A three-dimensional stratum model construction method, system, electronic device and storage medium

Abstract

The present invention provides a method, system, electronic device, and storage medium for constructing a three-dimensional stratum model. The method comprises acquiring multiple acquired borehole data to form an acquired borehole data sequence; inputting the acquired borehole data sequence into a trained borehole data prediction model to output multiple predicted borehole data; and constructing a three-dimensional stratum model based on the multiple acquired borehole data and the multiple predicted borehole data. The borehole data prediction model of the present invention adopts a LightGBM model. During the training process, interpolation prediction and model prediction are performed on the borehole data respectively. A mixed loss of the LightGBM model is jointly constructed based on the physical loss of the interpolation prediction and the predicted loss of the model prediction. The LightGBM model is jointly trained based on the mixed loss. The trained LightGBM model provides more accurate predictions for borehole data.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional model construction, and more specifically, to a three-dimensional stratum model construction method, system, electronic equipment and storage medium. Background Art

[0002] During power grid construction, the geological survey phase plays a crucial role influencing key project indicators, such as design and construction quality, and project economic benefits. The level of detail and accuracy of geological survey information significantly impacts the smooth implementation of construction projects. Accurate 3D geological information models are a prerequisite for engineering construction, especially infrastructure projects, and provide crucial guidance for engineering projects.

[0003] Processing borehole data to obtain information that can be used to describe actual stratigraphic formations is a major challenge in 3D modeling. Furthermore, the current modeling process remains complex and has a low degree of automation, significantly limiting the application of 3D models in engineering. Due to sampling limitations, geological borehole data is inherently sparse, resulting in uncertainty regarding stratigraphic thickness and position between boreholes. Furthermore, due to tectonic movement, stratigraphic formations often share similar spatial characteristics.

[0004] Traditional 3D modeling often uses surface fitting to construct parametric surfaces of stratigraphic structure or directly interpolates information from other stratigraphic layers, ignoring the mutual constraints between stratigraphic layers. This results in poor 3D model simulations. Traditional spatial interpolation methods only consider the interactions between adjacent boreholes within the same stratigraphic layer, without considering the influence of adjacent stratigraphic layers. Geological borehole data are typically high-dimensional, nonlinear, and spatially correlated. Machine learning offers advantages in processing this complex data, including nonlinear modeling capabilities, automatic feature extraction, high-dimensional data processing, adaptation to complex spatial relationships, robustness, real-time prediction capabilities, and scalability. However, both spatial interpolation and machine learning have limitations. While physical methods offer clear interpretability and theoretical foundations, they often struggle to address nonlinear problems in complex systems. Traditional numerical simulation methods also suffer from high computational costs and low efficiency. While machine learning excels at discovering patterns and relationships in data, its "black box" nature makes the models lack physical interpretability. Furthermore, its reliance on large amounts of high-quality data limits its application in areas where data acquisition is difficult. Summary of the Invention

[0005] The present invention aims to solve the technical problems existing in the prior art and provides a three-dimensional stratum model construction method, system, electronic equipment and storage medium to overcome the sparsity problem of drilling acquisition data in the traditional three-dimensional stratum model construction process.

[0006] According to a first aspect of the present invention, a method for constructing a three-dimensional stratum model is provided, comprising:

[0007] Acquire multiple acquisition drilling data to form an acquisition drilling data sequence;

[0008] Inputting the collected drilling data sequence into the trained drilling data prediction model to output a plurality of predicted drilling data;

[0009] constructing a three-dimensional stratum model based on the plurality of acquired drilling data and the plurality of predicted drilling data;

[0010] The drilling data prediction model is a LightGBM model, and the training of the LightGBM model includes:

[0011] Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence;

[0012] performing interpolation processing on the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data;

[0013] Inputting the sample set sequence into the LightGBM model and outputting a plurality of predicted drilling data;

[0014] Calculate the mixed loss of the LightGBM model according to the physical loss and the predicted loss, wherein the physical loss is calculated based on the plurality of interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the plurality of predicted drilling data and the corresponding actual drilling data;

[0015] The LightGBM model is adjusted according to the hybrid loss to obtain the trained LightGBM model.

[0016] According to a second aspect of the present invention, a three-dimensional stratum model construction system is provided, comprising:

[0017] An acquisition module, configured to acquire a plurality of collected drilling data to form a collected drilling data sequence;

[0018] A prediction module, configured to input the collected drilling data sequence into a trained drilling data prediction model and output a plurality of predicted drilling data;

[0019] A construction module, configured to construct a three-dimensional formation model based on the plurality of acquired drilling data and the plurality of predicted drilling data;

[0020] The drilling data prediction model is a LightGBM model, and the training of the LightGBM model includes:

[0021] Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence;

[0022] performing interpolation processing on the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data;

[0023] Inputting the sample set sequence into the LightGBM model and outputting a plurality of predicted drilling data;

[0024] Calculate the mixed loss of the LightGBM model according to the physical loss and the predicted loss, wherein the physical loss is calculated based on the plurality of interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the plurality of predicted drilling data and the corresponding actual drilling data;

[0025] The LightGBM model is adjusted according to the hybrid loss to obtain the trained LightGBM model.

[0026] According to a third aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein the processor is configured to implement the steps of a method for constructing a three-dimensional stratum model when executing a computer management program stored in the memory.

[0027] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer management program is stored. When the computer management program is executed by a processor, the steps of the three-dimensional stratum model construction method are implemented.

[0028] The present invention provides a three-dimensional stratum model construction method, system, electronic device and storage medium. The drilling data prediction model adopts the LightGBM model. During the training process, interpolation prediction and model prediction are performed on the drilling data respectively. The mixed loss of the LightGBM model is jointly constructed according to the physical loss of the interpolation prediction and the prediction loss of the model prediction. The LightGBM model is jointly trained based on the mixed loss. The trained LightGBM model can more accurately predict the drilling data. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A flow chart of a method for constructing a three-dimensional stratum model provided by the present invention;

[0030] Figure 2 This is a schematic diagram of the structure of the LightGBM model according to an embodiment of the present invention;

[0031] Figure 3 A schematic diagram of the effect of a three-dimensional stratum model constructed in an embodiment of the present invention;

[0032] Figure 4 A schematic structural diagram of a three-dimensional stratum model construction system provided by an embodiment of the present invention;

[0033] Figure 5 A schematic diagram of the hardware structure of a possible electronic device provided by the present invention;

[0034] Figure 6 A schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0036] Based on the defects in the background technology, the present invention proposes a method for constructing a three-dimensional stratigraphic model, which combines physical laws with machine learning and can give full play to the complementary advantages of the two. First, physical laws can provide prior knowledge for machine learning models, improve the physical interpretability of the models, and make their prediction results more scientifically meaningful. Secondly, models combined with physical laws can reduce dependence on data and improve the generalization ability of the models, which is particularly suitable for fields where data acquisition is difficult. In addition, this combination can also significantly improve computing efficiency. For example, in materials science and climate modeling, machine learning models combined with physical laws can quickly predict the properties of complex systems without the need for time-consuming numerical simulations. Therefore, the present invention proposes a physical data hybrid-driven borehole data prediction, which predicts the stratigraphic information at unknown boreholes while considering the multiple mutual influences of each stratum, and then realizes the rapid construction of a three-dimensional geological model of the power grid through Dynamo. The present invention provides a widely applicable method framework for solving the problems of insufficient samples in the geological field and continuous prediction of geological data.

[0037] Figure 1 A flow chart of a three-dimensional stratum model construction method provided by the present invention is as follows: Figure 1 As shown, the method includes:

[0038] Step 1: Acquire multiple acquisition drilling data to form an acquisition drilling data sequence.

[0039] It is understandable that the first step is to collect and pre-process the borehole data. Specifically, in 3D stratigraphic modeling, the stratigraphic interface is generally interpolated based on the borehole data. Correctly processing the borehole data is the premise and foundation for building an ideal 3D stratigraphic model. In the borehole histogram, the original borehole data mainly includes: plane coordinates, borehole depth, soil layer type, and the elevation values ​​of each soil layer boundary point. Data pre-processing is to convert the borehole data into a series of points with spatial attributes, and then use The elevation Z of the i-th borehole at the representative position (x, y) on the ground plane is denoted by . It should be noted that, considering the thickness of the ground, the elevation Z here is the depth of the bottom of the soil layer. The collected borehole data forms a collected borehole data sequence.

[0040] Step 2: input the collected drilling data sequence into the trained drilling data prediction model to output a plurality of predicted drilling data.

[0041] It should be noted that the collected borehole data is limited and sparse, and the amount of data used to model 3D stratigraphic formations is clearly insufficient. Therefore, it is necessary to expand the borehole data based on the already collected borehole data. This invention, based on machine learning, predicts some borehole data based on the collected borehole data sequence to obtain more borehole data, providing data support for the subsequent construction of the 3D stratigraphic model.

[0042] Specifically, one embodiment of the present invention is to construct a drilling data prediction model to predict drilling data based on sparse drilling data to obtain more drilling data. In one embodiment of the present invention, the drilling data prediction model is a LightGBM model. The LightGBM algorithm is an open source, efficient, distributed gradient boosting tree algorithm. Due to its fast speed, low memory consumption, and relatively high accuracy, it is widely used in classification and regression problems.

[0043] In machine learning, a learner whose generalization performance is slightly better than random guessing is called a weak learner. GBDT is an ensemble algorithm that treats regression decision trees as weak learners and continuously improves their performance through multiple rounds of iteration.

[0044] When training the LightGBM model, in the iteration of the gradient boosting tree GBDT of the LightGBM model, based on the strong learner in the previous round, the weak learner of this round is found through learning, and the weak learner of this round is superimposed on the strong learner in the previous round as the strong learner of this round. Through continuous iterative training and learning, the LightGBM model after training and learning is obtained.

[0045] In one embodiment of the present invention, the step of learning and finding a weak learner in this round based on a strong learner in the previous round, and superimposing the weak learner in this round on the strong learner in the previous round as a strong learner in this round includes:

[0046] Calculate the loss function of this round of strong learner:

[0047]

[0048] in, represents the loss function of the strong learner in this round, Represents the sample data of different batches input, N represents the existence of all sample batches, represents the strong learner in the previous round Represents The corresponding drilling actual collection data label, represents the calculation function of the loss function, Represents the weak learner of this round.

[0049] Calculate the negative gradient of the loss function of this round of strong learner:

[0050] in, It represents the negative gradient of the loss function of the current round of strong learner, which is used to fit the approximate value of the loss function of the current round of strong learner.

[0051] Gradient descent method is used to optimize the parameters of the LightGBM model:

[0052]

[0053] in, learning_r Represents the learning rate.

[0054] In the regression problem, the loss function of the strong learner in this round is approximately fitted based on the squared error MAE:

[0055]

[0056] With the goal of minimizing the loss function of the strong learner in this round, solve the weak learner in this round;

[0057] According to the strong learner in the previous round and the weak learner in this round, the strong learner in this round is obtained:

[0058] .

[0059] GBDT has become an algorithm widely used in many fields due to its efficiency and accuracy. However, as the data becomes more and more complex, the computational overhead is very high when processing big data, making it difficult to achieve a good balance between accuracy and efficiency. The introduction of the LightGBM model not only reduces the computational overhead and improves the computational efficiency of the model, but also achieves a higher accuracy while maintaining a higher computational efficiency.

[0060] Training the LightGBM model using the training dataset includes:

[0061] Step 21: Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence.

[0062] It is understandable that in the process of training the drilling data prediction model, some drilling data will be collected to form a collected drilling data set.

[0063] The architecture diagram of the LightGBM model can be found in Figure 2 The LightGBM model mainly consists of three parts: gradient-based single-side sampling GOSS, complementary feature compression EFB and histogram algorithm Histogram.

[0064] For gradient-based single-side sampling (GOSS), according to the definition of information gain, samples with large loss function gradients contribute significantly to information gain. Further learning of samples with small loss function gradients does not significantly improve the accuracy of the results. Therefore, to maintain the accuracy of the information gain estimate, when sampling instances, it is important to retain those with large gradients and randomly select samples with small gradients. By applying the GOSS algorithm, not only can learning accuracy be maintained, but the learning rate can also be significantly accelerated.

[0065] Therefore, when selecting training data samples from the sampled drilling data set, data samples with large gradients and a portion of samples with small gradients are selected to form a training data sample set. In one embodiment of the present invention, a sampled drilling data set is obtained, wherein the sampled drilling data set includes multiple groups of drilling data samples;

[0066] Calculating the gradient value of the loss function for each group of drilling data samples;

[0067] Arranging the plurality of groups of drilling data samples in descending order according to the gradient values;

[0068] Selecting a first preset proportion of the plurality of drilling data samples with the highest gradient values ​​as subset A, and the remaining drilling data samples as subset B;

[0069] Selecting a second preset proportion of the drilling data samples from the subset B as a subset C;

[0070] Merging the subset A and the subset C as a training data sample set D;

[0071] The LightGBM model is trained based on the training sample dataset D.

[0072] Specifically, arranging the plurality of groups of drilling data samples in descending order according to the gradient values ​​includes:

[0073]

[0074] in, 、 、....、 Represents the gradient value of the loss function of N groups of drilling data samples, 、 2. ... N represents the index of the drilling data sample;

[0075] The step of selecting a first preset proportion of the plurality of drilling data samples having the highest gradient values ​​as a subset A and the remaining drilling data samples as a subset B includes:

[0076] The first a% of the drilling data samples with large gradients are recorded as subset A, and the remaining (1-a%) of the drilling data samples with small gradients are recorded as subset B, wherein the subset A is represented as:

[0077]

[0078] in, 、 、....、 represents p groups of drilling data samples;

[0079] The subset B is represented as:

[0080]

[0081] =a%

[0082] Randomly extract b% of the drilling data samples from the subset B and record them as subset C. The weights of the drilling data samples in the subset C are assigned as , wherein the subset C is expressed as:

[0083]

[0084] q=(Np)×b%

[0085] The subset A and the subset C are combined as a training data sample set D, and the training data sample set D is expressed as:

[0086] D={A+C}.

[0087] Among them, GOSS can improve the generalization ability of the model by focusing on samples with larger gradients without losing accuracy.

[0088] Among them, the complementary feature compression (EFB) in the LightGBM model is an algorithm that can reduce the number of features of high-dimensional data and minimize the loss. High-dimensional data is usually very sparse, and many features are mutually exclusive, so these features can be merged. LightGBM uses (Histogram) to merge mutually exclusive features. Its solution is to discretize continuous attribute values ​​into m integers, and construct a histogram with a width of m during the discretization. When traversing the data, the discretized values ​​can be used as indices to accumulate statistics in the histogram, and then the optimal split point can be found by traversing the discrete values ​​of the histogram. In this way, the huge amount of unnecessary calculations is avoided, and the performance of the model is further accelerated.

[0089] Step 22: interpolate the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data.

[0090] It is understandable that step 21 selects a training data sample set from the collected drilling data set, that is, a sample set sequence, and performs interpolation processing based on multiple drilling data in the sample set sequence to obtain multiple interpolated drilling data after interpolation processing.

[0091] In one embodiment of the present invention, the Kriging spatial interpolation method is used for interpolation operations. Kriging interpolation is an unbiased optimal interpolation method based on spatial statistics theory. The core of this method lies in the determination of variograms, which describe the similarities or differences between boreholes at different locations in space. By calculating these variograms, the Kriging interpolation method can perform unbiased optimal estimation of the attribute values ​​of the interpolation points within a limited area. Compared with methods such as inverse distance weighted interpolation and natural neighbor interpolation, Kriging interpolation not only considers the distance relationship between boreholes, but also the correlation between them, thereby improving the accuracy and reliability of the interpolation results.

[0092] Ordinary Kriging (OK) interpolation is the basic form of kriging interpolation. It assumes the mean of the data is an unknown constant and solves for weight coefficients by minimizing the semivariance of the estimated error. This method is similar to inverse distance weighted interpolation, in that both methods estimate the value of the interpolated point by weighting all boreholes in space. However, the weight coefficients are not the inverse of the distance, but rather a set of optimal coefficients that minimize the variance between the predicted and actual values ​​at the interpolated point.

[0093] Assume that the point to be inserted There are n holes in the field, namely x1, x2, x3, …, x n , is the attribute value at the i-th drilling hole, then the attribute value at the point to be inserted is is the weighted sum of the attribute values ​​of n boreholes in the area, that is:

[0094]

[0095] In the formula is the weight coefficient of the i-th borehole. In order to meet the prediction value of the point to be inserted With actual value The variance Var between them is the smallest, and the expectation E is 0, that is:

[0096]

[0097]

[0098] According to the conditions of variance and expected value, the parameters in the Kriging spatial interpolation method are improved. Based on the improved Kriging spatial interpolation method, the borehole data in the sample set sequence are interpolated to obtain multiple interpolated borehole data.

[0099] Step 23: input the sample set sequence into the LightGBM model and output a plurality of predicted drilling data.

[0100] It can be understood that while the interpolation operation is performed in step 22, the sample set sequence is input into the LightGBM model for prediction. The LightGBM model predicts a larger number of drilling data based on the sparse drilling data sequence. The drilling data predicted by the LightGBM model is called predicted drilling data.

[0101] Step 24, calculate the mixed loss of the LightGBM model based on the physical loss and the predicted loss, wherein the physical loss is calculated based on the multiple interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the multiple predicted drilling data and the corresponding actual drilling data.

[0102] It's understandable that the loss function plays a crucial role in guiding the prediction bias of machine learning models, directly impacting the model's accuracy and efficiency. The model's goal is to achieve the highest possible fitting accuracy by minimizing the loss function. To enable machine learning models to fit data according to the physical laws of reasoning, the original loss function must be redesigned to incorporate information about physical laws. This physical guidance is then incorporated into the gradient calculation of the loss function, guiding the model toward the desired prediction direction and improving its prediction accuracy and generalization capabilities.

[0103] The LightGBM model supports custom loss functions and evaluation criteria during training. The loss function definition requires returning the calculation method for the first- and second-order derivatives of the loss function, while the evaluation criteria require calculating the true and predicted values ​​of the data. The loss function is used to learn the tree structure during training, while the evaluation criteria are often used to evaluate the effect on the validation set.

[0104] The structure of the hybrid loss function proposed in the embodiment of the present invention is shown in the following formula. It consists of two parts: physical loss and predicted loss. The hybrid loss is expressed as follows:

[0105]

[0106] in, represents the mixed loss of the LightGBM model, The actual data collected for the i-th borehole, represents the i-th interpolated drilling data, obtained by interpolation processing, The i-th predicted drilling data output by the LightGBM model, Indicates physical loss, represents the prediction loss, It represents the proportion of physical losses to mixed losses.

[0107] The physical loss represents the error between the current physical analysis value (i.e., interpolated data) and the true value; the prediction loss represents the LightGBM model loss function value, that is, the error between the model prediction value and the true value. λ represents the proportion of physical loss to the overall loss. When λ=0, the model prediction bias is entirely guided by data information; when λ is 1, the model prediction bias is entirely guided by physical laws. By adjusting the value of λ, loss functions with different biases can be obtained. Generally, the larger the value of λ, the closer the model prediction result is to the physical analysis value, and the more robust the model will be. The smaller the value of λ, the more the model focuses on the characteristics of the dataset itself, and the model may perform better on certain specific datasets.

[0108] Step 25: Adjust the LightGBM model according to the hybrid loss to obtain the trained LightGBM model.

[0109] In one embodiment of the present invention, the first-order gradient value and the second-order gradient value of the hybrid loss are calculated, wherein the calculation formulas for the first-order gradient value and the second-order gradient value returned by the hybrid loss function are as follows:

[0110]

[0111]

[0112] in, represents the first-order gradient value of the mixed loss, and hess represents the second-order gradient value of the mixed loss.

[0113] According to the first-order gradient value and the second-order gradient value, the parameters of the LightGBM model are adjusted until the loss function of the LightGBM model reaches a minimum value, and the trained LightGBM model is obtained.

[0114] After training the LightGBM model, the trained LightGBM model is evaluated. In order to verify the quality of the model, the output error of the prediction model is mainly measured by the model evaluation index. Common evaluation indicators include the coefficient of determination (R 2 ), root mean square error (RMSE), and maximum absolute error (MAE). For RMSE and MAE, the smaller the indicator values, the higher the model's prediction accuracy. For R2, the closer they are to 1, the higher the model's prediction accuracy. The indicator calculation method is shown in the following equation:

[0115]

[0116]

[0117]

[0118] The LightGBM model is evaluated based on the calculated evaluation indicators, and the LightGBM model can be adjusted until the LightGBM model is optimal. The drilling data is predicted based on the trained LightGBM model to obtain more drilling data.

[0119] Step 3: construct a three-dimensional stratum model based on the plurality of collected drilling data and the plurality of predicted drilling data.

[0120] It is understandable that the collected borehole data and the borehole data predicted by the LightGBM model are used to construct a three-dimensional stratigraphic model and realize the visualization of the borehole data.

[0121] Specifically, Dynamo is used as the basic platform for automated modeling. In this program, the file storage path is first obtained through the file path node, and the file path is input as a parameter to the file from path node to create a file object. The Data Import Excel node reads the Excel table data in the file object row by row to obtain the required borehole coordinates and elevation information of each stratum. The List.Transpose node implements the exchange of rows and columns of table data. If some rows or columns are inconsistent in length, null values ​​can be inserted into the result array as placeholders to make it rectangular in shape, avoiding the failure of subsequent nodes to read data. Cold Block can obtain point groups and terrain surfaces separately and name them. It should be noted that when naming point groups and surfaces, the names cannot be the same, otherwise the program will not be able to recognize them. The point group refers to the collection of elevation data of all boreholes in the same stratum surface. The Cogopoint Bygeometry node can create geometric space points with geological attributes from Dynamo points. Finally, the Tinsurface Bycogopointgroup node is used to create a triangulated surface from the geometric space points to obtain a three-dimensional stratum model. The effect diagram of constructing a three-dimensional stratum model can be seen in Figure 3 .

[0122] The following describes the method for constructing a three-dimensional stratum model provided by an embodiment of the present invention with reference to a specific embodiment.

[0123] 1. Drilling data collection.

[0124] This paper uses geological drilling data from a channel engineering project as an example to validate the proposed prediction method based on a multi-output support vector machine regression model. Drilling results from the study area reveal that the main strata explored in the region are (in descending order): silt, silty clay, muddy medium sand, and gravel, corresponding to four geological horizons. A total of 294 boreholes were collected in the study area, numbered ZK1–ZK294. Data preprocessing of the borehole histogram yields the following (unit: m):

[0125] Table 1 Geological drilling information data list (the first 40 as an example)

[0126]

[0127] It should be noted that in order to compare the prediction accuracy of drilling data prediction based on multi-output support vector machine regression with other methods, the embodiment of the present invention randomly selects 80% of the drilling points as the model training set and 20% of the drilling points as the model test set for model training and testing to verify the accuracy and feasibility of the method.

[0128] 2. LightGBM model prediction combined with Kriging interpolation.

[0129] Hyperparameters that influence Lightgbm model training efficiency and accuracy include the number of training runs, the learning rate, the maximum tree depth, and the regularization coefficient. Grid search ensures that the optimal parameter combination is found by exhaustively searching all possible parameter combinations within a specified range of candidate values. This comprehensiveness ensures that no possible optimal parameter combination is missed.

[0130] Grid search is relatively simple to implement; simply specify the candidate parameter value range and the evaluation metric. The program then automatically performs the search, eliminating the need for manual parameter tuning. The hyperparameter combinations obtained through grid search in this study are shown in the table below. For computational and time efficiency reasons, the optimal values ​​for the number of training runs and learning rate were separately grid-searched, and some hyperparameters were left at their default values. The model hyperparameters are shown in Table 2.

[0131] Table 2 List of model hyperparameters

[0132]

[0133] The LightGBM model's learning rate, number of training iterations, and the proportion of physical laws in the loss function are the most important core parameters. A grid search was used to refine the learning rate and number of training iterations. The model achieved the best prediction accuracy when the learning rate was 0.1 and the number of iterations was 200.

[0134] 3. Comparison of the accuracy of various methods.

[0135] To evaluate the accuracy and reliability of the proposed physical data hybrid prediction model, common model evaluation metrics were used to assess its output error. This paper selected the Kriging-LightGBM hybrid model (Kriging-LightGBM), LightGBM, and Kriging interpolation methods (Kriging) for prediction of borehole data. The specific calculation results are shown in Table 3 below:

[0136] Table 3 Summary of various machine learning prediction results

[0137]

[0138]

[0139]

[0140] From Table 3 above, we can see that the model with the best prediction effect is Kriging-LightGBM. 2As a result, the hybrid prediction method proposed in this invention can fully and effectively give play to the advantages of machine learning and physical laws, and the model prediction accuracy is better than that of any single method.

[0141] In general, the LightGBM model combined with Kriging interpolation proposed in this paper provides an efficient and highly interpretable solution for the accurate prediction of power grid geological drilling data. In the future, it is expected to further promote its application in intelligent geological exploration by combining spatiotemporal feature coding and more complex geophysical laws.

[0142] 4. Modeling results of the three-dimensional stratigraphic model based on the Dynamo platform.

[0143] Import the predicted drilling data based on the physical data hybrid-driven model into the Dynamo automatic modeling program and run it to obtain the following 3D model.

[0144] See also Figure 4 , is a three-dimensional stratum model construction system provided by an embodiment of the present invention, the construction system comprising:

[0145] An acquisition module 401 is used to acquire a plurality of collected drilling data to form a collected drilling data sequence;

[0146] A prediction module 402 is configured to input the collected drilling data sequence into a trained drilling data prediction model and output a plurality of predicted drilling data;

[0147] A construction module 403 is configured to construct a three-dimensional formation model based on the plurality of acquired drilling data and the plurality of predicted drilling data;

[0148] The drilling data prediction model is a LightGM model, and the training of the LightGM model includes:

[0149] Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence;

[0150] performing interpolation processing on the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data;

[0151] Inputting the sample set sequence into the LightGBM model and outputting a plurality of predicted drilling data;

[0152] Calculate the mixed loss of the LightGBM model according to the physical loss and the predicted loss, wherein the physical loss is calculated based on the plurality of interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the plurality of predicted drilling data and the corresponding actual drilling data;

[0153] The LightGBM model is adjusted according to the hybrid loss to obtain the trained LightGBM model.

[0154] It can be understood that the three-dimensional formation model construction system provided by the present invention corresponds to the three-dimensional formation model construction method provided in the aforementioned embodiments. The relevant technical features of the three-dimensional formation model construction system can refer to the relevant technical features of the three-dimensional formation model construction method, and will not be repeated here.

[0155] See also Figure 5 , Figure 5 Schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. Figure 5 As shown, an embodiment of the present invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, the steps of the three-dimensional formation model construction method are implemented.

[0156] See also Figure 6 , Figure 6 Schematic diagram of an embodiment of a computer-readable storage medium provided by the present invention. Figure 6 As shown, this embodiment provides a computer-readable storage medium 600 on which a computer program 611 is stored. When the computer program 611 is executed by a processor, the steps of the method for constructing a three-dimensional stratum model are implemented.

[0157] The embodiments of the present invention provide a three-dimensional stratigraphic model construction method, system, electronic device and storage medium, which efficiently capture the complex nonlinear relationship between drilling parameters through the LightGBM model, and explicitly model spatial autocorrelation using Kriging interpolation. It inherits the learning ability of the gradient boosting tree for high-dimensional features, and incorporates the spatial distribution laws of geological data. It can effectively improve the generalization performance of unexplored areas and is particularly suitable for processing spatial extrapolation prediction problems of sparse drilling data. Compared with traditional spatial interpolation methods, this method has achieved a certain degree of improvement in prediction accuracy.

[0158] Based on the Dynamo software platform, visual programming was performed and a three-dimensional stratigraphic model automatic construction program was built. The processed drilling data was imported to directly generate a three-dimensional stratigraphic model, which to a certain extent realized the automation of the three-dimensional stratigraphic model.

[0159] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0160] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0161] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, 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 computer, 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 flowcharts and / or block diagrams. 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.

[0162] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0164] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0165] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for constructing a three-dimensional stratum model, characterized in that: include: Acquire multiple acquisition drilling data to form an acquisition drilling data sequence; Inputting the collected drilling data sequence into the trained drilling data prediction model to output a plurality of predicted drilling data; constructing a three-dimensional stratum model based on the plurality of acquired drilling data and the plurality of predicted drilling data; The drilling data prediction model is a LightGBM model, and the training of the LightGBM model includes: Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence; Performing interpolation processing on the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data; wherein the interpolation operation on the collected drilling data is performed using a Kriging spatial interpolation method; Inputting the sample set sequence into the LightGBM model and outputting a plurality of predicted drilling data; Calculate the mixed loss of the LightGBM model according to the physical loss and the predicted loss, wherein the physical loss is calculated based on the plurality of interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the plurality of predicted drilling data and the corresponding actual drilling data; The LightGBM model is adjusted according to the hybrid loss to obtain the trained LightGBM model.

2. The method for constructing a three-dimensional stratum model according to claim 1, wherein: In the iteration of the gradient boosting tree GBDT of the LightGBM model, based on the strong learner in the previous round, the weak learner in this round is learned and found, and the weak learner in this round is superimposed on the strong learner in the previous round as the strong learner in this round. Through continuous iterative training and learning, the LightGBM model after training and learning is obtained.

3. The method for constructing a three-dimensional stratum model according to claim 2, wherein: The process of learning and finding a weak learner in this round based on the strong learner in the previous round, and superimposing the weak learner in this round on the strong learner in the previous round as the strong learner in this round includes: Calculate the loss function of this round of strong learner: Among them, L t Represents the loss function of this round of strong learner, x n Represents the sample data of different batches input, N represents the existence of all sample batches, F t-1 (x n ) represents the previous round of strong learner, y n Indicates that x n The corresponding drilling data label is actually collected, L(.) represents the calculation function of the loss function, h t (x n ) represents the weak learner of this round; Calculate the negative gradient of the loss function of this round of strong learner: Among them, r t Represents the negative gradient of the loss function of the current round of strong learner, which is used to fit the approximate value of the loss function of the current round of strong learner; Gradient descent method is used to optimize the parameters of the LightGBM model: F t (x n )=F t-1 (x n )-learning_r*r t Among them, learning_r represents the learning rate; In the regression problem, the loss function of the strong learner in this round is approximately fitted based on the squared error MAE: With the goal of minimizing the loss function of the strong learner in this round, solve the weak learner in this round; According to the strong learner in the previous round and the weak learner in this round, the strong learner in this round is obtained: F t (x n )=F t-1 (x n )+h t (x n )。 4. The method for constructing a three-dimensional stratum model according to claim 1 or 3, wherein: The acquiring of the acquired drilling data set and selecting a portion of the acquired drilling data from the acquired drilling data set as a sample set sequence includes: Acquire a collection of drilling data sets, wherein the collection of drilling data sets includes multiple sets of drilling data samples; Calculating the gradient value of the loss function for each group of drilling data samples; Arranging the plurality of groups of drilling data samples in descending order according to the gradient values; Selecting a first preset proportion of the plurality of drilling data samples with the highest gradient values ​​as subset A, and the remaining drilling data samples as subset B; Selecting a second preset proportion of the drilling data samples from the subset B as a subset C; Merging the subset A and the subset C as a training data sample set D; The LightGBM model is trained based on the training data sample set D.

5. The method for constructing a three-dimensional stratum model according to claim 4, wherein: Arranging the plurality of groups of drilling data samples in descending order according to the gradient values ​​includes: |g m1 |≥|g m2 |≥...≥|g mN | Among them, g m1 、g m2 、....、g mN represents the gradient value of the loss function of N groups of drilling data samples, m1, m2, ..., mN represent the index of the drilling data samples; The step of selecting a first preset proportion of the plurality of drilling data samples having the highest gradient values ​​as a subset A and the remaining drilling data samples as a subset B includes: The first a% of the drilling data samples with large gradients are recorded as subset A, and the remaining (1-a%) of the drilling data samples with small gradients are recorded as subset B, wherein the subset A is represented as: A={(x1,g m1 ),(x2,g m2 ),....,(x p ,g mp )} Among them, x1, x2, ..., x p represents p groups of drilling data samples; The subset B is represented as: B={(x p+1 ,g m(p+1) ),(x p+2 ,g m(p+2) ),....,(x N ,g mN )} b% of the drilling data samples are randomly selected from the subset B and recorded as subset C. The weights of the drilling data samples in the subset C are assigned as (1-a) / b, wherein the subset C is expressed as: q=(Np)×b% The subset A and the subset C are combined as a training data sample set D, and the training data sample set D is expressed as: D={A+C}.

6. The method for constructing a three-dimensional stratum model according to claim 1, wherein: The mixed loss of the LightGBM model is calculated based on the physical loss and the predicted loss, including: in, Represents the mixed loss of the LightGBM model, y i The actual data label collected for the i-th borehole, y phyi Represents the i-th interpolated drilling data, obtained according to the interpolation process, Huber is the i-th predicted drilling data output by the LightGBM model. phy Indicates physical loss, Huber pre represents the predicted loss, λ represents the proportion of physical loss to mixed loss; The adjusting the LightGBM model according to the hybrid loss to obtain the trained LightGBM model includes: The first-order gradient value and the second-order gradient value of the mixed loss are calculated respectively, where: Wherein, gard represents the first-order gradient value of the mixed loss, and hess represents the second-order gradient value of the mixed loss; The adjusting the LightGBM model according to the hybrid loss to obtain the trained LightGBM model includes: According to the first-order gradient value and the second-order gradient value, the parameters of the LightGBM model are adjusted until the loss function of the LightGBM model reaches a minimum value, and the trained LightGBM model is obtained.

7. The method for constructing a three-dimensional stratum model according to claim 1, wherein: The constructing of a three-dimensional stratum model based on the plurality of acquired drilling data and the plurality of predicted drilling data comprises: Construct a three-dimensional stratigraphic model based on the Dynamo automated basic modeling platform; The three-dimensional stratum model is constructed based on the Dynamo automated basic modeling platform, including: Get the file storage path through the file path node, and input the file storage path as a parameter to the filefrom path node to create a file object; The Excel table data in the file object is read row by row through the Data Import Excel node to obtain the required drilling coordinates and each stratum elevation information; Use the List.Transpose node to swap the rows and columns of the Excel spreadsheet data, and use the Cold Block node to obtain and name the point group and terrain surface respectively. The point group refers to the collection of elevation data of all boreholes in the same ground level. Use the Cogopoint Bygeometry node to create geometric space points with geological attributes from Dynamo points, and use the Tinsurface Bycogopointgroup node to create triangulated surfaces from the geometric space points to obtain a three-dimensional stratum model.

8. A three-dimensional stratum model construction system, characterized in that: include: An acquisition module, configured to acquire a plurality of collected drilling data to form a collected drilling data sequence; A prediction module, configured to input the collected drilling data sequence into a trained drilling data prediction model and output a plurality of predicted drilling data; A construction module, configured to construct a three-dimensional formation model based on the plurality of acquired drilling data and the plurality of predicted drilling data; The drilling data prediction model is a LightGBM model, and the training of the LightGBM model includes: Acquire a collection drilling data set, and select a portion of the collection drilling data from the collection drilling data set as a sample set sequence; Performing interpolation processing on the collected drilling data in the sample set sequence to obtain a plurality of interpolated drilling data; wherein the interpolation operation on the collected drilling data is performed using a Kriging spatial interpolation method; Inputting the sample set sequence into the LightGBM model and outputting a plurality of predicted drilling data; Calculate the mixed loss of the LightGBM model according to the physical loss and the predicted loss, wherein the physical loss is calculated based on the plurality of interpolated drilling data and the corresponding actual drilling data, and the predicted loss is calculated based on the plurality of predicted drilling data and the corresponding actual drilling data; The LightGBM model is adjusted according to the hybrid loss to obtain the trained LightGBM model.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the processor is used to implement the three-dimensional stratum model construction method according to any one of claims 1 to 7 when executing a computer management program stored in the memory.

10. A computer-readable storage medium, characterized in that A computer management program is stored thereon, and when the computer management program is executed by a processor, the three-dimensional stratum model construction method according to any one of claims 1 to 7 is implemented.

Images