geological model building module; The data acquisition module, the real borehole data modeling module, the virtual borehole modeling module and the geological model construction module realize three-dimensional geological modeling based on the three-dimensional geological modeling method described in any one of claims 1-8.

10. An electronic device, characterized in that: The system comprises at least one processor and a memory in communication with the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the three-dimensional geological modeling method according to any one of claims 1 to 8.

Images