Intelligent design method and system for roadbed and foundation treatment, medium and program product

Abstract

The invention relates to the field of intelligent design of road foundations, in particular to an intelligent design method and system for roadbed and foundation treatment, a medium and a program product. The method comprises the following steps: S1, obtaining a historical engineering case of roadbed and foundation treatment; s2, fusing geological parameters into the engineering design drawing in a color fusion mode; s3, generating an engineering case sample on the engineering design drawing fused with the geological parameters; s4, constructing a data set by using the engineering case samples, and training to obtain a foundation processing intelligent design model; s5, obtaining a to-be-designed engineering image; and processing the to-be-designed engineering image by using the foundation processing intelligent design model to obtain a recommendation scheme of the to-be-designed foundation processing scheme. According to the method, the foundation treatment scheme is represented in the form of image data, spatial information such as terrain and geology is comprehensively considered, training is performed based on the convolutional neural network to obtain the foundation treatment intelligent design model for intelligent design of the foundation treatment scheme, and the design efficiency of the foundation treatment scheme is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent design of road foundation, in particular to a road foundation treatment intelligent design method, system, medium and program product. BACKGROUND

[0002] In modern transportation infrastructure construction, roadbed engineering, as the basic structure supporting the upper structure of track, road, etc., is the key link to ensure the long-term stability, durability and operation safety of the entire transportation network. Its engineering content covers foundation treatment, retaining structure, slope protection and other interrelated design and construction subsystems, among which the design of foundation treatment scheme is the most complex and crucial. The scientificity and rationality of the scheme are directly related to the roadbed settlement control, bearing capacity guarantee and overall structure service performance. High-quality foundation treatment scheme design highly depends on the fine and quantitative characterization of the topographic relief characteristics, stratum spatial distribution law and continuous variation characteristics of rock-soil physical and mechanical parameters in the engineering area. When carrying out such design work, the traditional technical means mainly rely on the engineering experience of designers based on limited survey point data for manual judgment and analogy deduction, which not only requires high professional accomplishment and experience accumulation of technical personnel, but also is difficult to fully and accurately reflect the spatial variability of underground medium when facing complex geological conditions or large-scale engineering areas, resulting in certain subjectivity and uncertainty in the design scheme. In addition, this process usually needs repeated adjustment and checking, consuming a large amount of manpower, time and computing resources, with long design cycle and low efficiency, which is difficult to meet the current demand of large-scale, high-standard and rapid transportation infrastructure construction. SUMMARY

[0003] The purpose of the present application is to overcome the deficiencies of long design cycle and low efficiency in the prior art in the design of foundation treatment scheme, and to provide a road foundation treatment intelligent design method, system, medium and program product.

[0004] Firstly, the present application provides a road foundation treatment intelligent design method, comprising the following steps: S1, obtaining a historical engineering case of road foundation treatment; the historical engineering case at least includes an engineering design drawing; the engineering design drawing contains a foundation treatment scheme, topography, stratum spatial distribution information and geological parameters; S2, fusing the geological parameters into the engineering design drawing in a color fusion manner; the geological parameters at least include specific weight, bearing capacity and compression modulus; S3, on the engineering design drawing with fused geological parameters, combining the position of the roadbed surface, extracting engineering images according to a preset specification, and setting labels for the extracted engineering images to generate engineering case samples; wherein the labels include engineering type and design parameters; S4, constructing a dataset by using the engineering case sample, training a convolutional neural network, and obtaining an intelligent design model for foundation treatment; S5, fusing geological parameters to a roadbed engineering design drawing of a to-be-designed foundation treatment scheme, extracting image data of a preset specification at a treatment location to generate a to-be-designed engineering image, inputting the to-be-designed engineering image into the intelligent design model for foundation treatment to process the to-be-designed engineering image, and outputting a recommended scheme of the to-be-designed foundation treatment scheme by the intelligent design model for foundation treatment.

[0005] The intelligent design model for foundation treatment extracts a feature vector of an input image by taking CNN as a backbone network, and connects double-branch fully connected layers to be respectively used for engineering type classification prediction and design parameter regression prediction of the foundation treatment scheme. The intelligent design model for foundation treatment processes the input to-be-designed engineering image to output an engineering type and design parameters corresponding to the to-be-designed engineering image.

[0006] According to a preferred embodiment, the geological parameters are fused into the engineering design drawing by using color fusion in S2.

[0007] According to a preferred embodiment, a rectangular region is set in the engineering case sample to identify a roadbed location.

[0008] According to a preferred embodiment, the intelligent design model for foundation treatment includes a feature extractor and two parallel fully connected layers. The feature extractor is used to extract a feature vector of the to-be-designed engineering image. The feature extractor includes a convolutional layer, a max-pooling layer, a residual module, and a global average pooling layer. The convolutional layer is used to perform primary feature extraction on the input to-be-designed engineering image to capture local features by using a convolution kernel. The max-pooling layer is used to compress the spatial dimension of a feature map to enhance the position invariance of the features. The residual module is used to realize deep feature extraction by using a skip connection to avoid gradient vanishing. The residual module includes a plurality of residual blocks. The residual blocks are used to fuse an identity mapping and a convolution path to optimize the expression ability of deep features. The global average pooling layer is used to compress spatial features into a channel feature vector to retain key semantic information. Two fully connected layers are respectively connected to the output of the global average pooling layer. One of the fully connected layers is used to process the feature vector to output an engineering type prediction result of the to-be-designed engineering image. The other fully connected layer is used to process the same feature vector to output a design parameter prediction result of the to-be-designed engineering image.

[0009] According to a preferred embodiment, the foundation treatment intelligent design model is configured with a vector database. A feature extractor of the foundation treatment intelligent design model extracts features of the engineering case samples, obtains a feature vector of each sample, stores the feature vector in the vector database, and establishes a corresponding relationship between the feature vector and the engineering case sample.

[0010] According to a preferred embodiment, the road foundation treatment intelligent design method further comprises: S6, constructing a dataset using the engineering case samples, training an image segmentation network, and obtaining a foundation treatment range prediction model; S7, inputting the to-be-designed engineering image into the foundation treatment range prediction model for processing, and outputting a foundation treatment range prediction result by the foundation treatment range prediction model.

[0011] The application further provides a road foundation treatment intelligent design system, comprising an input unit, a processing unit and an output unit. The input unit is used for inputting a road foundation engineering design drawing of a to-be-designed foundation treatment scheme and geological parameters. The processing unit is used for obtaining a recommended scheme according to the road foundation treatment intelligent design method provided by the application. The output unit is used for outputting the recommended scheme.

[0012] The application further provides an electronic device. The electronic device comprises at least one processor and a memory connected with the at least one processor in communication. The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor to enable the at least one processor to execute the road foundation treatment intelligent design method provided by the application.

[0013] The application further provides a computer readable storage medium. The computer readable storage medium stores computer instructions for enabling a processor to execute the road foundation treatment intelligent design method provided by the application when the processor executes the computer instructions.

[0014] The application further provides a computer program product. The computer program product comprises a computer program. The computer program is executed by a processor to implement the road foundation treatment intelligent design method provided by the application.

[0015] Compared with the prior art, the application has the following beneficial effects: The application adopts the form of image data to represent the foundation treatment scheme, comprehensively considers spatial information such as topography and geology, and obtains a foundation treatment intelligent design model based on a convolutional neural network to perform intelligent design of the foundation treatment scheme; by converting complex topography and geological features into an image form, the method can more intuitively capture various details on site, and avoid important spatial variation information lost due to simplification in the traditional method. The foundation treatment intelligent design model based on deep learning can automatically learn rules from a large number of historical cases, identify key factors affecting the foundation treatment scheme, improve the design accuracy and efficiency of the foundation treatment scheme, and can provide a more reasonable and safe foundation treatment scheme, providing stronger technical support for the safety and stability of the roadbed engineering. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The figure is a flowchart of the road foundation treatment intelligent design method of a preferred embodiment of the application.

[0017] Figure 2 The figure is a schematic diagram of an engineering design drawing containing a foundation treatment scheme, topography and spatial distribution information of strata.

[0018] Figure 3 The figure is a schematic diagram after stratum region identification of the engineering design drawing.

[0019] Figure 4 The figure is a schematic diagram of an engineering design drawing fused with geological parameters.

[0020] Figure 5 The figure is a schematic diagram of an engineering image extraction.

[0021] Figure 6 The figure is a schematic diagram of an engineering case sample.

[0022] Figure 7 The figure is a mask sample schematic diagram of a design region of a foundation treatment scheme.

[0023] Figure 8 The figure is a schematic diagram of a convolutional neural network architecture.

[0024] Figure 9 The figure is a schematic diagram of an engineering image to be designed.

[0025] Figure 10 The figure is a schematic diagram of a recommended scheme output by a foundation treatment intelligent design model.

[0026] Figure 11 The figure is a schematic diagram of a foundation treatment scheme after foundation treatment range prediction. DETAILED DESCRIPTION

[0027] The application will be described in further detail below with reference to the embodiments. However, it should be understood that the scope of the above subject matter of the application is not limited to the following embodiments, and any technology achieved based on the content of the application falls within the scope of the application.

[0028] In the description of the embodiments of the application, the terms of orientation or positional relationship such as "upper", "lower", "left", "right", "center", "inner", "outer", etc. are expressed based on the orientation or positional relationship shown in the drawings or the orientation or positional relationship in which the product / device / apparatus of the application is usually placed, unless otherwise specified. These terms of orientation or positional relationship are only used for the convenience of describing the application or simplifying the description in the embodiments to facilitate the understanding of the application by the skilled person, and are not intended to indicate or imply that a specific device / component / element must have a specific orientation or be constructed and operated in a specific positional relationship, and therefore cannot be understood as a limitation on the application.

[0029] In addition, the terms "horizontal", "vertical", "suspended", "parallel", etc. do not mean that the corresponding device / component / element must be absolutely horizontal or vertical or suspended or parallel, but can be slightly inclined or deviated. For example, "horizontal" only means that its direction is relatively more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly inclined. Alternatively, it can be simplified to mean that the corresponding device / component / element is arranged in the direction of "horizontal", "vertical", "suspended", "parallel", etc., and can have an error / deviation of ±10% with respect to the corresponding direction, more preferably an error / deviation of ±8% or less, more preferably an error / deviation of ±6% or less, more preferably an error / deviation of ±5% or less, and more preferably an error / deviation of ±4% or less. As long as the corresponding device / component / element is within the error / deviation range, it can still achieve its role in the application scheme.

[0030] In addition, the terms "first", "second", "third", etc. in the terms are only used to distinguish the same or similar components for description, and should not be understood as emphasizing or implying the relative importance of the specific components.

[0031] In addition, in the description of the embodiments of the application, "several", "a plurality of", "several" represent at least 2. It can be 2, 3, 4, 5, 6, 7, 8, 9, etc. in any case, and even more than 9.

[0032] Furthermore, in the description of the technical solutions of the present application, unless otherwise explicitly specified / limited / limited, the terms "provided", "installed", "connected", "connected", "provided", "laid", "arranged" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or it can be integrally connected, which can be welding, riveting, bolting, screwing and other commonly used connection means in the art. The connection can be mechanical connection, electrical connection or communication connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the communication between two elements.

[0033] Embodiment 1 The present embodiment provides an intelligent design method for road foundation treatment. Referring to Figure 1 , preferably, the intelligent design method for road foundation treatment comprises the following steps: S1, obtaining historical engineering cases of road foundation treatment; the historical engineering cases at least include engineering design drawings; the engineering design drawings contain: foundation treatment scheme, topography, stratum spatial distribution information and geological parameters; S2, fusing the geological parameters into the engineering design drawings in a color fusion manner; the geological parameters at least include: specific gravity, bearing capacity and compression modulus; S3, on the engineering design drawings fused with the geological parameters, combining the position of the roadbed surface, extracting engineering images according to the preset specifications, and setting labels for the extracted engineering images to generate engineering case samples; wherein the labels include engineering type and design parameters; S4, constructing a data set using the engineering case samples, training a convolutional neural network, and obtaining an intelligent design model for foundation treatment; S5, fusing the geological parameters of the roadbed engineering design drawings of the to-be-designed foundation treatment scheme, and extracting image data of a preset specification at the proposed treatment position to generate a to-be-designed engineering image; inputting the to-be-designed engineering image into the intelligent design model for foundation treatment for processing, and the intelligent design model for foundation treatment outputs the recommended scheme of the to-be-designed foundation treatment scheme; Wherein, the intelligent design model for foundation treatment extracts the feature vector of the input image with CNN as the backbone network, and then connects the double-branch fully connected layer, which is respectively used for engineering type classification prediction and design parameter regression prediction of the foundation treatment scheme; The intelligent design model for foundation treatment processes the input to-be-designed engineering image, and outputs the engineering type and design parameters corresponding to the to-be-designed engineering image. The engineering type and design parameters of different foundation treatment schemes, for example: ① CFG pile: pile spacing D, pile length L; ② cement soil mixing pile: pile spacing D, pile length L; ③ excavation and replacement: replacement depth L.

[0034] The roadbed treatment intelligent design method provided by the embodiment adopts the form of image data to represent the roadbed treatment scheme, comprehensively considers spatial information such as topography and geology, and obtains a roadbed treatment intelligent design model based on a convolutional neural network to perform intelligent design of the roadbed treatment scheme; by converting complex topographic and geological features into image form, the method can more intuitively capture various details on site, avoiding important spatial variation information lost due to simplification in traditional methods. The roadbed treatment intelligent design model based on deep learning can automatically learn rules from a large number of historical cases, identify key factors affecting the roadbed treatment scheme, improve the design accuracy and efficiency of the roadbed treatment scheme, and provide a more reasonable and safe roadbed treatment scheme, providing stronger technical support for the safety and stability of roadbed engineering.

[0035] Embodiment 2 The embodiment is a further improvement of embodiment 1, and the repeated content will not be described again.

[0036] In the design of the roadbed treatment scheme, the basic data is mainly the geological profile formed by geological exploration. Such a profile is usually composed of a plurality of line segments representing the boundaries of strata and text information labeling the names of strata. The engineering design drawing is usually formed after subsequent design work based on the geological profile. The roadbed treatment scheme is usually presented in the form of an engineering design drawing, as shown in FIG. 1. Figure 2

[0037] Preferably, step S1, after obtaining the engineering design drawing of the historical engineering case, performs structured analysis on each stratum region in the engineering design drawing to realize stratum region identification. Preferably, the method for stratum region identification of the engineering design drawing in the embodiment can be the technical means disclosed in the invention patent “Stratum Region Approximate Identification Method and System in Geological Profile” (application number: CN202311583115.3), and the geological profile region identification method therein is used to process the engineering design drawing, identify the stratum region in the engineering design drawing, extract the boundaries of each region, and form color blocks by filling different colors to realize effective differentiation of strata.

[0038] Preferably, the engineering design drawing processed by the stratum region identification is as shown in FIG. 2. Figure 3 Preferably, Figure 3 Each color block in FIG. 2 represents a stratum region, and the upper surface of the topmost color block is the ground surface, and the boundary undulation represents the topographic undulation.

[0039] Referring to FIG. 3, the roadbed treatment intelligent design method provided by the embodiment comprises the following steps: Figure 3 ​, after the processing of step S1, the spatial distribution of the terrain and strata can be directly observed from the figure. In the design of the ground treatment scheme, the influencing factors of the ground treatment scheme not only include the spatial distribution of the terrain and strata, but also involve the geological parameters of each stratum. The main related geological parameters include but are not limited to the unit weight (γ), bearing capacity (f) and compression modulus (E) of the rock-soil mass, which are crucial to ensure the safety and stability of the engineering structure.

[0040] Preferably, the geological parameter fusion in step S2 adopts a color fusion method to fuse the geological parameters with the RGB values of the image.

[0041] The color of the image is usually composed of RGB 3-channel (or AGRB, 4-channel) data, and each pixel point contains 3 data values, each of which ranges from 0 to 255. Therefore, the geological parameters related to the ground treatment scheme design, such as unit weight, bearing capacity and compression modulus, can be integrated into the RGB color space of the image to realize the visualization of the geological parameters, and thus the spatial distribution of the geological parameters can be intuitively displayed in the form of color change.

[0042] Preferably, for Figure 3 any point in the stratum in the middle, it corresponds to a unique RGB value in the image, and a unique stratum region in the actual stratum, i.e. a unique geological parameter.

[0043] Preferably, the fusion method of the geological parameters and the RGB values includes one of the following methods: ① Normalization and direct mapping: normalizing the geological parameters and then linearly converting them to the RGB values of the corresponding positions; ② Weighted summation: defining an importance weight coefficient for each geological parameter, weighting and summing multiple geological parameters according to the weight coefficients and normalizing them to the interval [0, 1] to obtain a comprehensive parameter S, and then linearly and gradually mapping the comprehensive parameter S to the RGB color space to obtain the corresponding RGB values; ③ Color mapping: using different RGB values to represent different stratum types; or using different textures to represent different stratum regions.

[0044] The fusion of the geological parameters and the RGB color can be realized by various methods, including normalization and direct mapping, weighted summation and color mapping.

[0045] Preferably, the number of geological parameters involved in the present embodiment is limited, and the method of normalization and direct mapping is adopted: first, the geological parameters are normalized to the interval [0, 1], and then linearly converted to the corresponding RGB color values.

[0046] After the fusion of the geological parameters and the RGB color values is completed, the color of the original image will be updated to reflect the new geological information, as shown in Figure 4 .

[0047] Referring to Figure 4 The original color of the engineering design drawing is replaced by the RGB value after the fusion of the geological parameters, and the image color of the engineering design drawing will also change accordingly. Referring to Figure 4 Preferably, the spatial distribution of different color bands in the image represents the spatial distribution of the actual terrain and geology in the project, and the color RGB value of each point reflects the geological parameter value of the corresponding stratum rock mass.

[0048] This visual representation method greatly improves the efficiency of understanding and analyzing complex geological conditions and provides strong support for subsequent foundation treatment design.

[0049] The intelligent design model of foundation treatment learns in depth according to the engineering case samples generated according to the historical engineering cases of foundation treatment.

[0050] The engineering case samples for training the intelligent design model of foundation treatment are obtained through step S3.

[0051] Referring to Figure 5 Preferably, based on the design result data and combined with the specific location of the subgrade surface, the engineering image is extracted within a certain size range, and the engineering type corresponding to each extracted engineering image is labeled. For example, according to the information in the design result drawing, it can be judged that the foundation treatment project adopts CFG pile composite foundation. In order to define the influence range of the foundation treatment project, the center of the subgrade surface is taken as the reference point, the influence area width is set as W / 2 (extending from the reference point to both sides), the height is set as H (extending from the reference point upwards), and the depth is set as D (extending from the reference point downwards), forming a rectangular area with a size of Wx(H+D) (the specific values can be adjusted according to engineering experience and model training effect). The stratum data in this area will be extracted and stored in image format (such as bmp), and the file name will be used as the label of this sample, containing the related engineering type and related design parameters.

[0052] In the design of subgrade foundation treatment scheme, even in the case of the same topography and geological conditions, the foundation treatment scheme requirements of different railway grades also have significant differences. The difference of railway grade determines the diversity of required foundation treatment scheme types.

[0053] Preferably, a rectangular area is set in the engineering case sample to identify the engineering location. Preferably, the engineering case sample is processed according to the pre-set importance grading rules before being used to train the intelligent design model of foundation treatment; different importance grades correspond to different importance proportion coefficients. The RGB value in the rectangular area in the engineering case sample used to identify the engineering location is multiplied by the importance proportion coefficient, and then used to train the intelligent design model of foundation treatment.

[0054] Preferably, in order to clearly identify the location of the roadbed surface, a rectangular area with an initial color RGB value of (255, 255, 255) is added above the roadbed surface, and the specific size can be optimized later according to the training effect. In addition, considering that there are differences in the requirements of different railway grades for settlement control, the corresponding proportionality coefficient can be set according to the railway grade. For example, the proportionality coefficient of ballastless high-speed railway can be set to 1.0, while the proportionality coefficient of ballast high-speed railway can be set to 0.8, and the coefficients of other types of railways are adjusted according to their respective settlement control standards. By multiplying the RGB value of the rectangular area representing the engineering location by the importance proportionality coefficient, the final RGB value of the area is obtained, thereby reflecting the sensitivity of different railway grades to engineering locations.

[0055] The engineering case sample generated by considering the roadbed surface location and railway grade factors is shown in Figure 6 This processing method not only helps to improve the quality of sample data, but also enhances the understanding ability of the foundation treatment intelligent design model for different working conditions, providing more accurate support for subsequent foundation treatment intelligent design.

[0056] Preferably, in order to further clearly define the design area range of the foundation treatment scheme, the design area is processed according to the mask mode, i.e. the design area is marked as white (RGB value is 255, 255, 255), while the remaining non-concerned areas are set to black (RGB value is 0, 0, 0). The sample image output in this way can more intuitively show the specific impact range of engineering design, as shown in Figure 7 This processing method not only helps to improve the visualization effect of the data set, but also enhances the attention of the foundation treatment intelligent design model to specific areas during training, and improves the accuracy of the final analysis result.

[0057] The classic convolutional neural network model (CNN) can effectively extract image features and complete classification tasks through convolution, pooling and activation operations.

[0058] In the context of intelligent design of road foundation treatment, this capability can be translated into the recognition of engineering scheme types, i.e. to realize the core task of "engineering scheme selection".

[0059] However, actual engineering decision-making not only needs to determine which type of foundation treatment (such as CFG pile, gravel pile, etc.) to adopt, but also needs to obtain the corresponding design parameters (such as pile diameter, pile length, pile spacing, etc.) at the same time. Therefore, only outputting category information cannot support the complete intelligent design process. Therefore, the embodiment improves the traditional CNN architecture: taking CNN as the backbone network (Backbone) to construct a feature extractor to extract a high-dimensional semantic feature vector from the input engineering sample image; on this basis, a double-branch output head is introduced - one branch is a classification head for predicting the engineering type of foundation treatment; the other branch is a regression head for predicting the key design parameters matched with the type.

[0060] Referring to Figure 8 , the intelligent design model of foundation treatment involved in step S4 includes a feature extractor and two fully connected layers arranged in parallel; The feature extractor is configured to extract a feature vector of the to-be-designed engineering image. The feature extractor includes a convolutional layer, a max-pooling layer, a residual module, and a global average pooling layer. The convolutional layer is configured to perform primary feature extraction on the input to-be-designed engineering image by using a convolution kernel to capture local features. The max-pooling layer is configured to compress the spatial dimension of the feature map and enhance the position invariance of the features. The residual module is configured to implement deep feature extraction through a skip connection to avoid gradient vanishing. The residual module includes a plurality of residual blocks. The residual block is configured to fuse the identity mapping and the convolution path to optimize the expression ability of the deep features. The global average pooling layer is configured to compress the spatial features into a channel feature vector to retain key semantic information. The two fully connected layers are respectively connected to the output of the global average pooling layer. One of the fully connected layers is configured to process the feature vector to output an engineering type prediction result of the to-be-designed engineering image. The other fully connected layer is configured to process the same feature vector to output a design parameter prediction result of the to-be-designed engineering image.

[0061] Preferably, the intelligent design model of foundation treatment takes a standardized engineering image that integrates terrain, stratum, and geological parameters as input, and synchronously outputs the engineering type and the design parameters in an end-to-end manner, thereby providing complete and executable design basis for subsequent scheme recommendation and range prediction.

[0062] Preferably, the intelligent design model of foundation treatment is configured with a vector database. The feature extractor of the intelligent design model of foundation treatment extracts features from engineering case samples to obtain a feature vector of each sample, which is stored in the vector database. A correspondence between the feature vector and the engineering case sample is established.

[0063] Preferably, the feature extractor in the intelligent design model for foundation treatment is used to extract features from the samples in the foundation treatment scheme case dataset, obtaining the feature vector of each sample and storing it in a vector database. Simultaneously, a mapping relationship is established between the feature vectors and engineering case samples in the foundation treatment scheme case dataset to facilitate efficient subsequent retrieval and case backtracking.

[0064] The feature vectors extracted by the above-mentioned intelligent design model for foundation treatment take into account both project type and design parameters during training. Therefore, in the feature vector space, samples of different project types will form different clusters (with large distances between clusters), while samples within the same project type will form continuous or distributed trajectories within the cluster according to the changes in parameters (distinguished within the cluster).

[0065] Thanks to the joint optimization of engineering type classification and design parameter regression tasks during the training phase of the intelligent design model for foundation treatment, the extracted feature vectors possess a good semantic structure in the vector space: samples from different engineering types form clearly separated clusters (with large inter-cluster distances), while samples of the same type exhibit continuous or regular distribution patterns within their respective clusters based on differences in their design parameters (such as pile length, pile diameter, and spacing) (distinguished within each cluster). This structured feature representation lays a solid foundation for intelligent solution recommendation based on similarity matching.

[0066] After training the intelligent design model for foundation treatment is completed, intelligent design of the foundation treatment scheme can be achieved through step S5.

[0067] Preferably, step S5 includes: First, the geological regions of the roadbed engineering design drawings for the foundation treatment scheme to be designed are identified, and the geological parameters (such as unit weight, bearing capacity, compressive modulus, etc.) corresponding to each stratum are encoded into the RGB color channels of the image; Subsequently, combining the subgrade location and railway grade information, a standardized image region with the same size as the training samples is cropped from the location of the subgrade to be processed, generating the image of the project to be designed, such as... Figure 9 As shown; Input the image of the project to be designed into the intelligent design model for foundation processing.

[0068] The feature extractor of the intelligent design model for foundation treatment performs forward reasoning on the input image of the project to be designed, generating a corresponding high-dimensional feature vector. Then, it retrieves the Top K nearest neighbors of this feature vector from the pre-built vector database. The engineering cases corresponding to these samples represent the most similar historical design schemes under the current working conditions. Their engineering type and design parameters can be directly used as recommended schemes for the current project, such as... Figure 10 As shown.

[0069] Preferably, the foundation treatment intelligent design model extracts a feature vector of the current to-be-designed engineering image by using the feature extractor, and retrieves a closest feature vector in the vector database with a distance less than a tolerance from the feature vector. According to a corresponding relationship between the feature vector and the engineering case sample, the closest feature vector corresponding engineering case sample is determined, and the engineering case corresponding to the engineering case sample is the most similar engineering case to the current to-be-designed engineering image. The engineering type and the design parameter of the engineering case can be used as the engineering type and the design parameter of the current to-be-designed engineering image. If there is no vector satisfying the condition in the vector database, the engineering type and the design parameter predicted by the intelligent model are used as the engineering type and the design parameter of the current to-be-designed engineering image.

[0070] Preferably, the recommended scheme of the foundation treatment intelligent design model outputted can be the engineering type and the design parameter predicted by the foundation treatment intelligent design model, or the engineering type and the design parameter of the most similar engineering case matched from the vector database.

[0071] The training of the improved CNN model by the foundation treatment scheme case dataset in this embodiment can obtain the corresponding intelligent model and the feature extractor. In the feature space, samples of the same engineering type form a clustering cluster, and samples of different design parameters in the cluster can be further distinguished. This makes the feature vectors of more similar schemes (the same engineering type and similar design parameters) closer in the feature space, so that the optimal foundation treatment scheme retrieved is closer to the engineering demand.

[0072] Embodiment 3 This embodiment is a further improvement of embodiment 2, and the repeated contents will not be described again.

[0073] Preferably, for the composite foundation, only determining the engineering type and the key design parameter is not enough to form a complete intelligent design scheme, and the foundation treatment range also needs to be further determined. The foundation treatment range is usually determined by multiple factors such as the roadbed filling height, the settlement control standard, and the spatial form of the underlying stratum.

[0074] To realize the intelligent prediction of the treatment range, this embodiment introduces an image segmentation network (such as U-Net) to construct a foundation treatment range prediction model. Specifically, the standardized geological condition image fused with the terrain and the geological parameters, i.e., the engineering case sample, is used as the input, and the corresponding design region mask image (i.e., a binary image in which the treatment region is marked as white and the remaining regions are black) is used as the supervision label to train the image segmentation network end to end. After training, the obtained foundation treatment range prediction model can automatically output a high-precision foundation treatment range prediction result according to the input to-be-designed engineering image, and generate the corresponding mask image.

[0075] The foundation treatment range prediction model effectively converts complex engineering experience rules into data-driven spatial decision-making capabilities, providing key technical support for the automatic delineation of treatment areas in subsequent intelligent design.

[0076] Preferably, after the foundation treatment intelligent design model outputs the recommended scheme, the image of the to-be-designed engineering is input into the foundation treatment range prediction model, and the foundation treatment range prediction model generates a binary mask image of the treatment area after processing. Subsequently, the mask boundary is accurately extracted through an image contour extraction algorithm, and the contour is superimposed and drawn onto the original design map to form the foundation treatment area, as shown in Figure 11 The designer can make local fine-tuning according to the actual topography and geological conditions, and finally form a foundation treatment design scheme that meets the engineering requirements.

[0077] At this point, an end-to-end intelligent design closed loop that integrates multi-dimensional information such as roadbed location, railway grade, terrain undulation, and geological spatial distribution is achieved, significantly improving the efficiency, consistency, and scientificity of the foundation treatment scheme design.

[0078] Embodiment 4 The embodiment provides a deep learning-based support scheme intelligent design system. The roadbed foundation treatment intelligent design system comprises an input unit, a processing unit and an output unit. The input unit is used to input the roadbed engineering design map and geological parameters of the to-be-designed foundation treatment scheme. The processing unit is used to obtain the recommended scheme according to the roadbed foundation treatment intelligent design method of embodiment 1, embodiment 2 or embodiment 3. The output unit is used to output the recommended scheme.

[0079] Embodiment 5 The embodiment provides an electronic device. The electronic device comprises at least one processor and a memory connected with the at least one processor in communication. The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the roadbed foundation treatment intelligent design method of embodiment 1, embodiment 2 or embodiment 3.

[0080] Embodiment 6 The embodiment provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions for causing a processor to execute the roadbed foundation treatment intelligent design method of embodiment 1, embodiment 2 or embodiment 3 when executed.

[0081] Embodiment 7 The embodiment provides a computer program product. The computer program product comprises a computer program, and the computer program implements the roadbed foundation treatment intelligent design method of embodiment 1, embodiment 2 or embodiment 3 when executed by a processor.

[0082] The above merely describes preferred embodiments of the present application, and is not used to limit the present application, any modification, equivalent replacement and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for intelligent design of roadbed foundation treatment, characterized in that, Includes the following steps: S1. Obtain historical engineering cases of roadbed foundation treatment; the historical engineering cases include at least engineering design drawings; the engineering design drawings include: foundation treatment scheme, topography, spatial distribution information of strata and geological parameters; S2. The geological parameters are integrated into the engineering design drawings using a color blending method; the geological parameters include at least: unit weight, bearing capacity, and compressive modulus. S3. On the engineering design drawing that integrates geological parameters, and in conjunction with the location of the roadbed surface, extract engineering images according to preset specifications, and set labels for the extracted engineering images to generate engineering case samples; wherein, the labels include engineering type and design parameters; S4. Construct a dataset using the engineering case samples, train a convolutional neural network, and obtain an intelligent design model for foundation treatment. S5. Geological parameters are fused into the roadbed engineering design drawings of the foundation treatment scheme to be designed, and image data of preset specifications are extracted at the location to be treated to generate the engineering image to be designed; the engineering image to be designed is input into the intelligent design model for foundation treatment for processing, and the intelligent design model for foundation treatment outputs a recommended scheme for the foundation treatment scheme to be designed; Among them, the intelligent design model for foundation treatment uses CNN as the backbone network to extract feature vectors from the input image, and then connects them to a two-branch fully connected layer, which are used for engineering type classification prediction and design parameter regression prediction of foundation treatment schemes, respectively. The intelligent design model for foundation treatment processes the input image of the project to be designed and outputs the project type and design parameters corresponding to the image.

2. The intelligent design method for roadbed foundation treatment according to claim 1, characterized in that, The method of integrating the geological parameters into the engineering design drawing using color fusion in S2 includes: fusing the geological parameters with the RGB values ​​of the image.

3. The intelligent design method for roadbed foundation treatment according to claim 2, characterized in that, A rectangular area is set in the engineering case sample to mark the location of the roadbed.

4. The intelligent design method for roadbed foundation treatment according to claim 1, characterized in that, The intelligent design model for foundation treatment includes: a feature extractor and two parallel fully connected layers; The feature extractor is used to extract the feature vector of the engineering image to be designed; The feature extractor includes: a convolutional layer, a max pooling layer, a residual module, and a global average pooling layer; The convolutional layer is used to perform primary feature extraction on the input engineering image to be designed, and to capture local features through the convolutional kernel; The max pooling layer is used to compress the spatial dimension of the feature map and enhance the position invariance of the features; The residual module is used to extract deep features through skip connections to avoid gradient vanishing; wherein, the residual module includes several residual blocks; the residual blocks are used to optimize the expressive power of deep features by fusing identity mapping with convolutional paths; The global average pooling layer is used to compress spatial features into channel feature vectors while retaining key semantic information; Two fully connected layers are respectively connected to the output of the global average pooling layer. One fully connected layer is used to process the feature vector and output the prediction result of the engineering type of the engineering image to be designed; the other fully connected layer is used to process the same feature vector and output the prediction result of the design parameters of the engineering image to be designed.

5. The intelligent design method for roadbed foundation treatment according to claim 4, characterized in that, The intelligent design model for foundation treatment is equipped with a vector database; The feature extractor of the intelligent design model for foundation treatment extracts features from engineering case samples, obtains the feature vector of each sample, stores it in the vector database, and establishes a correspondence between the feature vector and the engineering case samples.

6. The intelligent design method for roadbed foundation treatment according to claim 1, characterized in that, Also includes: S6. Construct a dataset using the engineering case samples, train the image segmentation network, and obtain a foundation processing range prediction model. The image of the project to be designed is input into the foundation treatment range prediction model for processing, and the foundation treatment range prediction model outputs the foundation treatment range prediction result.

7. A roadbed foundation treatment intelligent design system, characterized in that, include: The input unit is used to input the subgrade engineering design drawings and geological parameters of the foundation treatment scheme to be designed; The processing unit obtains a recommended scheme based on the intelligent design method for roadbed foundation treatment as described in any one of claims 1 to 6; The output unit is used to output the recommended solution.

8. An electronic device, characterized in that, The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the roadbed infrastructure intelligent design method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that are used to cause a processor to execute the intelligent design method for roadbed infrastructure processing as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the intelligent design method for roadbed infrastructure processing as described in any one of claims 1 to 6.

Images