Training method and device of foundation pit supporting scheme generation model, equipment and medium

Abstract

The invention discloses a training method and device for a foundation pit supporting scheme generation model, equipment and a medium, and the method comprises the steps: obtaining a foundation pit CAD graph set and a generation graph corresponding to each CAD graph in the foundation pit CAD graph set, carrying out the preprocessing of each CAD graph in the foundation pit CAD graph set, and obtaining training data, the training data comprises a foundation pit outer contour diagram, a punching contour diagram, foundation pit description information and a generative diagram; performing data expansion processing on the training data to obtain an expanded image corresponding to each training data, the expanded image being a noise image; loading a pre-trained deep learning model, and performing fine tuning processing on the deep learning model based on the extended image to obtain a fine-tuned deep learning model; and performing model training on the fine-tuned deep learning model according to the training data and the extended image, and obtaining a trained foundation pit supporting scheme generation model after the training is completed. And the efficiency and the accuracy of generating the integrated support scheme graph are improved.

Description

Technical Field

[0001] This application relates to the field of civil engineering application technology, and in particular to a training method, apparatus, equipment and medium for generating a foundation pit support scheme model. Background Technology

[0002] In the field of civil engineering, foundation pit engineering is a crucial link in infrastructure construction, and the design of the internal support scheme for the foundation pit plays a vital role in ensuring its stability, safety, and construction feasibility. With the acceleration of urbanization, the scale of foundation pit projects is increasing, and the geological conditions and surrounding environment are becoming more complex and diverse.

[0003] Traditional design methods for internal support systems in foundation pits primarily rely on engineers' experience and expertise, involving manual calculations, reference to standards, and past engineering cases. This approach faces numerous challenges. Firstly, designers must expend significant time and effort processing massive amounts of data, including geological survey reports, detailed information on surrounding buildings, and precise measurements of the foundation pit's outline. Furthermore, the difficulty of manual design and calculation increases exponentially when dealing with complex pit shapes, such as irregular polygons or those with internal perforations, increasing the risk of human error. Secondly, traditional design methods are relatively inefficient, failing to generate multiple feasible design solutions for comparison and optimization within a short timeframe, thus unable to meet the demands of rapid construction and the need for highly refined design solutions in modern engineering projects. Summary of the Invention

[0004] The purpose of this application is to provide a training method, device, electronic device, and storage medium for generating a foundation pit support scheme model, so as to solve the technical problem that the related technologies have poor spraying effect and efficiency for UAVs.

[0005] In a first aspect, embodiments of this application provide a training method for a foundation pit support scheme generation model, including: Obtain the foundation pit CAD drawing set and the generated drawing corresponding to each CAD drawing in the foundation pit CAD drawing set, and preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data, wherein the training data includes the foundation pit outer contour drawing, the drilling contour drawing, the foundation pit description information and the generated drawing; The training data is subjected to data augmentation processing to obtain an augmented image corresponding to each training data, wherein the augmented image is a noisy image; Load a pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain a fine-tuned deep learning model, wherein the deep learning model is a Sable Diffusion model. The fine-tuned deep learning model is trained based on the training data and the expanded image, and a trained foundation pit support scheme generation model is obtained upon completion of training.

[0006] Secondly, embodiments of this application provide a training apparatus for generating a foundation pit support scheme model, comprising: The data acquisition module is used to acquire the foundation pit CAD drawing set and the generated drawing corresponding to each CAD drawing in the foundation pit CAD drawing set, and to preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data, wherein the training data includes the foundation pit outer contour drawing, the drilling contour drawing, the foundation pit description information and the generated drawing; The data processing module is used to perform data expansion processing on the training data to obtain an expanded image corresponding to each training data, wherein the expanded image is a noisy image; The fine-tuning module is used to load a pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain a fine-tuned deep learning model, wherein the deep learning model is a Sable Diffusion model. The model training module is used to train the fine-tuned deep learning model based on the training data and the extended image, and to obtain the trained foundation pit support scheme generation model when the training is completed.

[0007] Thirdly, embodiments of this application provide an electronic device, which includes a processor, a memory, and a computer program or embedded program stored in the internal storage of the memory and executable on the processor. When the processor executes the computer program or embedded program, it implements the steps in the training method for generating a foundation pit support scheme as described above.

[0008] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program or embedded program, which, when executed by a processor, implements the steps in the training method for generating a foundation pit support scheme as described above.

[0009] This application provides a training method, apparatus, electronic device, and storage medium for a foundation pit support scheme generation model. The method involves acquiring a foundation pit CAD drawing set and the corresponding generated image for each CAD drawing in the set. Each CAD drawing in the set is preprocessed to obtain training data, which includes the foundation pit outer contour map, drilling contour map, foundation pit description information, and the generated image. The training data is then expanded to obtain an expanded image for each training data point, where the expanded image is a noisy image. A pre-trained deep learning model is loaded, and fine-tuned based on the expanded image to obtain a fine-tuned deep learning model, where the deep learning model is a Sable Diffusion model. The fine-tuned deep learning model is trained using the training data and the expanded image, and a trained foundation pit support scheme generation model is obtained upon completion of training. By preprocessing the CAD drawings of the foundation pit to generate training data containing the outer contour map, the drilling contour map and descriptive information, and combining noise expansion and deep learning model fine-tuning, the intelligent generation of foundation pit support schemes is realized. This solves the problems of low efficiency and difficulty in integrating multimodal information by relying on manual data processing in traditional methods, and improves the efficiency and accuracy of scheme drawing generation. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the steps of a training method for generating a foundation pit support scheme model provided in an embodiment of this application. Figure 2 This is a schematic diagram of training data obtained by processing CAD drawings, provided in an embodiment of this application. Figure 3 This is a flowchart illustrating the steps for fine-tuning a deep learning model according to an embodiment of this application; Figure 4 This is a flowchart illustrating one step in obtaining a foundation pit support scheme generation model according to an embodiment of this application; Figure 5 This is a block diagram illustrating a scheme for generating a model based on a foundation pit support scheme, as provided in an embodiment of this application. Figure 6 This is a schematic diagram of a training device for generating a foundation pit support scheme model provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 8 This is another structural schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0012] It should be understood that the steps described in the method embodiments disclosed in this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0013] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0014] In related technologies, the traditional design of internal support schemes for foundation pits mainly relies on the experience and expertise of engineers, involving manual calculations, reference to specifications, and past engineering cases. This design approach faces numerous challenges. On the one hand, designers need to spend a significant amount of time and effort processing massive amounts of data, including geological survey reports, detailed information on surrounding buildings, and precise measurements of the foundation pit's outer contour. Furthermore, when dealing with complex foundation pit shapes, such as irregular polygons or those with internal perforations, the difficulty of manual design and calculation increases exponentially, making human error more likely. On the other hand, traditional design methods are relatively inefficient, making it difficult to generate multiple feasible design schemes for comparison and optimization in a short period, thus failing to meet the demands of rapid progress in modern engineering construction and the need for highly refined design solutions.

[0015] To address the technical problems existing in related technologies, this application provides a training method for a foundation pit support scheme generation model. Please refer to [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart illustrating the steps of a training method for generating a foundation pit support scheme model provided in this application embodiment, which includes steps 101 to 105.

[0016] Step 101: Obtain the foundation pit CAD drawing set and the corresponding generated drawing for each CAD drawing in the foundation pit CAD drawing set. Preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data, which includes the foundation pit outer contour drawing, drilling contour drawing, foundation pit description information, and generated drawing.

[0017] In one embodiment, a set of CAD drawings of the foundation pit and corresponding generated drawings are acquired. The CAD drawings are preprocessed to form training data containing the outer contour of the foundation pit, the drilling contour, and descriptive information. The descriptive information may include the area of ​​the foundation pit (square meters), the length of the foundation pit (meters), the width of the foundation pit (meters), and whether drilling has been performed on the foundation pit (yes or no). The obtained training data can be referenced... Figure 2 .

[0018] The foundation pit CAD drawing set refers to a collection of computer-aided design files containing complete foundation pit design information. Specifically, it can be implemented using AutoCAD format files, with each file corresponding to the geometric parameters and structural features of a specific foundation pit. Preprocessing includes text recognition to extract descriptive information, boundary recognition to generate outer contour maps, and region recognition to generate perforation contour maps. This can be achieved using OCR technology and image edge detection algorithms, converting unstructured drawings into standard training data.

[0019] Specifically, engineering parameters are extracted from the original CAD drawings using a text recognition module. Simultaneously, computer vision algorithms are used to separate the main outline of the foundation pit and the internal hole areas. The annotated drawings and generated drawings form a paired dataset. Therefore, the process of obtaining training data includes: performing text recognition on each CAD drawing in the foundation pit CAD drawing set, and obtaining the foundation pit description information corresponding to the foundation pit CAD drawing set according to the text recognition results and description rules; performing boundary recognition on each CAD drawing in the foundation pit CAD drawing set, and obtaining the foundation pit outer contour map set corresponding to the foundation pit CAD drawing set based on the regions obtained from region recognition; performing region recognition on each CAD drawing in the foundation pit CAD drawing set, and obtaining the drilling contour map set corresponding to the foundation pit CAD drawing set based on the drilling areas obtained from region recognition; and associating the foundation pit description information, the foundation pit outer contour map set, and the drilling contour map set based on map identifiers to obtain training data.

[0020] When preprocessing CAD drawings, information can be identified in various ways. For example, text recognition can be used to obtain the description information of the foundation pit in the CAD drawing, boundary recognition can be used to obtain the outer contour of the foundation pit in the CAD drawing, and region recognition can be used to obtain the contour of the drilling area in the CAD drawing. Then, the obtained multimodal data is associated with the generated drawing corresponding to the CAD drawing to obtain training data for model training.

[0021] Step 102: Perform data augmentation on the training data to obtain an augmented image corresponding to each training data point, wherein the augmented image is a noisy image.

[0022] In one embodiment, after preprocessing each CAD drawing in the foundation pit CAD drawing set to obtain training data, the training data can be extended to obtain images containing noise. Then, through training, the obtained foundation pit support scheme generation model can accurately generate a suitable scheme even when faced with images containing noise.

[0023] The data expansion process is achieved by adding Gaussian noise or random occlusion to the contour map, such as adding pixel perturbations to the outer contour edge of the pit to form training samples with anti-interference capabilities.

[0024] The process of expanding the training data includes: adding noise to each pit outline and each hole outline in the training data to obtain a noise image corresponding to each pit outline and each hole outline; and labeling the noise image in the generated image in the training data to obtain an expanded image corresponding to each training data.

[0025] Noise addition refers to injecting random interference signals into the outer contour map of the foundation pit and the drilling contour map. This can be achieved using Gaussian noise or salt-and-pepper noise algorithms, simulating measurement errors and image acquisition interference that may exist in actual engineering projects. Labeling refers to establishing a correspondence between the original generated image and the noise-processed image. This can be achieved by adding associated labels to the noisy image using image annotation tools, providing supervised data with noise features for subsequent model training.

[0026] Specifically, during the data expansion process, noise can be gradually added to the pit outline and the hole outline, making the original images (pit outline and hole outline) gradually become unclear until they become images with random noise. Then, the original generated images are used to label the noisy images to form labeled expanded training samples. By adding the constructed expanded images to the training data, the model can learn from noisy images during training, and thus obtain accurate generated images even for noisy images.

[0027] Step 103: Load the pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain the fine-tuned deep learning model, wherein the deep learning model is the Sable Diffusion model.

[0028] In one embodiment, when training data for model training is obtained after processing is completed, a pre-trained deep learning model is loaded. The deep learning model can be a Sable Diffusion model (hereinafter referred to as SD model). The deep learning model is fine-tuned based on the obtained extended image to obtain a fine-tuned deep learning model that can identify and process noise.

[0029] Furthermore, referring to Figure 3 , Figure 3 This is a flowchart illustrating a step for fine-tuning a deep learning model according to an embodiment of this application, wherein the step includes steps 201 to 205.

[0030] Step 301: Load the pre-trained deep learning model and perform segmentation processing on the extended image to obtain the first training set and the first test set; Step 302: Train and fine-tune the parameters of the deep learning model based on the first training set to obtain the fine-tuned deep learning model; Step 303: Test the fine-tuned deep learning model based on the first test set, and determine whether the fine-tuning is complete based on the accuracy obtained from the test. Step 304: When the accuracy is determined to be greater than the preset threshold, the fine-tuned deep learning model is obtained. Step 305: When the accuracy is determined to be less than or equal to a preset threshold, perform the step of segmenting the extended image.

[0031] Specifically, a pre-trained deep learning model is loaded, and the extended image is divided to obtain a first training set and a first test set. The deep learning model is trained and its parameters are fine-tuned based on the first training set to obtain a fine-tuned deep learning model. The fine-tuned deep learning model is tested based on the first test set, and the accuracy obtained from the test determines whether the fine-tuning is complete. When the accuracy is greater than a preset threshold, the fine-tuned deep learning model is obtained. When the accuracy is less than or equal to the preset threshold, the extended image division process is repeated.

[0032] In the processing, image segmentation refers to dividing the noisy training data into subsets for different purposes. This can be achieved through random or hierarchical partitioning, such as splitting the training and test sets in a 7:3 ratio, thus ensuring data independence during model training and validation. Fine-tuning the parameters on the first training set involves iteratively optimizing the model weights using the training data. This can be achieved using backpropagation combined with gradient descent, adjusting the learning rate to control the model's convergence speed. Testing on the first test set involves evaluating the model's performance on data not used in training, using metrics such as accuracy, recall, or F1 score. For example, a model is considered satisfactory when accuracy reaches 85%. The preset threshold is a predefined performance threshold for the model, which can be dynamically set according to actual engineering needs. For example, setting the threshold to a range of 80%-90% can help determine whether the data partitioning strategy needs to be re-optimized.

[0033] Specifically, after the pre-trained deep learning model is loaded, the expanded images are divided into a first training set and a first test set. The first training set is used to fine-tune the model's parameters, adjusting the model's internal weights through iterative training. The fine-tuned model is then tested on the first test set. Fine-tuning is considered complete when the test accuracy exceeds a preset threshold; otherwise, the data division and training process is repeated. This process uses a loop optimization mechanism to ensure the model maintains stable performance even under noisy data interference, while avoiding overfitting or underfitting problems caused by unreasonable data division.

[0034] Step 104: Train the fine-tuned deep learning model based on the training data and extended images, and obtain the trained foundation pit support scheme generation model upon completion of training.

[0035] In one embodiment, after fine-tuning the deep learning model based on the extended image, the fine-tuned deep learning model is trained based on the training data and the extended image, and a trained foundation pit support scheme generation model is obtained when the training is completed.

[0036] For example, during training, the fine-tuned deep learning model is trained based on multimodal data, wherein the multimodal data is the data obtained by preprocessing the CAD drawing as described above. By training based on multimodal data, the trained model can identify and analyze data of different modalities, so as to generate accurate generated maps for one or more modalities of data in the multimodal dataset.

[0037] Furthermore, referring to Figure 4 , Figure 4This is a flowchart illustrating one step in obtaining a foundation pit support scheme generation model according to an embodiment of this application. This step includes steps 301 to 303.

[0038] Step 401: Based on the training data and the extended image, generate input data for training the fine-tuned deep learning model, wherein the input data includes a combination of pit description information and at least one of pit outline map and drilling outline map, or at least one of pit outline map and drilling outline map. Step 402: Input the input data into the fine-tuned deep learning model to obtain the output result; Step 403: Compare and analyze the output results and the generated image, and adjust the parameters of the fine-tuned deep learning model based on the comparison and analysis results, so that a well-trained foundation pit support scheme generation model is obtained when training is completed.

[0039] Specifically, when training the fine-tuned deep learning model based on training data and extended images, input data for training the fine-tuned deep learning model is generated. The input data includes a combination of at least one of the foundation pit description information and the foundation pit outer contour map and the drilling contour map, or at least one of the foundation pit outer contour map and the drilling contour map. The input data is input into the fine-tuned deep learning model to obtain the output result. The output result and the generated map are compared and analyzed, and the parameters of the fine-tuned deep learning model are adjusted according to the obtained comparison and analysis results, until the trained foundation pit support scheme generation model is obtained when the training is completed.

[0040] The input data refers to the combination of multimodal information used for model training. Specifically, this can be achieved by concatenating textual descriptions with graphical features; for example, encoding the pit description information into a text vector and then overlaying it with the image feature matrix. Comparative analysis involves calculating the pixel-level difference between the generated image and the output result. This can be achieved using a structural similarity index or a mean squared error function as the evaluation criterion. Parameter tuning involves updating the model weights based on the difference through backpropagation, which can be implemented using a stochastic gradient descent optimization algorithm.

[0041] Specifically, during the model training phase, the input data is flexibly combined with pit description information, outer contour maps, and perforation contour maps to form multi-dimensional feature representations. For example, when dealing with pits with complex internal structures, both outer contour maps and perforation contour maps can be input simultaneously as spatial constraints. When dealing with standardized pits, only the outer contour map along with text description is required. After receiving the input data, the model gradually denoises and generates a support scheme diagram through the inverse operation of the diffusion process. The generated result is compared pixel-by-pixel with the original generated image, and the difference value is fed back to the network weight update module through the loss function, driving the model to learn the mapping relationship between different input combinations and the target output. Compared with existing technologies, traditional methods rely on manually drawing design diagrams and cannot dynamically adjust generation rules. In contrast, this solution constructs a composable input data system, enabling the model to adapt to pit morphologies of varying complexity. By combining the input of the hole contour map and the outer contour map, the spatial constraint relationship of the support structure can be accurately captured, which solves the problem of low accuracy of the generated scheme when the traditional design method is faced with multiple holes and irregular foundation pits. Through the flexible combination of multimodal input data, the model can adapt to different geological conditions and engineering needs, effectively avoiding the risk of support structure instability caused by human design negligence.

[0042] Furthermore, after training the deep learning model to obtain the foundation pit support scheme generation model, it can be used to generate scheme diagrams. At this time, the data input into the model can be CAD drawings, foundation pit outer contour diagrams, drilling contour diagrams, or combinations of multimodal data.

[0043] Therefore, when generating a scheme based on the foundation pit support scheme generation model, the process includes: receiving the input foundation pit CAD drawing to be processed, and preprocessing the CAD drawing to be processed to obtain the drawing information contained in the foundation pit CAD drawing to be processed, wherein the drawing information includes at least one of the foundation pit outer contour drawing and the drilling contour drawing; inputting the drawing information into the trained foundation pit support scheme generation model to output the target scheme drawing corresponding to the foundation pit CAD drawing to be processed.

[0044] The CAD drawing of the foundation pit to be processed refers to the original engineering drawing that requires support scheme design. It can be generated using two-dimensional or three-dimensional drawing files generated by AutoCAD or BIM software, and is used to provide basic morphological data of the foundation pit. Preprocessing refers to extracting effective information from the drawing through image analysis technology. As described above, it involves processing the CAD drawing of the foundation pit to be processed using methods such as text recognition algorithms, boundary recognition, and region recognition to obtain descriptive information, the outer contour drawing of the foundation pit, and the drilling contour drawing.

[0045] Specifically, after obtaining the CAD drawing of the foundation pit to be processed, text recognition processing is first performed to extract descriptive information such as the pit depth and soil layer parameters. Simultaneously, a boundary detection algorithm is used to extract the coordinate data of the pit's outer contour boundary. For complex foundation pit structures with internal holes, a region segmentation algorithm is used to identify the shape and location parameters of the perforated areas. The preprocessed data is converted into a standardized tensor format and input into a trained model. The model uses a decoder network to map the input features into a three-dimensional spatial distribution of the support structure, ultimately outputting a target scheme drawing containing the position and size parameters of the support beams and columns. When the input data contains perforation outlines, the model automatically adjusts the support layout to avoid the perforated areas and maintain structural stability.

[0046] Of course, when generating the scheme drawing, the input information can be either the outer contour drawing of the foundation pit or the drilling contour drawing, or both can be input together. The specific input depends on the operator's operation. That is, when generating the scheme drawing, you can input a CAD drawing, or you can directly input an outer contour drawing of the foundation pit or a drilling contour drawing to generate the corresponding scheme drawing.

[0047] In addition, when generating the scheme diagram, the lack of input information may lead to incomplete or problematic scheme diagrams. In this case, the accuracy of the scheme diagram generation can be improved by re-entering the corresponding information, including: when receiving the scheme adjustment instruction, receiving the input description information and inputting the description information and diagram information into the trained foundation pit support scheme generation model to output the updated scheme diagram of the foundation pit to be processed CAD drawing.

[0048] In other words, when the operators review the design drawings and determine that adjustments are needed, they can trigger adjustment commands through the interface and simultaneously input text descriptions containing the intention to modify the drawings. The model then fuses the original pre-processed pit outline and drilling outline with the newly added descriptive information, adjusting the support layout parameters while maintaining the spatial structural features to generate a new design drawing.

[0049] Furthermore, one can refer to Figure 5 When generating scheme diagrams based on the foundation pit support scheme generation model obtained after training, the model outputs corresponding scheme diagrams by inputting combination conditions under different combinations. Descriptive information is optional, and the combination conditions must include at least one of the foundation pit outer contour map and the drilling contour map. Furthermore, when regenerating scheme diagrams, the combination conditions can be updated, for example, by adding descriptive information. The specific update method depends on the previous combination conditions. For instance, if the combination conditions only include one of the foundation pit outer contour map and the drilling contour map, another image and / or descriptive information can be added. Similarly, if both the foundation pit outer contour map and the drilling contour map are included, descriptive information can be added.

[0050] Furthermore, the process of training the foundation pit support scheme generation model based on the above method and obtaining the scheme diagram based on the foundation pit support scheme generation model can be as follows: Step 1: Preprocessing of CAD drawings for the foundation pit support scheme.

[0051] By preprocessing the actual foundation pit support scheme drawings, the drawings are converted into three different inputs that the model can recognize and train. These inputs include an image of the foundation pit's outer contour, an image of the internal drilling holes, and an editable text description of the foundation pit. The text description includes information such as the foundation pit area (square meters), foundation pit length (meters), foundation pit width (meters), and whether the foundation pit has been drilled (yes or no).

[0052] Step 2: Based on the real information on the support schemes inside the foundation pit obtained in Step 1, construct an image dataset of the support schemes inside the foundation pit, and divide it into a training set and a test set.

[0053] Step 3: Build the model.

[0054] ControlNet technology is used to fine-tune the already constructed stable diffusion model. By incorporating image information of the pits, the model is fine-tuned to control the generated results, thereby improving the accuracy of the generated images and enabling more flexible and diverse input conditions.

[0055] Furthermore, the implementation method and model structure of step three are as follows: 1. Construct a Stable Diffusion model, using an editable text description of the foundation pit as the input and a diagram of the support scheme within the foundation pit as the generated image. The model includes two parts: diffusion and denoising. 2. In the diffusion part of the model, noise is gradually added to the data, making the original data increasingly unclear until it becomes random noise. This simulates the gradual degradation of data related to the foundation pit support scheme under the influence of natural or human factors. Through this process, the model can generate a series of data samples from the original data to a completely noisy state, which are then used to train the denoising model. Starting from the original data X0, noise is gradually added to produce X1, X2, ..., X... T Each step involves a transition probability q(x) t |x t -1), this probability describes how to transition from the previous state to the next noisier state; 3. In the denoising part of the model, the model is trained to recover the original foundation pit structure information from the noise-affected data and generate the support scheme within the foundation pit. The model first uses the noise data from the previous step and the conditional input information to predict the noise distribution information in the diffusion step, thereby obtaining the conditional probability of the denoising process. Finally, it gradually generates clearer images until the final scheme image is output. 4. Further fine-tune the Sable Diffusion model using ControlNet, locking and copying the original model structure and information parameters from the constructed stablediffusion model to the trainable copy on the ControlNet side; 5. By using Zero Convolution in ControlNet to connect the copied original model structure, the model can combine text and images as input conditions to obtain new outputs.

[0056] Step 4: Model training.

[0057] The model is trained using the Dropout strategy to ensure that it can generalize to various types of input information, thereby completing the construction and training of an automatic generation model for foundation pit support schemes based on a multimodal model.

[0058] Step 5: Experimental verification.

[0059] The accuracy of the model's output results was verified by changing different input conditions, where the input conditions were different combinations of the multimodal data obtained from the aforementioned preprocessing.

[0060] 1. Based on the established automatic generation method for foundation pit support schemes using a multimodal model, the model is validated using editable text descriptions of the foundation pit and images of its outer contour. The generated images of the foundation pit support schemes are used in the validation test set.

[0061] 2. Further verify the input conditions in step five. Using the editable text description information of the foundation pit, the image of the outer contour of the foundation pit, and the sketch information of the drilling inside the foundation pit as input, verify the generated image of the support scheme inside the foundation pit in the test set.

[0062] 3. Further verify the validity of the textual information of the input conditions in step five by changing different textual descriptions of the foundation pits to the model input, such as different foundation pit lengths and widths, to verify whether the model can understand the physical meaning expressed by the text.

[0063] Step 6: For models that have passed validation, input the corresponding data to output the corresponding solution diagram.

[0064] In summary, the above embodiments provide a training method for a foundation pit support scheme generation model. This method involves acquiring a foundation pit CAD drawing set and the corresponding generated image for each CAD drawing in the set. Each CAD drawing in the set is preprocessed to obtain training data, which includes the foundation pit outer contour map, drilling contour map, foundation pit description information, and the generated image. The training data is then expanded to obtain an expanded image for each training data point, where the expanded image is a noisy image. A pre-trained deep learning model is loaded, and fine-tuned based on the expanded image to obtain a fine-tuned deep learning model, where the deep learning model is a Sable Diffusion model. The fine-tuned deep learning model is trained using the training data and the expanded image, and a trained foundation pit support scheme generation model is obtained upon completion of training. By preprocessing the CAD drawings of the foundation pit to generate training data containing the outer contour map, the drilling contour map and descriptive information, and combining noise expansion and deep learning model fine-tuning, the intelligent generation of foundation pit support schemes is realized. This solves the problems of low efficiency and difficulty in integrating multimodal information by relying on manual data processing in traditional methods, and improves the efficiency and accuracy of scheme drawing generation.

[0065] Based on the method described in the above embodiments, this embodiment will further describe it from the perspective of the training device for generating the foundation pit support scheme model. The training device for generating the foundation pit support scheme model can be implemented as an independent entity or integrated into an electronic device, such as a terminal, which may include a mobile phone, a tablet computer, etc.

[0066] Please see Figure 6 , Figure 6 This is a schematic diagram of a training device for generating a foundation pit support scheme model provided in an embodiment of this application, as shown below. Figure 6 As shown in the embodiment of this application, the training device 600 for generating a foundation pit support scheme model includes: The data acquisition module 601 is used to acquire the foundation pit CAD drawing set and the generated drawing corresponding to each CAD drawing in the foundation pit CAD drawing set, and to preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data. The training data includes the foundation pit outer contour drawing, the drilling contour drawing, the foundation pit description information and the generated drawing. Data processing module 602 is used to perform data expansion processing on training data to obtain an expanded image corresponding to each training data, wherein the expanded image is a noisy image; The fine-tuning processing module 603 is used to load the pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain the fine-tuned deep learning model, wherein the deep learning model is the SableDiffusion model. The model training module 604 is used to train the fine-tuned deep learning model based on the training data and extended images, and to obtain the trained foundation pit support scheme generation model when the training is completed.

[0067] In one embodiment, the data acquisition module 601 is further configured to: For each CAD drawing in the foundation pit CAD drawing set, perform text recognition, and obtain the foundation pit description information corresponding to the foundation pit CAD drawing set according to the text recognition results and description rules; Boundary identification is performed on each CAD drawing in the foundation pit CAD drawing set, and the outer contour drawing set of the foundation pit corresponding to the foundation pit CAD drawing set is obtained based on the region obtained by region identification. For each CAD drawing in the foundation pit CAD drawing set, region identification is performed, and the corresponding drilling outline drawing set is obtained based on the drilling area obtained from the region identification. Training data is obtained by associating the foundation pit description information, the foundation pit outer contour map set, and the drilling contour map set based on the graph identifier.

[0068] In one embodiment, the data processing module 602 is further configured to: Noise is added to each outer contour map of the foundation pit and each hole contour map in the training data to obtain a noise image corresponding to each outer contour map of the foundation pit and each hole contour map. The generated image in the training data is used to label the noisy image, resulting in the expanded image corresponding to each training data point.

[0069] In one embodiment, the fine-tuning processing module 603 is further configured to: Load the pre-trained deep learning model and divide the extended image to obtain the first training set and the first test set; The deep learning model is trained and its parameters are fine-tuned based on the first training set to obtain the fine-tuned deep learning model. The fine-tuned deep learning model is tested based on the first test set, and the accuracy obtained from the test determines whether the fine-tuning is complete. When the accuracy is determined to be greater than a preset threshold, a fine-tuned deep learning model is obtained. When the accuracy is determined to be less than or equal to a preset threshold, the step is to perform segmentation processing on the expanded image.

[0070] In one embodiment, the model training module 604 is further configured to: Based on the training data and the extended image, input data is generated to train the fine-tuned deep learning model. The input data includes a combination of pit description information and at least one of the pit outer contour map and the hole drilling contour map, or at least one of the pit outer contour map and the hole drilling contour map. The input data is fed into the fine-tuned deep learning model to obtain the output result; The output results and the generated image are compared and analyzed. Based on the results of the comparison and analysis, the parameters of the fine-tuned deep learning model are adjusted until a well-trained foundation pit support scheme generation model is obtained when training is completed.

[0071] In one embodiment, the training device 600 for generating the foundation pit support scheme model further includes a scheme generation module, used for: The system receives an input CAD drawing of the foundation pit to be processed and preprocesses the CAD drawing to obtain the drawing information contained in the CAD drawing of the foundation pit to be processed, wherein the drawing information includes at least one of the outer contour drawing of the foundation pit and the drilling contour drawing. The drawing information is input into the trained foundation pit support scheme generation model to output the target scheme drawing corresponding to the CAD drawing of the foundation pit to be processed.

[0072] In one embodiment, the training device 600 for generating the foundation pit support scheme further includes a scheme update module, used for: When a scheme adjustment instruction is received, the input description information is received, and the description information and drawing information are input into the trained foundation pit support scheme generation model to output an updated scheme drawing of the CAD drawing of the foundation pit to be processed.

[0073] Additionally, please see Figure 7 , Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. For example... Figure 7 As shown, the electronic device 700 includes a processor 701 and a memory 702. The processor 701 and the memory 702 are electrically connected.

[0074] The processor 701 is the control center of the electronic device 700. It connects various parts of the electronic device through various interfaces and circuits. It runs or loads the application program stored in the internal flash memory of the memory 702, which can be an embedded program, and calls the data stored in the memory 702 to complete the training of the foundation pit support scheme generation model using the training method of the foundation pit support scheme generation model described above.

[0075] In this embodiment, the processor 701 in the electronic device 700 will load the instructions corresponding to the processes of one or more applications into the memory 702 according to the steps in the training method for the foundation pit support scheme generation model described above, and the processor 701 will run the applications stored in the memory 702 to realize the training and analysis of the foundation pit support scheme generation model.

[0076] The electronic device 700 can implement the steps of any embodiment of the training method for generating a foundation pit support scheme provided in the embodiments of this application. Therefore, it can achieve the beneficial effects that the training method for generating a foundation pit support scheme provided in the embodiments of this application can achieve. For details, please refer to the previous embodiments, which will not be repeated here.

[0077] Please see Figure 8 , Figure 8 This is another structural schematic diagram of the electronic device provided in the embodiments of this application, such as... Figure 8 As shown, Figure 8 A specific structural block diagram of the electronic device provided in the embodiments of this application is shown. The electronic device can be used to implement the training method of the foundation pit support scheme generation model provided in the above embodiments.

[0078] RF circuit 810 is used to receive and transmit electromagnetic waves, realizing the mutual conversion between electromagnetic waves and electrical signals, thereby enabling communication with communication networks or other devices. RF circuit 810 may include various existing circuit elements used to perform these functions, such as antennas, radio frequency transceivers, digital signal processors, encryption / decryption chips, subscriber identity modules (SIM cards), memory, etc. RF circuit 810 can communicate with various networks such as the Internet, corporate intranets, and wireless networks, or communicate with other devices via wireless networks. The aforementioned wireless networks may include cellular telephone networks, wireless local area networks (WLANs), or metropolitan area networks (MANs). The aforementioned wireless networks may use various communication standards, protocols, and technologies, including but not limited to Global System for Mobile Communication (GSM), Enhanced Data GSM Environment (EDGE), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Wireless Fidelity (Wi-Fi) (such as IEEE 802.11a, IEEE 802.11b, IEEE 802.11g, and / or IEEE 802.11n), Voice over Internet Protocol (VoIP), Worldwide Interoperability for Microwave Access (Wi-Max), other protocols for email, instant messaging, and short messages, and any other suitable communication protocols, including those that have not yet been developed.

[0079] The memory 820 can be used to store software programs and modules, such as the program instructions / modules corresponding to the training method of the foundation pit support scheme generation model in the above embodiment. The processor 880 executes various functional applications to realize the training method of the foundation pit support scheme generation model by running the dual-mode module and the program for controlling the dual-mode module stored in the internal flash of the memory 820. The program for controlling the dual-mode module can be an embedded program stored in the internal flash of the memory 820.

[0080] Memory 820 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, memory 820 may further include memory remotely located relative to processor 880, which can be connected to electronic device 800 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0081] The input unit 830 can be used to receive uploaded digital or character information, and to generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control. Specifically, the input unit 830 may include a touch-sensitive surface 831 and other input devices 832. The touch-sensitive surface 831, also known as a touch display or touchpad, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch-sensitive surface 831), and drive the corresponding connection device according to a pre-set program. Optionally, the touch-sensitive surface 831 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, sends it to the processor 880, and can receive and execute commands from the processor 880. In addition, the touch-sensitive surface 831 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch-sensitive surface 831, the input unit 830 may also include other input devices 832. Specifically, other input devices 832 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0082] Display unit 840 can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of electronic device 800. These graphical user interfaces can be composed of graphics, text, icons, video, and any combination thereof. Display unit 840 may include display panel 841, which may optionally be configured as an LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), or similar display panel. Further, touch-sensitive surface 831 may cover display panel 841. When touch-sensitive surface 831 detects a touch operation on or near it, it transmits the information to processor 880 to determine the type of touch event. Subsequently, processor 880 provides corresponding visual output on display panel 841 according to the type of touch event. Although in the figures, touch-sensitive surface 831 and display panel 841 are implemented as two separate components to achieve input and output functions, in some embodiments, touch-sensitive surface 831 and display panel 841 can be integrated to achieve input and output functions.

[0083] The electronic device 800 may also include at least one sensor 850, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor can adjust the brightness of the display panel 841 according to the ambient light level, and the proximity sensor can generate an interruption when the flip is closed or shut down. As a type of motion sensor, a gravity acceleration sensor can detect the magnitude of acceleration in various directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc. Other sensors that may be configured in the electronic device 800, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0084] Audio circuitry 860, speaker 861, and microphone 862 provide an audio interface between the user and electronic device 800. Audio circuitry 860 converts received audio data into electrical signals, which are then transmitted to speaker 861, where they are converted into sound signals for output. Conversely, microphone 862 collects sound signals, converts them into electrical signals, which are received by audio circuitry 860, converted back into audio data, and then processed by processor 880 before being transmitted via RF circuitry 810 to, for example, another terminal, or output to memory 820 for further processing. Audio circuitry 860 may also include an earphone jack to facilitate communication between peripheral headphones and electronic device 800.

[0085] Electronic device 800, through transmission module 870 (e.g., Wi-Fi module), can help users receive requests, send information, etc., providing users with wireless broadband internet access. Although transmission module 870 is shown in the figure, it is understood that it is not a necessary component of electronic device 800 and can be omitted as needed without changing the essence of the invention.

[0086] The processor 880 is the control center of the electronic device 800. It connects to various parts of the phone via various interfaces and lines, and performs various functions and processes data of the electronic device 800 by running or executing software programs and / or modules stored in the memory 820, and by calling data stored in the memory 820, thereby providing overall monitoring of the electronic device. Optionally, the processor 880 may include one or more processing cores; in some embodiments, the processor 880 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 880.

[0087] The electronic device 800 also includes a power supply 890 (such as a battery) that supplies power to various components. In some embodiments, the power supply may be logically connected to the processor 880 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. The power supply 890 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0088] Specifically, in this embodiment, the display unit of the electronic device is a touch screen display, and the mobile terminal also includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors to implement any step in the training method of the foundation pit support scheme generation model provided in the above embodiment.

[0089] In practice, the above modules can be implemented as independent entities or combined in any way to be implemented as the same or several entities. For the specific implementation of the above modules, please refer to the previous method implementation examples, which will not be repeated here.

[0090] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. Therefore, this application provides a storage medium storing multiple instructions that, when executed by a processor, can implement any step in the training method for the foundation pit support scheme generation model provided in the above embodiments.

[0091] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk, or optical disk, etc. Specifically, the storage medium may be internal flash memory, that is, the embedded program that controls the training method for implementing the above-mentioned foundation pit support scheme generation model can be stored in the internal flash memory.

[0092] Since the instructions stored in the storage medium can execute the steps in any embodiment of the training method for generating a foundation pit support scheme provided in this application, the beneficial effects that the training method for generating a foundation pit support scheme provided in this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0093] The training method, apparatus, electronic device, and storage medium for generating a foundation pit support scheme model provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application. Moreover, those skilled in the art can make several improvements and modifications without departing from the principles of this application, and these improvements and modifications are also considered to be within the protection scope of this application.

Claims

1. A training method for a foundation pit support scheme generation model, characterized in that, include: Obtain the foundation pit CAD drawing set and the generated drawing corresponding to each CAD drawing in the foundation pit CAD drawing set, and preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data, wherein the training data includes the foundation pit outer contour drawing, the drilling contour drawing, the foundation pit description information and the generated drawing; The training data is subjected to data augmentation processing to obtain an augmented image corresponding to each training data, wherein the augmented image is a noisy image; Load a pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain a fine-tuned deep learning model, wherein the deep learning model is a Sable Diffusion model. The fine-tuned deep learning model is trained based on the training data and the expanded image, and a trained foundation pit support scheme generation model is obtained upon completion of training.

2. The method as described in claim 1, characterized in that, The preprocessing of each CAD drawing in the foundation pit CAD drawing set to obtain training data includes: Text recognition is performed on each CAD drawing in the foundation pit CAD drawing set, and the text recognition results are used to obtain the foundation pit description information corresponding to the foundation pit CAD drawing set according to the description rules; Boundary identification is performed on each CAD drawing in the foundation pit CAD drawing set, and the foundation pit outer contour drawing set corresponding to the foundation pit CAD drawing set is obtained based on the region obtained by region identification. For each CAD drawing in the foundation pit CAD drawing set, region identification is performed, and the drilling outline drawing set corresponding to the foundation pit CAD drawing set is obtained based on the drilling area obtained by region identification. Training data is obtained by associating the foundation pit description information, the foundation pit outer contour map set, and the drilling contour map set based on graph identifiers.

3. The method as described in claim 1, characterized in that, The step of performing data augmentation on the training data to obtain the augmented image corresponding to the training data includes: Noise is added to each outer contour map of the foundation pit and each hole contour map in the training data to obtain a noise image corresponding to each outer contour map of the foundation pit and each hole contour map. The generated image in the training data is used to label the noisy image to obtain an expanded image corresponding to each training data point.

4. The method as described in claim 3, characterized in that, The process of loading a pre-trained deep learning model and fine-tuning the deep learning model based on the expanded image to obtain a fine-tuned deep learning model includes: Load the pre-trained deep learning model and perform segmentation processing on the extended image to obtain the first training set and the first test set; The deep learning model is trained and its parameters are fine-tuned based on the first training set to obtain the fine-tuned deep learning model. The fine-tuned deep learning model is tested based on the first test set, and the accuracy obtained from the test determines whether the fine-tuning is complete. When the accuracy is determined to be greater than a preset threshold, a fine-tuned deep learning model is obtained; When it is determined that the accuracy is less than or equal to a preset threshold, the step of segmenting the extended image is executed.

5. The method as described in claim 1, characterized in that, The step of training the fine-tuned deep learning model based on the training data and the expanded image, and obtaining the trained foundation pit support scheme generation model upon completion of training, includes: Based on the training data and the extended image, input data is generated to train the fine-tuned deep learning model, wherein the input data includes a combination of the pit description information and at least one of the pit outer contour map and the drilling contour map, or at least one of the pit outer contour map and the drilling contour map. The input data is fed into the fine-tuned deep learning model to obtain the output result; The output results and the generated graph are compared and analyzed, and the parameters of the fine-tuned deep learning model are adjusted according to the comparison and analysis results, so that a well-trained foundation pit support scheme generation model is obtained when training is completed.

6. The method as described in claim 1, characterized in that, The method further includes: The system receives an input CAD drawing of a foundation pit to be processed and preprocesses the CAD drawing to obtain the drawing information contained in the CAD drawing of the foundation pit to be processed, wherein the drawing information includes at least one of the outer contour drawing of the foundation pit and the drilling contour drawing. The image information is input into the trained foundation pit support scheme generation model to output the target scheme image corresponding to the CAD image of the foundation pit to be processed.

7. The method as described in claim 6, characterized in that, After inputting the diagram information into the trained foundation pit support scheme generation model to output the scheme diagram corresponding to the CAD diagram of the foundation pit to be processed, the method further includes: When a scheme adjustment instruction is received, the input description information is received, and the description information and the drawing information are input into the trained foundation pit support scheme generation model to output an updated scheme drawing of the CAD drawing of the foundation pit to be processed.

8. A training device for generating a foundation pit support scheme model, characterized in that, include: The data acquisition module is used to acquire the foundation pit CAD drawing set and the generated drawing corresponding to each CAD drawing in the foundation pit CAD drawing set, and to preprocess each CAD drawing in the foundation pit CAD drawing set to obtain training data, wherein the training data includes the foundation pit outer contour drawing, the drilling contour drawing, the foundation pit description information and the generated drawing; The data processing module is used to perform data expansion processing on the training data to obtain an expanded image corresponding to each training data, wherein the expanded image is a noisy image; The fine-tuning module is used to load a pre-trained deep learning model and fine-tune the deep learning model based on the extended image to obtain a fine-tuned deep learning model, wherein the deep learning model is a Sable Diffusion model. The model training module is used to train the fine-tuned deep learning model based on the training data and the extended image, and to obtain the trained foundation pit support scheme generation model when the training is completed.

9. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a computer program or embedded program stored in the internal storage of the memory and executable on the processor, wherein the processor executes the computer program or embedded program to implement the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program or embedded program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.

Images