Deep fracture surface reconstruction method based on microseismic signals

Abstract

The invention belongs to the technical field of deep rock mass engineering, and discloses a deep fracture surface reconstruction method based on micro-seismic signals, which comprises the following steps of: denoising and classifying a micro-seismic event to obtain more than one fracture surface point set based on a two-step DBSCAN (Density Based Spatial Clustering of Applications with Noise) algorithm; then, taking the normalized moment magnitude of each micro-seismic event as the weight of the micro-seismic event, and fitting the position of the micro-seismic event in each fracture surface point set by using a weighted least square method; and finally, based on a projection method, determining a fracture surface boundary obtained by fitting, and reconstructing a deep fracture surface. The method is a new method for reconstructing the fracture surface by using the position of the microseism event and the spatial distribution of the moment magnitude, the accuracy and effectiveness of fracture surface reconstruction can be improved, and the calculation cost can be remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of deep rock mass engineering, and relates to a deep fracture surface reconstruction method based on microseismic signals in the hydraulic fracturing and microseismic detection in deep resource mining. BACKGROUND

[0002] For deep unconventional shale gas mining, hydraulic fracturing technology greatly improves the efficiency and feasibility of unconventional reservoir mining. Due to the deep burial depth of deep reservoirs, conventional detection techniques cannot effectively obtain fracture information after fracturing. Microseismic monitoring is applied to the monitoring and analysis of the hydraulic fracturing process because it can obtain wave signals in thousands of meters deep. The microseismic monitoring method is based on the formation of shear fractures and tensile fractures by injecting fracturing fluid into the reservoir. When the formation is broken, seismic waves are generated. These seismic waves can be captured and recorded by geophones. By analyzing the seismic waves, the source location and intensity of the microseismic event can be obtained. Therefore, the extension range and extension direction of the fracture network can be determined by using the microseismic event, and the fracture morphology can be reconstructed, which can be used to evaluate the fracturing performance and optimize the design and decision-making. However, in actual monitoring, sensors are usually located on the ground and are subject to interference noise events caused by many factors. It is still a great challenge to reconstruct the deep fracture surface.

[0003] In deep shale reservoirs, most microseismic events are recorded under low signal-to-noise ratio (SNR) conditions, resulting in a large number of low-magnitude noise events in the interpretation results. According to the generation mechanism of noise events, these noise events can be divided into two categories: the first category is caused by environmental noise, which is usually random and unrelated to fracturing activities; the second category is caused by the slip of existing fractures far from the hydraulic fracture, which may reflect the stress disturbance in the reservoir. Although the first category of environmental noise can be eliminated by the signal-to-noise ratio limit, it is difficult to distinguish the second category of non-environmental noise from effective events. Traditional methods analyze all microseismic events directly to obtain the fracture structure, which leads to an overestimation of the number of fractures and misleading judgments of the fracture geometry.

[0004] In recent years, the density-based spatial clustering (DBSCAN) algorithm has been frequently proposed for removing noise events, which is also a powerful denoising technique. However, there are still two main problems in using DBSCAN combined with other methods to construct a microseismic event-based fracture configuration: (1) how to ensure that the fracture configuration is strongly related to the characteristics of the microseismic event; (2) in the analysis of microseismic events, although the parameters of the DBSCAN algorithm have a great influence on the clustering results, it is still difficult to quickly determine them.

[0005] In summary, how to realize fast and effective reconstruction of the fracture surface in microseismic detection based on the DBSCAN algorithm is still a technical problem to be solved. SUMMARY

[0006] The purpose of the present application aims at the technical problems existing in the prior art, and provides a deep fracture surface reconstruction method based on microseismic signals, which can effectively remove noise and realize effective reconstruction of the deep fracture surface.

[0007] Hydraulic fracturing greatly improves the permeability of the reservoir by creating fractures around the wellbore or establishing new hydraulic fractures connected to natural fractures, usually extending tens to hundreds of meters around the wellbore. Induced seismic activity is usually concentrated near the wellbore, but pressure disturbances generated during the fracturing process can propagate through highly permeable channels to more distant areas, possibly inducing earthquakes hundreds or even thousands of meters away from the wellbore. This increase in permeability is reflected not only in the spatial distribution of fractures but also in the spatial pattern of induced seismic activity. Therefore, locating the fracture surface through the spatial distribution of microseismic events is an effective method to determine the fracture geometry. This method is crucial for evaluating the fracture extension during hydraulic fracturing and managing and mitigating potential geological disasters.

[0008] The present application adopts normalized moment magnitude as the weight factor for constructing the fracture surface, and the moment magnitude of each microseismic event plays a key role in the final fracture surface geometry. Moment magnitude is a standard for measuring the energy released by a microseismic event. Magnitude M w <3> earthquakes, usually not felt on the surface, are called "microseismic events" or "microseismic activity". Most of these events are in the range of -3.0≤M w ≤0. The higher the moment magnitude, the larger the scale of rock damage, and the greater the likelihood of inducing major fractures. Previous studies have shown that using moment magnitude as a key parameter to determine the location of the fracture surface is a reasonable approach. It is crucial that, due to the complex mechanism of hydraulic fracturing-induced earthquakes, microseismic positioning alone cannot capture the multi-scale, complex geometry of the final fracture surface.

[0009] In view of this, the present application uses the spatial distribution of the positions and moment magnitudes of microseismic events to derive the geometry of the main fracture structure.

[0010] The present application provides a deep fracture surface reconstruction method based on microseismic signals, comprising the following steps: S1, based on a two-step DBSCAN algorithm, denoising and classifying microseismic events to obtain more than one fracture surface point set; S2, taking the normalized moment magnitude of each microseismic event as the weight of the microseismic event, using weighted least squares to fit the positions of the microseismic events in each fracture surface point set in step S1; S3, based on the projection method, determining the fracture surface boundary obtained by fitting, generating the reconstructed deep fracture surface; for any fracture surface point set classified in step S1, the following steps are performed: S31, projecting all microseismic event positions in the fracture surface point set to the coordinate plane with the largest projection area to obtain a projected point set; S32, identifying the boundary line of the projected point set; S33, projecting the boundary line of the projected point set to the fracture surface obtained by fitting in step S2 to obtain the fitted fracture surface boundary, and generating the reconstructed deep fracture surface.

[0011] In the above step S1, the DBSCAN algorithm is used twice for noise identification and removal, and the microseismic events are classified. Based on the first step DBSCAN algorithm, the noise points and isolated points far from the microseismic events are filtered out, and the remaining effective microseismic events are more concentrated and uniformly distributed in space; then, based on the second step DBSCAN algorithm, the denoised microseismic events are classified, and the noise points in the low-density area can be further identified and removed, improving the classification accuracy, especially for microseismic events with overlapping in space or time and complex structure.

[0012] The neighborhood search radius epsilion (Eps, ) and the minimum point threshold (MinPts) are two key hyperparameters. A larger can cluster events that are far apart in space but still belong to the same geological structure or source area, thereby preventing the deletion of potential valid signals; a smaller can more accurately identify the spatial relationship between microseismic events; therefore, the present application first uses a larger to identify and remove noise that exists alone in the microseismic events, and to cluster the microseismic events to effectively filter out noise points and isolated points far from the main event cluster; then a smaller is used to identify and remove noise in the fracture boundary, further cluster the microseismic events and obtain more than one fracture surface point set.

[0013] The present application further determines the neighborhood search radius and the minimum point threshold of the two-step DBSCAN algorithm by a parameter optimization method based on the contour coefficient.

[0014] Specifically, starting from the minimum value of the set MinPts value, the following iterative search is performed; select the and that maximize the contour coefficient S: For the current MinPts value, let k = MinPts-1, calculate the distance from each point to its k nearest neighbors, and sort the resulting distances in ascending order to obtain a k-distance curve that reflects the trend of spatial density variation, i.e., a k-distance distribution graph; then take the slope mutation point (i.e., the point at which the slope on the curve changes significantly, which can be determined by the difference quotient method or the derivative method) in the k-distance curve as the inflection point, and the distance corresponding to the inflection point is the corresponding ; According to the determined and the current MinPts value, all microseismic events are clustered, and the silhouette coefficient of each cluster category is calculated using the following formula : ; In the formula, is -1~1, denotes the average distance from point j to other points in the same cluster; denotes the average distance from point j to all points in the nearest cluster, and N is the number of all microseismic events.

[0015] The purpose of the above step S2 is to fit the positions of the microseismic events in each fracture surface point set to obtain the best fitting surface.

[0016] According to step S1, the coordinates of each microseismic event in each fracture surface point set are determined, and the positions of the microseismic events on each fracture surface are fitted by the weighted least squares method. The fitting surface control equation is as follows: ; where x and y represent the coordinates of the microseismic event in XOY, and z predicted represents the Z coordinate fitting value, , , represents the fitting parameter.

[0017] Further, the following fitting error is defined as the objective function J: ; In the formula, n is the number of microseismic events in the fracture surface, M wl,i is the weight of the i th microseismic event, z predicted,i represents the Z coordinate fitting value of the i th microseismic event, and z actual,i represents the actual value of the Z coordinate of the i th microseismic event. By minimizing the objective function, the best fitting fracture surface is obtained. Specifically, different values of a, b, and c are given to obtain different surface predicted values z predicted . When the fitting error between the actual value and the predicted value of the surface is minimized, i.e., the value of J is minimized, the best fitting fracture surface is obtained.

[0018] using normalized moment magnitude as the weight assigned to the i-th microseismic event, ; wherein, is the i-th moment magnitude, , is the seismic moment of the i-th microseismic event, , , , and respectively represent the thermal modulus of the fault medium, the fault slip area and the average fault slip displacement corresponding to the i-th microseismic event; The range value is -3-3.

[0019] The purpose of the above step S3 is to generate the boundary of each fracture surface.

[0020] In the above step S31, the fracture surface point set is projected on all microseismic event position points to obtain the maximum projection plane, which is realized by setting the maximum projection plane normal axis coordinate in the three-dimensional coordinates of all microseismic event positions to zero.

[0021] In the above step S32, the 2D Alpha-shape algorithm is used to identify the boundary of the projected point set.

[0022] In the above step S33, the boundary line coordinates of the projected point set are substituted into the fitting surface control equation to obtain the final boundary of the fracture surface, and then the reconstructed deep fracture surface is generated by combining the fracture surface fitting surface.

[0023] Compared with the prior art, the deep fracture surface reconstruction method based on microseismic signals provided by the present application has the following beneficial effects: (1) The present application first denoises and classifies the microseismic events based on a two-step DBSCAN algorithm, then fits each fracture surface obtained by clustering, and finally identifies the boundary line by two-step projection combined with the 2D Alpha-shape algorithm, and projects the boundary line back to the fitted fracture surface to obtain the reconstructed deep fracture surface, which not only improves the accuracy and effectiveness of the fracture surface reconstruction, but also significantly reduces the calculation cost; (2) The application of the two-step DBSCAN algorithm can obtain an accurate fracture surface area; (3) The present application introduces the moment magnitude as a weight factor in the process of fitting the fracture surface, so that the contribution of microseismic events of different magnitudes in the fracture surface reconstruction can reflect its physical meaning, thereby improving the accuracy and robustness of the fitting; (4) The present application can effectively optimize the fitting accuracy of the fracture surface geometry and data points by minimizing the objective function in the fitting process; (5) The application adaptively constructs the boundary of the data point set through the 2D Alpha-shape algorithm, solves the deficiency of the traditional convex hull method in processing irregular crack geometry, and thus improves the accuracy and adaptability of boundary extraction. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A flowchart of a deep crack face reconstruction method based on microseismic signals is provided for the embodiments of the application. Figure 2 A flowchart for determining the crack face boundary. Figure 3 A k-distance distribution diagram when the MinPts value is 44. Figure 4 A process diagram for determining the crack face boundary based on the 2D Alpha-shape algorithm; wherein (a) is to generate the maximum projection plane, (b) is to generate the boundary of the identified projection point set, and (c) is to generate the crack face boundary. Figure 5 A spatial distribution diagram of microseismic events recorded during hydraulic fracturing (scaled by moment magnitude): (a) is a map view of microseismic events, (b) is an east-west direction depth view, and (c) is a north-south direction depth view. Figure 6 A moment magnitude distribution diagram of microseismic events, wherein (a) is the microseismic event distribution before and after denoising using the first step DBSCAN; (b) is the microseismic event distribution corresponding to the three crack faces obtained by classification. Figure 7 A spatial distribution diagram of the geometric structure of the three reconstructed crack faces and the corresponding microseismic events. Figure 8 A comparison diagram of the Euclidean distance between the microseismic events and the crack data on the three crack faces. Figure 9 A comparison diagram of the strike and dip of the focal mechanism solution of the microseismic events and the crack data; wherein (a) and (d) correspond to the strike and dip of crack face 1; (b) and (e) correspond to the strike and dip of crack face 2; (c) and (f) correspond to the strike and dip of crack face 3. DETAILED DESCRIPTION

[0025] The technical solutions of the embodiments of the application will be described clearly and completely in combination with the drawings. Obviously, the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the application.

[0026] ​​Embodiment 1

[0027] The embodiment provides a deep fracture surface reconstruction method based on a microseismic signal, as shown in the formula: Figures 1-2 The method comprises the following steps: S1, based on a two-step DBSCAN algorithm, denoising and classifying microseismic events to obtain more than one fracture surface point set.

[0028] In this step, the microseismic events are formatted into a string, and then the DBSCAN algorithm (see Ester, M., Kriegel, H.P., Sander, J., Xiaowei, X., 1996. A density-based algorithm for discovering clusters in large spatial databases with noise. KDD-96 Proceedings. Second International Conference on Knowledge Discovery and Data Mining, 226-231) is used twice to identify and remove noise and cluster the microseismic events. Based on the first-step DBSCAN algorithm, the microseismic events are denoised, and noise points and isolated points far from the microseismic events are filtered out, and the remaining effective microseismic events are more concentrated and uniformly distributed in space; subsequently, based on the second-step DBSCAN algorithm, the denoised microseismic events are classified, and noise points in a low-density area can be further identified and removed, the clustering accuracy is improved, and the microseismic events with overlapping in space or time and complex structure have a good effect.

[0029] The neighborhood search radius epsilion (Eps, ) and the minimum point threshold (MinPts) are two key hyperparameters. A larger can cluster events that are far apart in space but still belong to the same geological structure or source area, thereby preventing the deletion of potential effective signals; a smaller can more accurately identify the spatial relationship between microseismic events; therefore, the present application first uses a larger to identify and remove noise that exists in isolation in the microseismic events, and at the same time, the microseismic events are clustered to effectively filter out noise points and isolated points far from the main event cluster; then a smaller is used to identify and remove noise in the fracture boundary, and the microseismic events are further clustered to obtain more than one fracture surface point set.

[0030] The embodiment determines the neighborhood search radius and the minimum point threshold of the two-step DBSCAN algorithm respectively by a parameter optimization method based on the silhouette coefficient . .

[0031] Specifically, starting from the minimum value of the set MinPts value, the iterative search is performed according to the following operations; the MinPts value that maximizes the silhouette coefficient S is selected and : For the current MinPts value, let k = MinPts-1, calculate the distance from each point to its k nearest neighbors, and sort the obtained distances in ascending order to obtain a k-distance curve reflecting the trend of spatial density change, that is, a k-distance distribution diagram; Figure 3 The k-distance distribution diagram when the MinPts value = 44 is given; then the inflection point in the k-distance curve is taken as the inflection point (such as 3436 in Figure 3 ), and the distance corresponding to the inflection point is the corresponding under the current MinPts value; According to the determined and the current MinPts value, all microseismic events are clustered, and the silhouette coefficient of each cluster category is calculated by the following formula : ; In the formula, is-1~1, , represents the average distance from point j to other points in the same cluster; represents the average distance from point j to all points in the nearest cluster, and N is the number of all microseismic events.

[0032] The values of and in the first-step DBSCAN algorithm and the second-step DBSCAN algorithm are optimized respectively by the above parameter optimization method based on the silhouette coefficient.

[0033] S2, taking the normalized magnitude of each microseismic event as the weight of the microseismic event, the positions of the microseismic events in each fracture surface point set in step S1 are fitted by using the weighted least squares method.

[0034] According to the coordinates of each microseismic event in each fracture surface point set determined in step S1, the positions of the microseismic events on each fracture surface are fitted by using the weighted least squares method, and the fitting surface control equation is as follows: ; Wherein, x, y represent the coordinates of the microseismic event in XOY, z predicted represents the fitting value of Z coordinate, a, b, c represent the fitting parameters.

[0035] The fitting error is defined as the objective function J: ; where n is the number of microseismic events in the fracture surface, M wl,i is the weight of the i-th microseismic event, z predicted,i represents the Z coordinate fitting value of the i-th microseismic event, z actual,i represents the Z coordinate actual value of the i-th microseismic event.

[0036] By introducing an optimization algorithm, the plane parameters are gradually adjusted, so that the value of the objective function gradually decreases. The objective function provides a clear evaluation standard for the optimization problem, which is used to measure the pros and cons of the model parameters. By optimizing the objective function, a set of parameters can be found to minimize the value of the objective function, so that the fitting effect or performance is optimal.

[0037] Specifically, different values of a, b, and c are given to obtain different surface prediction values z predicted When the fitting error between the actual value and the predicted value of the surface is minimized, that is, when the value of J is minimized, the fracture surface with the best fitting effect is obtained.

[0038] The moment magnitude can accurately represent the energy release of microseismic events, and provides important geophysical information in microseismic monitoring of deep shale gas development and exploitation. In this embodiment, the normalized moment magnitude is assigned as the weight to the i-th microseismic event, ; where is the i-th moment magnitude, , is the seismic moment of the i-th microseismic event, ), , , and represent the thermal modulus, fault slip area and average fault slip displacement of the i-th microseismic event corresponding to the fault medium, respectively. The range value is -3~3.

[0039] S3, based on the projection method, determines the boundary of the fitted fracture surface to generate a reconstructed deep fracture surface.

[0040] This embodiment proposes a method of applying 2D Alpha-shape algorithm and two-step projection to obtain the fracture surface boundary. For any fracture surface point set classified in step S1, as shown in Figure 4 , the following steps are performed: S31, project all microseismic event positions in the fracture surface point set to the coordinate plane with the largest projection area to obtain a projected point set.

[0041] Project all microseismic event position points in the fracture surface point set to obtain the maximum projection plane. Here, this is achieved by setting the maximum projection plane normal axis coordinate in the three-dimensional coordinates of all microseismic event positions to zero. As shown in FIG. 8 (a), when the projection area of the fracture surface in the XOY plane is the largest, only the Z coordinate of each microseismic event on the fracture surface needs to be set to 0. Figure 4

[0042] S32, identify the boundary line of the projected point set.

[0043] This step uses the 2D Alpha-shape algorithm to identify the boundary of the projected point set. Alpha-shape is a generalized convex hull algorithm for a given point set, which can more accurately represent the shape of the point set. In this embodiment, the Alpha-shape algorithm is used to identify the microseismic event point boundary on the two-dimensional plane after projection by rolling the edge according to the set radius a. For specific operations, see H. Edelsbrunner, D. Kirkpatrick and R. Seidel, "On the shape of a set of points in the plane," in IEEE Transactions on Information Theory, vol. 29, no. 4, pp. 551-559, July 1983, doi: 10.1109 / TIT.1983.1056714.

[0044] It is worth noting that although this method provides high accuracy for polygonal fracture configuration, careful consideration of the projection direction is essential. This step can ensure accurate representation of the fracture boundary, thereby minimizing distortion in the final result.

[0045] S33, project the boundary line of the projected point set to the fracture surface fitted in step S2 to obtain the fitted fracture surface boundary and generate the reconstructed deep fracture surface.

[0046] This step substitutes the boundary line coordinates of the projected point set into the fitted surface control equation to obtain the final boundary of the fracture surface, and then generates the reconstructed deep fracture surface in combination with the fracture surface fitting surface.

[0047] Application example

[0048] ​This application example compares the fracture network formed after multi-stage hydraulic fracturing in the A section of the Luzhou Oilfield with the fracture surface obtained by the deep fracture surface reconstruction method based on microseismic signals provided by the present invention. The above content is described in detail below.

[0049] This well section, located in the Luzhou Oilfield, ranges from 3,100 to 3,400 meters in depth and covers a horizontal length of approximately 1,550 meters. Multi-stage hydraulic fracturing was performed by maintaining high flow rates and continuous sand injection to create a fracture network and maximize reservoir permeability. During the stimulation treatment, microseismic events were recorded using surface geophones arranged in a radial configuration. This configuration consists of 1,312 geophone stations distributed across 12 survey lines.

[0050] like Figure 5 The figure shows the spatial distribution of microseismic events recorded during hydraulic fracturing (scaled by moment magnitude), which contains 4110 microseismic events. The range is between -2.72 and 2.19, and a large number of microseismic events are observed to occur in groups around well section A, while the density of microseismic events is relatively low away from well section A. In addition, a considerable number of microseismic events in the group show moment magnitude This observation suggests that the fracture plane may be roughly perpendicular to the Earth's surface.

[0051] According to the deep fracture surface reconstruction method based on microseismic signals given above, the hyperparameters of the two-step DBSCAN algorithm are first determined by the parameter optimization method based on the silhouette coefficient given above. and , as shown in Table 1.

[0052] Then, the hyperparameters shown in Table 1 are used. and , the microseismic events are denoised and classified according to the two-step DBSCAN algorithm given in step S1. In this application example, the minimum value of MinPts in the two-step DBSCAN algorithm is 2.

[0053] Table 1 DBSCAN parameter selection results .

[0054] Moment magnitude of microseismic events Distribution Figure 6 As shown, from Figure 6 It can be seen that the moment magnitude of most microseismic events is All of them are less than -1, and a considerable number of them are between -3 and -2. After denoising and classification by the two-step DBSCAN algorithm, the remaining microseismic events still show a similar moment magnitude distribution. Figure 6(a) shows that a considerable number of microseismic events (more than 1000 events) are filtered out in the denoised data set. The moment magnitudes of these filtered microseismic events are Between -3 and -2. Figure 6 (b) Shows the moment magnitude of the microseismic events corresponding to the three fracture surfaces. It can be seen that the moment magnitudes of most microseismic events corresponding to the three fracture surfaces are all lower than -1, and their distribution patterns are similar to Figure 6 The event distribution after the first step of DDSCAN denoising is similar as presented in (a).

[0055] By identifying the moment magnitude Using the normalized moment magnitude given above The normalized moment magnitude of each microseismic event is obtained by the calculation formula .

[0056] Then, according to step S2, the fracture surface with the best fitting effect is obtained by combining the fitting control equation and the objective function through the least squares method.

[0057] Then, according to the above step S3, the reconstructed deep fracture surface is generated by combining the 2D Alpha-shape algorithm and the two-step projection, as shown in Figure 7 shown. Figure 7 Three polygonal models representing the fracture geometry are presented.

[0058] Figure 8 The Euclidean distance comparison diagram between microseismic events on three fracture surfaces. Figure 8 The box plots and trend lines in the figure show that the majority of microseismic events on the three fracture planes are concentrated within 20 meters. Furthermore, the median distances for fracture planes 1, 2, and 3 are 12.17, 15.42, and 20.35 meters, respectively. 75% of the microseismic events on the three fracture planes fall within the interquartile range of 21.38, 23.22, and 37.59 meters, respectively, indicating a strong correlation between fracture plane geometry and microseismic events. Fracture planes 1 and 2 have 19 and 1 outliers, respectively, while fracture plane 3 has 77 outliers. The number of microseismic events on fracture planes 1, 2, and 3 is 632, 186, and 1403, respectively, with outlier proportions of 3.01%, 0.54%, and 5.49%, respectively.

[0059] right Figure 7Three groups of events around the fracture surface 1, the fracture surface 2 and the fracture surface 3, the focal mechanisms of three groups of microseismic events around the three fracture surfaces are calculated by the moment tensor inversion (P wave polarity inversion), and the specific calculation process is described in (Hardebeck, J.L., Shearer, P.M., 2002. A new method for determining first-motion focal mechanisms. BULLETIN OF THE SEISMOLOGICAL SOCIETY OF AMERICA 92(6), 2264-2276).

[0060] 66, 47 and 48 microseismic events with relatively small positioning errors are selected from the fracture surface 1, the fracture surface 2 and the fracture surface 3 respectively to calculate the strike and dip of the fracture surface structure, and the calculation formula is as follows: ; ; wherein, is the final calculated dip or strike of the fracture surface; is the dip or strike of the fracture surface of the i-th microseismic event solved based on the above focal mechanism, is the Euclidean distance from the i-th microseismic event to the fracture surface fitted in the step S2; is a hyperparameter for controlling the influence of distance on weight, and in the application example is set to 1.

[0061] The strike and dip of the fracture surface structure are calculated and shown in Table 2. The microseismic data is shown in Figure 9 , which gives the focal mechanism analysis and fracture surface structure analysis to facilitate the comparison of the results of the crack reconstruction method and the moment tensor inversion method about the spatial trend of crack extension. From Figure 9 , it can be seen that the focal mechanism results of the clusters around the fracture surface 1 and the fracture surface 2 are in good agreement. Although the strike data of the fracture surface 3 is consistent, the dip data shows greater variability and complexity. The strike of the fracture surface 1 is 38.41°, and the dip is 64.13°, which is very consistent with the dominant direction observed in the microseismic. The strike of the fracture surface 2 is 50.45°, and the dip is 68.07°, which is very consistent with the main direction of the microseismic. The strike of the fracture surface 3 is 19.26°, and the dip is 76.75°, which is very consistent with the strike result of the microseismic. In summary, the crack configuration direction obtained by the focal mechanism analysis is consistent with the main direction of the calculated fracture surface.

[0062] Table 2 Fracture structure: event count, strike and dip .

[0063] Therefore, in the present application, the DBSCAN method is used in two steps for denoising and classification, which avoids over-sensitivity to event data in a single processing step, thereby reducing the risk of over-fitting (i.e., over-reliance on details) or under-fitting (i.e., failure to identify key structures). After initial denoising, the data volume is reduced, thereby providing a more streamlined dataset for subsequent classification. This not only improves computational efficiency, but also reduces the risk of false classification caused by noise points in the classification process. The present application also uses an optimization method that combines a k-distance map with a contour coefficient-based parameter optimization method to select the hyperparameters of DBSCAN. Compared with the traditional grid search method, this method provides a clear indicator for evaluating the quality of clustering and reduces the computational cost. In addition, for datasets with an undetermined number of clusters, this method can improve the accuracy of clustering.

[0064] A key difference between the present application and previous studies is the use of a weighted least trimmed squares optimization algorithm based on moment magnitude to determine the spatial extension trend of the fracture structure. In contrast, previous studies often relied on the random sample consensus (RANSAC) method or simpler optimization techniques based on this method. The least squares minimization method is beneficial for more accurate fitting of major events, especially when these events exhibit different energy levels. The addition of moment magnitude weights allows the fitting process to better capture the impact of large-scale events on the model. In addition, compared with RANSAC, the least squares method generally has higher computational efficiency, especially when dealing with large datasets, as it does not require a large number of random samples and iterations. Most importantly, the performance of RANSAC is highly dependent on the selection of two key parameters: the maximum number of iterations and the tolerance range. Given that DBSCAN itself requires parameter adjustment, combining the two algorithms and selecting the best parameters becomes increasingly complex.

[0065] Moreover, in previous studies, fracture structures were often assumed to have various shapes, such as rectangular or tree-like fracture models, but in reality, the shape of a fracture is generally irregular. Traditional boundary extraction methods, such as convex hull methods, tend to oversimplify the surface topography of a fracture, while the two-step projection method combined with the 2D Alpha-shape algorithm used in the present application has better anti-interference ability when dealing with noisy data, ensuring that the fitting results accurately reflect the complexity of the actual geological structure.

[0066] Those skilled in the art will appreciate that the embodiments described herein are presented for purposes of illustration and that the inventive principles are not limited to these particular embodiments. Other variations and modifications can be made to the embodiments without departing from the spirit and scope of the inventive principles.

Claims

1. A method for reconstructing deep fracture surfaces based on microseismic signals, characterized in that: The following steps are involved: S1, based on the two-step DBSCAN algorithm, the microseismic events are denoised and classified to obtain more than one fracture surface point set; S2, using the normalized moment magnitude of each microseismic event as the weight of the microseismic event, and fitting the microseismic event position of each fracture surface point set in step S1 using the weighted least squares method; S3, based on the projection method, determine the boundary of the fitted fracture surface and generate a reconstructed deep fracture surface; for any fracture surface point set classified in step S1, perform the following operations: S31, projecting all microseismic event locations in the fracture surface point set onto the coordinate plane with the largest projection area to obtain a projection point set; S32, identifying the boundary line of the projection point set; S33, projecting the boundary line of the projection point set onto the fracture surface fitted in step S2, obtaining a fitted fracture surface boundary, and generating a reconstructed deep fracture surface.

2. The method for deep fracture surface reconstruction based on microseismic signals according to claim 1, characterized in that: In step S1, the microseismic events are denoised based on the first-step DBSCAN algorithm; and the denoised microseismic events are classified based on the second-step DBSCAN algorithm.

3. The method for deep fracture surface reconstruction based on microseismic signals according to claim 2, characterized in that: In step S1, the neighborhood search radius of the two-step DBSCAN algorithm is determined by the parameter optimization method based on the silhouette coefficient. And the minimum point threshold MinPts.

4. The method for deep fracture surface reconstruction based on microseismic signals according to claim 3, characterized in that: Start by setting the minimum value of MinPts and perform iterative search according to the following operations; select the one that maximizes the silhouette coefficient S. and : For the current MinPts value, let k=MinPts-1, calculate the distance from each point to its k nearest neighbor points, and sort the obtained distances from small to large to obtain the k-distance curve; then use the slope mutation point in the k-distance curve as the inflection point, and the distance corresponding to the inflection point is the corresponding distance under the current MinPts value. ; Based on the determined and the current MinPts value, cluster all microseismic events, and calculate the silhouette coefficient of each cluster category using the following formula: : ； Where, -1~1, Represents the average distance from point j to other points in the same cluster; represents the average distance from point j to all points in the nearest cluster, and N is the number of all microseismic events.

5. The method for deep fracture surface reconstruction based on microseismic signals according to claim 1, characterized in that: In step S2, the coordinates of each microseismic event in each fracture surface point set are determined according to step S1, and the positions of the microseismic events on each fracture surface are fitted by weighted least squares method. The fitting surface control equation is as follows: ； Among them, x and y represent the coordinates of the microseismic event in XOY, z predicted represents the Z coordinate fitting value, , , represents the fitting parameters.

6. The method for deep fracture surface reconstruction based on microseismic signals according to claim 5, characterized in that: Define the following fitting error as the objective function J: ； Where n is the number of microseismic events in the fracture surface, M wl,i is the weight of the i-th microseismic event, z predicted,i represents the Z coordinate fitting value of the i-th microseismic event, z actual,i Represents the actual value of the Z coordinate of the i-th microseismic event.

7. The method for deep fracture surface reconstruction based on microseismic signals according to claim 6, characterized in that: Use normalized moment magnitude As the weight assigned to the i-th microseismic event, ； Where, is the i-th moment magnitude, , is the seismic moment of the i-th microseismic event.

8. The method for deep fracture surface reconstruction based on microseismic signals according to claim 1, characterized in that: In step S32, a 2D alpha-shape algorithm is used to identify the boundary of the projected point set.

Images