Method for inhibiting formation thickness interpolation oscillation based on improved KNN

Abstract

The invention discloses a method for inhibiting formation thickness interpolation oscillation based on an improved KNN, and belongs to the technical field of geological modeling. Aiming at the problem that a traditional interpolation method is easy to generate a pseudo stratum without geological basis near a stratum pinching line, the method comprises the following steps: firstly, extracting a drilling position, a stratum existence label and thickness; then, on the basis of an improved KNN algorithm, intelligently partitioning a to-be-predicted region into an interpolation effective region and an interpolation ineffective region by calculating a spatial anisotropy weight and constructing a weighted distance formula; then, spatial interpolation is only carried out on the stratum thickness in the effective area, and the thickness is forcibly set to zero in the ineffective area; and finally fusing results of the two parts to generate a final thickness model. According to the method, interpolation oscillation is inhibited from the source, a clear stratum boundary is ensured, a model result is more geologically reasonable, and the reliability of resource assessment is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of geological engineering and computer-aided modeling technology, and specifically relates to a method for suppressing oscillations in formation thickness interpolation based on an improved KNN. Background Technology

[0002] In the field of geological modeling, modeling methods based on stratigraphic thickness interpolation have significant advantages over traditional elevation-based modeling. Elevation-based modeling directly interpolates the absolute spatial coordinates of stratigraphic interfaces, which conflates later tectonic deformation signals with original sedimentary features. In tectonically complex areas, strong tectonic dips can lead to severe distortion of the apparent stratigraphic thickness. In contrast, true thickness-based modeling first restores the original sedimentary thickness of the stratigraphy, resulting in better structural continuity.

[0003] However, deterministic interpolation methods, including Kriging, are mathematically based on constructing a statistically optimal fit surface for spatial data, which is continuous and smooth. However, this requirement for mathematical smoothness often leads to numerical oscillations at data boundaries or in sparse control point regions, resulting in severe distortion of geological characterization. When such surfaces are applied to geological attributes with strict physical boundaries, such as coal seam thickness, two types of problems may arise:

[0004] (1) The problem of negative thickness. At the pinch-out or erosion boundary of the coal seam, the actual thickness should be zero. However, due to the smoothness requirement of the interpolation function, it will produce a gradual transition when crossing the boundary, resulting in a negative thickness value calculated in the coal-free area, which violates the basic physical law that the formation thickness is non-negative.

[0005] (2) The problem of pseudo-strata without geological basis. Algorithms may generate coal seams with significant thickness out of thin air through mathematical extrapolation in paleohighlands or eroded areas not covered by sedimentary systems. Such pseudo-strata lack geological basis and will seriously affect the accurate evaluation of sedimentary paleogeography and coal resource potential.

[0006] To address this problem, there are two common methods: one is to force the region with negative thickness to have a thickness of 0, but this method cannot solve the problem (2); the other is to triangulate the sample points and constrain the coal seam thickness of the region based on the coal seam thickness of the sample points, but this method cannot be extrapolated and the interpolation oscillation outside the triangular mesh cannot be effectively constrained. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for suppressing formation thickness interpolation oscillations based on an improved KNN.

[0008] The technical solution adopted in this invention is: a method for suppressing formation thickness interpolation oscillations based on an improved KNN, the key technical points of which include the following steps:

[0009] S1: Data preparation: Acquire borehole data, extract the location information, target formation existence label and thickness of each borehole, and construct a sample feature set for spatial analysis and interpolation;

[0010] S2: Intelligent region discrimination: Based on the improved KNN algorithm, the study region is divided into interpolation effective region and interpolation invalid region by using the location information and existence label in the sample feature set;

[0011] S3: Zone thickness assignment: Within the interpolation effective region, based on the location information and thickness of the sample points with an existence label of 1 in the sample feature set, spatial interpolation is performed on the target formation thickness; within the interpolation invalid region, the target formation thickness is forcibly set to zero;

[0012] S4: Model Generation: The interpolation results of the effective interpolation region and the zero thickness results of the ineffective interpolation region are merged to generate the final thickness model.

[0013] In the above scheme, step S1, which involves acquiring borehole data and extracting the location information, target formation presence label, and thickness of each borehole, specifically includes:

[0014] For each borehole, its opening coordinates are obtained as the location information, and the opening coordinates include at least east and north coordinates;

[0015] Traverse its lithological stratification record to determine whether it contains the target stratum; if the target stratum exists, calculate its thickness and mark it as present with a label of 1; if the target stratum does not exist, set its thickness to zero and mark it as present with a label of 0.

[0016] In the above scheme, the division of the study area into an interpolation effective region and an interpolation invalid region in step S2 specifically includes:

[0017] S21: Based on the location information and existence labels of all samples in the sample feature set, calculate the Pearson correlation coefficient between the east and north coordinate directions and the existence labels, and calculate the normalized weights of the east and north directions accordingly.

[0018] S22: Construct the anisotropic distance formula and use the normalized weights to calculate the weighted Euclidean distance between the point to be predicted and each sample point;

[0019] S23: For each point to be predicted within the study area, use the weighted Euclidean distance to find its K nearest neighbor sample points, and calculate the probability of the existence of the strata at the point based on the existence labels of the K nearest neighbor samples.

[0020] S24: Based on a preset probability threshold, points with a probability of existence greater than or equal to the threshold are included in the interpolation effective area, and points with a probability less than the threshold are included in the interpolation invalid area.

[0021] In the above scheme, the spatial interpolation within the effective interpolation region in step S3 specifically includes:

[0022] S31: From the sample feature set, select sample points with an existence label of 1 to form an effective sample subset for thickness interpolation;

[0023] S32: Based on the location information and thickness value of each sample point in the effective sample subset, a spatial interpolation model is established using the Kriging interpolation method;

[0024] S33: For each point to be predicted that falls within the interpolation effective region, calculate its initial thickness estimate using the spatial interpolation model.

[0025] In the above scheme, step S3, which involves forcibly setting the target formation thickness to zero within the interpolation invalid region, specifically includes: for each point to be predicted that is assigned to the interpolation invalid region, regardless of the estimated value calculated based on any interpolation method, its target formation thickness value is directly assigned to zero.

[0026] The beneficial effects of this invention are as follows: This method for suppressing formation thickness interpolation oscillations based on an improved KNN algorithm intelligently pre-determines the spatial distribution range of the formation, dividing the study area into effective and ineffective interpolation regions, and forcibly assigning a thickness value of zero within the ineffective region. This fundamentally eliminates the negative thickness values ​​generated at formation pinch-out boundaries and the pseudo-formation problem extrapolated outside of formation-free regions caused by traditional smooth interpolation methods, ensuring that the thickness model strictly conforms to the basic physical law of formation non-negativity.

[0027] This method eliminates the need for subjective interpretation of pinch-out lines based on human experience. Instead, it leverages the anisotropic correlation coefficients of the sample point data to drive KNN classification, automatically generating stratigraphic zoning maps that conform to geostatistical laws. This method significantly improves the efficiency and objectivity of boundary determination, and the results are closer to geological reality.

[0028] The final thickness model maintains the rationality of spatial interpolation within the effective region and exhibits no oscillation noise in the invalid region. The model has clear boundaries and reliable data, providing an extremely accurate data foundation for subsequent coal resource reserve calculations, coalbed methane exploration and development, mine development design, and geological hazard prediction, effectively reducing decision-making risks caused by model distortion. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart of a method for suppressing formation thickness interpolation oscillations based on an improved KNN in an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram comparing the implementation of the present invention with traditional interpolation methods. Detailed Implementation

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the following description is provided in conjunction with the accompanying drawings. Figures 1-2 The present invention will be further described in detail below with reference to specific embodiments.

[0033] The method for suppressing formation thickness interpolation oscillations based on improved KNN used in this embodiment includes three core stages: data preprocessing, intelligent formation existence determination, and fusion of spatial thickness interpolation and physical constraints. These will be elaborated in detail below using borehole data structures.

[0034] S1 Data Preprocessing and Feature Extraction.

[0035] The input data is a set of multiple boreholes. The data structure of each borehole includes: project number, borehole coordinates E (east coordinate), borehole coordinates N (north coordinate), borehole coordinates R (elevation), and several lithological stratification records (from, to, lithology).

[0036] Table 1 shows...

[0037] Project No. Hole coordinates E Hole coordinate N Hole coordinates R Lithology ID1312 8000 5000 20 coal

[0038] S1.1: Determine the target formation: First, define the target formation to be modeled (e.g., "11# coal") based on the "lithology" description.

[0039] S1.2 Calculation of Formation Thickness and Existence Labels for Single Boreholes: For each borehole p i (based on its opening coordinates (E) i N i (Unique identifier), traversing all its lithological stratigraphic records:

[0040] If the record of the target stratum is found, its thickness T is calculated. i = "to" - "from", and assign it the stratigraphic existence label L. i=1. If no record of the target formation is found, the formation at the borehole is considered missing, and its thickness T is [missing information]. i =0, and assign the formation the existence label L. i =0.

[0041] Building the sample set: At this point, each borehole p... i They were all assigned a three-dimensional feature vector (E) i N i ,L i ,T i All the boreholes constituted the sample set for model training.

[0042] S2 is based on an improved KNN-based intelligent identification and partitioning of stratigraphic existence.

[0043] The study area was divided into "interpolation effective area" and "interpolation invalid area".

[0044] S2.1 Calculation of anisotropic weights based on Pearson correlation coefficient:

[0045] Calculate the Pearson correlation coefficient ρ between the eastern coordinate E and the label L. E,L And the Pearson correlation coefficient ρ between the north coordinate N and the label L N,L .

[0046]

[0047] Calculate the anisotropy weights. The larger the absolute value of the correlation coefficient, the greater the influence of that coordinate direction on determining the presence of strata.

[0048] ω E =|ρ E,L |,ω N =|ρ N,L |

[0049] Normalize the weights:

[0050]

[0051] S2.2 Constructing the anisotropic distance formula: For any grid point q(E) to be predicted within the study area... q N q ), which is related to sample point p i (E i N i The anisotropic weighted distance d ani Defined as:

[0052]

[0053] This formula ensures that the difference between two points contributes more to the "similarity" determination in coordinate directions with greater influence.

[0054] S2.3 Performs KNN classification and partitioning:

[0055] For each grid point q, use the above d ani Distance, find its K nearest borehole sample points N K (q).

[0056] Calculate the probability that the strata exist at point q:

[0057]

[0058] Set a probability threshold θ = 0.5 and divide the region accordingly:

[0059] Interpolation effective region Ω valid : The target stratum was determined to exist in the area.

[0060] Invalid interpolation region Ω invalid : The target strata in this area are determined to be missing.

[0061] S3 spatial thickness interpolation and oscillation suppression.

[0062] Initial thickness interpolation: using only all L i =1 borehole sample point and its thickness T i Using a selected spatial interpolation algorithm, such as Kriging, interpolation is performed on the entire study area to obtain the initial thickness value of each grid point q.

[0063] Results Fusion and Oscillation Suppression: The partitioning results from the second step are used as physical constraints and fused with the initial thickness interpolation results to generate the final thickness model T. final (q):

[0064]

[0065] Analysis of the effect of the method in this embodiment:

[0066] This embodiment takes the exploration of coal resources in a certain mining area as an example. The method of the present invention is used to construct a spatial thickness model of coal seam No. 11 in order to accurately delineate the coal seam boundary and assess the resource quantity.

[0067] Step 1: Data preparation and feature extraction.

[0068] Data from 125 boreholes in a mining area were collected. The data structure for each borehole is shown in Table 1, including the project number, borehole coordinates (East coordinate E, North coordinate N, elevation R), and multiple lithological stratification records (from, to, lithology).

[0069] Table 1 shows the original borehole data structure.

[0070]

[0071] For the target formation "11# coal", the stratigraphic records of all 125 boreholes were traversed:

[0072] For borehole ZK001: A target formation was found; thickness T was calculated. 001 =6.2-4.8=1.4m, mark the existence tag L 001 =1.

[0073] For borehole ZK002: If the target formation was not found, its thickness T is set. 002 =0m, and mark the existence of label L. 002 =0

[0074] This process is repeated for all boreholes. Ultimately, each borehole is represented as a sample point with a four-dimensional feature vector (E...). i N i ,L i ,T i ), where (E i N i L represents the borehole opening coordinates. i The formation has a label (value is 0 or 1), T i The thickness of the formation is represented by the actual thickness when it exists, and 0 when it is missing. All 125 sample points constitute the sample feature set used for subsequent analysis.

[0075] Step 2: Stratigraphic existence determination and partitioning based on improved KNN.

[0076] This step mainly involves intelligently dividing the entire study area, determined based on the coordinate range of all boreholes, into areas where coal seams may exist (i.e., effective interpolation areas) and areas where coal seams are not identified (i.e., invalid interpolation areas).

[0077] 1. Calculate the anisotropy weights. Based on the data from all 125 sample points in the sample feature set, calculate the Pearson correlation coefficient between the eastern coordinate E and the stratigraphic presence label L, and the correlation coefficient between the northern coordinate N and the stratigraphic presence label L. The calculations are as follows:

[0078] ρ E,L =0.68 (Correlation coefficient between eastern coordinates and stratigraphic existence)

[0079] ρ N,L =0.42 (Correlation coefficient between north coordinates and stratigraphic existence)

[0080] This indicates that in this region, the continuity of stratigraphic distribution in the east-west direction has a more significant impact on the presence of coal seams than in the north-south direction.

[0081] Calculate the normalized weights:

[0082]

[0083] 2. Construct anisotropic distances.

[0084] For any point q(E) to be predicted within the study area q N q ), which is related to sample point p i (E i N i The weighted distance is:

[0085]

[0086] This distance formula makes the difference in coordinates in the east-west direction (with higher weight) contribute more to the determination of the similarity between two points, and is more in line with the geological trend of the area.

[0087] 3. KNN classification and probability calculation.

[0088] The study area was discretized into a regular grid of 10m × 10m, resulting in 15,600 grid points. For each grid point, K = 8 (nearest neighbor number). For each grid point q, the 8 nearest neighbor sample points were found from the 125 sample points. The probability of the formation existing at point q was calculated: for example, for point q (3547000.00, 394500.00): among the 8 nearest neighbors found, 6 sample points L... j =1, confirming the existence of a coal seam, L of 2 sample points j =0, indicating a missing coal seam. Therefore:

[0089]

[0090] It is proven that each point q falls within the effective interpolation region, thus determining that a coal seam may exist at that location.

[0091] Step 3: Integration of partitioned spatial interpolation with physical constraints.

[0092] From 125 boreholes, 89 sample points containing the target formation were selected (L). i =1), forming a valid sample subset, these points contain true thickness information. A spatial interpolation algorithm using ordinary kriging is employed. Based on the positions (E,N) and thicknesses of 89 points in the valid sample subset, a kriging interpolation model is established. Using this model, interpolation calculations are performed on all points within the defined valid interpolation region to obtain the initial thickness value for each grid point within the study area.

[0093] The partitioning results are used as physical constraints and fused with the initial interpolation results to generate the final thickness model.

[0094] The final thickness model of the study area "11# coal" generated in this embodiment shows that the effective interpolation area is approximately 3.8 km². 2 The thickness ranges from 0.5 to 1.9 meters, consistent with geological understanding; the interpolation invalid area covers approximately 1.2 km². 2 The thickness is strictly zero; the boundary is clear: the transition between the effective zone and the ineffective zone is natural, eliminating the false thickness values ​​at the stratigraphic pinch-out boundary in traditional methods.

[0095] Comparative experiments have demonstrated that the stratigraphic model generated by the method in this embodiment has clear boundaries and accurate pinch-out line positions, completely eliminating the thickness values ​​of invalid zones. This provides an extremely reliable data foundation for subsequent engineering design and resource assessment, effectively reducing the technical risks caused by model distortion.

[0096] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for suppressing formation thickness interpolation oscillations based on an improved KNN, characterized in that, Includes the following steps: S1: Data preparation: Acquire borehole data, extract the location information, target formation existence label and thickness of each borehole, and construct a sample feature set for spatial analysis and interpolation; S2: Intelligent region discrimination: Based on the improved KNN algorithm, the study region is divided into interpolation effective region and interpolation invalid region by using the location information and existence label in the sample feature set; S3: Zone thickness assignment: Within the interpolation effective region, based on the location information and thickness of the sample points with an existence label of 1 in the sample feature set, spatial interpolation is performed on the target formation thickness; within the interpolation invalid region, the target formation thickness is forcibly set to zero; S4: Model Generation: The interpolation results of the effective interpolation region and the zero thickness results of the ineffective interpolation region are merged to generate the final thickness model.

2. The method for suppressing formation thickness interpolation oscillations based on improved KNN according to claim 1, characterized in that, The acquisition of borehole data in step S1, including extracting the location information, target formation presence label, and thickness of each borehole, specifically includes: For each borehole, its opening coordinates are obtained as the location information, and the opening coordinates include at least east and north coordinates; Traverse its lithological stratification record to determine whether it contains the target stratum; if the target stratum exists, calculate its thickness and mark it as present with a label of 1; if the target stratum does not exist, set its thickness to zero and mark it as present with a label of 0.

3. The method for suppressing formation thickness interpolation oscillations based on improved KNN according to claim 1, characterized in that, The division of the study area into an effective interpolation region and an ineffective interpolation region in step S2 specifically includes: S21: Based on the location information and existence labels of all samples in the sample feature set, calculate the Pearson correlation coefficient between the east and north coordinate directions and the existence labels, and calculate the normalized weights of the east and north directions accordingly. S22: Construct the anisotropic distance formula and use the normalized weights to calculate the weighted Euclidean distance between the point to be predicted and each sample point; S23: For each point to be predicted within the study area, use the weighted Euclidean distance to find its K nearest neighbor sample points, and calculate the probability of the existence of the strata at the point based on the existence labels of the K nearest neighbor samples. S24: Based on a preset probability threshold, points with a probability of existence greater than or equal to the threshold are included in the interpolation effective area, and points with a probability less than the threshold are included in the interpolation invalid area.

4. The method for suppressing formation thickness interpolation oscillations based on improved KNN according to claim 1, characterized in that, The spatial interpolation performed within the effective interpolation region in step S3 specifically includes: S31: From the sample feature set, select sample points with an existence label of 1 to form an effective sample subset for thickness interpolation; S32: Based on the location information and thickness value of each sample point in the effective sample subset, a spatial interpolation model is established using the Kriging interpolation method; S33: For each point to be predicted that falls within the interpolation effective region, calculate its initial thickness estimate using the spatial interpolation model.

5. The method for suppressing formation thickness interpolation oscillations based on improved KNN according to claim 1, characterized in that, The step S3, which involves forcibly setting the target formation thickness to zero within the interpolation invalid region, specifically includes: for each point to be predicted that is assigned to the interpolation invalid region, regardless of the estimated value calculated based on any interpolation method, its target formation thickness value is directly assigned to zero.

Images