Energy pile intelligent design method and electronic device

Abstract

The present application belongs to the field of civil engineering and energy technology, and discloses an energy pile intelligent design method and an electronic device. The present application firstly constructs an energy pile finite element simulation model and an energy pile prediction proxy model; then selects energy pile design parameters, assigns values to the energy pile design parameters, and obtains a plurality of energy pile design parameter groups; inputs the energy pile design parameter groups into the energy pile finite element simulation model for simulation, and forms a sample set together with the energy pile design parameter groups and the simulation results; trains the energy pile prediction proxy model using the sample set; selects an optimization target and a constraint condition, and iteratively calls the energy pile prediction proxy model using a multi-objective optimization algorithm until the energy pile design parameters are optimized through iterative convergence, and the energy pile design parameters are output as optimal design parameters. The present application solves technical problems such as complex multi-physical field coupling effect, large finite element simulation calculation amount, and low optimization efficiency through the ingenious design of the proxy model.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of civil engineering and energy technology, specifically to an intelligent design method and electronic device for energy piles based on artificial intelligence and finite element analysis. This invention addresses the technical challenges in the design process of energy piles, such as complex multi-physics coupling effects, large computational load of finite element simulation, and low optimization efficiency, and proposes a complete intelligent solution. Background Technology

[0002] As a renewable and clean energy technology, energy piles offer significant advantages in improving energy efficiency and generating better economic benefits, making them an important component of ground source heat pump systems. Unlike traditional pile foundation design methods, energy pile design requires consideration not only of the vertical bearing capacity of a single pile but also multiple factors such as heat exchange efficiency and thermo-mechanical coupling effects, significantly increasing design complexity. Currently, energy pile design primarily employs the traditional traversal enumeration method. Engineers set parameters based on experience and then perform finite element simulation verification. If the requirements are not met, parameters are manually adjusted and recalculated. This process typically requires more than 50 iterations, with each finite element simulation taking 2-4 hours, resulting in a design cycle of 2-3 weeks. In terms of finite element analysis, energy piles need to consider the full thermo-mechanical-hydraulic (THM) coupling effect. In actual engineering projects, energy piles are usually part of a pile group, making modeling and boundary condition setting extremely complex. To minimize the impact of assumed boundary conditions on numerical simulation results, the computational domain needs to be expanded, further increasing the computational load. When using finite element simulation software (such as COMSOL) to analyze the thermo-mechanical behavior of energy piles, it is necessary to establish a three-dimensional geometric model of the foundation and the energy pile, which consumes a lot of computational resources and is inefficient. Regarding optimization methods, existing technologies mostly employ single optimization algorithms with limited search space, making it difficult to obtain the global optimum. For complex load conditions, the load loading of the energy pile finite element model has multi-section and multi-directional characteristics, which severely restricts the iterative efficiency of the design optimization process. Traditional optimization methods require a complete finite element simulation for each parameter adjustment, resulting in high computational costs. Energy pile design also faces special problems such as thermal interference phenomena in cylindrical helical energy piles, further increasing the optimization difficulty. In the safety evaluation stage, existing technologies mainly rely on manual review by engineers, which is not only inefficient but also prone to inconsistencies in evaluation results due to subjective factors. The THM coupling behavior analysis of energy piles requires professional knowledge, which is difficult for ordinary designers to fully grasp, potentially leading to safety hazards in the design scheme. Furthermore, energy pile design must comply with multiple standards and specifications; manual verification is prone to overlooking details, affecting the compliance of the design scheme.

[0003] Existing patented technologies, such as CN109458757B's static drilling-based embedded energy pile, primarily focus on structural improvements and fail to address the issue of intelligent design methods. While CN101908172B proposes various intelligent proxy algorithms, these are applied to power market simulation and are not optimized for the specific design characteristics of energy piles. Some research attempts to combine parametric design with finite element analysis to automatically generate analysis models by changing parameters, but this is limited to simple structures and fails to address the complex multi-physics coupling issues of energy piles. Facing the design requirements of large-scale energy pile projects, there is an urgent need to develop an integrated system that combines parametric modeling, efficient simulation, intelligent optimization, and automatic evaluation to overcome existing technological bottlenecks. Summary of the Invention

[0004] This invention provides an intelligent design method and electronic device for energy piles, which solves the defects of low intelligence and low efficiency in the design of energy piles in the prior art, and realizes high-efficiency automatic design of energy piles.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In a first aspect, the present invention provides an intelligent design method for energy piles, comprising the following steps: Construct a finite element simulation model of the energy pile; Construct a predictive agent model for energy piles; Select the energy pile design parameters, assign values ​​to the energy pile design parameters, and obtain several sets of energy pile design parameters; The energy pile design parameter set is input into the energy pile finite element simulation model for simulation, and the energy pile design parameter set and simulation results together constitute a sample set; Train an energy pile prediction proxy model using a sample set; The optimization objective and constraints are selected, and a multi-objective optimization algorithm is used to iteratively call the energy pile prediction proxy model until the iterative process converges and optimizes the energy pile design parameters. The energy pile design parameters are then output as the optimal design parameters.

[0006] In the above-mentioned intelligent design method for energy piles, the step of constructing a finite element simulation model of the energy pile includes: The parameters of the energy pile are determined, including geometric parameters, material parameters, geological parameters, and load parameters; A web-based 3D engine is used as the front end to receive energy pile parameters, construct a geometric model of the energy pile including material parameters, and save it in an exchange data format. An application programming interface (API) is built as the backend. The API is used to call the simulation engine to receive and exchange data formats for energy pile simulation, thus completing the construction of the finite element simulation model of the energy pile.

[0007] In the above-mentioned intelligent design method for energy piles, the finite element simulation model of the energy pile also includes a cache database, which is used to receive and store the exchange data output by the simulation engine.

[0008] In the aforementioned intelligent design method for energy piles, the predictive surrogate model for energy piles includes any one or more combinations of models such as the Multilayer Perceptron (MLP) model, the Multitask Learning MLP model, the Physical Information Neural Network (PIN), the ResNet-34 model, and the attention-enhanced TabNet model. The Multilayer Perceptron (MLP) model is one of the common surrogate models in engineering design optimization, with common types including Response Surface Model (RSM), Kriging model, and Radial Basis Function (RBF). The surrogate model used in this paper is not limited to the Multilayer Perceptron (MLP) model.

[0009] In the above-mentioned intelligent design method for energy piles, the design parameters of the energy piles include pile foundation parameters and heat exchange tube parameters. The pile foundation parameters include any one or more of pile length, pile diameter, and concrete specific heat. The heat exchange tube parameters include any one or more of burial depth, U-tube spacing, and inlet flow velocity.

[0010] In the above-mentioned intelligent design method for energy piles, the optimization objectives include the amount of concrete used in the pile foundation, the safety factor of the pile foundation, and the heat exchange efficiency of the pile foundation.

[0011] In the above-mentioned intelligent design method for energy piles, the constraints include pile foundation constraints, heat exchanger tube constraints, and soil constraints. The pile foundation constraints include any one or more of the following: pile foundation axial compression ratio, temperature-force ratio, and pile side resistance ratio. The heat exchanger tube constraints include any one or more of the following: outlet temperature, minimum flow velocity, and internal pressure. The soil constraints include one or more of the following: maximum pressure and maximum shear force.

[0012] In the above-mentioned intelligent design method for energy piles, the step of assigning values ​​to the design parameters of the energy piles includes the following steps: Determine the design range for each parameter in the energy pile design parameter group; Within the design scope, a spatial filling method based on random numbers was used to sample and obtain several sets of energy pile parameters.

[0013] In the above-mentioned intelligent design method for energy piles, the space filling method based on random numbers includes any one or a combination of simple random sampling, Latin hypercube sampling, optimal Latin hypercube sampling, orthogonal array Latin hypercube design, and Sobol sequence sampling method.

[0014] In the above-mentioned intelligent design method for energy piles, the energy pile prediction agent model adopts the ResNet-34 model, and the ResNet-34 model is trained using an adaptive learning rate strategy. During the training process, physical equations are introduced as constraints to embed the loss function.

[0015] In the above-mentioned intelligent design method for energy piles, the training method of the energy pile prediction proxy model is as follows: Data preparation: Divide the sample set into a training set and a validation set; Model definition: The input layer is constructed by using a fully connected layer to map the input data to the input dimension required by ResNet. The main network of the ResNet-34 model is changed from a two-dimensional convolutional layer to a one-dimensional convolutional layer. The output of ResNet is mapped to the output data through the fully connected layer as the output layer. Loss function definition: The total loss function includes data loss and physical loss. Data loss is the mean square error between the predicted value and the simulation result. Physical loss includes the heat conduction equation and the elasticity equation. Training loop using the training set: Initialize the weight parameters, optimizer, loss function, and maximum number of iterations for the energy pile prediction agent model; Enter the outer loop, iterating for one training cycle (Epoch) each time; The training process uses mini-batch gradient descent, traversing the entire training set in each training cycle and processing a mini-batch of data each time. Set the model to training mode, then enter the inner loop, and divide each training epoch into several mini-batches; Within each mini-batch, forward propagation, loss function calculation, back propagation, and optimizer parameter update are performed sequentially; the inner loop exits after all mini-batches in the training set have been traversed in a training cycle. Set the model to evaluation mode, propagate forward on the validation set, and calculate the validation loss. Validate the model performance using the validation set. Stop training when the validation loss does not improve after several consecutive training epochs, or stop training when the maximum number of iterations for a training epoch is reached.

[0016] In the above-mentioned intelligent design method for energy piles, during the training loop using the training set, the gradient is clipped during backpropagation to prevent gradient explosion; during the training process, the weight of physical loss in the loss function calculation is increased intermittently.

[0017] In the above-mentioned intelligent design method for energy piles, the step of using a multi-objective optimization algorithm to iteratively call the energy pile prediction surrogate model until the iterative optimization of energy pile design parameters converges includes the following steps: Select optimization targets and energy pile design parameters; Define the range and type of variation for the energy pile design parameters; Initialize the parameters for the ant colony algorithm and the genetic algorithm respectively; The ant colony algorithm and genetic algorithm are iterated repeatedly until convergence, and the non-dominated solution set and the corresponding energy pile design parameters are output. Ant colony algorithm and genetic algorithm are preferred implementation schemes of this invention, and their combination with other biomimetic algorithms such as particle swarm optimization is not excluded.

[0018] In each iteration, the improved ant colony algorithm and the adaptive genetic algorithm are executed sequentially, and co-evolution is performed every few iterations; The improved ant colony algorithm introduces a dynamic pheromone update mechanism into the ant colony algorithm, which automatically adjusts the pheromone evaporation rate according to the search progress. The genetic algorithm employs an adaptive crossover mutation probability, which gradually decreases as the number of iterations increases. The co-evolution involves the improved ant colony algorithm and the adaptive genetic algorithm exchanging high-quality solutions every few iterations.

[0019] In the above-mentioned intelligent design method for energy piles, after outputting the energy pile design parameters as the optimal design parameters, it also includes structural safety calculation and intelligent evaluation of the scheme. The structural safety assessment includes preliminary assessment, detailed assessment, compliance verification, and sensitivity analysis. The intelligent evaluation of the proposed solution employs an evaluation method based on a large language model and retrieval enhancement generation technology.

[0020] In a second aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the above-described intelligent design method for energy piles.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention uses Vue.js as the front-end framework for parametric modeling and result rendering, FastAPI as the back-end service framework, Redis as the caching database, and Abaqus 2024 as the finite element analysis engine, achieving deep integration through the Python API. Data exchange between the front-end and back-end is conducted using JSON format to ensure efficient and reliable data transmission.

[0022] (1) Improved design efficiency This system reduces the number of finite element simulations from over 50 in traditional methods to 30-50 initial samples plus a small number of verification calculations through surrogate model technology. The design cycle is shortened from 2-3 weeks to 3-5 days, improving efficiency by over 60%. The surrogate model's prediction speed is 1000 times faster than direct finite element calculations, enabling large-scale parameter searches and solving the problem in traditional methods where "for complex load conditions, the multi-section, multi-directional characteristics of the finite element model's load application restrict the iterative efficiency of the design optimization process."

[0023] (2) Precise implementation of multiphysics coupling analysis The system incorporates a fully coupled THM finite element analysis module, accurately simulating the thermo-mechanical-hydraulic coupling behavior of energy piles under actual working conditions. This avoids the inefficiency caused by the traditional method's approach of "expanding the computational domain to minimize the impact of assumed boundary conditions on numerical simulation results." Adaptive mesh technology is employed to significantly reduce computational load while maintaining accuracy. Specific optimizations have been made for the thermal interference phenomenon of cylindrical helical energy piles, improving the accuracy of heat exchange efficiency prediction.

[0024] (3) Multi-algorithm collaborative optimization mechanism This paper innovatively combines an improved ant colony algorithm with an adaptive genetic algorithm to form a co-evolutionary mechanism, solving the problem that traditional single optimization algorithms are prone to getting trapped in local optima. Through techniques such as "effective optimization space identification and accelerated optimization," the global search capability is significantly improved. The surrogate model incorporates Physical Information Neural Network (PINN) technology, using physical laws as constraints to ensure the physical rationality of the optimization results and avoid the non-physical interpretations that may occur in purely data-driven models.

[0025] (4) Automation and refinement of structural safety assessment This system deeply integrates structural safety verification into the optimization process, achieving a seamless connection between "optimization and verification" and avoiding the tedious process of repeatedly adjusting parameters in traditional design. A dedicated verification module, specifically designed for energy piles, has developed a safety evaluation method under the thermo-mechanical coupling effect, solving the complex problem that, unlike traditional pile foundation design methods, it requires consideration of multiple factors beyond the calculation of the vertical bearing capacity of a single pile. The verification process is highly automated, capable of simultaneously verifying over 30 specification requirements with an accuracy rate exceeding 95%.

[0026] (5) Intelligent evaluation system based on RAG This system innovatively applies RAG technology to the field of engineering specification inspection, constructing a knowledge base containing over 300 specification clauses. Through "multi-agent technology," it achieves intelligent matching and interpretation of specification clauses. The evaluation process is objective and comprehensive, avoiding the subjectivity and omission risks of manual review, improving compliance inspection efficiency by over 80%. The system generates specific and actionable improvement suggestions, directly guiding design optimization and forming a true closed-loop design process.

[0027] (6) Modular architecture and scalability Employing a microservice architecture with front-end and back-end separation, each functional module is relatively independent yet closely collaborative, allowing for flexible configuration based on project needs. The system supports integration with BIM platforms, achieving seamless integration of energy pile design and building lifecycle management. The Redis database design ensures system stability under high concurrency scenarios, supporting the simultaneous processing of design tasks for 50+ projects.

[0028] (7) Knowledge accumulation and inheritance mechanism The system automatically archives design process data and results, forming a dedicated energy pile design knowledge base for the enterprise. By continuously accumulating historical cases, the system can continuously optimize the proxy model and evaluation rules, achieving systematic accumulation and reuse of design experience, effectively reducing reliance on the experience of individual experts, and improving the overall design level of the enterprise.

[0029] This invention, through the aforementioned technological innovations, systematically solves key problems in energy pile design, such as low efficiency, poor accuracy, difficulty in optimization, and subjective evaluation, achieving a fundamental shift from experience-driven to data-driven approaches and from manual trial and error to intelligent optimization. Compared to existing technologies, this system significantly improves design efficiency, solution quality, and compliance assurance, demonstrating outstanding innovation and practicality, and providing a scientific, efficient, and reliable intelligent solution for energy pile engineering design. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the intelligent design method for energy piles in Embodiment 1 provided by the present invention.

[0032] Figure 2 This is a schematic diagram of the training method of the energy pile prediction agent model in Embodiment 1 of the present invention.

[0033] Figure 3This is the training loop logic diagram of the energy pile prediction agent model in Embodiment 1 of the present invention, which utilizes the training set for training.

[0034] Figure 4 This is the logic diagram of the energy pile prediction agent model using a multi-objective optimization algorithm in Embodiment 1 provided by the present invention.

[0035] Figure 5 This refers to the iterative ant colony algorithm and genetic algorithm in Embodiment 1 provided by the present invention, until the convergence logic diagram is obtained.

[0036] Figure 6 This is a schematic diagram of the framework of an intelligent design system for energy piles provided by the present invention.

[0037] Figure 7 This is a schematic diagram of another intelligent design system framework for energy piles provided by the present invention.

[0038] Figure 8 This is a schematic diagram of the software technology architecture in Embodiment 1 provided by the present invention.

[0039] Figure 9 This is a schematic diagram of front-end and back-end data interaction in Embodiment 1 provided by the present invention.

[0040] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0041] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0042] In the description of this invention, it should be noted that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0043] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0044] This invention proposes an intelligent design system for energy piles. By constructing a modular architecture with separate front-end and back-end, it achieves intelligent and automated design throughout the entire energy pile design process. The system adopts a closed-loop optimization mechanism of "data-driven - model proxy - intelligent search - security verification - intelligent evaluation," significantly improving design efficiency and quality.

[0045] Example 1, such as Figure 1 As shown, this invention provides an intelligent design method for energy piles, comprising the following steps: S100, Construct a finite element simulation model of the energy pile; S200, Constructing an energy pile prediction proxy model; S300. Select the energy pile design parameters, assign values ​​to the energy pile design parameters, and obtain several energy pile design parameter groups. S400. Input the energy pile design parameter set into the energy pile finite element simulation model for simulation, and form a sample set together with the energy pile design parameter set and simulation results. S500, using a sample set to train an energy pile prediction agent model; S600: Select the optimization objective and constraints, and use a multi-objective optimization algorithm to iteratively call the energy pile prediction proxy model until the iterative optimization of the energy pile design parameters is converged, and output the energy pile design parameters as the optimal design parameters.

[0046] In step S100, the finite element simulation model of the energy pile includes a parametric modeling module and a finite element simulation analysis module. The specific construction process is as follows: S110. Determine the energy pile parameters, which include simulated pile foundation parameters, geological parameters, and simulated heat exchanger pipe parameters. The simulated pile foundation parameters include geometric parameters, material parameters, and load parameters. In some embodiments, geometric parameters include pile diameter, pile length, helix angle, heat exchange tube arrangement, etc. For circular piles, the pile diameter includes outer diameter and inner diameter; for square piles, the pile diameter means horizontal width and vertical width.

[0047] In some embodiments, material parameters include concrete strength grade, heat exchanger tube material, concrete specific heat, thermal conductivity, etc.

[0048] In some embodiments, geological parameters include soil layer distribution, thermal properties, etc.

[0049] In some embodiments, load parameters include vertical load, horizontal load, thermal load, etc.

[0050] The simulated heat exchanger tube parameters include any one or more of the following: heat exchanger tube type (single U, double U, W, spiral, etc.), burial depth, single tube spacing (e.g., U-tube spacing), heat exchanger tube thickness, heat exchanger tube material, inlet flow velocity, etc.

[0051] In some embodiments, the geological model adopts a layered modeling approach, automatically generating a three-dimensional geological model based on the input stratum thickness and material properties, with the model boundary dimensions being at least 20 times the maximum size of the pile foundation and more than twice the pile length.

[0052] S120 uses a web-based 3D engine as the front end to receive energy pile parameters, construct a geometric model of the energy pile including material parameters, and save it in an exchange data format to obtain the parametric modeling module.

[0053] In some embodiments, the parametric modeling module is developed based on the Vue3 framework, and the 3D engine can be any of the existing technologies, such as the Three.js graphics engine, the Babylon.js graphics engine, or the Cesium graphics engine; for example, the Three.js graphics engine, a standard library in the WebGL domain, is used.

[0054] In the specific implementation process, a parameter input interface can be created, adopting a tabular data input and drop-down selection interaction design to improve the user experience. After the user inputs the energy pile parameters, the 3D engine automatically generates a three-dimensional geometric model of the energy pile through a parametric algorithm and displays a model preview in real time. Specifically targeting the thermal interference phenomenon of spiral energy piles, a built-in thermal interference correction coefficient calculation model is implemented to automatically adjust the spacing of the heat exchange tubes.

[0055] In some embodiments, the exchanged data format can be JSON, glTF, or IFC. For example, in this embodiment, JSON format is used.

[0056] S130. Construct an application programming interface (FastAPI) as the backend, and use the application programming interface to call the simulation engine to receive and exchange data formats for energy pile simulation, thereby completing the construction of the energy pile finite element simulation model.

[0057] For example, the finite element simulation analysis module is developed based on the Python API of Abaqus 2024, achieving seamless integration with the FastAPI backend. The finite element simulation analysis module receives the geometric model and material parameters generated by the parametric modeling module, automatically constructs the finite element model, including mesh generation for the concrete pile, U-shaped heat exchange pipe, and layered geological body, as well as precise settings for the material constitutive model (such as the Cambridge model for soil). The simulation process considers the thermo-mechanical coupling effect, calculating the stress, deformation, and heat exchange efficiency of the pile foundation during heat exchange. Simulation results are cached through Redis for subsequent modules to access.

[0058] To address the fully coupled thermal-mechanical-hydraulic (THM) characteristics of energy piles, a coupled finite element model of the thermal-mechanical-hydraulic multiphysics field was established, considering pile-soil interaction, heat exchange processes, and moisture migration effects. To improve computational efficiency, adaptive meshing technology was employed, with finer meshes in key areas (such as the pile-soil interface and around heat exchange pipes) and coarser meshes in other areas. The finite element simulation analysis module automatically converted the geometric model generated by the parametric modeling module into an Abaqus-recognizable input file and called the Abaqus kernel for calculation. During the calculation, the system monitored the calculation progress and resource usage in real time, automatically adjusting the calculation strategy or prompting the user when insufficient computing resources were detected. The calculation results, including key indicators such as temperature field distribution, stress-strain state (pil axial compression, pile side resistance), displacement field, and heat exchange efficiency, were returned to the front end in structured JSON format.

[0059] In some embodiments, the energy pile finite element simulation model further includes a cache database, which is used to receive and store the exchange data output by the simulation engine. Specifically, the data output by the finite element simulation analysis module is returned to the front end in JSON format and stored in the cache database. The cache database can be any one of Redis, MongoDB, or PostgreSQL databases; in this embodiment, it is a Redis database.

[0060] In step S200, the energy pile prediction agent model includes any one or more combined models among the following: multilayer perceptron (MLP) model, multi-task learning MLP model, physical information neural network, ResNet-34 model, and attention enhancement model TabNet.

[0061] For example, the energy pile prediction proxy model is an improved ResNet-34 model, which is modified from the standard ResNet-34 model as follows: The input layer is constructed by mapping the input data to the required input dimensions of ResNet using a fully connected layer. For example, if there are a total of 7 energy pile design parameters, then the input dimension of this fully connected layer is 7. The main network of the ResNet-34 model is changed from a two-dimensional convolutional layer (2D) to a one-dimensional convolutional layer (1D), including convolutional layers, pooling layers, and batch normalization layers, all of which are changed to one-dimensional (1D), and the residual blocks are also adjusted accordingly to 1D. Parameters such as convolutional kernel size and stride are adjusted accordingly (e.g., 3x3 convolution is changed to 3, stride 2 is changed to 2, etc.). The output of ResNet is mapped to the output data as the output layer through the fully connected layer. If the output result has 4 parameters, then the output dimension of this fully connected layer is 4.

[0062] Here are some specific modification steps, as an example: Convolutional layers: Replace all 2D convolutional layers Conv2d with 1D convolutional layers Conv1d, and change the kernel size from 2D (e.g., 3x3) to 1D (e.g., 3).

[0063] Batch normalization: Replace BatchNorm2d with BatchNorm1d.

[0064] Pooling layer: Replace MaxPool2d with MaxPool1d, and adjust the kernel size and stride from two dimensions to one dimension.

[0065] Adjust the residual blocks: The convolutional layers in the residual blocks and the convolutional layers in the shortcut connections should be changed to 1D.

[0066] Original ResNet-34 structure: Initial convolutional layer: 7x7 convolution, stride 2, padding 3, output channels 64, then batch normalization and ReLU, followed by 3x3 max pooling (stride 2).

[0067] Four stages (layer1, layer2, layer3, layer4), each stage consisting of multiple residual blocks: Layer 1: 3 residual blocks, each block has two 3x3 convolutions, and the output channels are 64 (Note that downsampling is not performed in the first stage because the shortcut connection of the first residual block may use 1x1 convolutions to adjust the dimensions, but since the input and output channels are the same here, it may not be necessary). Layer 2: 4 residual blocks, each block has two 3x3 convolutions. The first residual block halves the feature map size and doubles the number of channels (128). Therefore, the shortcut connection uses 1x1 convolutions to adjust the channels and stride.

[0068] layer 3: 6 residual blocks, 256 channels; layer4: 3 residual blocks, 512 channels; Global average pooling is used, followed by output from a fully connected layer.

[0069] Improved ResNet-34 model modified to 1D: Initial convolutional layer: Change the kernel size to 7, stride to 2, padding to 3 (note that the padding for one-dimensional convolution is 3 on each side, so that the output length is about half the input length).

[0070] Pooling layer: kernel size 3, step size 2, fill size 1.

[0071] Convolutional layers in residual blocks: use 3-dimensional convolutions, padded with 1 (keeping the length constant unless stride > 1).

[0072] In addition, downsampling in the residual block is achieved by adjusting the convolution stride, while 1x1 convolutions may be used in the shortcut connection to adjust the number of channels and length.

[0073] The improved ResNet-34 model employs an adaptive learning rate strategy during training and introduces Physical Information Neural Network (PINN) technology, embedding physical laws such as heat conduction equations and elasticity equations as constraints into the loss function.

[0074] In step S300, the energy pile design parameters include pile foundation parameters and heat exchanger tube parameters. The pile foundation parameters include any one or more of the following: pile length, pile diameter, concrete specific heat, thermal conductivity coefficient, graphite and other additives. The heat exchanger tube parameters include any one or more of the following: heat exchanger tube type (single U, double U, W, spiral, etc.), burial depth, single tube spacing, heat exchanger tube thickness, heat exchanger tube material, inlet flow velocity, etc.

[0075] In step S300, the design parameters of the energy piles are assigned values ​​to obtain several sets of energy pile design parameters, including the following steps: S310. Determine the design range of each parameter in the energy pile design parameter group; For example, the pile foundation parameters include pile length, pile diameter, and concrete specific heat, while the heat exchange tube parameters include burial depth, U-tube spacing, and inlet flow velocity. The design range is determined as follows: Pile length: lower limit 5m, upper limit 20m; The pile diameter has a lower limit of 0.5m and an upper limit of 1.2m. Specific heat of concrete, lower limit 0.3 Upper limit 0.8 ; Burial depth: lower limit 1m, upper limit 5m; The spacing between U-tubes is 0.1m at the lower limit and 0.5m at the upper limit. Inlet flow rate: lower limit 0.5 m / s, upper limit 2 m / s; S320. Within the design scope, a spatial filling method based on random numbers is used to sample and obtain several sets of energy pile parameters.

[0076] The space filling method based on random numbers includes any one or a combination of several of the following: simple random sampling, Latin hypercube sampling, optimal Latin hypercube sampling, orthogonal array Latin hypercube design, and Sobol sequence sampling method.

[0077] For example, Latin hypercube sampling is used to generate 30-50 representative sample points, i.e., 30-50 representative energy pile parameter sets. Latin hypercube sampling can achieve efficient, uniform, and unbiased exploration of multi-parameter, high-dimensional design spaces with relatively few sample points. This is crucial for computationally expensive numerical simulations (such as finite element analysis of energy piles), significantly reducing the cost of finite element analysis. Within a limited simulation budget (30-50 simulations), it maximizes the acquisition of information about the parameter space, ensuring the reliability of analysis results (such as sensitivity ranking).

[0078] In step S400, the simulation results include at least the optimization objective and constraints. The process of inputting the energy pile design parameter set into the energy pile finite element simulation model is implemented using web technology. A parameter input interface is created using web technology, and a tabular data input and drop-down selection interactive design is adopted. The input data is modeled by the 3D engine of the energy pile finite element simulation model, and the simulation engine is called to receive and exchange data formats for energy pile simulation.

[0079] The optimization objectives include maximizing heat exchange efficiency, minimizing concrete usage, and minimizing carbon emissions; the constraints include geometric constraints (such as pile diameter range), physical constraints (such as maximum deformation limit), and material constraints (such as strength requirements).

[0080] For example, the optimization objectives include at least concrete volume, pile foundation safety factor, and heat exchange efficiency.

[0081] The constraints include the axial compression ratio of the pile foundation, the temperature-force ratio, the pile side resistance ratio, the outlet temperature (U-tube), the minimum flow velocity (inlet), the pressure inside the pipe, the maximum pressure, and the maximum shear force.

[0082] In step S500, such as Figure 2 As shown, the training method for the energy pile prediction proxy model is as follows: S510. Data preparation: Divide the sample set into a training set and a validation set; S520, Model Definition: Change the ResNet-34 model from a 2D network to a 1D network; details are as follows: The input layer is constructed by mapping the input data to the required input dimension of ResNet using a fully connected layer. The main network of the ResNet-34 model is changed from a two-dimensional convolutional layer to a one-dimensional convolutional layer. The output of ResNet is mapped to the output data through the fully connected layer as the output layer. S530. Definition of loss function: The total loss function includes data loss and physical loss. Data loss is the mean square error between the predicted value and the simulation result. Physical loss includes the heat conduction equation and the elasticity equation. S540. Perform a training loop using the training set until the model converges, and output the trained model; the loop flowchart is as follows. Figure 3 As shown, the specific steps are as follows: S541. Initialization: Initialize the weight parameters of the energy pile prediction agent model, and initialize the optimizer, loss function, and maximum number of iterations, etc.

[0083] S542, Enter the outer loop, iterating one training cycle (Epoch) each time; S543. Set the model to training mode and then enter the inner loop. Each training epoch is divided into several small batches (for example, each batch size is 16). S544. Within each mini-batch, perform forward propagation, loss function calculation (see step S530 for details), backpropagation, and parameter update (through the optimizer) in sequence. After traversing all mini-batches in the training set within a training cycle, exit the inner loop. S545. Set the model to evaluation mode, propagate forward on the validation set, and calculate the validation loss (calculation method is the same as step S544). Validate the model performance using the validation set. Stop training if the validation loss does not improve after several consecutive training cycles, such as 20 consecutive cycles; or stop training when the maximum number of iterations in the training cycle is reached.

[0084] In some embodiments, the learning efficiency is adjusted every 10 training cycles, for example, the initial learning rate is set to 0.001 and decays by 10% every 50 epochs.

[0085] In step S510, in some embodiments, the training set is 80%, the validation set is 20%, or the validation set is 10%, and an additional 10% test set is set. After training, the model performance is tested using the test set. If the model accuracy meets the requirements, the model parameters with the minimum validation loss are used as the final training model parameters; otherwise, the energy pile prediction agent model needs to be retrained.

[0086] In step S530, the total loss function ,in, For data loss, For physical loss, The physical loss weights are used; the data loss is the mean squared error (MSE).

[0087] Physical losses include the heat conduction equation and the elasticity equation. The heat conduction equation is essentially a manifestation of the law of conservation of energy. For energy piles, one-dimensional or two-dimensional models are usually considered. Taking one-dimensional (along the pile depth direction z) as an example, its governing equations are as follows: Formula (1) The temperature field of the pile body is a function of position z and time t, and it is also one of the core variables that the neural network needs to predict. : The partial derivative of temperature with respect to time (transient term), representing the rate of change of temperature at a certain point with time. If steady-state conditions are considered, this term is zero.

[0088] : The second partial derivative of temperature with respect to spatial position z (diffusion term), representing the net heat inflow due to thermal conduction.

[0089] : Thermal conductivity coefficient of the pile material.

[0090] : Volumetric heat capacity of the pile material (density ρ × specific heat capacity c).

[0091] Internal heat source: For energy piles, this represents the heat source generated by the heat exchange between the circulating fluid and the pile body.

[0092] In this invention, the internal heat source term It is not an externally given constant, but rather determined by design parameters (which determine the heat transfer coefficient and geometry) and the local pile temperature predicted by the network. Coupling terms that are jointly determined. By embedding the law of convective heat transfer into the physical loss function, a deep integration of the heat-fluid boundary conditions and the heat conduction equation of the pile body is achieved.

[0093] The process of converting the heat conduction equation into physical loss is as follows: During training, the energy pile prediction proxy model receives input energy pile design parameters (such as location z, time t, material properties, boundary condition parameters, etc.) and outputs the predicted temperature. .

[0094] Temperature is calculated using the automatic differentiation function of a deep learning framework. First-order partial derivative with respect to t ( ) and the second-order partial derivative with respect to z ( ).

[0095] Substitute T and its derivative into formula (1) to calculate the residual: Formula (2) The physical loss of heat conduction is the mean of this residual over a large number of configuration points: Formula (3) The equations of elasticity are based on the law of conservation of momentum. For energy piles, the focus is typically on thermal stress and thermal deformation caused by temperature changes. The governing equations consist of three parts: equilibrium equations, geometric equations, and constitutive equations.

[0096] The equilibrium equations describe the balance relationship of forces within the pile. For a one-dimensional model, the equations are as follows: Formula (4) Axial stress on the pile cross section is another prediction target of the energy pile prediction proxy model.

[0097] f: Volume force (such as gravity), which is usually negligible or considered a constant in a one-dimensional model.

[0098] The geometric equations (strain-displacement relationship) describe the relationship between deformation (strain) and displacement. The equations are in the following form: Formula (5) u The axial displacement of the pile body is the prediction target of the energy pile prediction proxy model.

[0099] The axial strain of the pile can be obtained from the displacement u through automatic differentiation.

[0100] The constitutive equation (stress-strain-temperature relationship) relates the mechanical response (stress, strain) to the thermal response (temperature). For linear elastic materials, considering thermal effects, the most commonly used is the Duhamel-Neumann constitutive relation, with the following equation form: Formula (6) E: Elastic modulus of the pile material.

[0101] : Coefficient of thermal expansion of the pile material.

[0102] The current temperature T relative to a certain reference temperature The change in (such as initial ground temperature).

[0103] The method for converting the elasticity equations into physical losses is as follows: Displacement field predicted using an energy pile predictive proxy model and temperature field .

[0104] Strain and stress are calculated using automatic differentiation: (Predicted strain is obtained from geometric equations); (The “predicted stress” is obtained from the constitutive equation); Calculate the residuals: Predict stress Substituting into the equilibrium equation, we obtain the equilibrium equation residuals. ; If the energy pile predictive proxy model directly predicts stress at the same time Then we can add a residual to ensure that it satisfies the constitutive relation. The constitutive equation residual is: This provides stronger physical constraints for the model.

[0105] The physical loss in elasticity is a combination of these residuals: ,in It is the weighting coefficient.

[0106] It should be noted that the validation set can also be set to 10%, and the remaining 10% to be used as the test set. After training, the trained energy pile prediction proxy model can be tested using the test set. During the testing process, the coefficient of determination is calculated. ,when When the coefficient of determination is 0.95, the model accuracy is considered to meet the requirements. The formula for calculating the coefficient of determination is as follows: Formula (7) in, For the sum of squared residuals, , The actual value (label in the test set). This is a predicted value; For the total sum of squares, ), This is the average value.

[0107] In step S600, the optimization objectives include maximizing heat exchange efficiency, minimizing concrete usage, and minimizing carbon emissions; the constraints include geometric constraints (such as pile diameter range), physical constraints (such as maximum deformation limit), and material constraints (such as strength requirements).

[0108] The optimization objectives include the amount of concrete used in the pile foundation, the safety factor of the pile foundation, and the heat exchange efficiency.

[0109] The constraints include pile foundation constraints, heat exchanger tube constraints, and soil constraints. The pile foundation constraints include any one or more of the following: pile foundation axial compression ratio, temperature-force ratio, and pile side resistance ratio. The heat exchanger tube constraints include any one or more of the following: outlet temperature, minimum flow velocity, and tube pressure. The soil constraints include one or more of the following: maximum pressure and maximum shear force.

[0110] Select the optimization objective and constraints, and use a multi-objective optimization algorithm to iteratively call the energy pile predictive proxy model until the iterative process converges and optimizes the energy pile design parameters. The output energy pile design parameters are then considered the optimal design parameters. The flowchart is as follows: Figure 4 As shown, the specific steps include: S610. Select optimization objectives and energy pile design parameters; S620. Define the range and type of variation of energy pile design parameters; S630. Initialize the parameters for the ant colony algorithm and the genetic algorithm respectively; S640, iterative ant colony algorithm and genetic algorithm, until convergence, output non-dominated solution set and corresponding energy pile design parameters; In each iteration, the improved ant colony algorithm and the adaptive genetic algorithm are executed sequentially, and co-evolution is performed every few iterations; The improved ant colony algorithm introduces a dynamic pheromone update mechanism into the ant colony algorithm, which automatically adjusts the pheromone evaporation rate according to the search progress. The genetic algorithm employs an adaptive crossover mutation probability, which gradually decreases as the number of iterations increases. The co-evolution involves the improved ant colony algorithm and the adaptive genetic algorithm exchanging high-quality solutions every few iterations.

[0111] In step S610, an exemplary energy pile design parameter includes a design parameter vector consisting of pile length, pile diameter, concrete specific heat, embedment depth, U-tube spacing, and inlet flow velocity. The optimization objectives are chosen to be maximizing heat exchange efficiency, minimizing concrete usage, and minimizing carbon emissions, denoted as follows: .

[0112] In step S620, the range of variation for the energy pile design parameters is set according to relevant specifications or experience. For example, the design range is determined as follows: Pile length: lower limit 5m, upper limit 20m; The pile diameter has a lower limit of 0.5m and an upper limit of 1.2m. Specific heat of concrete, lower limit 0.3 Upper limit 0.8 ; Burial depth: lower limit 1m, upper limit 5m; The spacing between U-tubes is 0.1m at the lower limit and 0.5m at the upper limit. Inlet flow rate: lower limit 0.5 m / s, upper limit 2 m / s; Variation types include discrete and continuous. Continuous variables are usually encoded using real numbers, meaning the actual value of the variable is used directly. In genetic algorithms, real number encoding can be used directly, or binary encoding can be used (but modern algorithms mostly use real number encoding). In ant colony algorithms, continuous variables need to be discretized, for example, by dividing the variable range into multiple bins, each bin representing a state.

[0113] Discrete variables directly use discrete values, which can be integers (e.g., material grade) or categories (e.g., layout). In genetic algorithms, discrete variables can be represented directly by integers or symbols, or by binary encoding (e.g., using several bits of binary digits to represent several discrete options). In ant colony algorithms, discrete variables are naturally suitable, and each discrete value can serve as a state.

[0114] In step S630, the ant colony algorithm (ACO) parameters include the number of ants m, the pheromone concentration (m / m), and the ant colony density (m / m). ), Importance of pheromones Importance of Heuristic Information pheromone volatilization rate (Dynamically updated, initial value) ), pheromone constant Q (Used for updating pheromones), maximum number of iterations .

[0115] Genetic algorithm (AGA) parameters include population size (NP) and crossover probability. (Adaptive, initial value) Probability of mutation (Adaptive, initial value) ) and maximum number of iterations .

[0116] Initialization is performed to generate an initial solution, as follows: Ant colony algorithm: Randomly generate m ants (solution), for example, pheromone concentration is... =100, pheromone evaporation rate =0.1.

[0117] Genetic Algorithm: Randomly generate NP=50 individuals, crossover rate =0.8, variability =0.1.

[0118] Set the maximum number of iterations For example, set it to 200 times.

[0119] In step S640, the ant colony algorithm and genetic algorithm are iterated repeatedly until convergence, outputting the non-dominated solution set and the corresponding energy pile design parameters. The flowchart is as follows. Figure 5 As shown, the specific steps are as follows: S641. Set the iteration variable t=1, with a maximum value of t. Initialize the Pareto front; S642, Ant Colony Algorithm Steps: (1) Calculate the dynamic volatility ; (2) Ant path construction (differentiation of continuous / discrete variables. For continuous variables, probability-based sampling can be used, such as using a Gaussian distribution, with the mean determined by pheromone and heuristic information). (3) Evaluate the solution quality using a surrogate model; (4) Pheromone volatilization and enhancement / renewal, volatilization rate ; (5) Record the Pareto front of the current iteration; S643, Genetic Algorithm Steps: (1) Selection: Select the parent generation based on fitness (objective function and constraint handling).

[0120] (2) Crossover: Adaptive crossover probability is used. Or in a similar manner, the crossover probability gradually decreases as iterations proceed.

[0121] (3) Mutation: Adaptive mutation probability is used. Or similarly, as iterations proceed, the mutation probability gradually increases. Typically, to maintain diversity, the mutation probability may not continuously decrease; this depends on the design. A common adaptive strategy is to adjust the crossover and mutation probabilities according to the relationship between individual fitness and the average fitness of the population, such as using a linear variation related to the number of iterations.

[0122] (4) Use a proxy model to evaluate the objective function and the degree of constraint violation of the new population.

[0123] (5) Environmental selection: Selecting the next generation of population.

[0124] (6) Record the Pareto front of the current iteration.

[0125] S644, Co-evolution: If t is a multiple of 10, then information exchange between algorithms is performed.

[0126] (1) Elite solution quality assessment and screening; (2) Select a certain number of solutions (such as the top 10% of solutions) from the Pareto front of the ant colony algorithm to replace the poor solutions (such as the bottom 10% of solutions) in the genetic algorithm population, and perform pheromone matrix enhancement.

[0127] (3) Select a certain number of solutions from the Pareto front of the genetic algorithm to replace the poor solutions in the ant colony algorithm and perform population replacement.

[0128] S645. Update the Pareto optimal frontier, which includes non-dominated sorting, crowding calculation and frontier solution set maintenance. S646. Algorithm convergence test: If one of the following conditions is met, stop the loop and output the final non-dominated solution set (Pareto front). Condition 1: The number of iterations t reaches the set maximum number of iterations. ; Condition 2: Pareto front convergence is determined when the objective function improves by less than 0.1% in a number of consecutive iterations (e.g., 20 iterations). Condition 3: The quality of the solution meets the requirements.

[0129] like Figure 6 As shown, the present invention uses the above method to obtain an intelligent design system for energy piles, including... The finite element simulation model of the energy pile is used to obtain simulation results by performing simulation analysis based on the parameters of the energy pile. The energy pile prediction proxy model is used to make predictions based on the energy pile design parameters and output the prediction results. The energy pile design parameter optimization module is used to iteratively call the energy pile predictive proxy model using a multi-objective optimization algorithm based on the selected optimization objective and constraints, until the iterative optimization of the energy pile design parameters is converged, and the energy pile design parameters are output as the optimal design parameter set.

[0130] For example, the finite element simulation model of the energy pile includes a parametric modeling module and a finite element simulation analysis module. The parametric modeling module is developed based on the Vue3 framework, and the 3D engine can be any of the existing technologies, such as the Three.js graphics engine, the Babylon.js graphics engine, or the Cesium graphics engine; for example, the Three.js graphics engine, a standard library in the WebGL domain, is used.

[0131] The finite element simulation analysis module is developed based on the Python API of Abaqus 2024, achieving seamless integration with the FastAPI backend. The module receives the geometric model and material parameters generated by the parametric modeling module, automatically constructs the finite element model, including mesh generation for the concrete pile, U-shaped heat exchange pipe, and layered geological body, as well as precise settings for the material constitutive model (such as the Cambridge model for soil). The simulation process considers thermo-mechanical coupling effects, calculating the stress, deformation, and heat exchange efficiency of the pile foundation during heat exchange. Simulation results are cached in Redis for subsequent modules to access.

[0132] like Figure 7 As shown, this invention provides another intelligent design system for energy piles, which further includes, based on the above system, The structural safety calculation module is used to perform structural safety calculations on the design parameter set output by the energy pile design parameter optimization module to assess whether it meets the design requirements. If it does not meet the requirements, it needs to be flushed and optimized through the energy pile design parameter optimization module until the requirements are met. During the re-optimization process, new optimization targets can be added based on the unmet requirements.

[0133] Structural safety assessment includes preliminary assessment, detailed assessment, compliance verification, and sensitivity analysis, as detailed below: Preliminary assessment: Based on simplified formulas, quickly evaluate the feasibility of the proposed solutions and screen out those that clearly do not meet the requirements.

[0134] Detailed calculation: For schemes that pass the preliminary calculation, the pile foundation-specific finite element submodule is called for detailed analysis, focusing on checking the pile stress, displacement and stability under the thermo-mechanical coupling effect.

[0135] Standard Compliance Verification: Based on standards such as the *Code for Design of Building Foundations* (GB50007) and the *Technical Code for Ground Source Heat Pump System Engineering* (GB50366), the system automatically verifies whether the design scheme meets all requirements. The calculations include key indicators such as pile foundation deformation, stress, axial force, pile-soil friction, and pile end bearing capacity along the depth direction, generating data tables and graphs. The calculation process strictly follows standards such as the *Code for Design of Pile Foundations* to ensure the design scheme meets safety requirements. The calculation results are returned to the front end in JSON format and visualized by the results rendering and display module.

[0136] Sensitivity analysis: Perform ±10% disturbance analysis on key parameters to assess the robustness of the proposed solution. The results are output in a structured report, clearly identifying any non-compliance items and providing improvement suggestions.

[0137] The intelligent evaluation module is used for evaluation based on large language models and retrieval enhancement generation technology; the evaluation method based on large language models and retrieval enhancement generation technology is detailed in the intelligent evaluation steps of Example 2.

[0138] This module innovatively applies large language models and RAG (Retrieval Enhanced Generation) technology to engineering scheme evaluation, enabling automatic evaluation of design schemes: Standards and Norms Knowledge Base: This base collects and organizes relevant domestic and international standards and norms, including over 300 specific clauses, and constructs a structured knowledge graph containing information such as clause content, applicable conditions, and calculation formulas. It also vectorizes and stores standard documents such as the "Code for Design of Pile Foundations."

[0139] Prompt word engineering: Design multi-level prompt word templates, including basic information extraction templates, specification matching templates, problem diagnosis templates, and improvement suggestion generation templates. The basic information extraction template format is: "Extract key parameters from the following energy pile design schemes: {parameter list}, design conditions: {condition list}, performance indicators: {indicator list}". Design multi-role prompt word templates (experts, construction engineers, etc.) to support scheme evaluation from different perspectives.

[0140] Specifically, the prompt template is as follows: I am an energy pile design expert, construction engineer, structural design engineer, energy solution engineer, etc., and I am currently conducting energy pile scheme optimization, scheme design, scheme safety calculation, etc. The type of energy pile I designed is cast-in-place pile, precast pipe pile, the diameter or width of the pile foundation is 1m, 2m, etc., the inner diameter is 0.5m, 1m, etc., the information of the U-shaped energy transducer is the diameter 0.08m, the wall thickness is 0.01m, the arrangement is single U, double U, triple U, etc., the geological information is [first layer, clay, bearing capacity 200KPa, skin friction 20KPa, elastic modulus 200000MPa, ...] ...}, the standard based on the safety calculation is {《Pile Foundation Design Code》, ...}, the calculation result is that the maximum bearing capacity of the pile foundation is 4000KN, the maximum deformation is 0.02m, please conduct an evaluation according to the relevant standards.

[0141] RAG retrieval mechanism: Employing a hybrid retrieval strategy combining BM25 and vector retrieval, it performs semantic understanding of user queries and uses RAG technology to retrieve relevant specification clauses from the knowledge base based on input design parameters and calculation results. Vector retrieval uses the Sentence-BERT model to generate text vectors with a dimension of 768.

[0142] Evaluation Report Generation: Based on the search results and design solutions, LLM generates a comprehensive evaluation report that includes compliance scores, problem diagnoses, and specific improvement suggestions. The report adopts a three-part structure of "problem-basis-suggestion" to ensure the traceability and actionability of the evaluation results. It supports integration with multiple large language models such as Deepseek, QWEN, KIMI, and GPT, improving evaluation accuracy through comparative analysis.

[0143] The results rendering and display module is used to display the design parameter groups passed by the structural safety calculation module and the intelligent evaluation module in a three-dimensional visualization.

[0144] The results rendering and display module is responsible for presenting the design results to users in a 3D visualization, including the geometry of the model and 3D cloud maps showing the data on the geometry. 3D rendering technology is an indispensable element in multimedia exhibition halls, not only enhancing the display effect but also bringing a brand-new interactive experience to the audience. The rendered display also includes curves of pile foundation deformation, stress, axial force, pile-soil skin friction, and pile end support force along the depth direction, as well as dynamic process diagrams of feasible solutions iterating towards the Pareto front during multi-objective optimization.

[0145] In this embodiment of the invention, the software technical architecture is as follows: Figure 8 As shown in the diagram, the front-end and back-end data interaction is as follows: Figure 9 As shown, a front-end and back-end separation architecture is adopted. The front-end is based on the Vue3 framework for user interaction, while the back-end provides RESTful API services through FastAPI. A Redis database serves as middleware, storing model data, simulation results, and intermediate calculation results. A Hash data structure is used to store the parameterized model, and a Sorted Set is used to store candidate solutions during the optimization process. Data is transferred between the front-end and back-end via JSON format. Key data structures are strictly defined to ensure data consistency. The system workflow forms a closed loop: parameter input → modeling → simulation → optimization → verification → evaluation → parameter adjustment, until the optimal solution is obtained.

[0146] After outputting the energy pile design parameters as the optimal design parameters, the following steps are also included: S700, structural safety calculation, including preliminary calculation, detailed calculation, standard compliance verification, and sensitivity analysis; input the optimized output energy pile design parameters into the structural safety calculation module and generate a geometric model; perform preliminary calculation to quickly screen obviously unqualified schemes; for schemes that pass the preliminary calculation, call the finite element simulation analysis module for detailed analysis; automatically verify whether each indicator meets the requirements according to the standard, generate a detailed calculation report, and mark the unqualified items and safety margins.

[0147] S800 intelligent evaluation of solutions adopts an evaluation method based on large language models and retrieval enhancement generation technology.

[0148] The design parameters and structural safety calculation results are input into the intelligent evaluation module. Prompt words are generated based on the prompt word template. At the same time, the RAG submodule retrieves relevant clauses from the standard library (BM25 weight 0.4, vector retrieval weight 0.6) and inputs them into the large language model. The large language model generates an evaluation report based on the search results, adopting a "problem-basis-suggestion" structure. The report includes a compliance score (0-100 points), a list of key issues, and specific improvement suggestions. The system automatically extracts the key parameter adjustment directions from the improvement suggestions to provide guidance for the next round of optimization.

[0149] Intelligent evaluation of the solution includes: S810, Standards and Norms Knowledge Base: This base collects and organizes relevant domestic and international standards and norms, including over 300 specific clauses, and constructs a structured knowledge graph containing information such as clause content, applicable conditions, and calculation formulas. It also constructs a standards and norms knowledge base, storing standard documents such as the "Code for Design of Pile Foundations" in vectorized form.

[0150] S820, Prompt Word Engineering: Design multi-level prompt word templates, including basic information extraction templates, specification matching templates, problem diagnosis templates, and improvement suggestion generation templates. The basic information extraction template format is: "Extract key parameters from the following energy pile design schemes: {parameter list}, design conditions: {condition list}, performance indicators: {indicator list}". Design multi-role prompt word templates (experts, construction engineers, etc.) to support scheme evaluation from different perspectives.

[0151] Specifically, the prompt template is as follows: I am an energy pile design expert, construction engineer, structural design engineer, energy solution engineer, etc., and I am currently conducting energy pile scheme optimization, scheme design, scheme safety calculation, etc. The type of energy pile I designed is cast-in-place pile, precast pipe pile, the diameter or width of the pile foundation is 1m, 2m, etc., the inner diameter is 0.5m, 1m, etc., the information of the U-shaped energy transducer is the diameter 0.08m, the wall thickness is 0.01m, the arrangement is single U, double U, triple U, etc., the geological information is [first layer, clay, bearing capacity 200KPa, skin friction 20KPa, elastic modulus 200000MPa, ...] ..., the standard based on the safety calculation is the "Pile Foundation Design Code", the calculation result is that the maximum bearing capacity of the pile foundation is 4000KN, the maximum deformation is 0.02m, please conduct an evaluation according to the relevant standards.

[0152] S830 and RAG retrieval mechanisms: A hybrid retrieval strategy combining BM25 and vector retrieval is employed. Semantic understanding of user queries is performed, and RAG technology is used to retrieve relevant specification clauses from the knowledge base based on input design parameters and calculation results. Vector retrieval uses the Sentence-BERT model to generate text vectors with a dimension of 768.

[0153] S840. Evaluation Report Generation: Based on the search results and design scheme, LLM generates a comprehensive evaluation report including compliance scores, problem diagnoses, and specific improvement suggestions. The report adopts a three-part structure of "problem-basis-suggestion" to ensure the traceability and actionability of the evaluation results. It supports integration with multiple large language models such as Deepseek, QWEN, KIMI, and GPT, and improves the accuracy of the evaluation through comparative analysis.

[0154] In this embodiment, the system can automatically adjust the optimization parameter range based on the safety calculation and intelligent evaluation results; incorporate key improvement suggestions into the optimization objective function to form new constraints; and return to step S600 for a new round of optimization, with the maximum number of iterations set to 5. When the convergence condition or the maximum number of iterations is reached, the optimization process terminates, forming a closed loop. The system automatically saves the optimization results of each round, supporting comparative analysis of historical solutions.

[0155] S900, Results Output and Display The system generates a final design report, including a 3D model, key parameters, performance indicators, evaluation results, and charts. It uses Three.js to provide an interactive 3D display of the energy pile model, supporting rotation, scaling, and sectioning. It also offers multi-format export functionality (PDF report, STEP model, Excel data, etc.). The system automatically archives design process data, forming a knowledge base for future project reference.

[0156] An exemplary complete process is as follows: The first step is system initialization.

[0157] Start the Vue frontend service, loading the Three.js rendering engine and UI component library; start the FastAPI backend service, establishing a connection to the Redis database (configuring the connection pool size to 100); load the Abaqus 2024 computing environment and verify the availability of the Python API; initialize the specification standard knowledge base, loading the pre-trained ResNet proxy model and large language model. The second step is parameter input and modeling.

[0158] Users input basic parameters through the front-end interface: geological conditions (soil layer distribution, thermal properties, etc.), load conditions (vertical load 800-2000kN, horizontal load 50-150kN, thermal load 30-60W / m, etc.), and performance requirements (heat exchange efficiency ≥45W / m, safety factor ≥2.0, etc.). The system automatically calculates recommended parameter ranges and provides parameter rationality checks; the parametric modeling module calls the Three.js API to generate a 3D model of the energy pile based on the input parameters; the model data is stored in JSON format, containing complete geometric information and attribute data.

[0159] The third step is to generate finite element samples.

[0160] Based on the input parameter range, the system generates 30-50 representative sample points using Latin hypercube sampling. These sample points are then sent to the finite element simulation module via the FastAPI interface. The finite element simulation module calls the Abaqus Python API to perform a fully coupled THM simulation. The calculation results (temperature field, stress field, displacement field, etc.) are returned in structured JSON format and stored in Redis. For samples that fail to be calculated, the system automatically adjusts the parameters and recalculates to ensure data integrity. The fourth step is to build the proxy model.

[0161] Obtain the sample dataset from Redis and perform data preprocessing (normalization, outlier handling, etc.); initialize the ResNet-34 neural network and configure the input and output layer dimensions; use 80% data for training and 20% data for validation, set the batch size to 16, and the maximum training epochs to 200; introduce a physical constraint loss term to ensure that the surrogate model conforms to basic physical laws; after the model training is completed, calculate the R² value on the test set (target > 0.95) to evaluate the model accuracy.

[0162] Step 5: Intelligent parameter optimization.

[0163] Based on the surrogate model, an ant colony optimization algorithm (pheromone quantity 100, volatility rate 0.1) and a genetic algorithm (population size 50, crossover rate 0.8, mutation rate 0.1) are initialized. The two algorithms run in parallel, exchanging optimal solutions every 10 iterations. The optimization objective is a multi-objective function: maximizing heat transfer efficiency, minimizing material usage, and satisfying safety factor constraints. The system monitors the optimization process in real time and dynamically adjusts the algorithm parameters. Convergence is determined when the objective function improves by less than 0.1% after 20 consecutive iterations, and the optimal parameter combination is output.

[0164] Step 6: Safety verification The optimized parameters are input into the structural safety calculation module to generate a geometric model; a preliminary calculation is performed to quickly screen obviously unqualified schemes; for schemes that pass the preliminary calculation, the pile foundation-specific finite element submodule is called for refined analysis; based on the specifications and standards, the various indicators are automatically verified to see if they meet the requirements, and a detailed calculation report is generated, marking the unmet items and safety margins.

[0165] Step 7: Intelligent evaluation of the solution.

[0166] The design parameters and calculation results are input into the intelligent evaluation module, which generates prompts based on the prompt template. Simultaneously, the RAG submodule retrieves relevant clauses from the standard library (BM25 weight 0.4, vector retrieval weight 0.6) and inputs them into the large language model. The large language model generates an evaluation report based on the search results, using a "problem-basis-suggestion" structure. The report includes a compliance score (0-100 points), a list of key issues, and specific improvement suggestions. The system automatically extracts the key parameter adjustment directions from the improvement suggestions to provide guidance for the next round of optimization.

[0167] Step 8: Closed-loop optimization iteration Based on the safety calculations and intelligent evaluation results, the system automatically adjusts the optimization parameter range; incorporates key improvement suggestions into the optimization objective function to form new constraints; and returns to step 4 for a new round of optimization, with an upper limit of 5 iterations. The optimization process terminates when the convergence condition or the upper limit of iterations is reached. The system automatically saves the optimization results of each round and supports comparative analysis of historical solutions. Step 9: Output and Presentation of Results The system generates a final design report, including a 3D model, key parameters, performance indicators, evaluation results, and charts. It utilizes Three.js to provide an interactive 3D display of the energy pile model, supporting rotation, scaling, and sectioning. Multiple export formats are supported (PDF report, STEP model, Excel data, etc.). The system automatically archives design process data, creating a knowledge base for future project reference. Compared with the prior art, the main technical advantages of the present invention are as follows: (1) Improved design efficiency This system reduces the number of finite element simulations from over 50 in traditional methods to 30-50 initial samples plus a small number of verification calculations through surrogate model technology. The design cycle is shortened from 2-3 weeks to 3-5 days, improving efficiency by over 60%. The surrogate model's prediction speed is 1000 times faster than direct finite element calculations, enabling large-scale parameter searches and solving the problem in traditional methods where "for complex load conditions, the multi-section, multi-directional characteristics of the finite element model's load application restrict the iterative efficiency of the design optimization process."

[0168] (2) Precise implementation of multiphysics coupling analysis The system incorporates a fully coupled THM finite element analysis module, accurately simulating the thermo-mechanical-hydraulic coupling behavior of energy piles under actual working conditions. This avoids the inefficiency caused by the traditional method's approach of "expanding the computational domain to minimize the impact of assumed boundary conditions on numerical simulation results." Adaptive mesh technology is employed to significantly reduce computational load while maintaining accuracy. Specific optimizations have been made for the thermal interference phenomenon of cylindrical helical energy piles, improving the accuracy of heat exchange efficiency prediction.

[0169] (3) Multi-algorithm collaborative optimization mechanism This paper innovatively combines an improved ant colony algorithm with an adaptive genetic algorithm to form a co-evolutionary mechanism, solving the problem that traditional single optimization algorithms are prone to getting trapped in local optima. Through techniques such as "effective optimization space identification and accelerated optimization," the global search capability is significantly improved. The surrogate model incorporates Physical Information Neural Network (PINN) technology, using physical laws as constraints to ensure the physical rationality of the optimization results and avoid the non-physical interpretations that may occur in purely data-driven models.

[0170] (4) Automation and refinement of structural safety assessment This system deeply integrates structural safety verification into the optimization process, achieving a seamless connection between "optimization and verification" and avoiding the tedious process of repeatedly adjusting parameters in traditional design. A dedicated verification module, specifically designed for energy piles, has developed a safety evaluation method under the thermo-mechanical coupling effect, solving the complex problem that, unlike traditional pile foundation design methods, it requires consideration of multiple factors beyond the calculation of the vertical bearing capacity of a single pile. The verification process is highly automated, capable of simultaneously verifying over 30 specification requirements with an accuracy rate exceeding 95%.

[0171] (5) Intelligent evaluation system based on RAG This system innovatively applies RAG technology to the field of engineering specification inspection, constructing a knowledge base containing over 300 specification clauses. Through "multi-agent technology," it achieves intelligent matching and interpretation of specification clauses. The evaluation process is objective and comprehensive, avoiding the subjectivity and omission risks of manual review, improving compliance inspection efficiency by over 80%. The system generates specific and actionable improvement suggestions, directly guiding design optimization and forming a true closed-loop design process.

[0172] (6) Modular architecture and scalability Employing a microservice architecture with front-end and back-end separation, each functional module is relatively independent yet closely collaborative, allowing for flexible configuration based on project needs. The system supports integration with BIM platforms, achieving seamless integration of energy pile design and building lifecycle management. The Redis database design ensures system stability under high concurrency scenarios, supporting the simultaneous processing of design tasks for 50+ projects.

[0173] (7) Knowledge accumulation and inheritance mechanism The system automatically archives design process data and results, forming a dedicated energy pile design knowledge base for the enterprise. By continuously accumulating historical cases, the system can continuously optimize the proxy model and evaluation rules, achieving systematic accumulation and reuse of design experience, effectively reducing reliance on the experience of individual experts, and improving the overall design level of the enterprise.

[0174] This invention, through the aforementioned technological innovations, systematically solves key problems in energy pile design, such as low efficiency, poor accuracy, difficulty in optimization, and subjective evaluation, achieving a fundamental shift from experience-driven to data-driven approaches and from manual trial and error to intelligent optimization. Compared to existing technologies, this system significantly improves design efficiency, solution quality, and compliance assurance, demonstrating outstanding innovation and practicality, and providing a scientific, efficient, and reliable intelligent solution for energy pile engineering design.

[0175] Figure 10 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840. The processor 810, communications interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions from the memory 830 to execute the intelligent design method for energy piles.

[0176] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0177] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intelligent design method for energy piles provided by the above methods.

[0178] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the intelligent design method for energy piles provided by the methods described above.

[0179] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0180] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A smart design method for energy piles, characterized in that, Includes the following steps: Construct a finite element simulation model of the energy pile; Construct a predictive agent model for energy piles; Select the energy pile design parameters, assign values ​​to the energy pile design parameters, and obtain several sets of energy pile design parameters; The energy pile design parameter set is input into the energy pile finite element simulation model for simulation, and the energy pile design parameter set and simulation results together constitute a sample set; Train an energy pile prediction proxy model using a sample set; The optimization objective and constraints are selected, and a multi-objective optimization algorithm is used to iteratively call the energy pile prediction proxy model until the iterative process converges and optimizes the energy pile design parameters. The energy pile design parameters are then output as the optimal design parameters.

2. The intelligent design method for energy piles according to claim 1, characterized in that, The steps for constructing the finite element simulation model of the energy pile include: The parameters of the energy pile are determined, including geometric parameters, material parameters, geological parameters, and load parameters; A web-based 3D engine is used as the front end to receive energy pile parameters, construct a geometric model of the energy pile including material parameters, and save it in an exchange data format. An application programming interface (API) is built as the backend. The API is used to call the simulation engine to receive and exchange data formats for energy pile simulation, thus completing the construction of the finite element simulation model of the energy pile.

3. The intelligent design method for energy piles according to claim 2, characterized in that, The energy pile finite element simulation model also includes a cache database, which is used to receive and store the exchange data output by the simulation engine.

4. The intelligent design method for energy piles according to claim 1, characterized in that, The energy pile prediction agent model includes any one or more combinations of the following models: Multilayer Perceptron (MLP) model, Multi-task Learning MLP model, Physical Information Neural Network, ResNet-34 model, and Attention Enhancement Model TabNet.

5. The intelligent design method for energy piles according to claim 1, characterized in that, The energy pile design parameters include pile foundation parameters and heat exchanger tube parameters. The pile foundation parameters include any one or more of pile length, pile diameter, and concrete specific heat. The heat exchanger tube parameters include any one or more of burial depth, U-tube spacing, and inlet flow velocity.

6. The intelligent design method for energy piles according to claim 1, characterized in that, The optimization objectives include the amount of concrete used in the pile foundation, the safety factor of the pile foundation, and the heat exchange efficiency.

7. The intelligent design method for energy piles according to claim 1, characterized in that, The constraints include pile foundation constraints, heat exchanger tube constraints, and soil constraints. The pile foundation constraints include any one or more of the following: pile foundation axial compression ratio, temperature-force ratio, and pile side resistance ratio. The heat exchanger tube constraints include any one or more of the following: outlet temperature, minimum flow velocity, and tube pressure. The soil constraints include one or more of the following: maximum pressure and maximum shear force.

8. The intelligent design method for energy piles according to claim 1, characterized in that, Assigning values ​​to the design parameters of the energy pile includes the following steps: Determine the design range for each parameter in the energy pile design parameter group; Within the design scope, a spatial filling method based on random numbers was used to sample and obtain several sets of energy pile parameters.

9. The intelligent design method for energy piles according to claim 4, characterized in that, The energy pile prediction proxy model adopts the ResNet-34 model and uses an adaptive learning rate strategy to train the ResNet-34 model. During the training process, physical equations are introduced as constraints to embed the loss function.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the intelligent design method for energy piles as described in any one of claims 1 to 9.

Images