Construction method of elevation map of bearing stratum, terminal equipment and storage medium

Abstract

The invention provides a construction method of an elevation map of a bearing stratum, terminal equipment and a storage medium. The method comprises the following steps: collecting elevation points of the bearing stratum surveyed in each drilling point; dividing the construction site into a plurality of equidistant original planes according to the elevation points; selecting part of the original plane as a target plane according to the elevation point; fitting a first target curve on the target plane according to the elevation point; extracting basic sequence features from the first target curve on the target plane and the vector input backbone structure from the target plane to the blank plane; respectively inputting the basic sequence features into the plurality of branch structures to generate a plurality of candidate curves and a plurality of confidence coefficients; for a blank plane adjacent to the target plane, screening out one candidate curve as a second target curve according to the elevation point and the confidence coefficient; and fitting an elevation map of the bearing stratum for the construction site according to the first target curve and the second target curve. And the accuracy of the elevation map of the bearing stratum is improved.

Description

Technical Field

[0001] The embodiments of the present application belong to the technical field of deep learning, and in particular, relate to a method for constructing an elevation map of a bearing layer, a terminal device, and a storage medium. Background Art

[0002] When constructing some buildings, such as building a low-rise building next to a high-rise building, in addition to driving engineering piles to support the building and provide compressive resistance, anti-floating anchor rods must also be driven in to prevent the building from having an unstable foundation due to the buoyancy underneath.

[0003] Before drilling anti-floating anchors, it is usually necessary to sample and drill the bearing layer, and use a linear method to construct an elevation map of the bearing layer, so that the anti-floating anchor drilling construction can be carried out according to the elevation.

[0004] However, the elevation map of the linearly constructed bearing layer has a low accuracy and a certain deviation from the actual bearing layer, which increases the workload of drilling construction. Summary of the invention

[0005] In view of this, an embodiment of the present application provides a method for constructing an elevation map of a bearing layer, a terminal device, and a storage medium, so as to improve the accuracy of constructing an elevation map of a bearing layer when resisting the construction of floating anchor rods.

[0006] A first aspect of an embodiment of the present application provides a method for constructing an elevation map of a bearing layer, comprising:

[0007] When performing advance drilling operations on multiple drilling points in the construction site, collecting the elevation points of the bearing layer surveyed in each of the drilling points;

[0008] Dividing the construction site into a plurality of equally spaced original planes according to the elevation points;

[0009] Under the condition that the elevation point is projected onto the original plane closest to the elevation point, a portion of the original plane is selected as the target plane according to the elevation point;

[0010] On the target plane, fitting a first target curve according to the elevation points;

[0011] Loading a sequence prediction network trained on a benchmark height map; the sequence prediction network includes a backbone structure and multiple branch structures;

[0012] The first target curve on the target plane and the vector pointing from the target plane to a blank plane are input into the backbone structure to extract basic sequence features; the blank plane is the original plane adjacent to the target plane;

[0013] Inputting the basic sequence features into the plurality of branch structures respectively to generate a plurality of candidate curves and a plurality of confidences;

[0014] For the blank plane adjacent to the target plane, one of the candidate curves is selected as the second target curve according to the elevation point and the confidence level;

[0015] An elevation map of the bearing layer is fitted to the construction site according to the first target curve and the second target curve.

[0016] A second aspect of an embodiment of the present application provides a device for constructing an elevation map of a bearing layer, comprising:

[0017] The elevation point acquisition module is used to acquire the elevation points of the bearing layer surveyed at each of the drilling points when performing advance drilling operations on multiple drilling points in the construction site;

[0018] An original plane division module, used to divide the construction site into a plurality of original planes with equal spacing according to the elevation points;

[0019] A target plane selection module, for selecting part of the original plane as the target plane according to the elevation point, under the condition that the elevation point is projected onto the original plane closest to the elevation point;

[0020] A first target curve fitting module, used for fitting a first target curve on the target plane according to the elevation points;

[0021] A sequence prediction network loading module is used to load a sequence prediction network trained on a benchmark height map; the sequence prediction network includes a backbone structure and multiple branch structures;

[0022] A basic sequence feature extraction module is used to extract basic sequence features by inputting the first target curve on the target plane and the vector pointing from the target plane to a blank plane into the backbone structure; the blank plane is the original plane adjacent to the target plane;

[0023] A candidate curve generation module, used for inputting the basic sequence features into the plurality of branch structures to generate a plurality of candidate curves and a plurality of confidences;

[0024] A second target curve screening module, for screening one of the candidate curves as a second target curve according to the elevation point and the confidence level for the blank plane adjacent to the target plane;

[0025] An elevation map fitting module is used to fit the elevation map of the bearing layer to the construction site according to the first target curve and the second target curve.

[0026] A third aspect of an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for constructing an elevation map of the bearing layer as described in the first aspect above is implemented.

[0027] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing an elevation map of the bearing layer as described in the first aspect above is implemented.

[0028] A fifth aspect of the embodiments of the present application provides a computer program product. When the computer program product is run on a computer, the computer executes the method for constructing an elevation map of the bearing layer described in the first aspect.

[0029] In this embodiment, when performing advance drilling operations on multiple drilling points in a construction site, the elevation points of the bearing layer surveyed at each drilling point are collected; the construction site is divided into multiple equally spaced original planes according to the elevation points; under the condition that the elevation points are projected onto the nearest original plane, part of the original plane is selected as the target plane according to the elevation points; on the target plane, a first target curve is fitted according to the elevation points; a sequence prediction network trained on the elevation map is loaded; the sequence prediction network includes a backbone structure and multiple branch structures; the first target curve on the target plane and a vector pointing from the target plane to a blank plane are input into the backbone structure to extract basic sequence features; the blank plane is an original plane adjacent to the target plane; the basic sequence features are respectively input into the multiple branch structures to generate multiple candidate curves and multiple confidence levels; for the blank plane adjacent to the target plane, one of the candidate curves is selected as the second target curve according to the elevation points and the confidence level; the elevation map of the bearing layer of the construction site is fitted according to the first target curve and the second target curve. This embodiment can use real elevation points on the target plane to fit a more confident first target curve, fit candidate curves on an adjacent blank plane based on the first target curve, use the elevation points as prior knowledge to correct the confidence of the candidate curve, and thus select a more confident second target curve. The first target curve and the second target curve are combined to provide abundant points to fit the elevation map of the bearing layer, which can effectively improve the accuracy of the elevation map of the bearing layer, reduce the deviation between the elevation map and the actual bearing layer, provide accurate data support for the construction of anti-floating anchors, and reduce the engineering workload of drilling construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or prior art descriptions. Obviously, the drawings described below are only some embodiments of the present application, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 It is a schematic diagram of a method for constructing an elevation map of a bearing layer provided in an embodiment of the present application;

[0032] Figure 2 It is a schematic diagram of the structure of a sequence prediction network provided in an embodiment of the present application;

[0033] Figure 3 It is a structural schematic diagram of a first convolution block structure, a second convolution block structure, a third convolution block structure and a fourth convolution block structure provided in an embodiment of the present application;

[0034] Figure 4 It is a structural schematic diagram of a first decoding block structure, a second decoding block structure and a third decoding block structure provided in an embodiment of the present application;

[0035] Figure 5 It is a schematic diagram of a device for constructing an elevation map of a bearing layer provided in an embodiment of the present application;

[0036] Figure 6 It is a schematic diagram of a terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In the following description, specific details such as specific system structures, technologies, etc. are proposed for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from hindering the description of the present application.

[0038] The technical solution of the present application is described below through specific embodiments.

[0039] Reference Figure 1 , shows a schematic diagram of a method for constructing an elevation map of a bearing layer provided in an embodiment of the present application, which may specifically include the following steps:

[0040] Step 101: When performing advance drilling operations on multiple drilling points in a construction site, the elevation points of the bearing layer surveyed at each drilling point are collected.

[0041] In this embodiment, the designer of the building can set multiple sampling points for construction engineering piles and anchor points for construction anti-floating anchors according to the conditions of the construction site of the building, and construct a real-world coordinate system for the construction site, such as a Cartesian coordinate system, in which drilling points (coordinate points) for construction engineering piles and anchor points (coordinate points) for construction anti-floating anchors are screened out.

[0042] To facilitate subsequent operations, the coordinate information of the sampling points and anchor points in this coordinate system can be converted into actual coordinate information such as longitude and latitude.

[0043] During the construction of engineering piles, advance drilling is performed on the drilling point (i.e. the actual position in the construction site), and the elevation points of the bearing layer surveyed during the advance drilling operation are recorded.

[0044] Step 102: Divide the construction site into a plurality of equally spaced original planes according to the elevation points.

[0045] In this embodiment, the construction site can be divided into multiple equally spaced original planes with reference to the distribution information of the elevation points to obtain original planes with higher density, so that the original planes are closer to the distribution information of the elevation points and the elevation points are concentrated on certain original planes as much as possible.

[0046] In one embodiment of the present application, step 102 may include the following steps:

[0047] Step 1021, vertically project all elevation points onto the horizontal plane to obtain reference points.

[0048] In this embodiment, any horizontal plane can be selected in the three-dimensional space, and all elevation points can be vertically projected onto the horizontal plane, and the elevation points projected onto the horizontal plane are recorded as reference points.

[0049] Step 1022: Add a minimum circumscribed rectangle to the reference point on the horizontal plane as a restriction interval.

[0050] On the horizontal plane, add a minimum enclosing rectangle to all reference points, and the minimum enclosing rectangle is recorded as the restriction interval.

[0051] Step 1023: Under the constraint of the restricted interval, construct the normal vector according to the distribution of the reference points.

[0052] Since the minimum circumscribed rectangle can characterize the overall directionality of the reference point to a certain extent, the restriction interval can be used as a constraint to construct the normal vector according to the distribution of the reference point, so that the normal vector can more accurately characterize the overall directionality of the reference point.

[0053] In a specific implementation, for two adjacent corner points of the restriction interval, multiple line segments may be generated with the corner points as endpoints and at fixed angles between the two restriction lines.

[0054] Among them, the two restriction lines are the edges of the restriction area and intersect at the corner point to realize the range constraint of the normal vector.

[0055] For each line segment, a histogram is constructed for the first distance between the reference point and the line segment; wherein each histogram has a plurality of intervals, and each interval records the reference point whose first distance is located in the interval.

[0056] The total number of reference points is counted in the highest K intervals (ie, the intervals with the largest number of reference points), where K is a positive integer, so that the line segment with the highest total number is converted into a normal vector.

[0057] Step 1024: Divide the construction site into a plurality of original planes that are equally spaced and parallel to the normal vector.

[0058] In the three-dimensional space representing the construction site, multiple equally spaced original planes parallel to the normal vector are divided, that is, the distance between two adjacent original planes is equal, and each original plane is parallel to the normal vector and does not intersect each other.

[0059] Step 103: Under the condition that the elevation point is projected onto the nearest original plane, a portion of the original plane is selected as the target plane according to the elevation point.

[0060] In this embodiment, the distance between the elevation point and each original plane can be calculated, the distance between the elevation point and each original plane can be compared, and the elevation point can be projected onto the original plane with the closest (i.e., smallest) distance so that the original plane contains the elevation point.

[0061] At this time, you can select part of the original plane as the target plane based on whether the original plane contains elevation points.

[0062] In the specific implementation, the number of elevation points contained in each original plane can be counted, and the K original planes with the highest number of points can be screened out as the target plane. That is, the target plane is the original plane with denser elevation points, which can effectively improve the confidence of fitting the first target curve on the target plane.

[0063] Step 104: On the target plane, fit the first target curve according to the elevation points.

[0064] On the target plane, a curve may be fitted according to the elevation points included in the target plane, which is recorded as a first target curve, so that the first target curve represents the change trend of the bearing layer on the target plane.

[0065] In a specific implementation, in different types of soil structures, there are certain differences in the variation trends of the bearing layers. Therefore, the variation trends of the bearing layers can be fitted in advance for different types of soil structures to obtain a polynomial function.

[0066] Then, the type of soil structure of the current construction site can be queried, thereby loading the polynomial function trained on the type, and using the polynomial function to fit the first target curve passing through the elevation points included in the target plane on the target plane to improve the accuracy of the first target curve.

[0067] Step 105: Load the sequence prediction network trained on the elevation map.

[0068] In this embodiment, a sequence prediction network can be pre-trained based on deep learning and the height map, so that the sequence prediction network can predict the change curve of the bearing layer on other planes based on the change curve of the bearing layer on a certain plane.

[0069] In practical applications, anti-floating anchors are constructed according to the elevation points in the elevation map of the bearing layer, and whether the bearing layer has been reached is determined based on the drilling pressure. If the actual elevation of the bearing layer is different from the expected elevation in the elevation map, the elevation map of the bearing layer will be corrected based on the actual elevation of the bearing layer. Therefore, the corrected elevation map of the bearing layer can be used as a sample to perform supervised training on the sequence prediction network.

[0070] Among them, Figure 2 As shown, the sequence prediction network includes a backbone structure Backbone and multiple branch structures Branch. The backbone structure Backbone extracts the common features of the sequence (i.e., the first target curve), and the multiple branch structures Branch can predict new sequences (i.e., candidate curves) based on the common features. Different branch structures Branch have the same structure but different parameters, and can provide new sequences (i.e., candidate curves) with certain differences.

[0071] Since the geological structure may have small-scale mutations in a local range, different branch structures can provide new sequences (i.e., candidate curves) with differences in the local range, which can reduce the requirements for the generalization of the sequence prediction model, reduce the structural complexity of the sequence prediction model, and speed up the operation of the sequence prediction model, thereby improving the performance of the sequence prediction model.

[0072] Step 106: Input the first target curve on the target plane and the vector pointing from the target plane to the blank plane into the backbone structure to extract basic sequence features.

[0073] In this embodiment, each target plane and other original planes except the target plane are traversed, and other original planes except the target plane can be distributed to the target planes adjacent to it (ie, neighboring) by an even distribution method.

[0074] The so-called equal division means that the other original planes between the two target planes are distributed to the two target planes according to the principle of nearest neighbors. For example, some original planes are distributed as Plane1, Plane2, Plane3, Plane4, and Plane5, among which Plane1 and Plane5 are target planes, then Plane2 is the original plane adjacent to Plane1, Plane4 is the original plane adjacent to Plane5, and Plane3 is randomly assigned to the original plane adjacent to Plane1 or Plane5.

[0075] For the convenience of distinction, other original planes adjacent to the target plane are recorded as blank planes, that is, the blank plane is the original plane adjacent to the target plane.

[0076] The first target curve on the target plane and the vector (including direction, modulus and other information) pointing from the target plane to the blank plane are combined into a new sequence, which is then input into the backbone structure. The backbone structure extracts common features from the new sequence and records them as basic sequence features.

[0077] In one embodiment of the present application, Figure 2 As shown, the backbone structure includes a first gated recurrent unit GRU_1, a second gated recurrent unit GRU_2, a first convolutional block structure ConvBlock_1, a second convolutional block structure ConvBlock_2, a third convolutional block structure ConvBlock_3, a fourth convolutional block structure ConvBlock_4 and a feature pyramid network (FeaturePyramidNetwork, FPN).

[0078] Among them, Figure 3 As shown, the first convolution block structure ConvBlock_1, the second convolution block structure ConvBlock_2, the third convolution block structure ConvBlock_3 and the fourth convolution block structure ConvBlock_4 have the same structure, and all include the first depth-separable convolution layer DSC_1, the first batch normalization layer BN_1, the first activation layer Act_1, the second depth-separable convolution layer DSC_2, the second batch normalization layer BN_2 and the second activation layer Act_2, which can reduce the cost of building and training the sequence prediction network.

[0079] In this embodiment, step 106 may include the following steps:

[0080] Step 1061: Input a first target curve on the target plane and a vector pointing from the target plane to the blank plane into a first gated recurrent unit for processing to obtain a first backbone sequence feature.

[0081] In this embodiment, the first target curve on the target plane and the vector (including direction, modulus and other information) pointing from the target plane to the blank plane are combined into a new sequence, and the new sequence is input into the first gated recurrent unit GRU_1 to extract time series features, which are recorded as the first backbone sequence features.

[0082] Step 1062: Input the first backbone sequence feature into the second gated recurrent unit for processing to obtain a second backbone sequence feature.

[0083] In this embodiment, the first backbone sequence feature is input into the second gated recurrent unit GRU_2 to continue extracting the time series feature, which is recorded as the second backbone sequence feature.

[0084] The first gated recurrent unit GRU_1 and the second gated recurrent unit GRU_2 are both gated recurrent units (GRU). GRU has two gates, namely the reset gate and the update gate. The reset gate determines how to combine new input information with the previous memory, and the update gate defines the amount of previous memory saved to the current time step, which can handle long dependency problems in sequences.

[0085] The structures of the first gated recurrent unit GRU_1 and the second gated recurrent unit GRU_2 are relatively simple and can meet the requirements of extracting common features. The training speed is faster and the risk of overfitting can be reduced.

[0086] Step 1063: Input the second backbone sequence features into the first convolutional block structure for processing to obtain third backbone sequence features.

[0087] In this embodiment, the second backbone sequence features are input into the first convolution block structure ConvBlock_1 to extract spatial features, thereby obtaining the third backbone sequence features.

[0088] Furthermore, in the first convolution block structure ConvBlock_1, the second backbone sequence feature is input into the first depth-separable convolution layer DSC_1 to extract the first convolution feature A1, the first batch normalization layer BN_1 performs a batch normalization operation (BatchNormalization, BN) on the first convolution feature to obtain the second convolution feature A2, the first activation layer Act_1 uses activation functions such as ReLU (Rectified LinearUnit, linear rectifier function) to perform an activation operation on the second convolution feature to obtain the third convolution feature A3, the third convolution feature is input into the second depth-separable convolution layer DSC_2 to extract the fourth convolution feature A4, the second batch normalization layer BN_2 performs a batch normalization operation on the fourth convolution feature to obtain the fifth convolution feature A5, and the second activation layer Act_2 uses activation functions such as ReLU to perform an activation operation on the fifth convolution feature to obtain the third backbone sequence feature.

[0089] Among them, the first depthwise separable convolution layer DSC_1 and the second depthwise separable convolution layer DSC_2 both belong to the depthwise separable convolution (DSC) layer, and their convolution calculation is divided into two parts. First, spatial convolution (depthwise convolution) is performed on the channel (depth) respectively, and the output is spliced, and then the unit convolution kernel is used to perform channel convolution (pointwise convolution) to obtain features.

[0090] Step 1064: Input the third backbone sequence feature into the second convolutional block structure for processing to obtain a fourth backbone sequence feature.

[0091] In this embodiment, the third backbone sequence features are input into the second convolution block structure ConvBlock_2 to extract spatial features, thereby obtaining the fourth backbone sequence features.

[0092] Furthermore, in the second convolution block structure ConvBlock_2, the third backbone sequence feature is input into the first depthwise separable convolution layer DSC_1 to extract the first convolution feature B1, the first batch normalization layer BN_1 performs a batch normalization operation on the first convolution feature to obtain the second convolution feature B2, the first activation layer Act_1 uses activation functions such as ReLU to perform an activation operation on the second convolution feature to obtain the third convolution feature B3, the third convolution feature is input into the second depthwise separable convolution layer DSC_2 to extract the fourth convolution feature B4, the second batch normalization layer BN_2 performs a batch normalization operation on the fourth convolution feature to obtain the fifth convolution feature B5, and the second activation layer Act_2 uses activation functions such as ReLU to perform an activation operation on the fifth convolution feature to obtain the fourth backbone sequence feature.

[0093] Step 1065: Input the second backbone sequence features into the third convolutional block structure for processing to obtain the fifth backbone sequence features.

[0094] In this embodiment, the second backbone sequence features are input into the third convolution block structure ConvBlock_3 to extract spatial features, thereby obtaining the fifth backbone sequence features.

[0095] Furthermore, in the third convolution block structure ConvBlock_3, the second backbone sequence feature is input into the first depthwise separable convolution layer DSC_1 to extract the first convolution feature C1, the first batch normalization layer BN_1 performs a batch normalization operation on the first convolution feature to obtain the second convolution feature C2, the first activation layer Act_1 uses activation functions such as ReLU to perform an activation operation on the second convolution feature to obtain the third convolution feature C3, the third convolution feature is input into the second depthwise separable convolution layer DSC_2 to extract the fourth convolution feature C4, the second batch normalization layer BN_2 performs a batch normalization operation on the fourth convolution feature to obtain the fifth convolution feature C5, the second activation layer Act_2 uses activation functions such as ReLU to perform an activation operation on the fifth convolution feature to obtain the fifth backbone sequence feature.

[0096] Step 1066: Input the fifth backbone sequence feature into the feature pyramid network for processing to obtain the sixth backbone sequence feature.

[0097] In this embodiment, the fifth backbone sequence feature is input into the feature pyramid network FPN to perform a pooling operation to obtain the sixth backbone sequence feature.

[0098] Generally speaking, the semantic feature information of the low-level in the sequence is relatively less, but the location of the feature is clear; the semantic feature information of the high-level is relatively rich, but the feature location is relatively rough. The feature pyramid network FPN fuses the two, and uses the high-resolution features of the low-level features and the high-semantic features of the high-level features. By fusing the features of these different layers, the quality of the features is improved, thereby improving the accuracy of the predicted sequence, especially the accuracy of small targets in the sequence.

[0099] Step 1067: concatenate the third backbone sequence feature, the fourth backbone sequence feature, and the sixth backbone sequence feature into a seventh backbone sequence feature.

[0100] In this embodiment, the third backbone sequence feature, the fourth backbone sequence feature and the sixth backbone sequence feature are concatenated into the seventh backbone sequence feature, so that the feature is rich in shallow information and deep information, improves the quality of the feature, and helps to improve the accuracy of the predicted sequence.

[0101] Step 1068: Input the seventh backbone sequence feature into the fourth convolutional block structure for processing to obtain the basic sequence feature.

[0102] In this embodiment, the seventh backbone sequence feature is input into the fourth convolution block structure ConvBlock_4 to extract spatial features and obtain basic sequence features.

[0103] Furthermore, in the fourth convolution block structure ConvBlock_4, the second backbone sequence feature is input into the first depth-separable convolution layer DSC_1 to extract the first convolution feature D1, the first batch normalization layer BN_1 performs a batch normalization operation on the first convolution feature to obtain the second convolution feature D2, the first activation layer Act_1 uses activation functions such as ReLU to perform an activation operation on the second convolution feature to obtain the third convolution feature D3, the third convolution feature is input into the second depth-separable convolution layer DSC_2 to extract the fourth convolution feature D4, the second batch normalization layer BN_2 performs a batch normalization operation on the fourth convolution feature to obtain the fifth convolution feature D5, and the second activation layer Act_2 uses activation functions such as ReLU to perform an activation operation on the fifth convolution feature to obtain the basic sequence feature.

[0104] Step 107: Input the basic sequence features into multiple branch structures respectively to generate multiple candidate curves and multiple confidence levels.

[0105] If the prediction sequence is regarded as a multi-classification task, the basic sequence features can be input into multiple branch structures respectively, and each branch structure Branch is classified according to the basic sequence features to generate candidate curves and confidence levels.

[0106] In one embodiment of the present application, Figure 2 As shown, each branch structure Branch includes a first decoding block structure DecBlock_1, a second decoding block structure DecBlock_2, a third decoding block structure DecBlock_3 and a fully connected layer (Fully Connected Layer, FC).

[0107] Among them, Figure 4 As shown, the first decoding block structure DecBlock_1, the second decoding block structure DecBlock_2 and the third decoding block structure DecBlock_3 have the same structure, and all include a third gated recurrent unit GRU_3, a transpose convolution layer (Transpose Convolution, TC) and a convolution layer (Convolution, Conv).

[0108] In this embodiment, step 107 may include the following steps:

[0109] Step 1071: In each branch structure, input the basic sequence feature into the first decoding block structure to perform a decoding operation to obtain a first branch sequence feature.

[0110] If the extraction of common sequence features is regarded as a coding operation, then in each branch structure Branch, the basic sequence features can be input into the first decoding block structure DecBlock_1 to perform a decoding operation to obtain the first branch sequence features to restore the size of the features.

[0111] Furthermore, in the first decoding block structure DecBlock_1, the basic sequence features are input into the third gated recurrent unit GRU_3 to extract the time series features, which are recorded as the first decoding features E1. The first decoding features are input into the transposed convolution layer TC for upsampling to obtain the second decoding features E2. The second coded features are input into the convolution layer Conv to perform the convolution operation to obtain the first branch sequence features.

[0112] Step 1072: Input the first branch sequence feature into the second decoding block structure to perform a decoding operation to obtain a second branch sequence feature.

[0113] In this embodiment, the first branch sequence feature is input into the second decoding block structure DecBlock_2 to continue to perform the decoding operation, obtain the second branch sequence feature, and continue to restore the size of the feature.

[0114] Furthermore, in the second decoding block structure DecBlock_2, the basic sequence features are input into the third gated recurrent unit GRU_3 to extract the time series features, which are recorded as the first decoding features F1. The first decoding features are input into the transposed convolution layer TC for upsampling to obtain the second decoding features F2. The second coded features are input into the convolution layer Conv to perform the convolution operation to obtain the second branch sequence features.

[0115] Step 1073: Input the second branch sequence feature into the third decoding block structure to perform a decoding operation to obtain a third branch sequence feature.

[0116] In this embodiment, the second branch sequence feature is input into the third decoding block structure DecBlock_3 to continue to perform the decoding operation, obtain the third branch sequence feature, and continue to restore the size of the feature.

[0117] Furthermore, in the third decoding block structure DecBlock_3, the basic sequence features are input into the third gated recurrent unit GRU_3 to extract the time series features, recorded as the first decoding features G1, the first decoding features are input into the transposed convolution layer TC for upsampling Upsampling to obtain the second decoding features G2, and the second coded features are input into the convolution layer Conv to perform the convolution operation to obtain the third branch sequence features.

[0118] Step 1074: Input the third branch sequence features into the fully connected layer and map them into candidate curves and confidences.

[0119] In this embodiment, the third branch sequence feature is input into the fully connected layer FC, and the fully connected layer FC maps the third branch sequence feature into a candidate curve and a confidence level.

[0120] Step 108: For the blank plane adjacent to the target plane, select one candidate curve as the second target curve according to the elevation point and the confidence level.

[0121] In this embodiment, for each blank plane adjacent to the target plane, the elevation points on the blank plane can be used as prior knowledge and combined with the confidence to screen out one of the candidate curves as the second target curve, thereby improving the accuracy of the second target curve.

[0122] In one embodiment of the present application, step 108 may include the following steps:

[0123] Step 1081: On each blank plane, calculate the second distance between the elevation point and each candidate curve.

[0124] On each blank plane, multiple branch structures Branch output multiple candidate curves for the blank plane. At this time, each elevation point on the blank plane can be traversed to calculate the second distance between the elevation point and each candidate curve.

[0125] Generally, the second distance is usually the minimum distance between the elevation point and the tangent line of the candidate curve.

[0126] Step 1082: For the same candidate curve on the blank plane, calculate the sum of the second distances as the total distance.

[0127] For the same candidate curve on the same blank plane, the sum of the second distances corresponding to all elevation points on the blank plane and the candidate curve may be calculated and recorded as the total distance.

[0128] Step 1083: For the candidate curves output on all blank planes of the same branch structure, the average value of the total distances is calculated as the loss value.

[0129] Within the range of all blank planes, the average value of the total distances between candidate curves output from different blank planes for the same branch structure is calculated and recorded as the loss value.

[0130] Step 1084: Correct the confidence of the candidate curve output by the branch structure according to the loss value.

[0131] In this embodiment, the loss value can characterize the matching degree between the branch structure and the geological structure of the current construction site to a certain extent. Therefore, the loss value can be used to correct the confidence of the candidate curve output by the corresponding branch structure to improve the accuracy of the confidence.

[0132] Exemplarily, the loss value may be substituted into the following formula to correct the confidence of the candidate curve output by the branch structure:

[0133]

[0134] Among them, Score new Score is the corrected confidence of the candidate curve output by the branch structure. old is the confidence of the candidate curve corrected by the branch structure output, tanh is the hyperbolic tangent function, L is the loss value, α and β are weights, α+β=1, which is the regulatory effect.

[0135] In this example, is a penalty item. The larger the loss value, the lower the matching degree between the branch road structure and the geological structure of the current construction site, and the smaller the penalty item, thereby reducing the confidence. Conversely, the smaller the loss value, the higher the matching degree between the branch road structure and the geological structure of the current construction site, and the larger the penalty item, thereby increasing the confidence.

[0136] Step 1085: When the correction is completed, select the candidate curve with the highest confidence on each blank plane adjacent to the target plane as the second target curve.

[0137] If the confidence correction is completed, the confidences of the candidate curves can be compared on the blank planes adjacent to the target plane, and the candidate curve with the highest confidence can be selected as the second target curve.

[0138] Step 109: fitting an elevation map of the bearing layer of the construction site according to the first target curve and the second target curve.

[0139] In this embodiment, the first target curve and the second target curve contain relatively dense points. The first target curve and the second target curve can be input into industrial software, and these points can be used to fit the elevation map (three-dimensional map) of the bearing layer of the construction site.

[0140] In this embodiment, when performing advance drilling operations on multiple drilling points in a construction site, the elevation points of the bearing layer surveyed at each drilling point are collected; the construction site is divided into multiple equally spaced original planes according to the elevation points; under the condition that the elevation points are projected onto the nearest original plane, part of the original plane is selected as the target plane according to the elevation points; on the target plane, a first target curve is fitted according to the elevation points; a sequence prediction network trained on the elevation map is loaded; the sequence prediction network includes a backbone structure and multiple branch structures; the first target curve on the target plane and a vector pointing from the target plane to a blank plane are input into the backbone structure to extract basic sequence features; the blank plane is an original plane adjacent to the target plane; the basic sequence features are respectively input into the multiple branch structures to generate multiple candidate curves and multiple confidence levels; for the blank plane adjacent to the target plane, one of the candidate curves is selected as the second target curve according to the elevation points and the confidence level; the elevation map of the bearing layer of the construction site is fitted according to the first target curve and the second target curve. This embodiment can use real elevation points on the target plane to fit a more confident first target curve, fit candidate curves on an adjacent blank plane based on the first target curve, use the elevation points as prior knowledge to correct the confidence of the candidate curve, and thus select a more confident second target curve. The first target curve and the second target curve are combined to provide abundant points to fit the elevation map of the bearing layer, which can effectively improve the accuracy of the elevation map of the bearing layer, reduce the deviation between the elevation map and the actual bearing layer, provide accurate data support for the construction of anti-floating anchors, and reduce the engineering workload of drilling construction.

[0141] It should be noted that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0142] Reference Figure 5 , shows a schematic diagram of a device for constructing an elevation map of a bearing layer provided in an embodiment of the present application, which may specifically include the following modules:

[0143] The elevation point acquisition module 501 is used to acquire the elevation points of the bearing layer surveyed at each drilling point when performing advance drilling operation on multiple drilling points in the construction site;

[0144] The original plane division module 502 is used to divide the construction site into a plurality of original planes with equal spacing according to the elevation points;

[0145] A target plane selection module 503 is used to select part of the original plane as the target plane according to the elevation point under the condition that the elevation point is projected onto the original plane closest to the elevation point;

[0146] A first target curve fitting module 504, configured to fit a first target curve on the target plane according to the elevation points;

[0147] The sequence prediction network loading module 505 is used to load the sequence prediction network trained on the elevation map; the sequence prediction network includes a backbone structure and multiple branch structures;

[0148] A basic sequence feature extraction module 506 is used to extract basic sequence features by inputting the first target curve on the target plane and the vector pointing from the target plane to a blank plane into the backbone structure; the blank plane is the original plane adjacent to the target plane;

[0149] A candidate curve generating module 507, used for inputting the basic sequence features into the plurality of branch structures to generate a plurality of candidate curves and a plurality of confidences;

[0150] A second target curve screening module 508 is used to screen out one of the candidate curves as a second target curve according to the elevation point and the confidence level for the blank plane adjacent to the target plane;

[0151] The elevation map fitting module 509 is used to fit the elevation map of the bearing layer for the construction site according to the first target curve and the second target curve.

[0152] In one embodiment of the present application, the original plane division module 502 is further used to:

[0153] Project all the elevation points vertically onto a horizontal plane to obtain a reference point;

[0154] Adding a minimum circumscribed rectangle to the reference point on a horizontal plane as a restriction interval;

[0155] Under the constraint of the restriction interval, constructing a normal vector according to the distribution of the reference points;

[0156] The construction site is divided into a plurality of original planes which are equally spaced and parallel to the normal vector.

[0157] In one embodiment of the present application, the original plane division module 502 is further used to:

[0158] For two adjacent corner points of the restriction interval, a plurality of line segments are generated between two restriction lines with the corner points as endpoints; the two restriction lines are the sides of the restriction area and intersect at the corner points;

[0159] For each of the line segments, construct a histogram for the first distance between the reference point and the line segment; each of the histograms has a plurality of intervals;

[0160] Counting the total number of the reference points in the highest K intervals; K is a positive integer;

[0161] Convert the line segment with the highest total number into a normal vector;

[0162] The target plane selection module 503 is also used for:

[0163] Counting the number of the elevation points contained in each of the original planes;

[0164] The K original planes with the highest number of points are selected as target planes.

[0165] In one embodiment of the present application, the first target curve fitting module 504 is further configured to:

[0166] Query the type of soil structure of the construction site;

[0167] loading a polynomial function trained on the type;

[0168] On the target plane, a first target curve passing through the elevation points is fitted using the polynomial function.

[0169] In one embodiment of the present application, the backbone structure includes a first gated recurrent unit, a second gated recurrent unit, a first convolutional block structure, a second convolutional block structure, a third convolutional block structure, a fourth convolutional block structure and a feature pyramid network;

[0170] The first convolution block structure, the second convolution block structure, the third convolution block structure and the fourth convolution block structure all sequentially include a first depth-separable convolution layer, a first batch normalization layer, a first activation layer, a second depth-separable convolution layer, a second batch normalization layer and a second activation layer;

[0171] The basic sequence feature extraction module 506 is also used for:

[0172] Inputting the first target curve on the target plane and the vector pointing from the target plane to the blank plane into the first gated recurrent unit for processing to obtain a first backbone sequence feature;

[0173] Inputting the first backbone sequence feature into the second gated recurrent unit for processing to obtain a second backbone sequence feature;

[0174] Inputting the second backbone sequence feature into the first convolution block structure for processing to obtain a third backbone sequence feature;

[0175] Inputting the third backbone sequence feature into the second convolution block structure for processing to obtain a fourth backbone sequence feature;

[0176] Inputting the second backbone sequence feature into the third convolution block structure for processing to obtain a fifth backbone sequence feature;

[0177] Inputting the fifth backbone sequence feature into the feature pyramid network for processing to obtain a sixth backbone sequence feature;

[0178] splicing the third backbone sequence feature, the fourth backbone sequence feature and the sixth backbone sequence feature into a seventh backbone sequence feature;

[0179] The seventh backbone sequence feature is input into the fourth convolutional block structure for processing to obtain a basic sequence feature.

[0180] In one embodiment of the present application, each of the branch structures includes a first decoding block structure, a second decoding block structure, a third decoding block structure and a fully connected layer; the first decoding block structure, the second decoding block structure and the third decoding block structure all include a third gated recurrent unit, a transposed convolutional layer and a convolutional layer;

[0181] The candidate curve generation module 507 is also used for:

[0182] In each of the branch structures, the basic sequence feature is input into the first decoding block structure to perform a decoding operation to obtain a first branch sequence feature;

[0183] Inputting the first branch sequence feature into the second decoding block structure to perform a decoding operation to obtain a second branch sequence feature;

[0184] Inputting the second branch sequence feature into the third decoding block structure to perform a decoding operation to obtain a third branch sequence feature;

[0185] The third branch sequence features are input into the fully connected layer and mapped into candidate curves and confidences.

[0186] In one embodiment of the present application, the second target curve screening module 508 is further used to:

[0187] On each of the blank planes, calculating a second distance between the elevation point and each of the candidate curves;

[0188] For the same candidate curve on the blank plane, calculating the sum of the second distances as the total distance;

[0189] For the candidate curves outputted on all the blank planes of the same branch structure, an average value of the total distances is calculated as a loss value;

[0190] Correcting the confidence of the candidate curve output by the branch structure according to the loss value;

[0191] When the correction is completed, the candidate curve with the highest confidence is selected on each of the blank planes adjacent to the target plane as the second target curve.

[0192] In one embodiment of the present application, the second target curve screening module 508 is further used to:

[0193] Substitute the loss value into the following formula to correct the confidence of the candidate curve output by the branch structure:

[0194]

[0195] Among them, Score new Score is the confidence score of the candidate curve after correction output by the branch structure. old is the corrected confidence of the candidate curve output by the branch structure, tanh is the hyperbolic tangent function, L is the loss value, and α and β are weights.

[0196] An embodiment of the present application provides a device for constructing an elevation map of a bearing layer. By using the device, each step in the aforementioned method embodiments can be implemented.

[0197] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment part.

[0198] Reference Figure 6 , shows a schematic diagram of a terminal device provided in an embodiment of the present application. Figure 6 As shown, the terminal device 400 in the embodiment of the present application includes: a processor 410, a memory 420, and a computer program 421 stored in the memory 420 and executable on the processor 410. When the processor 410 executes the computer program 421, the steps of each embodiment of the method for constructing the elevation map of the bearing layer are implemented. Alternatively, when the processor 410 executes the computer program 421, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0199] Exemplarily, the computer program 421 may be divided into one or more modules / units, which are stored in the memory 420 and executed by the processor 410 to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which may be used to describe the execution process of the computer program 421 in the terminal device 400.

[0200] The terminal device 400 may include, but is not limited to, a processor 410 and a memory 420. Those skilled in the art will appreciate that Figure 6 It is only an example of the terminal device 400 and does not constitute a limitation on the terminal device 400. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device 400 may also include input and output devices, network access devices, buses, etc.

[0201] The processor 410 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0202] The memory 420 may be an internal storage unit of the terminal device 400, such as a hard disk or memory of the terminal device 400. The memory 420 may also be an external storage device of the terminal device 400, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 400. Further, the memory 420 may also include both an internal storage unit of the terminal device 400 and an external storage device. The memory 420 is used to store the computer program 421 and other programs and data required by the terminal device 400. The memory 420 may also be used to temporarily store data that has been output or is to be output.

[0203] An embodiment of the present application also discloses a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for constructing the elevation map of the bearing layer as described in the aforementioned embodiments.

[0204] The embodiment of the present application further discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for constructing an elevation map of the bearing layer as described in the aforementioned embodiments is implemented.

[0205] The embodiment of the present application further discloses a computer program product. When the computer program product is run on a computer, the computer is enabled to execute the method for constructing the elevation map of the bearing layer described in the aforementioned embodiments.

[0206] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application is described in detail with reference to the above-mentioned embodiments, a person skilled in the art should understand that the technical solutions described in the above-mentioned embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for constructing an elevation map of a bearing layer, characterized in that: include: When performing advance drilling operations on multiple drilling points in the construction site, collecting the elevation points of the bearing layer surveyed in each of the drilling points; Dividing the construction site into a plurality of equally spaced original planes according to the elevation points; Under the condition that the elevation point is projected onto the original plane closest to the elevation point, a portion of the original plane is selected as the target plane according to the elevation point; On the target plane, fitting a first target curve according to the elevation points; Loading a sequence prediction network trained on a benchmark height map; the sequence prediction network includes a backbone structure and multiple branch structures; The first target curve on the target plane, the vector from the target plane to the blank plane is input into the backbone structure to extract basic sequence features; the blank plane is the original plane adjacent to the target plane; Inputting the basic sequence features into the plurality of branch structures respectively to generate a plurality of candidate curves and a plurality of confidences; For the blank plane adjacent to the target plane, one of the candidate curves is selected as the second target curve according to the elevation point and the confidence level; An elevation map of the bearing layer is fitted to the construction site according to the first target curve and the second target curve.

2. The method according to claim 1, characterized in that The step of dividing the construction site into a plurality of equally spaced original planes according to the elevation points includes: Project all the elevation points vertically onto a horizontal plane to obtain a reference point; Adding a minimum circumscribed rectangle to the reference point on a horizontal plane as a restriction interval; Under the constraint of the restriction interval, constructing a normal vector according to the distribution of the reference points; The construction site is divided into a plurality of original planes which are equally spaced and parallel to the normal vector.

3. The method according to claim 2, characterized in that The constructing of the normal vector according to the distribution of the reference points under the constraint of the restriction interval includes: For two adjacent corner points of the restriction interval, a plurality of line segments are generated between two restriction lines with the corner points as endpoints; the two restriction lines are the edges of the restriction area, and intersect at the corner points; For each of the line segments, construct a histogram for the first distance between the reference point and the line segment; each of the histograms has a plurality of intervals; Counting the total number of the reference points in the highest K intervals; K is a positive integer; Convert the line segment with the highest total number into a normal vector; The step of selecting a portion of the original plane as a target plane according to the elevation point includes: Counting the number of the elevation points contained in each of the original planes; The K original planes with the highest number of points are selected as target planes.

4. The method according to claim 1, characterized in that: The step of fitting a first target curve on the target plane according to the elevation points includes: Query the type of soil structure of the construction site; loading a polynomial function trained on the type; On the target plane, a first target curve passing through the elevation points is fitted using the polynomial function.

5. The method according to any one of claims 1 to 4, characterized in that The backbone structure includes a first gated recurrent unit, a second gated recurrent unit, a first convolutional block structure, a second convolutional block structure, a third convolutional block structure, a fourth convolutional block structure and a feature pyramid network; The first convolution block structure, the second convolution block structure, the third convolution block structure and the fourth convolution block structure all sequentially include a first depth-separable convolution layer, a first batch normalization layer, a first activation layer, a second depth-separable convolution layer, a second batch normalization layer and a second activation layer; The step of inputting the first target curve on the target plane and the vector pointing from the target plane to the blank plane into the backbone structure to extract basic sequence features comprises: Inputting the first target curve on the target plane and the vector pointing from the target plane to the blank plane into the first gated recurrent unit for processing to obtain a first backbone sequence feature; Inputting the first backbone sequence feature into the second gated recurrent unit for processing to obtain a second backbone sequence feature; Inputting the second backbone sequence feature into the first convolution block structure for processing to obtain a third backbone sequence feature; Inputting the third backbone sequence feature into the second convolution block structure for processing to obtain a fourth backbone sequence feature; Inputting the second backbone sequence feature into the third convolution block structure for processing to obtain a fifth backbone sequence feature; Inputting the fifth backbone sequence feature into the feature pyramid network for processing to obtain a sixth backbone sequence feature; splicing the third backbone sequence feature, the fourth backbone sequence feature and the sixth backbone sequence feature into a seventh backbone sequence feature; The seventh backbone sequence feature is input into the fourth convolutional block structure for processing to obtain a basic sequence feature.

6. The method according to claim 5, characterized in that Each of the branch structures includes a first decoding block structure, a second decoding block structure, a third decoding block structure and a fully connected layer; the first decoding block structure, the second decoding block structure and the third decoding block structure all include a third gated recurrent unit, a transposed convolutional layer and a convolutional layer; The step of inputting the basic sequence features into the plurality of branch structures to generate a plurality of candidate curves and a plurality of confidences comprises: In each of the branch structures, the basic sequence feature is input into the first decoding block structure to perform a decoding operation to obtain a first branch sequence feature; Inputting the first branch sequence feature into the second decoding block structure to perform a decoding operation to obtain a second branch sequence feature; Inputting the second branch sequence feature into the third decoding block structure to perform a decoding operation to obtain a third branch sequence feature; The third branch sequence features are input into the fully connected layer and mapped into candidate curves and confidences.

7. The method according to claim 6, characterized in that The step of selecting one of the candidate curves as the second target curve according to the elevation point and the confidence level for the blank plane adjacent to the target plane includes: On each of the blank planes, calculating a second distance between the elevation point and each of the candidate curves; For the same candidate curve on the blank plane, calculating the sum of the second distances as the total distance; For the candidate curves outputted on all the blank planes of the same branch structure, calculating the average value of the total distances as a loss value; Correcting the confidence of the candidate curve output by the branch structure according to the loss value; When the correction is completed, the candidate curve with the highest confidence is selected on each of the blank planes adjacent to the target plane as the second target curve.

8. The method according to claim 7, characterized in that The step of correcting the confidence of the candidate curve output by the branch structure according to the loss value includes: Substitute the loss value into the following formula to correct the confidence of the candidate curve output by the branch structure: Among them, Score new Score is the confidence score of the candidate curve after correction output by the branch structure. old is the corrected confidence of the candidate curve output by the branch structure, tanh is the hyperbolic tangent function, L is the loss value, and α and β are weights.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for constructing the elevation map of the bearing layer as described in any one of claims 1-8 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing an elevation map of the bearing layer as described in any one of claims 1 to 8 is implemented.

Images