BIM prestressed pipe pile assembly optimization system based on multi-source geological data

Abstract

The invention relates to the technical field of intelligent design of geotechnical engineering, and discloses a BIM prestressed pipe pile assembly optimization system based on multi-source geological data. The system comprises a data preprocessing module, a feature extraction module, a feature sensing module, a feature fusion module and a BIM pile matching generation module. The system performs standardization processing on the original geological data; clustering analysis is carried out from a historical engineering case library, and geological mode features with a space influence radius are extracted; current site data is coded through a multi-level feature perception architecture, and features of different levels are fused by using a bidirectional attention mechanism; and mapping the fused comprehensive geological features to a BIM environment, and automatically driving to generate an optimized pile matching scheme. According to the method, intelligent reuse of engineering geological knowledge and deep fusion of multi-scale geological features are realized, a prestressed pipe pile design scheme highly adaptive to complex site conditions can be automatically and accurately generated, and the design efficiency and the scientificity of the scheme are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent design technology for geotechnical engineering, specifically to a BIM-based prestressed pipe pile configuration optimization system based on multi-source geological data. Background Technology

[0002] Currently, the application of Building Information Modeling (BIM) technology in pile foundation engineering is largely limited to the 3D visualization and clash detection of established design schemes. The design process itself still heavily relies on the personal experience of engineers, making judgments based on limited current site survey reports. Existing BIM platforms lack the ability to deeply mine and structurally reuse the inherent correlation between massive, multi-source historical engineering geological data and successful pile placement schemes, resulting in a single and fragmented source of knowledge for design decisions.

[0003] In utilizing geological data for intelligent decision support, existing methods often rely on direct statistical analysis of current site stratigraphic parameters or simple comparisons with a few similar cases. These methods struggle to systematically extract and quantify the common patterns of economically sound and practically validated pile configuration patterns under different geological conditions. Furthermore, conventional data processing methods typically treat geological features as homogeneous or independent attributes superimposed, failing to effectively express the differentiated and nonlinear comprehensive impacts of geological features at different scales and with varying influence ranges on design parameters such as pile length and spacing. This results in generated pile configuration schemes that are often conservative or insufficiently adapted to the actual geology of the site. Summary of the Invention

[0004] The purpose of this invention is to provide a BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a BIM-based prestressed concrete pipe pile allocation optimization system based on multi-source geological data, the system comprising: The data preprocessing module receives the original exploration information set of the target site from the geological data source, performs data integrity and consistency verification, and generates a standardized multi-source geological dataset. The feature extraction module calls the stored engineering case library, performs clustering on historical prestressed pipe pile schemes based on lithology and mechanical properties, forms multiple pipe pile scheme clusters with different geological backgrounds, extracts the core geological model of each pipe pile scheme cluster and calculates its spatial influence radius, and outputs the geological model feature vector and feature influence domain. The feature perception module constructs a multi-level feature perception architecture based on the dimensions of the feature vectors of the multiple geological models. The standardized multi-source geological dataset is fed into the multi-level feature perception architecture. At each level, the dataset is spatially divided and feature-encoded according to the corresponding feature influence domain, and the hierarchical geological feature code is output. The feature fusion module, through its built-in feature cross-fusion unit, performs bidirectional attention weighting and splicing on the hierarchical geological feature codes from different levels to generate a fused comprehensive geological feature representation of the site. The BIM pile generation module maps the comprehensive geological features of the site to the BIM model environment, drives the pile components in the model, and generates a set of optimized pile configuration schemes based on the specifications, length and layout parameters of the prestressed pipe piles.

[0006] Preferably, the core geological model of each pipe pile scheme cluster is extracted and its spatial influence radius is calculated, including: For each of the aforementioned pipe pile schemes, a geological sequence sample set is constructed with pile location coordinates and stratum parameter sequences as input; Initialize a deep embedding network, which includes a stacking module for sequence modeling and a pooling module for feature condensation; The geological sequence sample set is input into the deep embedding network, and the stacking module captures the long-range dependencies between stratigraphic parameters in the sample, outputting high-dimensional sequence features. The high-dimensional sequence features are input into the pooling module, and through adaptive weighted convergence operation, a dense vector representing the common geological laws of the pipe pile scheme cluster is extracted as the core geological model. After extracting the core geological model, for each geological sequence sample in the pipe pile scheme cluster, the feature space distance between its original sequence features and the core geological model is calculated. The distribution of the feature spatial distances of all samples is statistically analyzed, and the values ​​corresponding to the preset quantiles of the distance distribution are defined as the feature influence domain of the core geological model.

[0007] Preferably, the standardized multi-source geological dataset is fed into the multi-level feature-aware architecture, and at each level, the dataset is spatially divided and feature-encoded according to the corresponding feature influence domain, including: The perception range of each level of the multi-level feature perception architecture is matched with a feature influence domain. At a specific level of the multi-level feature perception architecture, the value of the feature influence domain associated with the specific level is obtained, and the value is used as the spatial neighborhood radius. Taking each exploration point or grid point in the target site as the center and the radius of the spatial neighborhood as the side length, a cubic spatial region is defined in the standardized multi-source geological dataset, and all geological data points in the cubic spatial region are taken as the neighborhood data block of the current center point. All data points in the neighboring data block are organized into a three-dimensional tensor with a channel structure according to their spatial location and geological attribute type. A three-dimensional convolution kernel group is used to perform convolution operations on the three-dimensional tensor to extract local geological structure features within the spatial neighborhood of the neighborhood data block; After processing all exploration points or grid points, all extracted local geological structural features are rearranged according to the spatial position of their center points to form the hierarchical geological feature code of the specific level. The hierarchical geological feature code is a feature map corresponding to the site spatial coordinates.

[0008] Preferably, the hierarchical geological feature codes from different levels are subjected to bidirectional attention weighting and splicing through a built-in feature cross-fusion unit, including: The feature cross-fusion unit receives hierarchical geological feature codes from at least two different levels of the multi-level feature perception architecture; The hierarchical geological feature codes from the coarser sensing level, i.e., those with a larger feature influence domain, are upsampled to make their spatial resolution consistent with that of the hierarchical geological feature codes from the finer sensing level. The coarser-level feature encoding and the finer-level feature encoding after upsampling are concatenated along the feature channel dimension to form a combined feature tensor. The combined feature tensor is input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights; Calculate the dot product similarity between the query feature map and the key feature map at each spatial location, and process it through a normalization function to generate a spatial attention weight map; The combined feature tensor is reweighted in a channel-like manner using the spatial attention weight map to highlight the feature responses of important spatial locations. The reweighted feature tensor is input into a fusion convolutional layer for dimensionality reduction and fusion, and finally the fused comprehensive geological feature expression of the site is output.

[0009] Preferably, the combined feature tensor is input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights, including: The two independent convolutional layers have the same kernel size but each has its own independent trainable weight parameters; After one of the convolutional layers processes the combined feature tensor, the output feature map is defined as the query feature map, which is used to characterize the feature pattern for which information needs to be obtained at each position. After another convolutional layer processes the combined feature tensor, the output feature map is defined as the key feature map, which is used to characterize the feature patterns that can provide information at each location; The query feature map and the key feature map have the exact same spatial size and number of feature channels to facilitate point-by-point similarity matching calculations.

[0010] Preferably, the stored engineering case library is invoked, and historical prestressed concrete pipe pile schemes are clustered based on lithology and mechanical properties, including: All historical prestressed concrete pipe pile scheme records are retrieved from the engineering case library. Each record contains a sequence of strata lithology descriptions and corresponding soil mechanical parameter sequences for its construction site. The lithological description sequence of the strata is digitally encoded, and the lithological categories are transformed into multidimensional unique heat vectors. At the same time, the sequence of soil mechanical parameters is standardized to eliminate dimensional differences. The processed lithology vector sequence and mechanical parameter sequence of each historical prestressed pipe pile scheme are connected in depth to form a high-dimensional geological feature representation of the historical prestressed pipe pile scheme. A density-based clustering algorithm is used to find regions with dense sample distribution in the space formed by the high-dimensional geological feature representation; All historical prestressed pipe pile schemes located in the same dense area are grouped into a set, each set constituting a pipe pile scheme cluster, and each pipe pile scheme cluster is assigned a unique cluster identifier.

[0011] Preferably, a density-based clustering algorithm is used to find densely distributed regions of samples within the space constituted by the high-dimensional geological feature representation, including: Set the neighborhood search radius parameter and the minimum number of neighborhood samples parameter for the density-based clustering algorithm; In the high-dimensional geological feature representation space, a historical prestressed pipe pile scheme sample that has not been accessed is randomly selected as the starting point; Using the starting point as the center, search for all other sample points within the hypersphere defined by the neighborhood search radius parameter; If the number of sample points found is greater than or equal to the minimum neighborhood sample number parameter, then the starting point is determined to be the core point, and based on the sample points, its neighborhood is recursively expanded, and all sample points that can be reached through density are marked as the same cluster. If the number of sample points found is less than the minimum neighborhood sample number parameter, the starting point is temporarily marked as a noise point. The process of selecting a starting point, searching for neighbors, and marking is repeated until all historical prestressed concrete pipe pile scheme samples have been visited, ultimately completing the division of multiple pipe pile scheme clusters.

[0012] Preferably, the system receives the original survey information set of the target site from a geological data source and performs data integrity and consistency checks, including: Establish a standard geological data field list, which specifies the data items that must be included and their format; Each data record in the received set of original exploration information is compared item by item with the list of standard geological data fields to identify missing data fields, data fields with incorrect formats, and data fields whose values ​​are significantly outside the reasonable range, and a data anomaly report is generated. For missing data fields, based on their spatial location, the Kriging interpolation algorithm is used to fill in the missing data by using similar data from surrounding valid exploration points. For data fields with incorrect formats, convert them to the standard format according to the predefined data conversion rule library; For data fields whose values ​​clearly exceed the reasonable range, mark them as data to be reviewed and trigger the manual review process. After manual review, they will either be replaced with corrected values ​​or removed according to technical specifications. After all verification and processing steps are completed, the standardized multi-source geological dataset is output.

[0013] Preferably, the comprehensive geological features of the site are mapped to the BIM model environment, driving the pile arrangement components in the model to generate a set of optimized pile arrangement schemes based on the specifications, length, and layout parameters of the prestressed concrete pipe piles, including: In the BIM model environment, a predefined component family library for prestressed pipe piles is provided, which contains various pipe pile family types with different diameters, wall thicknesses, concrete strength grades, and prestressed steel reinforcement configurations. The comprehensive geological features of the site are input into a parameterized rule inference engine, which has a pre-set set of mapping rules from geological features to pile bearing capacity and settlement features. The parametric rule reasoning engine, based on the comprehensive geological characteristics of the site and combined with the structural design load requirements, infers the required single pile bearing capacity characteristic value and estimated settlement at each proposed pile location. Based on the single pile bearing capacity characteristic value and the estimated settlement, the system automatically matches the pipe pile family type that meets the conditions in the component family library, and sorts the various feasible pipe pile specifications and length combinations based on the preset economic and construction convenience weights. Based on the sorting results, a variety of different pipe pile layout plans and cross-sectional views are generated. Each layout plan details the specifications, length, quantity, pile top elevation, and layout spacing of the pipe piles used, forming a set of optimized pile arrangement schemes.

[0014] Preferably, the parametric rule inference engine, based on the comprehensive geological characteristics of the site and combined with the structural design load requirements, infers the required single pile bearing capacity characteristic value and estimated settlement at each proposed pile location, including: The parameterized rule reasoning engine loads the comprehensive geological feature expression of the site, which includes the characteristic values ​​of soil compression modulus, internal friction angle, cohesion, and standard penetration test blow count at each simulated pile location. Meanwhile, the parametric rule inference engine reads the structural design load requirements from an external interface, which include vertical axial force, horizontal force, and bending moment. The parameterized rule reasoning engine calls the built-in pile bearing capacity calculation formula, takes the soil layer parameters and structural load at the proposed pile location as input variables, iteratively calculates the pile end resistance and pile side resistance required to meet the bearing capacity requirements, and then back-calculates the required single pile bearing capacity characteristic value. The parameterized rule inference engine calls the built-in layered summation method settlement calculation formula, uses the characteristic value of the soil layer compression modulus and the estimated additional stress at the pile end to calculate the settlement of the pile foundation under different load levels, and obtains the estimated settlement. The calculated characteristic value of the single pile bearing capacity and the estimated settlement are compared and verified with the allowable value in the design specifications to ensure that they meet the safety requirements.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By accessing a historical engineering case library, the system clusters pipe pile schemes based on geotechnical mechanics indicators and extracts the core geological patterns of each cluster. Simultaneously, it calculates the spatial influence radius, transforming tacit engineering knowledge into explicit, quantifiable spatial characteristics. This enables the system to automatically identify matching patterns between current site geological conditions and historically successful schemes, eliminating absolute reliance on engineers' personal experience and memory. It provides spatially validated design references for pile installation, validated by numerous examples, thus improving the reliability and economy of the scheme.

[0016] A multi-level feature perception architecture is constructed, and a bidirectional attention mechanism is used to fuse feature codes from different levels, enabling the system to understand the multi-layered influence of complex geological conditions. It can dynamically assess and integrate the importance of geological information at different scales, from local details to regional macroscopic data, rather than processing it equally or linearly. This allows the generated comprehensive geological feature representation of the site to more accurately reflect the heterogeneity of the geological environment and the interaction of various factors. The pile arrangement schemes driving the BIM model have a higher degree of coupling with the actual geology of the site in terms of pile length optimization, bearing layer selection, and pile density, reducing design deviations. Attached Figure Description

[0017] Figure 1 This is a timeline diagram of the BIM-based prestressed pipe pile configuration optimization system based on multi-source geological data described in this invention. Figure 2 A flowchart for core geological model extraction and influence radius calculation; Figure 3 A flowchart for feature cross-fusion; Figure 4 A bar chart comparing quality indicators before and after geological data standardization; Figure 5 Efficiency analysis diagram for each processing stage of the BIM prestressed pipe pile allocation optimization system. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1This invention provides a BIM-based prestressed concrete pipe pile configuration optimization system based on multi-source geological data. The system includes a data preprocessing module, a feature extraction module, a feature perception module, a feature fusion module, and a BIM configuration generation module. The data preprocessing module receives the original survey information set of the target site from geological data sources, performs data integrity and consistency checks, and generates a standardized multi-source geological dataset. The feature extraction module calls a stored engineering case library and performs clustering on historical prestressed concrete pipe pile schemes based on stratigraphic lithology and mechanical properties, forming multiple pipe pile scheme clusters with different geological backgrounds. It extracts the core geological model of each pipe pile scheme cluster and calculates its spatial influence radius, outputting the geological model feature vector and feature influence domain. The feature perception module constructs a multi-level feature perception architecture based on the dimensions of multiple geological model feature vectors. The standardized multi-source geological dataset is fed into the multi-level feature perception architecture. At each level, the dataset is spatially divided and feature-encoded according to the corresponding feature influence domain, outputting hierarchical geological feature codes. The feature fusion module uses a built-in feature cross-fusion unit to perform bidirectional attention weighting and splicing on the hierarchical geological feature codes from different levels, generating a fused comprehensive geological feature representation of the site. The BIM pile generation module maps the comprehensive geological feature representation of the site to the BIM model environment, drives the pile component in the model, and generates a set of optimized pile allocation schemes based on the specifications, length, and layout parameters of the prestressed pipe piles.

[0020] In one embodiment of the present invention, see [reference] Figure 2 For each pile foundation cluster, a geological sequence sample set is constructed, using pile location coordinates and stratigraphic parameter sequences as input. A deep embedding network is initialized, containing a stacking module for sequence modeling and a pooling module for feature condensation. The geological sequence sample set is input into the deep embedding network, and the stacking module captures the long-range dependencies between stratigraphic parameters in the samples, outputting high-dimensional sequence features. The high-dimensional sequence features are then input into the pooling module, and an adaptive weighted pooling operation is used to extract dense vectors representing the common geological patterns of the pile foundation cluster, serving as the core geological model. After extracting the core geological model, for each geological sequence sample in the pile foundation cluster, the feature space distance between its original sequence features and the core geological model is calculated. The distribution of the feature space distances of all samples is statistically analyzed, and the values ​​corresponding to the preset quantiles of the distance distribution are defined as the feature influence domain of the core geological model.

[0021] Each level of the multi-level feature perception architecture has a perception range matched with a feature influence domain. The numerical value of the feature influence domain associated with that level is obtained at a specific level of the architecture, and this value is used as the spatial neighborhood radius. A cubic spatial region is delineated in the standardized multi-source geological dataset, centered on each exploration point or grid point within the target site and with the spatial neighborhood radius as its side length. All geological data points within this cubic spatial region are considered as the neighborhood data block of the current center point. All data points in the neighborhood data block are organized into a three-dimensional tensor with a channel structure according to their spatial location and geological attribute type. A three-dimensional convolution kernel group is used to perform convolution operations on the three-dimensional tensor to extract local geological structural features within the spatial neighborhood of the neighborhood data block. After processing all exploration points or grid points, all extracted local geological structural features are rearranged according to the spatial location of their center points to form a hierarchical geological feature code for a specific level. This hierarchical geological feature code is a feature map corresponding to the site's spatial coordinates.

[0022] In practical implementation, a geological sequence sample set is constructed for each prestressed concrete pipe pile scheme cluster, with pile location coordinates and stratum parameter sequences as input. This sample set contains site exploration data corresponding to all historical prestressed concrete pipe pile schemes within the cluster. Each sample data consists of three-dimensional spatial coordinates and soil layer parameter vectors arranged in depth order. A deep embedding network is initialized, containing a stacking module for sequence modeling and a pooling module for feature condensation. The stacking module consists of multiple bidirectional long short-term memory (LSTM) network layers, while the pooling module is designed as a multi-head attention convergence layer with learnable parameters. The geological sequence sample set is input into the deep embedding network. The stacking module captures the long-range dependencies between stratum parameters in the samples, and the bidirectional LSM network layers process the sequence in both forward and backward directions, outputting high-dimensional contextual features for each soil layer location, thus forming high-dimensional sequence features. These high-dimensional sequence features are then input into the pooling module, where an adaptive weighted convergence operation extracts dense vectors representing the common geological patterns of the prestressed concrete pipe pile scheme cluster, serving as the core geological model. The calculation process of the adaptive weighted convergence operation can be described as follows:

[0023] in: This represents the final core geological model vector. This represents the high-dimensional sequence feature vector corresponding to the location of the i-th soil layer. The weighting coefficients, calculated using an attention mechanism, reflect the importance of this soil layer feature to the overall model. This represents the total number of layers in the stratigraphic sequence.

[0024] After extracting the core geological model, for each geological sequence sample in the pipe pile scheme cluster, the feature space distance between its original sequence features and the core geological model is calculated. The feature space distance uses Euclidean distance as a metric, and the normalized high-dimensional sequence feature vector and the core geological model vector are calculated in the feature space. The distribution of the feature space distances of all samples is statistically analyzed, and the value corresponding to the preset quantile of the distance distribution is defined as the feature influence domain of the core geological model. The preset quantile is usually selected as the 95th percentile, meaning that the feature space distance of 95% of the samples in the cluster is less than or equal to this value. This value defines the effective influence range of the core geological model in the feature space.

[0025] In practical implementation, the perception range of each level of the multi-level feature perception architecture is matched with a feature influence domain. This means that different levels in the architecture focus on different spatial scales of geological structures. Shallower levels correspond to larger feature influence domains and are responsible for perceiving large-scale macroscopic geological trends, while deeper levels correspond to smaller feature influence domains and are responsible for perceiving fine local geological changes. At a specific level of the multi-level feature perception architecture, the numerical value of the feature influence domain associated with that level is obtained, and this value is used as the spatial neighborhood radius. This spatial neighborhood radius is a specific unit of length, such as 15 meters, 8 meters, or 3 meters, which defines the three-dimensional spatial range examined when performing feature extraction at that level. Using each exploration point or grid point within the target site as the center and the spatial neighborhood radius as the side length, a cubic spatial region is delineated in the standardized multi-source geological dataset. All geological data points within this cubic spatial region are considered as neighborhood data blocks of the current center point. Each neighborhood data block contains all soil layer information and attribute data within a certain spatial range around the center point.

[0026] All data points in the neighboring data block are organized into a three-dimensional tensor with a channel structure according to their spatial location and geological attribute type. The three spatial dimensions of the three-dimensional tensor correspond to the east-west, north-south, and depth directions, while the channel dimensions correspond to different geological attributes, such as soil layer type encoding, compression modulus, and internal friction angle. A three-dimensional convolution kernel group is used to perform convolution operations on the three-dimensional tensor to extract local geological structure features within the spatial neighborhood of the neighboring data block. The three-dimensional convolution kernel slides along the three spatial dimensions, and its parameters are learned through training, enabling it to identify local stratigraphic combination patterns or attribute spatial distribution patterns. After processing all exploration points or grid points, all extracted local geological structure features are rearranged according to the spatial location of their center points to form a hierarchical geological feature encoding at a specific level. The hierarchical geological feature encoding is a feature map corresponding to the site's spatial coordinates. Each pixel in the feature map is associated with a feature vector, which encodes the geological structure information within a specific spatial neighborhood radius centered on that point.

[0027] In some embodiments, when constructing a geological sequence sample set, the stratigraphic parameter sequence includes not only the original geotechnical test parameters but also derived parameters converted from field tests to ensure the integrity of the sequence information. In some embodiments, the number of stacked modules in the deep embedding network is set according to the average complexity of geological sequences in the historical engineering case library. For case libraries with frequent stratigraphic changes, deeper stacked modules are used to enhance sequence modeling capabilities. It can be understood that the determination of the feature influence domain depends on the density of the sample distribution within the pipe pile scheme cluster. A pipe pile scheme cluster with high internal consistency has a concentrated feature spatial distance distribution, resulting in a smaller calculated feature influence domain value, and vice versa.

[0028] In one embodiment of the present invention, see [reference] Figure 3 The feature cross-fusion unit receives hierarchical geological feature codes from at least two different levels of a multi-level feature perception architecture. The hierarchical geological feature codes from the coarser perception level (i.e., those with a larger feature influence domain) are upsampled to maintain spatial resolution consistent with those from the finer perception level. The upsampled coarser and finer hierarchical feature codes are concatenated along the feature channel dimension to form a combined feature tensor. This combined feature tensor is then input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights. The dot product similarity between the query and key feature maps at each spatial location is calculated and processed using a normalization function to generate a spatial attention weight map. The spatial attention weight map is used to reweight the combined feature tensor in a channel-wise manner, highlighting the feature responses at important spatial locations. The reweighted feature tensor is then input into a fusion convolutional layer for dimensionality reduction and fusion, ultimately outputting a fused comprehensive geological feature representation of the site.

[0029] Two independent convolutional layers have the same kernel size but each has independent trainable weight parameters. One convolutional layer processes the combined feature tensor, and its output feature map is defined as the query feature map, which represents the feature pattern from which information is to be obtained at each location. The other convolutional layer processes the combined feature tensor, and its output feature map is defined as the key feature map, which represents the feature pattern from which information is provided at each location. The query feature map and the key feature map have the exact same spatial dimensions and number of feature channels to facilitate pointwise similarity matching calculations.

[0030] In practical implementation, the feature cross-fusion unit receives hierarchical geological feature codes from at least two different levels of the multi-level feature perception architecture. For example, the outputs from a coarser perception level with a feature influence domain of 12 meters and a finer perception level with a feature influence domain of 3 meters are received. These two levels of hierarchical geological feature codes have different resolutions in the spatial dimension. The hierarchical geological feature codes from the coarser perception level (i.e., the level with a larger feature influence domain) are upsampled to match the spatial resolution of the hierarchical geological feature codes from the finer perception level. In specific implementation, the hierarchical geological feature codes from the 12-meter feature influence domain have a 5x5 grid spatial resolution on the plane, while the hierarchical geological feature codes from the 3-meter feature influence domain have a 20x20 grid spatial resolution. A bilinear interpolation algorithm is then used to upsample the 5x5 feature map from the coarser level to 20x20, ensuring that both have the same size in the spatial dimension.

[0031] In practice, the upsampled coarser-level feature encoding and the finer-level feature encoding are concatenated along the feature channel dimension to form a combined feature tensor. Assuming the upsampled coarser-level feature encoding has 64 feature channels and the finer-level feature encoding has 128 feature channels, the resulting combined feature tensor will have 192 feature channels, while maintaining a spatial size of 20x20. This combined feature tensor is then input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights. The dot product similarity between the query feature map and the key feature map at each spatial location is calculated, and a spatial attention weight map is generated through a normalization function. The calculation process of the spatial attention weight map can be described as follows:

[0032] in: This represents the attention weight of position p to position q. It is the feature vector mapped to the query feature at position p. It is the feature vector of the key feature at position q. It is the dimension of the feature vector. The function ensures that the sum of all weights is 1. A channel-wise reweighting of the combined feature tensor is performed using a spatial attention weight map to highlight the feature responses of important spatial locations. Specifically, the spatial attention weight map is multiplied element-wise with the combined feature tensor. The spatial attention weight map is broadcast to each channel of the combined feature tensor, thereby enhancing the signal intensity of the areas of interest and suppressing the signal of unimportant areas. The reweighted feature tensor is then input into a fusion convolutional layer for dimensionality reduction and fusion, ultimately outputting a fused comprehensive geological feature representation of the site. The fusion convolutional layer is typically a 1x1 convolutional kernel that reduces the dimensionality of 192 feature channels and fuses them into, for example, 256 channels while maintaining a spatial resolution of 20x20. This 256-channel feature map is the comprehensive geological feature representation of the site that integrates multi-scale geological information.

[0033] In some embodiments, two independent convolutional layers have the same kernel size but each has independent trainable weight parameters. The kernel size can be set to 3x3, and the number of output feature channels of each convolutional layer is set to be the same as the number of channels of the input combined feature tensor, i.e., 192 channels, to maintain dimensionality consistency in subsequent calculations. In some embodiments, the feature map output by one convolutional layer after processing the combined feature tensor is defined as the query feature map, which is used to represent the feature patterns for which information needs to be obtained at each position. The feature map output by the other convolutional layer after processing the combined feature tensor is defined as the key feature map, which is used to represent the feature patterns for which information can be provided at each position. It can be understood that the query feature map and the key feature map have exactly the same spatial size and number of feature channels to facilitate pointwise similarity matching calculation. Dimensionality consistency between the two is a necessary condition for calculating dot product similarity. Optionally, the calculation of the spatial attention weight map can be simplified to only calculating the attention of each position itself, that is, mainly focusing on the dependencies between different channels within the feature map. In this case, the calculation scope of the attention weight is simplified from all position pairs to within a single position.

[0034] In one embodiment of the present invention, all historical prestressed concrete pipe pile scheme records are read from an engineering case database. Each record contains a lithological description sequence of its construction site and a corresponding soil mechanical parameter sequence. The lithological description sequence is digitally encoded, converting the lithology category into a multidimensional unique vector. Simultaneously, the soil mechanical parameter sequence is standardized to eliminate dimensional differences. The processed lithological vector sequence and mechanical parameter sequence of each historical prestressed concrete pipe pile scheme are connected in depth to form a high-dimensional geological feature representation of the historical prestressed concrete pipe pile scheme. A density-based clustering algorithm is used to find densely distributed regions of samples in the space constituted by the high-dimensional geological feature representation. All historical prestressed concrete pipe pile schemes located in the same dense region are grouped into a set, and each set constitutes a pipe pile scheme cluster. A unique cluster identifier is assigned to each pipe pile scheme cluster.

[0035] To optimize the density-based clustering algorithm, a neighborhood search radius parameter and a minimum neighborhood sample number parameter are set. An unvisited historical prestressed concrete pipe pile scheme sample is randomly selected as the starting point in the high-dimensional geological feature representation space. All other sample points are searched within a hypersphere defined by the neighborhood search radius parameter, centered on the starting point. If the number of found sample points is greater than or equal to the minimum neighborhood sample number parameter, the starting point is designated as the core point, and its neighborhood is recursively expanded based on this sample point, marking all sample points reachable by density into the same cluster. If the number of found sample points is less than the minimum neighborhood sample number parameter, the starting point is temporarily marked as a noise point. This process of selecting the starting point, searching the neighborhood, and marking is repeated until all historical prestressed concrete pipe pile scheme samples have been visited, ultimately completing the partitioning of multiple pipe pile scheme clusters.

[0036] In practice, all historical prestressed concrete pipe pile scheme records are retrieved from the engineering case library. For example, an engineering case library containing 300 completed project schemes contains a sequence of strata lithology descriptions and corresponding soil mechanical parameter sequences for the construction site. A typical record contains 10 soil layers at different depths, each with a lithology description of "silty clay" or "medium-coarse sand", as well as corresponding mechanical parameters such as compression modulus, cohesion, and internal friction angle. The stratigraphic lithology description sequence is digitally encoded, and the lithology category is transformed into a multidimensional unique vector. For example, the eight different lithologies appearing in the site are mapped to an 8-dimensional vector, where "silty clay" corresponds to the vector [1,0,0,0,0,0,0,0] and "medium-coarse sand" corresponds to the vector [0,1,0,0,0,0,0,0]. At the same time, the soil mechanical parameter sequence is standardized to eliminate dimensional differences. For example, all compression modulus values ​​are transformed into a distribution with a mean of 0 and a standard deviation of 1 by subtracting the mean and then dividing by the standard deviation.

[0037] In practical implementation, the processed lithology vector sequence and mechanical parameter sequence of each historical prestressed concrete pipe pile scheme are connected in depth to form a high-dimensional geological feature representation of the historical prestressed concrete pipe pile scheme. For a scheme containing 10 soil layers, the feature of each soil layer is formed by connecting an 8-dimensional lithology unique thermal vector and a vector containing 3 standardized mechanical parameters, resulting in an 11-dimensional soil layer feature vector. The entire sequence of 10 layers constitutes a 110-dimensional feature vector as the high-dimensional geological feature representation of the scheme. A density-based clustering algorithm is used to find densely distributed regions of samples in the space formed by the high-dimensional geological feature representation. All historical prestressed concrete pipe pile schemes located in the same dense region are grouped into a set, and each set constitutes a pipe pile scheme cluster. Each pipe pile scheme cluster is assigned a unique cluster identifier, such as "Cluster_01" and "Cluster_02".

[0038] In practical implementation, a neighborhood search radius parameter and a minimum number of neighboring samples parameter are set for the density-based clustering algorithm. For example, the neighborhood search radius parameter Eps is set to 0.35, and the minimum number of neighboring samples parameter MinPts is set to 5. An unvisited historical prestressed concrete pipe pile scheme sample is randomly selected as the starting point in the high-dimensional geological feature representation space, for example, a 110-dimensional feature vector P1. All other sample points are searched within the hypersphere defined by the neighborhood search radius parameter, centered on the starting point. The Euclidean distance between the starting point P1 and all other sample points in the engineering case library in the high-dimensional space is calculated using the following formula:

[0039] in: This represents the distance between sample x and sample y. This represents the z-th eigenvalue of sample x. This represents the z-th eigenvalue of sample y. It is the total dimension of the feature vectors, counting all that satisfy... Conditional sample point set If the number of found sample points is greater than or equal to the minimum neighborhood sample number parameter, the starting point is determined as the core point, and its neighborhood is recursively expanded based on the sample points. All sample points reachable through density are marked as the same cluster, for example... If there are 6 points {P2, P3, P4, P5, P6}, and this number is greater than MinPts=5, then P1 is marked as the core point and a new cluster C1 is created. Next, the neighborhoods of points P2, P3, etc., are checked, and points with reachable density are added to cluster C1. If the number of found sample points is less than the minimum neighborhood sample size parameter, the starting point is temporarily marked as a noise point, for example, another sample point Q1. If there are only 3 points (less than 5), Q1 is marked as a noise point. However, the noise point is later reclassified because it is included in the neighborhood of other core points. The process of selecting the starting point, finding the neighborhood, and marking is repeated until all historical prestressed concrete pipe pile scheme samples have been visited, thus completing the division of multiple pipe pile scheme clusters. For example, 300 samples are eventually divided into 5 main pipe pile scheme clusters and a few noise points.

[0040] In some embodiments, the standardization of soil mechanical parameter sequences employs a global statistical approach based on the case library. This involves calculating the overall mean and standard deviation of a specific mechanical parameter across all soil layers for all schemes in the engineering case library, and then uniformly standardizing all data. In some embodiments, besides using one-heat coding, the digital encoding of lithological description sequences employs ordered numerical coding for lithologies exhibiting clear hierarchical or progressive engineering properties to preserve their sequential information in the feature space. It is understood that the value of the neighborhood search radius parameter Eps directly affects the clustering results. An excessively large Eps value can lead to multiple geological patterns that should be separated being merged into a single cluster, while an excessively small Eps value can cause a continuous geological pattern to be split into multiple small clusters and generate too many noise points. It is also understood that the setting of the minimum neighborhood sample number parameter MinPts is related to the overall size of the engineering case library. For a large case library containing thousands of records, MinPts can be increased accordingly to form more statistically significant clusters, while for a smaller case library, a lower MinPts value needs to be set to avoid failing to form effective clusters.

[0041] Optionally, when calculating high-dimensional spatial distances, different weights can be assigned to different feature dimensions. For example, mechanical parameters are considered more discriminative than lithology codes, thus assigning a higher weight coefficient to the mechanical parameter dimension in the distance calculation formula. Optionally, after clustering, each formed cluster of pipe pile schemes can be manually reviewed and labeled. For example, clusters mainly containing schemes in soft soil layers can be labeled as "soft soil zone pile foundation model," and clusters mainly containing schemes in gravel layers can be labeled as "hard soil zone pile foundation model." These labels can be used for semantic interpretation in subsequent scheme recommendations.

[0042] In one embodiment of the invention, a standard geological data field list is established, specifying the required data items and their formats. Each data record in the received raw exploration information set is compared item by item with the standard geological data field list to identify missing data fields, incorrectly formatted data fields, and data fields whose values ​​significantly exceed reasonable ranges, generating a data anomaly report. For missing data fields, Kriging interpolation is used based on their spatial location, employing similar data from surrounding valid exploration points for spatial interpolation to fill in the gaps. For incorrectly formatted data fields, they are converted to the standard format according to a predefined data conversion rule library. For data fields whose values ​​significantly exceed reasonable ranges, they are marked as pending review data, triggering a manual review process. After manual review, the data is either replaced with corrected values ​​or removed according to technical specifications. After all verification and processing steps are completed, a standardized multi-source geological dataset is output.

[0043] In practice, a standard geological data field list is established. This list specifies the required data items and their formats. For example, the list stipulates that each borehole record must include "Breakhole Number," "X-coordinate," "Y-coordinate," "Borehead Elevation," "Sequence," "Bottom Depth," "Soil Name," "Natural Unit Weight," "Cohesion c," and "Internal Friction Angle φ (degrees)," with the "Internal Friction Angle φ" field being a floating-point number. Each data record in the received raw exploration information set is compared item by item with the standard geological data field list to identify missing data fields, incorrectly formatted data fields, and data fields with values ​​significantly exceeding reasonable ranges, generating a data anomaly report.

[0044] A specific example scenario involves a site exploration project comprising 15 boreholes. The original exploration information came from different exploration units, resulting in differences in data format and completeness. Refer to Table 1 for data comparison; Table 1 illustrates some typical data anomalies.

[0045] Table 1: Anomaly Identification Table for Original Survey Data

[0046] In practical implementation, for missing data fields, the Kriging interpolation algorithm is used based on their spatial location to fill in the gaps using similar data from surrounding valid exploration points. For example, if the cohesion data for the second soil layer in borehole BK03 is missing, the system will extract the cohesion data and three-dimensional coordinates of the second soil layer from other complete boreholes in the site, and estimate the cohesion value of the soil layer in BK03 using the Kriging interpolation model. For data fields with incorrect formats, they are converted to the standard format according to a predefined data conversion rule library. For example, the original value of the "Natural Specific Gravity" field in borehole BK11 is the string "18.5". The predefined data conversion rule library contains conversion rules for Chinese numeric expressions. This rule recognizes the "dot" in the string as a decimal separator and performs the conversion. The conversion process can be performed using the following formula:

[0047] in: This represents the converted floating-point value. Represents the original string value. This function parses a string and extracts its numerical values. After applying this rule, the field value is converted to the numerical value 18.5. Data fields with values ​​significantly exceeding a reasonable range are marked as pending review, triggering a manual review process. For example, in borehole BK05, the original value of the "cohesion c" field is -5 kPa. Since cohesion cannot be negative physically, the system will mark it as an anomaly based on a predefined reasonable range, generate a pending review task, and notify the relevant engineer. After reviewing the original record, the engineer finds an input error and enters a corrected value of 10 kPa to replace the anomaly. After all verification and processing steps are completed, a standardized multi-source geological dataset is output. This standardized multi-source geological dataset is a standardized dataset where all fields are complete, formatted uniformly, and all numerical values ​​have passed validity checks.

[0048] In some embodiments, the field definitions of the standard geological data field list can be extended to in-situ test data items such as dynamic drilling blow counts and wave velocity test values ​​to accommodate more complex engineering analysis needs. In some embodiments, if a certain parameter of a soil layer is missing in all boreholes due to site conditions, the system will use regional empirical values ​​or statistical averages of similar strata to globally fill the missing parameter and add a data source marker. It is understood that malformed data fields include not only mixed numerical and character data, but also inconsistent units, incorrect delimiters, and other forms. The predefined data conversion rule base needs to cover these common patterns. It is understood that the criteria for determining the reasonable range of numerical values ​​are based not only on physical common sense but also closely integrated with geotechnical engineering investigation specifications. For example, for "internal friction angle φ", the system sets different reasonable upper and lower limits according to soil type; the internal friction angle range for silt is set to 10 degrees to 30 degrees. Optionally, the predefined data conversion rule base supports regular expression matching and replacement, which can efficiently handle non-standard data formats containing various variations.

[0049] See Figure 4 This is a bar chart comparing the quality indicators of geological data before and after standardization. All indicators showed a significant improvement in compliance rates after standardization, with most approaching 100%. Before standardization, compliance rates were concentrated in the 60%-75% range, indicating significant data quality deficiencies. After standardization, the data's completeness, consistency, and validity all reached high levels, demonstrating the value of the data preprocessing process. This type of chart is typically used in data quality assessment reports for geotechnical engineering and geological exploration projects, visually showcasing the effects of data preprocessing and providing a reliable data foundation for subsequent BIM piling and geological analysis.

[0050] In one embodiment of the present invention, a predefined component family library of prestressed pipe piles is established within the BIM model environment. This library contains various pipe pile family types with different diameters, wall thicknesses, concrete strength grades, and prestressed steel reinforcement configurations. The comprehensive geological features of the site are input into a parametric rule-based reasoning engine. This engine contains a pre-defined set of mapping rules from geological features to pipe pile bearing capacity and settlement characteristics. Based on the comprehensive geological features and structural design load requirements, the parametric rule-based reasoning engine infers the required single-pile bearing capacity characteristic value and estimated settlement for each proposed pile location. Based on the single-pile bearing capacity characteristic value and estimated settlement, the engine automatically matches suitable pipe pile family types from the component family library and sorts various feasible combinations of pipe pile specifications and lengths based on preset weights for economic efficiency and ease of construction. The sorting results generate various different pipe pile layout plans and cross-sectional views. Each layout scheme details the specifications, lengths, quantities, pile top elevations, and spacing of the pipe piles used, forming a set of optimized pile configuration schemes.

[0051] The parametric rule-based inference engine loads a comprehensive geological feature representation of the site, which includes characteristic values ​​of soil compression modulus, internal friction angle, cohesion, and standard penetration test blow count at each proposed pile location. Simultaneously, the engine reads structural design load requirements from an external interface, including vertical axial force, horizontal force, and bending moment. Using the built-in pile bearing capacity calculation formula, the engine iteratively calculates the pile end resistance and pile side resistance required to meet the bearing capacity requirements, and then back-calculates the required single pile bearing capacity characteristic value. The engine also uses the built-in layered summation method settlement calculation formula, leveraging the soil compression modulus characteristic value and the estimated additional pile end stress, to calculate the pile foundation settlement under different load levels, obtaining the estimated settlement. Finally, the calculated single pile bearing capacity characteristic value and estimated settlement are compared with the allowable values ​​in the design specifications to ensure compliance with safety requirements.

[0052] In practical implementation, a predefined component family library for prestressed pipe piles is established within the BIM model environment. This library contains various pipe pile family types with different diameters, wall thicknesses, concrete strength grades, and prestressed steel reinforcement configurations. For example, pipe pile diameters can include 400 mm, 500 mm, and 600 mm; wall thicknesses can be 95 mm, 110 mm, and 130 mm; concrete strength grades can be C60 or C80; and prestressed steel reinforcement configurations can be PHC-AB-400-95-10 and PHC-AB-500-110-12. The comprehensive geological features of the site are input into a parametric rule inference engine, which contains a pre-defined set of mapping rules from geological features to pipe pile bearing capacity and settlement characteristics.

[0053] In practical implementation, the parametric rule inference engine, based on the comprehensive geological characteristics of the site and the structural design load requirements, infers the required single pile bearing capacity characteristic value and estimated settlement for each proposed pile location. For a proposed pile location located at the center of the site, the comprehensive geological characteristics of the site give the following values ​​for the compression modulus Es of the soil layers from top to bottom: 6.5 MPa, 8.2 MPa, 12.0 MPa, and 25.0 MPa; the internal friction angle φ: 15 degrees, 18 degrees, 22 degrees, and 35 degrees; the cohesion c: 25 kPa, 28 kPa, 5 kPa, and 0 kPa; and the standard penetration test blows N: 8 blows, 11 blows, 20 blows, and 45 blows. Simultaneously, the parametric rule inference engine reads the structural design load requirements from the external interface. These requirements include a vertical axial force Fk of 1200 kN, a horizontal force Hk of 80 kN, and a bending moment Mk of 150 kN·m. The parametric rule inference engine calls the built-in pile foundation bearing capacity calculation formula, taking the soil parameters and structural load at the proposed pile location as input variables. It iteratively calculates the pile end resistance and pile side resistance required to meet the bearing capacity requirements, and then back-calculates the required single pile bearing capacity characteristic value. The calculation process follows the standard formula:

[0054] in: This represents the standard value of the vertical ultimate bearing capacity of a single pile. This represents the standard value of the total limiting side resistance. This represents the standard value of the total limiting end resistance. Indicates the circumference of the pile. This represents the standard value of the ultimate lateral resistance of the i-th layer of soil along the pile. This represents the thickness of the i-th soil layer through which the pile penetrates. This represents the standard value of the extreme end resistance. This represents the pile tip area. The parametric rule inference engine calls the built-in layered summation method settlement calculation formula, using the characteristic value of soil layer compression modulus and the estimated additional stress at the pile tip to calculate the settlement of the pile foundation under different load levels, obtaining the estimated settlement. The calculated characteristic value of single pile bearing capacity and the estimated settlement are compared and verified with the allowable values ​​in the design code to ensure that they meet safety requirements. For example, the calculated characteristic value of single pile vertical bearing capacity Ra must be greater than or equal to 800 kN, and the calculated value of the final settlement at the center point of the pile foundation must be less than or equal to 50 mm.

[0055] Based on the characteristic value of single pile bearing capacity and estimated settlement, the system automatically matches suitable pipe pile family types from the component family library. It then sorts various feasible combinations of pipe pile specifications and lengths based on preset weights for economy and ease of construction. For example, for the aforementioned proposed pile location, the system calculates that Ra ≥ 800 kN and settlement must be controlled. By traversing the component family library, it matches three pipe pile types: PHC-AB-400-95-12, PHC-AB-500-110-10, and PHC-AB-600-130-10, all of which meet the bearing capacity and settlement requirements. The system then performs a comprehensive scoring and sorting of these three schemes according to preset weights. Based on the sorting results, various different pipe pile layout plans and cross-sectional views are generated. Each layout plan details the specifications, length, quantity, pile top elevation, and layout spacing of the pipe piles used, forming a set of optimized pile arrangement schemes. For example, Scheme 1 recommends using PHC-AB-500-110-10 pipe piles throughout the site, with a pile length of 26 meters, a total of 150 piles, and a pile spacing of 1.8 meters. Scheme 2 recommends using PHC-AB-600-130-10 pipe piles with a pile length of 24 meters in areas with higher loads, and using PHC-AB-500-110-10 pipe piles with a pile length of 26 meters in other areas, in a mixed layout.

[0056] In some embodiments, the pre-set mapping rule set in the parametric rule inference engine can be dynamically switched according to local regulations or enterprise standards. For example, a subset of rules including negative skin friction calculation can be enabled in coastal soft soil areas, while a subset of rules for rock-socketed pile calculation can be enabled in rock areas. In some embodiments, the pipe pile family type in the component family library is associated with dynamic attributes such as market unit price, local inventory, and delivery cycle, in addition to static parameters, making the economic evaluation in scheme ranking more real-time and accurate. It can be understood that the calculation of the characteristic value of single pile bearing capacity is an iterative process. The parametric rule inference engine will first assume a pile diameter and pile length, calculate its bearing capacity and settlement, and if the requirements are not met, automatically adjust the pile diameter or pile length parameters and recalculate until the minimum specification or optimal specification combination that meets all constraints is found. Optionally, in addition to text and drawing lists, the generated pile layout optimization scheme can also directly drive BIM software to generate a three-dimensional pile foundation layout model. Each pipe pile in the model carries complete specification attributes and can be correlated and checked with the superstructure model. In practice, the generated pile layout optimization scheme directly drives the BIM software through automated scripts or plugins integrated into the BIM model environment, thereby dynamically generating a three-dimensional pile foundation layout model. Based on the specifications, lengths, and layout parameters of the pipe piles determined in the pile layout scheme, the model automatically instantiates the corresponding pipe pile family type from the predefined component family library and assigns a complete set of attributes to each pipe pile component, including diameter, wall thickness, concrete strength grade, and prestressed steel reinforcement configuration specifications. At the same time, using the model coordination and conflict detection functions built into the BIM platform, the pile foundation layout model is correlated with the superstructure model, automatically identifying and highlighting spatial location conflicts or design mismatches, ensuring seamless connection and overall coordination between the pile foundation design and the superstructure in three-dimensional space.

[0057] See Figure 5 This is an efficiency analysis chart of the various processing stages of a BIM-based prestressed concrete pipe pile configuration optimization system. The improvement rate of data integrity and consistency continuously increases with the progress of the processing stages, eventually approaching 98% in the "BIM Pile Configuration Generation" stage; the improvement rate is lower in the early stages, while the quality optimization effect is more significant in the later stages. Processing time increases linearly with the progress of the stages, with the "BIM Pile Configuration Generation" stage taking the longest (approximately 360 seconds); the increase in time is positively correlated with the quality improvement rate, reflecting the higher complexity of data processing in the later stages. This type of chart is used for performance evaluation of engineering systems, reflecting both the data quality optimization effect of each stage and revealing the changing patterns of processing efficiency. It can help technicians to specifically optimize the algorithm efficiency of later stages (such as BIM Pile Configuration Generation) and balance the relationship between quality and time consumption.

[0058] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A BIM-based prestressed concrete pipe pile configuration optimization system based on multi-source geological data, characterized in that, The system includes: The data preprocessing module receives the original exploration information set of the target site from the geological data source, performs data integrity and consistency verification, and generates a standardized multi-source geological dataset. The feature extraction module calls the stored engineering case library, performs clustering on historical prestressed pipe pile schemes based on lithology and mechanical properties, forms multiple pipe pile scheme clusters with different geological backgrounds, extracts the core geological model of each pipe pile scheme cluster and calculates its spatial influence radius, and outputs the geological model feature vector and feature influence domain. The feature perception module constructs a multi-level feature perception architecture based on the dimensions of the feature vectors of the multiple geological models. The standardized multi-source geological dataset is fed into the multi-level feature perception architecture. At each level, the dataset is spatially divided and feature-encoded according to the corresponding feature influence domain, and the hierarchical geological feature code is output. The feature fusion module, through its built-in feature cross-fusion unit, performs bidirectional attention weighting and splicing on the hierarchical geological feature codes from different levels to generate a fused comprehensive geological feature representation of the site. The BIM pile generation module maps the comprehensive geological features of the site to the BIM model environment, drives the pile components in the model, and generates a set of optimized pile configuration schemes based on the specifications, length and layout parameters of the prestressed pipe piles.

2. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 1, characterized in that, Extract the core geological model for each pipe pile scheme family and calculate its spatial influence radius, including: For each of the aforementioned pipe pile schemes, a geological sequence sample set is constructed with pile location coordinates and stratum parameter sequences as input; Initialize a deep embedding network, which includes a stacking module for sequence modeling and a pooling module for feature condensation; The geological sequence sample set is input into the deep embedding network, and the stacking module captures the long-range dependencies between stratigraphic parameters in the sample, outputting high-dimensional sequence features. The high-dimensional sequence features are input into the pooling module, and through adaptive weighted convergence operation, a dense vector representing the common geological laws of the pipe pile scheme cluster is extracted as the core geological model. After extracting the core geological model, for each geological sequence sample in the pipe pile scheme cluster, the feature space distance between its original sequence features and the core geological model is calculated. The distribution of the feature spatial distances of all samples is statistically analyzed, and the values ​​corresponding to the preset quantiles of the distance distribution are defined as the feature influence domain of the core geological model.

3. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 2, characterized in that, The standardized multi-source geological dataset is fed into the multi-level feature-aware architecture. At each level, the dataset is spatially divided and feature-encoded according to the corresponding feature influence domain, including: The perception range of each level of the multi-level feature perception architecture is matched with a feature influence domain. At a specific level of the multi-level feature perception architecture, the value of the feature influence domain associated with the specific level is obtained, and the value is used as the spatial neighborhood radius. Taking each exploration point or grid point in the target site as the center and the radius of the spatial neighborhood as the side length, a cubic spatial region is defined in the standardized multi-source geological dataset, and all geological data points in the cubic spatial region are taken as the neighborhood data block of the current center point. All data points in the neighboring data block are organized into a three-dimensional tensor with a channel structure according to their spatial location and geological attribute type. A three-dimensional convolution kernel group is used to perform convolution operations on the three-dimensional tensor to extract local geological structure features within the spatial neighborhood of the neighborhood data block; After processing all exploration points or grid points, all extracted local geological structural features are rearranged according to the spatial position of their center points to form the hierarchical geological feature code of the specific level. The hierarchical geological feature code is a feature map corresponding to the site spatial coordinates.

4. The BIM-based prestressed pipe pile configuration optimization system based on multi-source geological data as described in claim 3, characterized in that, The system utilizes a built-in feature cross-fusion unit to perform bidirectional attention weighting and concatenation on the hierarchical geological feature codes from different levels, including: The feature cross-fusion unit receives hierarchical geological feature codes from at least two different levels of the multi-level feature perception architecture; The hierarchical geological feature codes from the coarser sensing level, i.e., those with a larger feature influence domain, are upsampled to make their spatial resolution consistent with that of the hierarchical geological feature codes from the finer sensing level. The coarser-level feature encoding and the finer-level feature encoding after upsampling are concatenated along the feature channel dimension to form a combined feature tensor. The combined feature tensor is input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights; Calculate the dot product similarity between the query feature map and the key feature map at each spatial location, and process it through a normalization function to generate a spatial attention weight map; The combined feature tensor is reweighted in a channel-like manner using the spatial attention weight map to highlight the feature responses of important spatial locations. The reweighted feature tensor is input into a fusion convolutional layer for dimensionality reduction and fusion, and finally the fused comprehensive geological feature expression of the site is output.

5. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 4, characterized in that, The combined feature tensor is input into two independent convolutional layers to generate query feature maps and key feature maps for calculating attention weights, including: The two independent convolutional layers have the same kernel size but each has its own independent trainable weight parameters; After one of the convolutional layers processes the combined feature tensor, the output feature map is defined as the query feature map, which is used to characterize the feature pattern for which information needs to be obtained at each position. After another convolutional layer processes the combined feature tensor, the output feature map is defined as the key feature map, which is used to characterize the feature patterns that can provide information at each location; The query feature map and the key feature map have the exact same spatial size and number of feature channels to facilitate point-by-point similarity matching calculations.

6. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 1, characterized in that, The system calls upon a stored engineering case library and performs clustering on historical prestressed concrete pipe pile schemes based on lithology and mechanical properties, including: All historical prestressed concrete pipe pile scheme records are retrieved from the engineering case library. Each record contains a sequence of strata lithology descriptions and corresponding soil mechanical parameter sequences for its construction site. The lithological description sequence of the strata is digitally encoded, and the lithological categories are transformed into multidimensional unique heat vectors. At the same time, the sequence of soil mechanical parameters is standardized to eliminate dimensional differences. The processed lithology vector sequence and mechanical parameter sequence of each historical prestressed pipe pile scheme are connected in depth to form a high-dimensional geological feature representation of the historical prestressed pipe pile scheme. A density-based clustering algorithm is used to find regions with dense sample distribution in the space formed by the high-dimensional geological feature representation; All historical prestressed pipe pile schemes located in the same dense area are grouped into a set, each set constituting a pipe pile scheme cluster, and each pipe pile scheme cluster is assigned a unique cluster identifier.

7. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 6, characterized in that, A density-based clustering algorithm is used to find densely distributed regions of samples within the space defined by the high-dimensional geological feature representation, including: Set the neighborhood search radius parameter and the minimum number of neighborhood samples parameter for the density-based clustering algorithm; In the high-dimensional geological feature representation space, a historical prestressed pipe pile scheme sample that has not been accessed is randomly selected as the starting point; Using the starting point as the center, search for all other sample points within the hypersphere defined by the neighborhood search radius parameter; If the number of sample points found is greater than or equal to the minimum neighborhood sample number parameter, then the starting point is determined to be the core point, and its neighborhood is recursively expanded based on the sample points, marking all sample points that can be reached through density as the same cluster. If the number of sample points found is less than the minimum neighborhood sample number parameter, the starting point is temporarily marked as a noise point. The process of selecting a starting point, searching for neighbors, and marking is repeated until all historical prestressed concrete pipe pile scheme samples have been visited, ultimately completing the division of multiple pipe pile scheme clusters.

8. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 1, characterized in that, Receive the original exploration information set of the target site from the geological data source, and perform data integrity and consistency checks, including: Establish a standard geological data field list, which specifies the data items that must be included and their format; Each data record in the received set of original exploration information is compared item by item with the list of standard geological data fields to identify missing data fields, data fields with incorrect formats, and data fields whose values ​​are significantly outside the reasonable range, and a data anomaly report is generated. For missing data fields, based on their spatial location, the Kriging interpolation algorithm is used to fill in the missing data by using similar data from surrounding valid exploration points. For data fields with incorrect formats, convert them to the standard format according to the predefined data conversion rule library; For data fields whose values ​​clearly exceed the reasonable range, mark them as data to be reviewed and trigger the manual review process. After manual review, they will either be replaced with corrected values ​​or removed according to technical specifications. After all verification and processing steps are completed, the standardized multi-source geological dataset is output.

9. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 1, characterized in that, The comprehensive geological features of the site are mapped to the BIM model environment, driving the pile configuration components in the model to generate a set of optimized pile configuration schemes based on the specifications, length, and layout parameters of the prestressed concrete pipe piles, including: In the BIM model environment, a predefined component family library for prestressed pipe piles is provided, which contains various pipe pile family types with different diameters, wall thicknesses, concrete strength grades, and prestressed steel reinforcement configurations. The comprehensive geological features of the site are input into a parameterized rule inference engine, which has a pre-set set of mapping rules from geological features to pile bearing capacity and settlement features. The parametric rule reasoning engine, based on the comprehensive geological characteristics of the site and combined with the structural design load requirements, infers the required single pile bearing capacity characteristic value and estimated settlement at each proposed pile location. Based on the single pile bearing capacity characteristic value and the estimated settlement, the system automatically matches the pipe pile family type that meets the conditions in the component family library, and sorts the various feasible pipe pile specifications and length combinations based on the preset economic and construction convenience weights. Based on the sorting results, a variety of different pipe pile layout plans and cross-sectional views are generated. Each layout plan details the specifications, length, quantity, pile top elevation, and layout spacing of the pipe piles used, forming a set of optimized pile arrangement schemes.

10. The BIM-based prestressed pipe pile allocation optimization system based on multi-source geological data as described in claim 9, characterized in that, The parametric rule inference engine, based on the comprehensive geological characteristics of the site and combined with structural design load requirements, infers the required single pile bearing capacity characteristic value and estimated settlement at each proposed pile location, including: The parameterized rule reasoning engine loads the comprehensive geological feature expression of the site, which includes the characteristic values ​​of soil compression modulus, internal friction angle, cohesion, and standard penetration test blow count at each simulated pile location. Meanwhile, the parametric rule inference engine reads the structural design load requirements from an external interface, which include vertical axial force, horizontal force, and bending moment. The parameterized rule reasoning engine calls the built-in pile foundation bearing capacity calculation formula, takes the soil layer parameters and structural load at the proposed pile location as input variables, iteratively calculates the pile end resistance and pile side resistance required to meet the bearing capacity requirements, and then back-calculates the required single pile bearing capacity characteristic value. The parameterized rule inference engine calls the built-in layered summation method settlement calculation formula, uses the characteristic value of the soil layer compression modulus and the estimated additional stress at the pile end to calculate the settlement of the pile foundation under different load levels, and obtains the estimated settlement. The calculated characteristic value of the single pile bearing capacity and the estimated settlement are compared and verified with the allowable value of the design code to ensure that they meet the safety requirements.

Images