Excavation volume index calculation method based on multi-source data characteristics

Abstract

The invention provides a multi-source data feature-based excavation volume index calculation method, which comprises the steps of training a slope excavation engineering quantity calculation model by using a training set containing historical data of a reference slope and a to-be-measured slope, and obtaining topographic features and similarity weights of all slope sections, the data volume of the reference slope being greater than the data volume of the to-be-measured slope; training an excavation volume prediction model based on historical data of the reference slope; in combination with the topographic feature similarity weight and historical data of the side slope to be measured, obtaining an excavation volume prediction result and geologic features of each side slope section through the prediction model; and calculating the excavation volume index of each slope section according to the prediction result, the geologic features and the topographic features, and summarizing to obtain the total excavation engineering quantity of the to-be-detected slope. According to the invention, the precision and efficiency of engineering quantity calculation can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geotechnical engineering, and more particularly, to a method for calculating an excavation volume index based on multi-source data features. BACKGROUND

[0002] In slope excavation engineering, accurate calculation of engineering quantity is crucial for engineering cost control, construction schedule arrangement, and construction safety. Traditional methods for calculating slope excavation engineering quantity mainly rely on manual measurement and simple geometric calculation. Such methods are not only inefficient, but also susceptible to human factors, making it difficult to ensure the accuracy of the calculation results. With the development of computer technology, some methods based on BIM have been introduced into the calculation of slope excavation engineering quantity. These methods have improved the efficiency and accuracy of calculation to some extent, but still have some limitations. For example, these methods usually require a large number of input parameters and are highly sensitive to parameters. When there are errors in the input parameters, the deviation of the calculation results can be large. In addition, existing methods for calculating slope excavation engineering quantity have obvious limitations in dealing with complex geological conditions and topographic features, resulting in a certain deviation between the calculation results and the actual situation.

[0003] In the implementation process of the embodiments of the present application, at least the following problems or defects exist in the prior art: the traditional method relies on manual measurement and simple geometric calculation, which is inefficient and susceptible to human factors, and the accuracy is difficult to guarantee; the existing method based on BIM improves the efficiency and accuracy, but requires a large number of input parameters and is sensitive to parameters, and the complex geological conditions and topographic features are not considered, resulting in a deviation of the results. SUMMARY

[0004] The present application provides a method for calculating an excavation volume index based on multi-source data features, comprising: Based on the training set including historical excavation volume data and topographic related data of reference slope and to-be-measured slope, a slope excavation engineering quantity calculation model is trained, and the topographic features and topographic feature similarity weight of each slope section of the reference slope and the to-be-measured slope are obtained; wherein the historical data of the reference slope is more than the to-be-analyzed to-be-measured slope; Based on the historical excavation volume data of the reference slope, an excavation volume prediction model is trained; Based on the topographic feature similarity weight and the historical excavation volume data of the to-be-measured slope, the excavation volume prediction model is used to obtain the excavation volume prediction result and geological feature of each slope section of the to-be-measured slope; Based on the excavation volume prediction result, geological feature and topographic feature of each slope section, the excavation volume index of the corresponding slope section is obtained, and the excavation volume indexes of each slope section are summarized to obtain the slope excavation engineering quantity of the to-be-measured slope.

[0005] Further, the excavation amount index of the corresponding slope section is calculated based on the following formula:

[0006] wherein, is a weight factor; PV is the excavation amount prediction result of the slope section; GF is the geological feature value; TF is the topographic feature value; SW is the topographic feature similarity weight; is a vector representation of the topographic feature; is a vector representation of the geological feature; e is a natural constant; is a vector norm.

[0007] Further, the slope excavation engineering quantity calculation model includes a graph structure, a feature network, and a weighted network; the slope excavation engineering quantity calculation model is trained, including: Based on the training set, a plurality of reference slope graphs and a plurality of to-be-measured slope graphs are obtained using the graph structure; Based on the plurality of reference slope graphs and the plurality of to-be-measured slope graphs, slope section division is performed respectively, feature learning of each slope section is performed using the feature network, and comprehensive node embedding features of each slope section of the reference slope and the to-be-measured slope are obtained; Based on the comprehensive node embedding features of the reference slope and the to-be-measured slope respectively, the slope section weight is learned using the weighted network and the weighted network loss function, and the similarity weight of each slope section of the reference slope and the to-be-measured slope is obtained; After the training is completed, the comprehensive node embedding features and the similarity weight obtained by the slope excavation engineering quantity calculation model based on the training set are the topographic feature and the topographic feature similarity weight of each slope section of the reference slope and the to-be-measured slope.

[0008] Further, obtaining a plurality of reference slope graphs and to-be-measured slope graphs includes: Based on the geographical positions of the reference slope and the to-be-measured slope respectively, a position adjacency matrix is constructed; Based on the geological stratum relationship of the reference slope and the to-be-measured slope respectively, a stratum adjacency matrix is constructed; Based on the displacement monitoring points of the reference slope and the to-be-measured slope respectively, a deformation adjacency matrix is constructed; wherein the displacement monitoring point is a slope deformation data point obtained through a sensor network; Based on the excavation profile data of the reference slope and the to-be-measured slope respectively, a profile similarity matrix of the source slope section and the destination slope section is constructed; Based on the above-mentioned various adjacency matrices, a plurality of reference slope graphs and to-be-measured slope graphs are obtained.

[0009] Further, the feature network sequentially performs multi-level feature alignment and feature learning of each slope section, including: The multi-view attention network is used to process the reference slope image and the to-be-detected slope image, to obtain a node embedding matrix corresponding to each image structure; The node embedding matrix of the reference slope and the to-be-detected slope is updated and fused respectively by using the self-attention mechanism of the fusion layer, to obtain updated node embedding features, and the comprehensive node embedding features are output by linear transformation; The feature learning of each slope section is sequentially performed by using regional feature alignment, node feature alignment and edge feature alignment.

[0010] Further, the regional feature alignment comprises: The cosine similarity of the feature vectors of the corresponding type of slope sections of the reference slope and the to-be-detected slope is calculated; The matching slope section pairs are screened based on a similarity threshold; The feature alignment training is performed by minimizing the Euclidean distance loss function of the matching slope section pairs.

[0011] Further, the Euclidean distance loss function is expressed as:

[0012] Wherein, N is the number of matching slope section pairs; is the feature vector of the i th matching slope section in the reference slope; is the feature vector of the i th matching slope section in the to-be-detected slope.

[0013] Further, the node feature alignment comprises: The Pearson correlation coefficient of the node feature vector of the displacement monitoring point is calculated; The feature of the monitoring point under the same geological condition is made consistent by maximizing the correlation coefficient loss function; The network parameters are updated based on the feature difference gradient.

[0014] Further, the edge feature alignment comprises: The edge feature vector of the stratum connection relationship between the slope sections is extracted; The similarity score of the edge features of the reference slope and the to-be-detected slope is calculated by bilinear transformation; The edge feature optimization is performed based on the score to construct a triple loss function.

[0015] Further, the weighted network loss function is:

[0016] Wherein, is the predicted excavation volume vector of the model; V is the actually observed excavation volume vector; is the covariance matrix of the excavation volume observation error; is the excavation volume index of the reference slope; is an excavation quantity index of the to-be-detected slope; is a regularization coefficient; is a similarity weight vector; is a similarity weight vector; is an L2 norm.

[0017] The above-mentioned embodiments of the present application have at least the following beneficial effects: 1. By fusing the topographic feature similarity weight of the reference slope and the to-be-detected slope, and combining the historical excavation quantity data to train the prediction model, the problem of estimation deviation of the traditional method caused by insufficient data of the to-be-detected slope is effectively solved, and the accuracy and reliability of the prediction result are significantly improved.

[0018] 2. The graph structure, feature network and weighted network are used to construct the slope excavation quantity calculation model, efficient feature extraction and alignment of the slope section multi-source data are realized, the problem that the slope features are difficult to be quantitatively expressed under complex geological conditions is solved, and the adaptability of the model to the engineering practice is enhanced.

[0019] 3. Based on the multi-factor calculation framework of the excavation quantity index, the interactive influence of the prediction result, the geological feature and the topographic feature is comprehensively considered, the limitation of single index estimation is overcome, the quantity calculation result is more in line with the actual construction demand, and a scientific basis is provided for the optimization design of the slope engineering. BRIEF DESCRIPTION OF DRAWINGS

[0020] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description read in conjunction with the accompanying drawings, in which several embodiments of the present application are shown. In the drawings: Figure 1 A flowchart of an excavation quantity index calculation method based on multi-source data features provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0021] The principles and spirits of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are given only to enable those skilled in the art to better understand and implement the present application, and do not limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0022] Those skilled in the art know that the embodiments of the present application can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present application can be embodied in the following forms: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0023] It should be noted that the number of any elements in the drawings is used for illustration and not limitation, and any naming is only used for differentiation and does not have any limiting meaning.

[0024] Reference will be made to the following Figure 1 , Figure 1 The flowchart of the excavation volume index calculation method based on multi-source data characteristics provided by an embodiment of the present application is shown in Figure Figure 1 The excavation volume index calculation method based on multi-source data characteristics includes the following steps: S1, based on the training set including the historical excavation volume data and the topographic related data of the reference slope and the to-be-measured slope, a slope excavation engineering quantity calculation model is trained, and the topographic features and the topographic feature similarity weight of each slope section of the reference slope and the to-be-measured slope are obtained; wherein the historical data amount of the reference slope is more than that of the to-be-analyzed to-be-measured slope; S2, an excavation volume prediction model is trained based on the historical excavation volume data of the reference slope; S3, based on the topographic feature similarity weight and the historical excavation volume data of the to-be-measured slope, the excavation volume prediction model is used to obtain the excavation volume prediction result and the geological feature of each slope section of the to-be-measured slope; S4, based on the excavation volume prediction result, the geological feature and the topographic feature of each slope section, the excavation volume index of the corresponding slope section is obtained, and the excavation volume indexes of each slope section are summarized to obtain the slope excavation engineering quantity of the to-be-measured slope.

[0025] It should be noted that the excavation volume index calculation method based on multi-source data characteristics proposed by the present application aims to train a model that can accurately predict the excavation engineering quantity of the to-be-measured slope by comprehensively considering the historical excavation volume data and the topographic related data of the reference slope and the to-be-measured slope. The core of this method is to use historical data and topographic information to obtain the topographic features and the topographic feature similarity weight of each slope section through model training, and then to accurately predict the slope excavation engineering quantity. The advantage of this method is that it can fully utilize existing data resources, improve the accuracy and efficiency of prediction through model training, and at the same time consider the complex geological conditions and topographic features of the slope.

[0026] In particular, the reference slopes refer to those slopes that have completed excavation and have detailed historical data, including excavation volume, geological conditions, topographic features, etc. The to-be-measured slopes refer to those slopes that need to predict the excavation volume, usually with less historical data. The historical excavation volume data refers to the data recorded by the reference slopes during the excavation process, such as earthwork volume, etc. These data are used to train the model so that it can learn the relationship between the excavation volume and various factors. The topographic related data includes the slope of the slope, the slope direction, the elevation, etc. These data are used to describe the topographic features of the slope. The topographic feature similarity weight is obtained by comparing the topographic features of the reference slope and the to-be-measured slope, and is used to measure the similarity between the two. The excavation volume prediction model is trained based on the historical excavation volume data of the reference slope, and is used to predict the excavation volume of each slope section of the to-be-measured slope. The excavation volume index is a comprehensive index calculated according to the excavation volume prediction results, geological features and topographic features of each slope section, and is used to measure the excavation volume of the slope section.

[0027] Preferably, in order to further refine the operation steps of this calculation method, the following aspects can be considered. First, when constructing the training set, the historical excavation volume data and topographic related data of the reference slope and the to-be-measured slope need to be collected and organized. These data can be obtained through field measurement, geological exploration and geographic information system (GIS), etc. Second, when training the excavation volume prediction model, machine learning algorithms such as decision tree, random forest or neural network, etc. can be used. The input parameters include the historical excavation volume data and topographic features of the reference slope, and the output is the excavation volume prediction result. When calculating the topographic feature similarity weight, similarity calculation methods such as cosine similarity or Euclidean distance, etc. can be used. Finally, when summarizing the excavation volume index of each slope section, the total excavation volume of the to-be-measured slope can be obtained by weighted summation according to the excavation volume prediction results, geological features and topographic features of each slope section. Through these refined operation steps, the accuracy and reliability of the slope excavation volume calculation can be further improved, providing strong support for engineering decision-making.

[0028] In some embodiments, the excavation volume index of the corresponding slope section is calculated based on the following formula:

[0029] wherein, is a weight factor; PV is the excavation volume prediction result of the slope section; GF is the geological feature value; TF is the topographic feature value; SW is the topographic feature similarity weight; is a vector representation of the topographic feature; is a vector representation of the geological feature; e is a natural constant; is a vector norm.

[0030] It should be noted that the present application adopts a calculation method considering multiple factors when calculating the slope excavation quantity. This method determines the excavation quantity index of each slope section through a specific formula, which combines the excavation quantity prediction result, geological characteristics, topographic characteristics, and topographic characteristic similarity weight of the slope section. The core of this method is to adjust the importance of each factor in the calculation through the weight factor, thereby obtaining a comprehensive excavation quantity index. This index can more accurately reflect the actual excavation demand of the slope section, and then the excavation quantity of the entire slope to be measured is obtained by aggregating the indices of each slope section. The advantage of this method is that it can comprehensively consider multiple factors, improve the accuracy and reliability of the excavation quantity calculation.

[0031] Specifically, each parameter in this formula has a clear meaning and physical background. The excavation quantity prediction result is the estimated value of the excavation quantity of the slope section based on historical data and model prediction, which reflects the estimated amount of earthwork needed under current conditions. The geological characteristic value is a numerical value calculated based on the geological conditions of the slope, which includes the comprehensive influence of rock type, rock layer inclination, groundwater level, etc. The topographic characteristic value is calculated based on the topographic conditions of the slope, such as slope, aspect, elevation, etc. The topographic characteristic similarity weight is a weight value obtained by comparing the topographic characteristics of the reference slope and the slope to be measured, used to measure the similarity between the two. The weight factor is a parameter determined by experience or optimization method, used to balance the contribution of different factors in the calculation. The vector norm is used to calculate the difference between two vectors, usually using Euclidean distance or other appropriate norm form. These parameters together form the basis for calculating the excavation quantity index, and by reasonably setting these parameters, an accurate excavation quantity index can be obtained.

[0032] Preferably, in order to further refine the calculation process of the excavation quantity index, the following aspects can be considered. First, when determining the weight factor, the specific value of these factors can be determined through analysis of historical data and optimization of the model. For example, the model performance under different combinations of weight factors can be evaluated through cross-validation and other methods, so as to select the optimal weight factor. Second, when calculating the geological characteristic value and the topographic characteristic value, various methods can be used to extract and quantify these characteristics. For example, for geological characteristics, geological exploration data and laboratory test results can be combined to calculate; for topographic characteristics, geographic information system GIS data and remote sensing images can be used to extract. In addition, when calculating the vector difference, appropriate calculation methods such as Euclidean distance, Manhattan distance, etc. can be selected according to specific application scenarios. Through these refined operation steps, the accuracy and reliability of the excavation quantity index calculation can be further improved, providing strong support for accurate prediction of slope excavation quantity.

[0033] In some embodiments, the slope excavation engineering quantity calculation model comprises a graph structure, a feature network, and a weighted network; the slope excavation engineering quantity calculation model is trained, comprising: Based on the training set, a plurality of reference slope graphs and a plurality of to-be-measured slope graphs are obtained respectively by using the graph structure; Based on the plurality of reference slope graphs and the plurality of to-be-measured slope graphs, slope section division is performed respectively, feature learning of each slope section is performed by using the feature network, and comprehensive node embedding features of each slope section of the reference slope and the to-be-measured slope are obtained; Based on the comprehensive node embedding features of the reference slope and the to-be-measured slope respectively, slope section weights are learned by using the weighted network and a weighted network loss function, and similarity weights of each slope section of the reference slope and the to-be-measured slope are obtained; After training, the comprehensive node embedding features and the similarity weights obtained by using the slope excavation engineering quantity calculation model based on the training set are the terrain features and the terrain feature similarity weights of each slope section of the reference slope and the to-be-measured slope.

[0034] It should be noted that the slope excavation engineering quantity calculation model mentioned in the present application is a comprehensive system that realizes accurate prediction of slope excavation engineering quantity through multiple steps. This model includes three main parts: graph structure, feature network, and weighted network. The graph structure is used to construct the graph representation of the slope, the feature network is used to learn the features of the slope section, and the weighted network is used to learn the weight of the slope section. Through these steps, the model can obtain the terrain features and the terrain feature similarity weights of each slope section of the reference slope and the to-be-measured slope, thereby providing a basis for the calculation of excavation engineering quantity. The advantage of this method is that it can fully utilize historical data and terrain information, and improve the accuracy and efficiency of prediction through model training.

[0035] Specifically, the graph structure is implemented by constructing multiple adjacency matrices, including position adjacency matrix, stratum adjacency matrix, deformation adjacency matrix and profile similarity matrix. The position adjacency matrix is constructed based on the geographical position information of the slope, reflecting the spatial relationship between the points of the slope; the stratum adjacency matrix is constructed according to the geological stratum relationship, reflecting the connection between different strata; the deformation adjacency matrix is constructed based on the data of displacement monitoring points, which are obtained through a sensor network and can reflect the deformation of the slope in real time; the profile similarity matrix is calculated according to the excavation profile data, reflecting the similarity between different slope sections. The feature network uses a multi-view attention network to process these graph structures to obtain the node embedding matrix corresponding to each graph structure, and then updates and fuses these node embedding features through the self-attention mechanism of the fusion layer to finally output the comprehensive node embedding features. The weighted network learns the slope section weights based on these comprehensive node embedding features through the weighted network loss function to obtain the similarity weights of each slope section of the reference slope and the slope to be measured. These weights and features together constitute the output of the model, providing a basis for subsequent excavation engineering quantity calculation.

[0036] Preferably, in order to further refine the construction process of the slope excavation engineering quantity calculation model, the following aspects can be considered. First, in the construction of the graph structure, the position adjacency matrix can be constructed by the latitude and longitude information of the slope, and the specific parameters can include the coordinates of each monitoring point; the stratum adjacency matrix can be constructed according to the stratum distribution information in the geological exploration data, and the specific parameters can include the dip angle and thickness of the stratum; the deformation adjacency matrix is constructed based on the data of displacement monitoring points, and the specific parameters can include the displacement amount and direction of the monitoring points. Second, in the feature network, the multi-view attention network can be set with multiple views, each view focusing on different feature dimensions such as terrain, geology and deformation, and automatically learning the importance of different views through the attention mechanism. In the weighted network, the design of the loss function is crucial, which not only includes the difference between the predicted excavation amount and the actual observed value, but also includes a regularization term for the similarity weight to prevent overfitting of the model. Through these refined operation steps, the performance and prediction accuracy of the model can be further improved, providing strong support for accurate prediction of slope excavation engineering quantity.

[0037] In some embodiments, obtaining a plurality of reference slope graphs and a slope to be measured graph comprises: constructing a position adjacency matrix based on the geographical position of the reference slope and the slope to be measured, respectively; constructing a stratum adjacency matrix based on the geological stratum relationship of the reference slope and the slope to be measured, respectively; constructing a deformation adjacency matrix based on the displacement monitoring points of the reference slope and the slope to be measured, respectively; wherein the displacement monitoring points are slope deformation data points obtained through a sensor network; Based on the excavation profile data of the reference slope and the to-be-tested slope respectively, a profile similarity matrix of the source slope section and the destination slope section is constructed; Based on the above-mentioned multiple adjacency matrices, multiple reference slope maps and to-be-tested slope maps are obtained.

[0038] It should be noted that the construction of the reference slope map and the to-be-tested slope map mentioned in the present application is based on multiple adjacency matrices. These adjacency matrices include a position adjacency matrix, a rock stratum adjacency matrix, a deformation adjacency matrix, and a profile similarity matrix. The position adjacency matrix reflects the geographical position relationship of each point of the slope, the rock stratum adjacency matrix reflects the connection relationship between different rock strata, the deformation adjacency matrix reflects the relationship between the slope deformation monitoring points, and the profile similarity matrix reflects the similarity between the slope sections. Through these adjacency matrices, multiple reference slope maps and to-be-tested slope maps can be constructed, providing a basis for subsequent feature learning and weight learning. The advantage of this method is that it can comprehensively consider multiple features of the slope, improving the accuracy and reliability of the model.

[0039] Specifically, the position adjacency matrix is constructed based on the geographical position information of the slope, which reflects the spatial position relationship between each point of the slope. For example, the matrix elements can represent the distance or position correlation between two monitoring points through the latitude and longitude information of each monitoring point on the slope. The rock stratum adjacency matrix is constructed according to the geological rock stratum relationship, which reflects the connection between different rock strata. For example, the distribution and mutual relationship of rock strata can be determined through geological exploration data, and the matrix elements represent the connection strength or similarity between two rock strata. The deformation adjacency matrix is constructed based on the displacement monitoring point data, which is obtained through a sensor network and can reflect the deformation of the slope in real time. The matrix elements can represent the displacement correlation or deformation trend similarity between two monitoring points. The profile similarity matrix is calculated according to the excavation profile data, which reflects the similarity between different slope sections. For example, the similarity can be calculated by comparing the profile shape, slope, and other characteristics of different slope sections, and the matrix elements represent the similarity value between two slope sections. These adjacency matrices together constitute the basis of the slope map, providing rich information for subsequent feature learning.

[0040] Preferably, in order to further refine the construction process of the reference slope graph and the to-be-measured slope graph, the following aspects can be considered. First, when constructing the position adjacency matrix, high-precision geographic information system (GIS) data can be used to obtain accurate position information of each point on the slope. For example, the latitude and longitude coordinates of each monitoring point can be determined through satellite remote sensing data or field measurement data, and the matrix elements can be calculated based on these coordinates. Second, when constructing the rock layer adjacency matrix, geological exploration reports and laboratory test results can be combined to determine the distribution and properties of the rock layers. For example, detailed information of the rock layers can be obtained through drilling sampling and rock physical property testing, and the matrix can be constructed accordingly. When constructing the deformation adjacency matrix, a high-precision sensor network can be used to monitor the deformation of the slope in real time. For example, multiple displacement sensors can be installed, data can be collected regularly, and matrix elements can be calculated through data processing algorithms. Finally, when constructing the profile similarity matrix, image processing and pattern recognition techniques can be used to compare the profile features of different slope sections. For example, edge features, texture features, and other features of the profile can be extracted to calculate the similarity, and the matrix can be constructed accordingly. Through these refined operation steps, the construction quality of the slope graph can be further improved, providing more reliable data support for accurate prediction of the slope excavation engineering quantity.

[0041] In some embodiments, the feature network sequentially performs multi-level feature alignment and feature learning of each slope section, including: The multi-view attention network is used to process the reference slope graph and the to-be-measured slope graph to obtain a node embedding matrix corresponding to each graph structure. The self-attention mechanism of the fusion layer is used to update and fuse the node embedding matrices of the reference slope and the to-be-measured slope respectively to obtain updated node embedding features, and the comprehensive node embedding features are output after linear transformation. The regional feature alignment, node feature alignment, and edge feature alignment are sequentially performed for feature learning of each slope section.

[0042] It should be noted that the feature network mentioned in the present application is used to process the reference slope graph and the to-be-measured slope graph, and performs multi-level feature alignment to learn the features of each slope section. The core of the feature network is to use the multi-view attention network to process the graph structure to obtain a node embedding matrix corresponding to each graph structure, and then update and fuse these node embedding features through the self-attention mechanism of the fusion layer, and finally output the comprehensive node embedding features. This process involves regional feature alignment, node feature alignment, and edge feature alignment to ensure the accuracy and comprehensiveness of feature learning. The advantage of this method is that it can fully utilize multi-view information, automatically learn the importance of different views through the attention mechanism, and thus improve the effect of feature learning.

[0043] Specifically, the multi-view attention network is a network structure that can process multi-source data. It captures different feature dimensions of the slope map through multiple views. Each view can focus on different features, such as terrain, geology, and deformation. The node embedding matrix is obtained after processing the graph structure through the network, which contains the feature representation of each node in the graph. The self-attention mechanism of the fusion layer is used to update and fuse these node embedding features. By calculating the attention weight between different nodes, it automatically learns which node features are more important. The integrated node embedding feature is the final feature representation after processing by the fusion layer. It integrates the information of multiple views and can more comprehensively reflect the characteristics of the slope section. The regional feature alignment refers to calculating the cosine similarity of the feature vectors of the corresponding type of slope sections between the reference slope and the to-be-measured slope. It filters the matching slope section pairs and performs feature alignment training by minimizing the Euclidean distance loss function of the matching slope section pairs. The node feature alignment is to calculate the Pearson correlation coefficient of the node feature vectors of the displacement monitoring points. It maximizes the correlation coefficient loss function to make the features of the monitoring points under the same geological conditions consistent. It also updates the network parameters based on the feature difference gradient. The edge feature alignment includes extracting the edge feature vectors of the rock layer connection relationship between the slope sections. It calculates the similarity score of the edge features between the reference slope and the to-be-measured slope through bilinear transformation. It also constructs a three-tuple loss function based on the score to optimize the edge features. These steps together ensure the accuracy and comprehensiveness of feature learning.

[0044] Preferably, in order to further refine the operation steps of the feature network, the following aspects can be considered. First, when constructing the multi-view attention network, multiple views can be set, each view focusing on different feature dimensions. For example, one view can focus on terrain features such as slope and slope direction; another view can focus on geological features such as rock type and rock layer inclination; and another view can focus on deformation features such as displacement and deformation rate. The input parameters of each view can include corresponding sensor data or geographic information system (GIS) data. Second, when calculating the node embedding matrix, the graph neural network (GNN) method can be used to update the feature representation of each node by aggregating the information of neighboring nodes. In the fusion layer, the self-attention mechanism can update the node features by calculating the attention scores between nodes, and the specific parameters can include attention weights and learning rates. In the regional feature alignment, a similarity threshold can be set to filter matching slope segment pairs, for example, only when the cosine similarity of the feature vectors of two slope segments is higher than a certain threshold, they are considered to be matched. In the node feature alignment, the correlation between monitoring points can be measured by calculating the Pearson correlation coefficient, and the correlation coefficient loss function is maximized by optimization algorithm. In the edge feature alignment, the similarity score of edge features can be calculated by bilinear transformation, and the edge features are optimized by triple loss function. Through these refined operation steps, the performance of the feature network can be further improved, providing more reliable feature support for accurate prediction of slope excavation engineering quantity.

[0045] In some embodiments, the regional feature alignment comprises: calculating the cosine similarity of the feature vectors of the corresponding type of slope segments of the reference slope and the to-be-measured slope; filtering matching slope segment pairs based on a similarity threshold; training the feature alignment by minimizing the Euclidean distance loss function of the matching slope segment pairs.

[0046] It should be noted that the regional feature alignment mentioned in the present application is an important step in the feature network, which aims to calculate the cosine similarity of the feature vectors of the corresponding type of slope segments of the reference slope and the to-be-measured slope, filter out matching slope segment pairs, and train the feature alignment by minimizing the Euclidean distance loss function of the matching slope segment pairs. The core of this process lies in using cosine similarity to measure the directional similarity of two feature vectors, and using the Euclidean distance loss function to optimize the feature alignment, so as to ensure that the correspondence between the reference slope and the to-be-measured slope in the feature space is as consistent as possible. The advantage of this method is that it can effectively handle high-dimensional feature data and improve the accuracy of feature alignment by optimizing the loss function.

[0047] Specifically, the cosine similarity is a measure of the similarity between two vectors in terms of their directions, with a value range of -1 to 1. The closer the value is to 1, the more similar the directions of the two vectors are. In the present application, the cosine similarity is used to compare the feature vectors of corresponding type of slope segments between the reference slope and the to-be-measured slope, so as to screen out matched slope segment pairs. The similarity threshold is a preset parameter used to determine which slope segment pairs are considered to be matched. For example, the similarity threshold can be set to 0.8, and only when the cosine similarity of the feature vectors of two slope segments is greater than or equal to 0.8, they are considered to be matched. The Euclidean distance loss function is a commonly used loss function for measuring the difference between two vectors, and the smaller the value, the closer the two vectors are. In the present application, by minimizing the Euclidean distance loss function of the matched slope segment pairs, the feature alignment can be optimized so that the features of the reference slope and the to-be-measured slope are as consistent as possible in the feature space. Specific parameter settings include learning rate, iteration number, etc. of the loss function, which can be adjusted according to actual data and model performance.

[0048] Preferably, in order to further refine the operation steps of regional feature alignment, the following aspects can be considered. First, when calculating the cosine similarity, the feature vectors can be normalized to ensure fair comparison between different feature vectors. Normalization can be achieved by dividing each feature vector by its norm, which can eliminate the influence of feature vector length on similarity calculation. Second, when setting the similarity threshold, a suitable threshold can be selected by analyzing the distribution of the data and the performance of the model. For example, the optimal similarity threshold can be determined by cross-validation method to ensure the performance of the model on different data sets. Finally, when optimizing the Euclidean distance loss function, gradient descent algorithm or other optimization algorithms can be used to update the model parameters. The specific steps include calculating the gradient of the loss function, updating the model parameters according to the gradient, and repeating this process until the loss function converges. Through these refined operation steps, the accuracy and reliability of regional feature alignment can be further improved, providing stronger support for accurate prediction of slope excavation engineering quantity.

[0049] In some embodiments, the Euclidean distance loss function is represented as:

[0050] where N is the number of matched slope segment pairs; is the feature vector of the i-th matched slope segment in the reference slope; is the feature vector of the i-th matched slope segment in the to-be-measured slope.

[0051] It should be noted that the Euclidean distance loss function mentioned in the present application is a key step in the optimization of the regional feature alignment process. The loss function calculates the Euclidean distance between the feature vectors of the matching slope segment pairs in the reference slope and the to-be-measured slope, and takes the sum of the squares of these distances as the loss value, thereby measuring the quality of feature alignment. By minimizing this loss function, the model can learn how to adjust the feature vectors so that the features of the reference slope and the to-be-measured slope are as close as possible in the feature space. The advantage of this method is that it can directly quantify the differences between feature vectors and reduce these differences through the optimization process, thereby improving the accuracy of feature alignment.

[0052] Specifically, the number N of matching slope segment pairs in the Euclidean distance loss function refers to the total number of matching slope segment pairs selected by cosine similarity between the reference slope and the to-be-measured slope. The feature vectors of each matching slope segment pair are the feature representations of the corresponding slope segments in the reference slope and the to-be-measured slope. The loss function calculates the sum of the squares of the Euclidean distances between these feature vectors, i.e., the sum of the squares of the differences between each feature vector. In practical applications, parameters such as learning rate and iteration number can be set to optimize this loss function. The learning rate determines the step size of each parameter update, and the iteration number determines the total duration of the optimization process. The setting of these parameters needs to be adjusted according to the specific data set and model performance to ensure that the model converges to the optimal solution within a limited number of iterations.

[0053] Preferably, in order to further refine the optimization process of the Euclidean distance loss function, the following aspects can be considered. First, when calculating the Euclidean distance, the feature vectors can be normalized to ensure fair comparison between different feature vectors. Normalization can be achieved by dividing each feature vector by its norm, which can eliminate the influence of feature vector length on distance calculation. Second, when setting the learning rate, adaptive learning rate algorithms such as Adam or RMSprop can be used. These algorithms can automatically adjust the learning rate according to the size of the gradient, thereby improving the efficiency and stability of the optimization process. Finally, in the optimization process, the early stopping mechanism can be used to prevent overfitting of the model. Early stopping will stop training when the loss on the validation set no longer decreases, thereby ensuring that the performance of the model on the training set and the validation set is balanced. Through these refined operation steps, the optimization effect of the Euclidean distance loss function can be further improved, providing more reliable feature alignment support for accurate prediction of slope excavation engineering quantities.

[0054] In some embodiments, the node feature alignment includes: calculating the Pearson correlation coefficient of the displacement monitoring point node feature vector; aligning the monitoring point features under the same geological conditions by maximizing the correlation coefficient loss function; updating network parameters based on feature difference gradient.

[0055] It should be noted that the node feature alignment mentioned in the present application is an important step in the feature network, and the purpose is to make the features of the monitoring points under the same geological conditions consistent by calculating the Pearson correlation coefficient of the node feature vectors of the displacement monitoring points. This process is realized by maximizing the correlation coefficient loss function, and the network parameters are updated based on the feature difference gradient. The core of node feature alignment is to measure the linear correlation between the feature vectors of the monitoring points by using the Pearson correlation coefficient, and to adjust the network parameters by optimizing the loss function, so that the features of the monitoring points under the same geological conditions are as consistent as possible. The advantage of this method is that it can effectively handle the linear relationship of the monitoring point features, and improve the accuracy of feature alignment through the optimization process.

[0056] Specifically, the Pearson correlation coefficient is a measure of the linear correlation between two vectors, and its value ranges from -1 to 1. The closer the value is to 1 or -1, the stronger the linear correlation between the two vectors. In the present application, the Pearson correlation coefficient is used to compare the node feature vectors of the displacement monitoring points, thereby measuring the linear correlation between these monitoring points. The correlation coefficient loss function is an optimization objective, which aims to maximize the Pearson correlation coefficient between the feature vectors of the monitoring points, so that the features of the monitoring points under the same geological conditions are as consistent as possible. The feature difference gradient refers to the gradient information calculated according to the difference between the feature vectors during the optimization process, which is used to update the network parameters. Specific parameter settings include learning rate, number of iterations, etc., which can be adjusted according to actual data and model performance.

[0057] Preferably, in order to further refine the operation steps of node feature alignment, the following aspects can be considered. First, when calculating the Pearson correlation coefficient, the feature vectors can be standardized to ensure fair comparison between different feature vectors. Standardization can be achieved by subtracting the mean of each feature vector and dividing by its standard deviation, which can eliminate the dimension and scale differences between feature vectors. Second, when setting the learning rate, adaptive learning rate algorithms such as Adam or RMSprop can be used. These algorithms can automatically adjust the learning rate according to the size of the gradient, thereby improving the efficiency and stability of the optimization process. Finally, in the optimization process, early stopping mechanism can be used to prevent model overfitting. Early stopping mechanism will stop training when the loss on the validation set no longer decreases, thereby ensuring that the performance of the model on the training set and the validation set reaches a good balance. Through these refined operation steps, the accuracy and reliability of node feature alignment can be further improved, providing stronger support for accurate prediction of slope excavation engineering quantity.

[0058] In some embodiments, the edge feature alignment includes: extracting an edge feature vector of the rock stratum connection relationship between the edge slope segments; calculating a similarity score of the edge features of the reference edge slope and the to-be-measured edge slope through bilinear transformation; constructing a triplet loss function based on the score to optimize the edge features.

[0059] It should be noted that the edge feature alignment mentioned in the present application is an important step in the feature network, and its purpose is to extract an edge feature vector of the rock stratum connection relationship between the edge slope segments, calculate a similarity score of the edge features of the reference edge slope and the to-be-measured edge slope through bilinear transformation, and then optimize the edge features based on the score by constructing a triplet loss function. The core of this process is to use the edge feature vector to describe the connection relationship between the edge slope segments, and measure the similarity between the reference edge slope and the to-be-measured edge slope through the similarity score. By optimizing the triplet loss function, the model can learn how to adjust the edge features so that the edge features of the reference edge slope and the to-be-measured edge slope are as close as possible in the feature space. The advantage of this method is that it can effectively handle the connection relationship between the edge slope segments and improve the accuracy of edge feature alignment through the optimization process.

[0060] Specifically, the edge feature vector is obtained by extracting the rock stratum connection relationship between the edge slope segments, which describes the geological structure and connection mode between the edge slope segments. Bilinear transformation is a method for calculating the similarity between two vectors, which measures the relationship between two vectors through a learnable weight matrix. In the present application, bilinear transformation is used to calculate the similarity score of the edge features of the reference edge slope and the to-be-measured edge slope, thereby quantifying the similarity between the two edge slope segments. The triplet loss function is a commonly used optimization objective, which optimizes model parameters by comparing the similarity between a positive sample, a negative sample and an anchor point. In the present application, the triplet loss function is used to optimize the edge features so that the edge features of the reference edge slope and the to-be-measured edge slope are as close as possible in the feature space. Specific parameter settings include learning rate, number of iterations, etc., which can be adjusted according to actual data and model performance.

[0061] Preferably, to further refine the operation steps of edge feature alignment, the following aspects can be considered. First, when extracting the edge feature vector, geological exploration data and rock distribution information can be combined to extract the rock connection relationship between each edge slope section. For example, the edge feature vector can be constructed by analyzing the dip angle, thickness and contact relationship of the rock layer. Second, when calculating the similarity score, a bilinear transformation model can be trained to learn the optimal weight matrix. This model can be trained by a backpropagation algorithm, with input parameters including edge feature vectors and corresponding labels. Finally, when optimizing the triplet loss function, the loss function can be constructed by setting appropriate positive samples, negative samples and anchor points. For example, a slope section of a reference slope can be selected as an anchor point, a similar slope section as a positive sample, and a dissimilar slope section as a negative sample. Through these refined operation steps, the accuracy and reliability of edge feature alignment can be further improved, providing stronger support for accurate prediction of edge slope excavation quantities.

[0062] In some embodiments, the weighted network loss function is:

[0063] wherein, is the model predicted excavation quantity vector; V is the actual observed excavation quantity vector; is the covariance matrix of the excavation quantity observation error; is the excavation quantity index of the reference slope; is the excavation quantity index of the to-be-measured slope; is the regularization coefficient; is the similarity weight vector; is the similarity weight vector; is the L2 norm.

[0064] It should be noted that the weighted network loss function mentioned in the present application is a key part of the optimization of the slope excavation quantity calculation model. This loss function takes into account the difference between the model predicted excavation quantity and the actual observed value, the difference between the excavation quantity index of the reference slope and the to-be-measured slope, and the regularization term of the similarity weight. By minimizing this loss function, the model can learn the optimal weights and parameters, thereby improving the accuracy and reliability of the excavation quantity prediction. The advantage of this method is that it can consider multiple factors at the same time, prevent model overfitting through the regularization term, and ensure the generalization ability of the model on different data sets.

[0065] Specifically, the model-predicted excavation volume vector in the weighted network loss function refers to the set of excavation volumes predicted by the model based on the input data, while the actually observed excavation volume vector refers to the set of excavation volumes obtained through actual measurement. The difference between the two vectors is measured by the covariance matrix of the excavation volume observation error, which reflects the distribution of the observation error. The excavation index is a comprehensive index calculated based on the characteristics of the slope section, used to measure the excavation difficulty or engineering quantity of the slope section. The similarity weight vector represents the degree of similarity between the reference slope and the to-be-measured slope, and its complexity is controlled by the regularization term to prevent overfitting of the model. The regularization coefficient is a parameter used to balance the importance of different loss terms and can be adjusted according to the performance of the model and the characteristics of the data.

[0066] Preferably, in order to further refine the construction and optimization process of the weighted network loss function, the following aspects can be considered. First, when constructing the loss function, the specific form and calculation method of each loss term can be defined in detail. For example, the difference between the model-predicted excavation volume and the actual observed value can be measured by the mean square error MSE, i.e. calculating the sum of squares of the difference between the two vectors. The difference between the excavation index of the reference slope and the to-be-measured slope can be measured by the L2 norm, i.e. calculating the square root of the sum of squares of the difference between the two vectors. The regularization term can adopt L2 regularization, i.e. calculating the sum of squares of the similarity weight vector. Secondly, in the optimization process, gradient descent algorithm or other optimization algorithms can be used to update the model parameters. The specific steps include calculating the gradient of the loss function, updating the model parameters according to the gradient, and repeating this process until the loss function converges. Finally, the optimal regularization coefficient can be selected through cross-validation and other methods to ensure the performance of the model on different data sets. Through these refined operation steps, the optimization effect of the weighted network loss function can be further improved, providing more reliable model support for the accurate prediction of slope excavation engineering quantity.

[0067] The above-mentioned various embodiments of the present application have the following beneficial effects: 1. By fusing the terrain feature similarity weight of the reference slope and the to-be-measured slope, and combining the historical excavation volume data to train the prediction model, the problem of engineering quantity estimation deviation caused by insufficient data of the to-be-measured slope in traditional methods is effectively solved, and the accuracy and reliability of the prediction result are significantly improved.

[0068] 2. The graph structure, feature network and weighted network are used to construct the slope excavation engineering quantity calculation model, realizing efficient feature extraction and alignment of multi-source data of the slope section, solving the problem of difficult quantization of slope features under complex geological conditions, and enhancing the adaptability of the model to engineering practice.

[0069] 3、Based on the excavation amount index multi-factor calculation framework, the interaction of the prediction results, the geological characteristics and the topographic features is comprehensively considered, the limitation of single index estimation is overcome, the engineering quantity calculation result is more close to the actual construction demand, and a scientific basis is provided for the optimization design of the slope engineering.

[0070] Further, the storage medium of the embodiments of the present application stores program instructions capable of implementing all the methods described above, wherein the program instructions can be stored in the storage medium in the form of a software product, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media capable of storing program codes, or a computer, a server, a mobile phone, a tablet, and other terminal devices.

[0071] The above description is only some of the preferred embodiments of the present application and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features and the technical features disclosed in the embodiments of the present application (but not limited to) having similar functions.

Claims

1. A method for calculating an excavation volume index based on multi-source data features, characterized in that, The method comprises the following steps: training a slope excavation engineering quantity calculation model based on a training set comprising historical excavation quantity data and terrain-related data of a reference slope and a to-be-measured slope, and obtaining terrain features and terrain feature similarity weights of each slope section of the reference slope and the to-be-measured slope; wherein the historical data of the reference slope is more than that of the to-be-measured slope to be analyzed; training an excavation quantity prediction model based on historical excavation quantity data of the reference slope; obtaining excavation quantity prediction results and geological features of each slope section of the to-be-measured slope by using the excavation quantity prediction model based on the terrain feature similarity weights and the historical excavation quantity data of the to-be-measured slope; obtaining an excavation quantity index of a corresponding slope section based on the excavation quantity prediction results, the geological features and the terrain features of each slope section, and summarizing the excavation quantity indexes of each slope section to obtain the slope excavation engineering quantity of the to-be-measured slope.

2. The method according to claim 1, wherein, The excavation quantity index of the corresponding slope section is calculated based on the following formula: wherein, is a weight factor; PV is the prediction result of the excavation amount of the slope section; GF is a geological feature value; TF is a topographic feature value; and SW is a topographic feature similarity weight; is a vector representation of the topographic feature; is a vector representation of the geological feature; and e is a natural constant; is a vector norm. 3.The method of claim 2, wherein, The slope excavation engineering quantity calculation model comprises a graph structure, a feature network and a weighted network; training the slope excavation engineering quantity calculation model comprises: obtaining a plurality of reference slope graphs and a plurality of to-be-measured slope graphs by using the graph structure based on the training set; dividing the slope sections based on the plurality of reference slope graphs and the plurality of to-be-measured slope graphs, and learning the features of each slope section by using the feature network to obtain comprehensive node embedding features of each slope section of the reference slope and the to-be-measured slope; learning the slope section weights by using the weighted network and the weighted network loss function based on the comprehensive node embedding features of the reference slope and the to-be-measured slope respectively to obtain the similarity weights of each slope section of the reference slope and the to-be-measured slope; after the training is completed, the comprehensive node embedding features and the similarity weights obtained by using the slope excavation engineering quantity calculation model based on the training set are the terrain features and the terrain feature similarity weights of each slope section of the reference slope and the to-be-measured slope.

4. The method of claim 3, wherein, obtaining a plurality of the reference slope graphs and the to-be-measured slope graphs comprises: constructing a position adjacency matrix based on the geographical positions of the reference slope and the to-be-measured slope respectively; constructing a rock stratum adjacency matrix based on the geological rock stratum relationships of the reference slope and the to-be-measured slope respectively; constructing a deformation adjacency matrix based on the displacement monitoring points of the reference slope and the to-be-measured slope respectively; wherein the displacement monitoring points are slope deformation data points obtained through a sensor network; constructing a profile similarity matrix of source slope sections and destination slope sections based on excavation profile data of the reference slope and the to-be-measured slope respectively; obtaining a plurality of the reference slope graphs and the to-be-measured slope graphs based on the above-mentioned various adjacency matrices.

5. The method of claim 4, wherein, The feature network sequentially performs multi-level feature alignment and feature learning of each slope section, comprising: processing the reference slope graph and the to-be-measured slope graph by using a multi-view attention network to obtain a node embedding matrix corresponding to each graph structure; updating and fusing the node embedding matrices of the reference slope and the to-be-measured slope by using a self-attention mechanism of a fusion layer to obtain updated node embedding features, and outputting comprehensive node embedding features through linear transformation; performing feature learning of each slope section in sequence by using regional feature alignment, node feature alignment and edge feature alignment.

6. The method of claim 5, wherein, The regional feature alignment comprises: Calculate the cosine similarity of the feature vectors of the corresponding type of slope segments of the reference slope and the to-be-measured slope; Filter the matching slope segment pairs based on a similarity threshold value; Perform feature alignment training by minimizing the Euclidean distance loss function of the matching slope segment pairs.

7. The method of claim 6, wherein, The Euclidean distance loss function is expressed as: Wherein, N is the number of matching slope section pairs; is the feature vector of the i th matching slope section in the reference slope; is the feature vector of the i th matching slope section in the slope to be measured. 8.The method of claim 5, wherein, The node feature alignment includes: Calculate the Pearson correlation coefficient of the node feature vectors of the displacement monitoring points; Make the features of the monitoring points under the same geological conditions consistent by maximizing the correlation coefficient loss function; Update the network parameters based on the feature difference gradient. 9.The method of claim 5, wherein, The edge feature alignment includes: Extract the edge feature vectors of the rock layer connection relationship between the slope segments; Calculate the similarity score of the edge features of the reference slope and the to-be-measured slope by bilinear transformation; Perform edge feature optimization based on the score to construct a triple loss function.

10. The method of claim 3-9, wherein, The weighted network loss function is: where, is the vector of predicted excavation quantities for the model; V is the vector of actual observed excavation quantities; is the covariance matrix of the excavation quantity observation error; is the vector of excavation quantity indices for the reference slope; is the vector of excavation quantity indices for the slope under test; is the regularization coefficient; is the vector of similarity weights; is the vector of similarity weights; is the L2 norm.

Images