Federal learning-based energy underground subway station design method

Abstract

The present application relates to geothermal energy development and utilization technical field, specifically to a kind of energy underground subway station design method based on federal learning, through parameter collection, local client local model update, calculation result is uploaded to central server, central server carries out update calculation, issue to local client local model, update iteration until global model robustness, the design result of each module is calculated, and heat exchange efficiency and thermal stress are predicted.The energy underground subway station design method proposed in the present application can fully consider the coordination between indoor user end, unit equipment selection and three kinds of energy station structure pipe end in station, and the design scheme obtained can greatly meet the energy-saving and emission-reducing design concept, with low calculation cost, the influence of heat exchange efficiency and heat exchange process on station structure can be accurately evaluated, and compared with traditional test method and field monitoring method, the cost of manpower and material resources can be greatly reduced.

Description

Technical Field

[0001] This invention relates to the field of geothermal energy development and utilization technology, specifically to a design method for underground subway stations based on federated learning. Background Technology

[0002] The energy consumption of the environmental control system in underground subway stations accounts for as much as 30-40% of the total energy consumption of urban rail transit, indicating significant energy-saving potential. Shallow geothermal energy, as a widely distributed and easily developed renewable low-grade heat energy source, is mainly used for building heating and cooling. Replacing traditional air conditioning systems with shallow geothermal energy can reduce the energy consumption of subway station environmental control systems by about one-third.

[0003] Energy-powered underground subway stations utilize an emerging technology that extracts shallow geothermal energy for indoor temperature control by embedding heat exchange pipes within the station structure to form an energy station structure (energy tunnel, energy base slab, energy support piles). This technology eliminates the need for additional drilling and heat exchange wells, resulting in low initial investment costs. Furthermore, it imbues traditional underground structures with new functions, aligning with the development concept of integrated underground space utilization. Underground structures are typically permanent structures designed for a 100-year lifespan, and their safety and stability determine the service life of underground stations. Therefore, assessing the impact of the heat exchange process on the mechanical properties of underground structures is an indispensable aspect of the design process.

[0004] The Chinese utility model patent application number is CN202120598233.1, entitled "A Composite Heat Pump System Applicable to Subway Stations". It establishes a composite heat pump system consisting of a ground source heat pump system and a water loop heat pump system. In summer, the ground source heat pump system provides cooling for the station's public areas and management rooms, while in winter, the water loop heat pump system provides cooling.

[0005] Chinese invention patent application number CN201210585495.X, entitled "Ground Source Heat Pump Air Conditioning System for Subway Stations", describes a system consisting of a ground source heat pump unit, a buried pipe heat exchanger, a water storage tank, station air conditioning terminals, and building air conditioning terminals.

[0006] Chinese invention patent application number CN201410077507.7, entitled "A Test Device and Test Method for Friction Force of Energy Pile-Soil Contact Surface", discloses a test device for friction force of energy pile-soil contact surface consisting of a circulation control system, a loading system, a pressurization system and a measurement system.

[0007] Chinese invention patent application number CN201810818755.0, entitled "Energy Pile Test Monitoring System and Test Method", discloses a test monitoring system including an energy pile body, a loading device, a pile top displacement testing device, a circulating temperature control device, a data acquisition device, and a geotechnical thermal response testing device, which can perform more comprehensive testing and analysis of the thermodynamic parameters of energy piles.

[0008] Chinese invention patent application number CN201611087188.3, ​​entitled "A Method for Calculating Axial Force under Thermal Coupling of Energy Piles", discloses a one-dimensional finite difference axial force calculation method for energy piles based on the load transfer method under thermal coupling.

[0009] The limited number of energy pile thermal response analysis methods mentioned above are difficult to apply to energy station structures (energy support piles, energy tunnels, energy base slabs) that are in direct contact with air in a semi-airborne state. Therefore, a more targeted and easily applicable calculation method is needed to accurately consider the impact of long-term temperature loads on the mechanical performance of station structures in practical engineering designs.

[0010] Currently, there are only a few studies on the structural composition of ground source heat pump systems in subway stations, lacking research on the composition and design calculations of energy-powered underground subway station systems consisting of energy station structures (energy support piles, energy tunnels, and energy foundation slabs). Traditional architectural design methods mainly use numerical simulation software for single-module design calculations, and the coordination between different modules still requires human experience to consider, resulting in design schemes with low versatility and a low degree of digitalization in the design process. The design of energy-powered underground subway stations requires a high degree of coordination between multiple modules to achieve the design goals of energy conservation and emission reduction. Traditional design methods are difficult to meet the design requirements and have high labor costs.

[0011] Machine learning is a scientific technology that enables computers to automatically analyze and extract patterns from a type of data by establishing appropriate algorithms, and then use these patterns to predict unknown data. Therefore, this paper considers introducing machine learning concepts into the design calculations of an energy-powered underground subway station to evaluate the impact of the heat exchange process under semi-open conditions on the station structure, as well as the overall design of an energy-powered underground subway station composed of multiple modules. Summary of the Invention

[0012] To address the problems existing in the prior art, this invention provides a design method for underground subway stations based on federated learning, comprising the following steps:

[0013] D1: Collect and organize station data, including structural information parameters, environmental information parameters, foundation information parameters, system equipment parameters, and energy station structural parameters. Perform necessary data preprocessing, establish local client-side local models, analyze sample data characteristics, and allocate them to the corresponding local client-side local models. The local client-side local models include station load calculation modules, unit equipment selection modules, energy support pile buried pipe design modules, energy tunnel buried pipe design modules, and energy base plate buried pipe design modules.

[0014] D2: Each local client updates its local model and uploads the calculation results to the central server;

[0015] D3: The central server uses its private key to decrypt and securely aggregate the encrypted and de-identified parameters uploaded by each local client's local model, and then updates the global shared model.

[0016] D4: The central server generates a public key for encrypting data transmission from the global shared model and distributes it to each local client model. Each local client model updates the iteration results of other related client models based on the global shared model as the new sample data attribute parameters.

[0017] D5: Repeat steps D2-D4 iteratively until the global shared model is robust. The local models of each local client calculate the matching design parameters in the energy underground subway station based on the global shared model. The design parameters include: station indoor load end, heat pump unit model, circulating pump model, energy support pile geometry and buried pipe parameters, energy tunnel geometry and buried pipe parameters, and energy base plate geometry and buried pipe parameters.

[0018] Preferably, the station load calculation module updates the local model by including the following steps:

[0019] D2.1.1: Decrypt the public key of the encrypted data using the private key, update the design parameters of the relevant clients of the unit equipment end, energy support pile buried pipe end, energy tunnel buried pipe end, and energy base plate buried pipe end according to the global shared model of the central server, and use them as the new sample data attribute parameters. Based on the sample data, use the ambient temperature, ambient humidity, passenger flow behavior, indoor equipment behavior, lighting behavior, ventilation behavior, station equipment heat, enclosure structure material, enclosure structure size, and enclosure structure thermal property parameters as the sample attribute space set, and use the station's heating and cooling load as the learning target.

[0020] D2.1.2: Introducing slack variables , and penalty coefficient A nonlinear segmentation support vector classifier considering soft margins is constructed to represent the relationship between the station load prediction sample attribute space and the station load. The loss function represents the model's prediction accuracy and its inherent stability, and its specific expression is as follows: , and obtain its corresponding conditional extremum function;

[0021] D2.1.3: The above conditional extremum function is transformed into a multivariate function for solution using the Lagrangian function, allowing the Lagrangian function to be applied to the optimization objective. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which can transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region;

[0022] D2.1.4: Processing the inner product of mapping functions in the computation model using kernel functions;

[0023] D2.1.5; Optimize the parameters of the support vector regression model using grid search, particle swarm optimization, PSO algorithm, and genetic algorithm: insensitive loss function. Penalty coefficient Hyperparameters in the kernel function: , , , , ;

[0024] D2.1.6: Input the target basic feature parameters of the station, use the trained support vector regression model to predict and calculate the annual hourly cooling and heating load of the target station, and upload the station cooling and heating load results to the server in the form of encrypted and desensitized parameters in the form of public key, so as to prepare for the global shared model iteration of federated learning on the central server.

[0025] Preferably, the unit equipment selection module updates the local model by including the following steps:

[0026] D2.2.1: Decrypt the public key of the encrypted data using the private key, update the iteration results of the relevant clients of the station load calculation module, energy support pile buried pipe design module, energy tunnel buried pipe design module, and energy base plate buried pipe design module according to the global shared model of the central server, and use them as the attribute parameters of the new sample data. Calculate the information gain of all attributes of the training samples according to the information entropy, and sort all attributes according to the information gain.

[0027] D2.2.2: Find the attributes with information gain higher than the average level from the candidate attributes, and select the attribute with the highest gain rate to predict as the branch attribute of the decision tree;

[0028] D2.2.3: Treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the REP method to prune the decision tree, and iterate to form a decision tree. The calculated unit parameters are converted into encrypted parameters using the public key and uploaded to the central server for global shared model iteration of federated learning on the central server.

[0029] Preferably, the energy support pile buried pipe design module updates the local model, including the following steps;

[0030] D2.3.1: Use the private key to decrypt the encrypted iteration results of the station load calculation module and the unit equipment selection module related clients of the global shared model update of the central server, and combine the energy support pile buried pipe parameters of the local client as new sample data attribute parameters. Use the Bootstrap sampling method to randomly generate a sample subset from the sample as a training sample of one of the decision tree models. Repeat the sampling k times to form k decision tree training samples.

[0031] D2.3.2: Train decision trees based on a subset of attributes from k training samples to form k independent random decision trees;

[0032] D2.3.3: Vote on the pipe laying schemes predicted by the k decision trees, and use the voting result as the optimal pipe laying scheme of the random forest;

[0033] D2.3.4: Convert the optimal buried pipe parameters of the energy support pile into encrypted parameters using the public key and upload them to the central server for global shared model iteration of federated learning on the central server side;

[0034] Preferably, both the energy tunnel buried pipe design module and the energy base plate buried pipe design module calculate the energy tunnel buried pipe design parameters and energy base plate buried pipe design parameters that meet the load requirements of the station equipment area, unit equipment parameters and thermal characteristics based on the random forest local model, and upload them to the central processing unit in encrypted form as the optimal solution;

[0035] Preferably, the method further includes predicting the heat exchange efficiency of the designed energy-powered underground subway station structure, including the following steps:

[0036] S1: Collection of training sample data and corresponding tag data on the heat exchange efficiency of the energy station structure. The training sample data includes geometric and material parameters of the energy station structure, geometric parameters of buried pipes, thermal properties of soil and rock, state parameters of heat exchange circulating fluid, ambient temperature, and wind speed. The tag data includes relevant information reflecting the heat exchange performance of the energy tunnel, energy base plate, and energy support piles.

[0037] S2: Perform quality analysis and data cleaning on the training sample data, extract the distribution pattern of the data, normalize the data, evaluate the correlation between training sample data of different labels, divide the training set and prediction set required for supervised learning, and distinguish between the three types of energy station structures: energy support piles, energy tunnels, and energy base plates.

[0038] S3: Establish a predictive model for the structural heat exchange efficiency of the energy station based on the sample attribute information of the training set, and train it.

[0039] S4: Input the sample attribute information of the prediction set. The sample attributes include: geometric parameters of the buried pipes of the energy station structure, thermal properties of the soil and rock, and state parameters of the heat exchange circulating fluid. Calculate the heat exchange efficiency prediction result corresponding to the prediction set based on the heat exchange efficiency prediction model of the energy station structure.

[0040] S5: Compare the prediction results with the labels corresponding to the prediction set samples, calculate the model prediction error, and evaluate the model accuracy;

[0041] S6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform the prediction calculation of the heat exchange efficiency of the station structure.

[0042] Preferably, the geometric parameters of the energy station structure are heat exchange tube length, tube spacing, tube diameter, and tube layout classification; the material parameters of the energy station structure are heat exchange medium thermal conductivity, heat exchange medium temperature, heat exchange medium flow velocity, concrete thermal conductivity, and concrete temperature; the geotechnical thermal property parameters are geotechnical thermal conductivity, groundwater depth, groundwater flow velocity, and geotechnical temperature; the state parameters of the heat exchange circulating fluid are circulating fluid flow velocity, density, specific heat, and inlet temperature; the heat exchange efficiency of the energy station structure is the comprehensive thermal conductivity, heat exchange per meter, heat extraction, and heat exchange power; the data cleaning objects include incomplete data, erroneous data, duplicate data, and abnormal data; the data cleaning content includes identifying invalid values, outliers, and missing values, and processing invalid values, outliers, and missing values.

[0043] Preferably, in step S3, the simulated annealing method is used to improve the support vector regression algorithm to train the prediction model for the heat transfer efficiency of the energy station structure. The specific steps are as follows:

[0044] S3.1: To resolve the relationship between sample attribute data and sample label data, slack variables are introduced. , and penalty coefficient Constructing a nonlinear segmentation support vector classifier that considers soft margins allows the model's prediction accuracy and stability to be improved through a loss function. This indicates that the corresponding conditional extremum function can be obtained;

[0045] S3.2: Transform the above conditional extremum function into a multivariate function using the Lagrangian function for solution, and let the Lagrangian function be the optimum. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region;

[0046] S3.3: Using kernel functions to process and calculate the inner product of mapping functions in multivariate functions;

[0047] S3.4: Optimization of model parameters in support vector regression model using piecewise simulated annealing: insensitive loss function Penalty coefficient C, hyperparameters in the kernel function , , , , ;

[0048] Preferably, it also includes predicting the thermal stress of the energy station structure near the designed underground subway station, including the following steps;

[0049] T1: Collection of sample data and corresponding label data. The training sample dataset includes the geometric parameters of the buried pipes of the energy station structure, the material parameters of the energy station structure, the geotechnical parameters, and the state parameters of the heat exchange circulating fluid. The label data includes relevant information reflecting the thermal stress of the energy station structure: thermally induced warping deformation, thermally induced strain, thermally induced bending moment, and thermally induced additional settlement. The sample data and corresponding label data are classified according to energy base plate, energy tunnel, and energy support pile.

[0050] T2: Sample data preprocessing, extraction of data distribution patterns, data normalization, evaluation of correlation between sample data of different labels, quality analysis and data cleaning of sample data, definition of state vector X and target vector Y, and sorting of sample attribute variables according to correlation through sensitivity analysis, taking the top-ranked attributes as elements of state parameters. The state vector at time t can be expressed as:

[0051] ;

[0052] T3: Use the Euclidean distance metric to measure the distance between each state vector in the state space, reflecting the degree of correlation between each state vector X and the target vector Y in the state space;

[0053] T4: Input the sample attribute information of the prediction set. The sample attributes include: buried pipe parameters and soil and rock mechanical parameters of the energy station structure; form a state vector from the sample attributes of the training set, search for the K nearest neighbors of the current vector in the training set to form a K nearest neighbor prediction model, and use the K nearest neighbor prediction model to predict the corresponding thermal stress prediction result of the set. The expression of the K nearest neighbor prediction model is: ;

[0054] T5: Evaluate the prediction results of thermal stress in the energy station structure, compare the prediction results of the prediction set with the labels corresponding to the samples in the prediction set to form a confusion matrix, calculate the sensitivity, accuracy, and precision based on the confusion matrix, and further evaluate the model prediction error and model precision;

[0055] T6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform thermal stress prediction calculation of the energy station structure.

[0056] Preferably, the geometric parameters of the buried pipes in the energy station structure are: heat exchange pipe length, pipe spacing, pipe diameter, pipe layout classification, and burial depth; the material parameters of the energy station structure are: concrete thermal conductivity, concrete temperature, concrete elastic modulus, and concrete expansion coefficient; the geotechnical parameters are: soil elastic modulus, groundwater depth, groundwater flow velocity, and soil thermal expansion coefficient; the state parameters of the heat exchange circulating fluid are: heat exchange medium thermal conductivity, specific heat, temperature, and flow velocity; the data cleaning objects include erroneous data, abnormal data, incomplete data, and duplicate data; the data cleaning content includes: identifying invalid values, outliers, and missing values; and processing outliers, invalid values, and missing values.

[0057] The energy-efficient underground subway station design method proposed in this invention can fully consider the coordination between the station's indoor user end, unit equipment selection, and the buried pipe ends of the three types of energy station structures. The resulting design scheme can greatly conform to the energy-saving and emission-reduction design concept. The support vector regression algorithm predicts the heat exchange efficiency of the designed energy station. Compared with traditional numerical simulation methods, it has a lower technical threshold and lower computational cost. By using the k-nearest neighbor regression training model to predict the thermal stress of the designed energy station, the impact of the heat exchange process on the station structure can be accurately assessed. Compared with traditional experimental methods and on-site monitoring methods, it can significantly reduce the input of human and material resources. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the federated learning framework structure in the energy underground subway station design method based on federated learning of the present invention.

[0059] Figure 2 This is a schematic diagram of the support vector regression prediction local model structure of the station load calculation module in the energy underground subway station design method based on federated learning of the present invention.

[0060] Figure 3 This is a schematic diagram of the support vector regression prediction local model process of the station load calculation module in the energy underground subway station design method based on federated learning of the present invention.

[0061] Figure 4 This is a schematic diagram of the decision tree local model structure of the unit equipment selection module in the energy underground subway station design method based on federated learning of the present invention.

[0062] Figure 5 This is a schematic diagram of the decision tree partial model flow of the unit equipment selection module in the energy underground subway station design method based on federated learning of the present invention.

[0063] Figure 6 This is a schematic diagram of the random forest local model structure of the energy support pile buried pipe design module in the energy underground subway station design method based on federated learning of the present invention.

[0064] Figure 7 This is a schematic diagram of the random forest local model process in the energy support pile buried pipe design module of the energy underground subway station design method based on federated learning in this invention.

[0065] Figure 8 This is a schematic diagram of the random forest local model structure of the energy tunnel buried pipe design module in the energy underground subway station design method based on federated learning of the present invention.

[0066] Figure 9 This is a schematic diagram of the random forest local model process in the energy tunnel buried pipe design module of the energy underground subway station design method based on federated learning in this invention.

[0067] Figure 10 This is a schematic diagram of the random forest local model structure of the energy foundation pipe design module in the energy underground subway station design method based on federated learning of the present invention.

[0068] Figure 11 This is a schematic diagram of the random forest local model process in the energy foundation pipe design module of the energy underground subway station design method based on federated learning in this invention.

[0069] Figure 12 This is a schematic diagram illustrating the calculation process of heat exchange efficiency of energy station structure in a federated learning-based energy underground subway station design method of the present invention.

[0070] Figure 13 This is a schematic diagram of the support vector regression model structure for the heat exchange efficiency of an energy station structure in the energy underground subway station design method based on federated learning of this invention.

[0071] Figure 14 This is a schematic diagram of the support vector regression algorithm improved by simulated annealing in the energy underground subway station design method based on federated learning of the present invention.

[0072] Figure 15 This is a schematic diagram of the K-nearest neighbor regression calculation process for the thermal stress of an energy station structure in a federated learning-based energy underground subway station design method of the present invention. Detailed Implementation

[0073] like Figure 1 As shown, this invention provides a method for designing underground subway stations based on federated learning, specifically including the following steps:

[0074] D1: Collect and organize station data, including structural information parameters, environmental information parameters, foundation information parameters, system equipment parameters, and energy station structural parameters, through numerical simulation calculations, on-site monitoring technology, questionnaire survey technology, or image intelligent recognition technology. Perform necessary data preprocessing, establish a local client-side local model, analyze sample data characteristics, and allocate them to the corresponding local client-side local modules. The local client-side local model includes a station load calculation module, a unit equipment selection module, an energy support pile burial pipe design module, an energy tunnel burial pipe design module, and an energy foundation slab burial pipe design module. The station structural parameters include building enclosure structural materials, building dimensions, passenger behavior within the station, equipment behavior within the station, lighting behavior within the station, ventilation behavior within the station, air conditioning design parameters, and building materials; the environmental information parameters include outdoor ambient temperature, outdoor ambient humidity, station design temperature, and station design humidity; the system equipment parameters include heat pump cooling capacity, heat pump heating capacity, heat pump flow rate, circulating pump flow rate, and circulating pump head; the energy station structural parameters include energy support pile dimensions, buried pipe type, buried pipe length, heat exchange medium flow rate, heat exchange efficiency, thermal stress, and energy tunnel buried pipe dimensions;

[0075] D2: Each local client model updates the parameters of other related client models as new sample data attribute parameters based on the global shared model of the central server, updates its own model, and uploads the parameters to the central server. This includes the following steps:

[0076] Local Client 1: The station load calculation module uses outdoor environmental parameters, passenger information, station equipment information, and station structure information as sample attribute data to establish a system such as... Figure 2-3 The station load calculation module shown supports vector regression to predict local models, calculates station load, and uploads the encrypted data to the central processing unit. Its key feature is that it calculates the heating and cooling loads of the subway station based on environmental parameters and building characteristics, and updates the local model using the following steps:

[0077] D2.1.1: Decrypt the public key of the encrypted data using the private key. Update the design parameters of the relevant clients for the unit equipment end, energy support pile buried pipe end, energy tunnel buried pipe end, and energy base plate buried pipe end according to the global shared model of the central server, and use them as the new sample data attribute parameters. Based on the sample data, use the ambient temperature, ambient humidity, passenger flow behavior, indoor equipment behavior, lighting behavior, ventilation behavior, station equipment heat, building envelope material, building envelope dimensions, and building envelope thermal properties as the sample attribute space set. This generates sample data containing multiple features. The station's heating and cooling loads were used as the learning target. ;

[0078] D2.1.2: Introducing slack variables , and penalty coefficient A nonlinear segmentation support vector classifier considering soft margins is constructed to represent the relationship between the station load prediction sample attribute space and the station load. The loss function represents the model's prediction accuracy and its inherent stability, and its specific expression is as follows: , and obtain its corresponding conditional extremum function;

[0079] D2.1.3: The above conditional extremum function is transformed into a multivariate function for solution using the Lagrangian function, allowing the Lagrangian function to be applied to the optimization objective. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which can transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region;

[0080] D2.1.4 using Gaussian kernel function Processing the inner product of mapping functions in the computational model Inner products can also be processed using linear kernels, polynomial kernels, sigmoid kernels, and combined kernel functions;

[0081] D2.1.5; Optimize vector regression model parameters using grid search, particle swarm optimization, PSO algorithm, and genetic algorithm: insensitive loss function. Penalty coefficient Hyperparameters in the kernel function: , , , , ;

[0082] D2.1.6: Input the target basic feature parameters of the station, use the trained support vector regression model to predict and calculate the hourly cooling and heating load of the target station throughout the year, and upload the station's cooling and heating load results to the server in the form of encrypted and anonymized parameters in public key form, for the global shared model iteration of federated learning on the central server side, and the client... The relevant data is represented as follows: It can be encrypted as ;

[0083] Local Client 2: Unit Equipment Selection Module. This module combines station load conditions and heat extraction information from buried pipes at the three types of energy station structures, using unit parameters as sample attributes to establish... Figures 4-5The decision tree model for unit selection shown determines the parameters of the heat pump, cooling pump, and chiller pump and uploads them to the central processing unit with encryption. These unit parameters include: the cooling method of the heat pump unit, the cooling capacity of the heat pump unit, the flow rate of the heat pump unit, the flow rate of the chiller pump, the head of the chiller pump, and the flow rate and head information of the chiller pump. The formation of the decision tree model is based on at least one of the following algorithms: ID3 algorithm, C4.5 algorithm, or CART algorithm. Updating the local model includes the following steps:

[0084] D2.2.1: Decrypt the public key of the encrypted data using the private key. Update the iteration results of the relevant clients of the station load calculation module, energy support pile buried pipe design module, energy tunnel buried pipe design module, and energy foundation buried pipe design module according to the global shared model of the central server, and use them as the attribute parameters of the new sample data. Calculate the information gain of all attributes of the training samples according to the information entropy, and sort all attributes according to the information gain. The method for calculating the information entropy is as follows:

[0085] The total information entropy of a given sample S: ;

[0086] Attribute A has K distinct attribute values, therefore S is divided into K subsets. Calculate the information entropy of attribute A for a given sample S. ;

[0087] Calculate the information gain of S based on attribute A. ;

[0088] Split information of computed attributes , and information gain , ;

[0089] D2.2.2: Find the attributes with information gain higher than the average level from the candidate attributes, and select the attribute with the highest gain rate to predict as the branch attribute of the decision tree;

[0090] D2.2.3: Treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the REP, PEP or MEP method to prune the decision tree, and iterate to form a decision tree. The calculated unit parameters are converted into encrypted parameters using the public key and uploaded to the central server for global shared model iteration of federated learning on the central server.

[0091] Local Client 3: Energy Support Pile Pipe Design Module. Based on the station load and unit equipment parameters iterated by the central server, the module establishes a design on the client side using the energy support pile parameters as sample attributes. Figures 6-7The energy support pile buried pipe design module, using a random forest local model, calculates the optimal buried pipe design parameters that meet the load requirements of the station equipment area, unit equipment parameters, and thermal characteristics. These parameters are then encrypted and uploaded to the central processing unit. The energy support pile parameters include: heat exchange pipe material, buried pipe length, buried pipe spacing, energy support pile dimensions, and heat exchange fluid velocity. Updating the local model includes the following steps:

[0092] D2.3.1: Use the private key to decrypt the encrypted iteration results of the station load calculation module and the unit equipment selection module related clients of the global shared model update of the central server, and combine the energy support pile buried pipe parameters of the local client as new sample data attribute parameters. Use the Bootstrap sampling method to randomly generate a sample subset from the sample as a training sample of one of the decision tree models. Repeat the sampling k times to form k decision tree training samples.

[0093] D2.3.2: Train decision trees based on a subset of attributes from k training samples to form k independent random decision trees;

[0094] D2.3.3: Vote on the pipe laying schemes predicted by the k decision trees, and use the voting result as the optimal pipe laying scheme of the random forest;

[0095] D2.3.4: Convert the optimal buried pipe parameters of the energy support pile into encrypted parameters using the public key and upload them to the central server for global shared model iteration in federated learning on the central server side.

[0096] Local Client 4: Energy Tunnel Pipeline Design Module. Based on the station load and unit equipment parameters iterated by the central server, the module establishes a design on the client side using energy tunnel parameters as sample attributes. Figures 8-9 The illustrated random forest local model for energy tunnel pipe laying calculates the optimal pipe laying scheme based on the load requirements of the station office area, unit equipment parameters, and thermal characteristics. This optimal scheme is then encrypted and uploaded to the central processing unit. The key feature is that the energy tunnel parameters include: heat exchanger pipe material, pipe length, pipe spacing, energy tunnel geometry, and heat exchange fluid velocity. Specific steps can be found in the energy support pile pipe laying design module.

[0097] Local Client 5: Energy Base Plate Buried Pipe Design Module. Based on the station load and unit equipment parameters iterated by the central server, the module establishes a design on the client side using the energy base plate parameters as sample attributes. Figures 10-11The illustrated random forest local model for energy base plate buried pipes calculates the optimal buried pipe design parameters that meet the load requirements of the station's public area, unit equipment parameters, and thermal characteristics. These parameters are then encrypted and uploaded to the central processing unit. The key feature is that the energy base plate parameters include: heat exchanger pipe material, buried pipe length, buried pipe spacing, energy base plate geometry, and heat exchange fluid velocity. Specific steps can be found in the energy support pile buried pipe design module.

[0098] D3: The central server uses its private key to decrypt and securely aggregate the encrypted and anonymized parameters uploaded by each local client's local model, and then updates the globally shared model. The central server then decodes the encrypted and anonymized parameters.

[0099] Update the global model weights for M clients according to the following formula: ,in These are the global model parameters for the nth round. For the nth round, the client i uploads the client sub-model weights to the server. After each round of model weight updates, the central server calculates the global model error and accuracy. The central server can also control the transmission speed and the shutdown of model training.

[0100] D4: The central server generates a public key for encrypting data transmission from the global shared model and distributes it to each local client model. Each local client model updates the iteration results of other related client models based on the global shared model as the new sample data attribute parameters. For example, the load prediction support vector regression model of the station load module updates the unit equipment parameters obtained by the matching unit equipment selection client, the buried pipe design parameters obtained by the energy support pile module, the buried pipe parameters obtained by the energy tunnel buried pipe module, and the buried pipe parameters obtained by the energy base plate buried pipe module based on the global shared model.

[0101] D5: Repeat steps D2-D4 iteratively until the global shared model is robust. The local client calculates the matching design parameters in the energy underground subway station based on the global shared model. The design parameters include: station indoor load end, heat pump unit model, circulating pump model, energy support pile geometry and buried pipe parameters, energy tunnel geometry and buried pipe parameters, and energy base plate geometry and buried pipe parameters.

[0102] like Figures 12-14 It also includes predicting the structural heat exchange efficiency of the designed energy-efficient underground subway stations, including the following steps:

[0103] S1: Collection of training sample data and corresponding tag data for the heat exchange efficiency of the energy station structure. The training sample data includes the geometric and material parameters of the energy station structure, the geometric parameters of the buried pipes, the thermal properties of the soil and rock, the state parameters of the heat exchange circulating fluid, the ambient temperature, and the wind speed. The tag data includes relevant information reflecting the heat exchange performance of the energy tunnel, the energy base plate, and the energy support piles. The geometric parameters of the energy station structure are the length of the heat exchange pipes, the pipe spacing, the pipe diameter, and the pipe layout classification. The material parameters of the energy station structure are the thermal conductivity of the heat exchange medium, the temperature of the heat exchange medium, the flow velocity of the heat exchange medium, the thermal conductivity of the concrete, and the temperature of the concrete. The thermal properties of the soil and rock are the thermal conductivity of the soil and rock, the depth of the groundwater, the flow velocity of the groundwater, and the temperature of the soil and rock. The state parameters of the heat exchange circulating fluid are the flow velocity, density, specific heat, and inlet temperature of the circulating fluid. The heat exchange efficiency of the energy station structure is the comprehensive thermal conductivity, the heat exchange per meter, the heat extraction, and the heat exchange power.

[0104] S2: Perform quality analysis and data cleaning on the training sample data, extract the distribution patterns of the data, perform data normalization, evaluate the correlation between training sample data of different labels, and divide the training set and prediction set required for supervised learning. It is necessary to distinguish between three types of energy station structures: energy support piles, energy tunnels, and energy base plates. The data cleaning objects include incomplete data, erroneous data, duplicate data, and anomalous data. The data cleaning content includes identifying invalid values, outliers, and missing values, and processing invalid values, outliers, and missing values. The specific data cleaning operation process is as follows: 1. Calculate the maximum, minimum, median, and upper and lower quartiles of the dataset based on the box plot. Outliers or invalid values ​​can be identified based on the upper and lower quartiles of the box plot. 2. Correct and fill invalid values, outliers, and missing values ​​according to the mean, median, and mode of the attribute values.

[0105] S3: Establish a prediction model for the heat transfer efficiency of the energy station structure based on the sample attribute information of the training set, and train it. Use simulated annealing to improve the support vector regression training of the prediction model for the heat transfer efficiency of the energy station structure. The specific steps are as follows:

[0106] S3.1: To resolve the relationship between sample attribute data and sample label data, slack variables are introduced. , and penalty coefficient Constructing a nonlinear segmentation support vector classifier that considers soft margins allows the model's prediction accuracy and stability to be improved through a loss function. This indicates that the corresponding conditional extremum function is obtained, and the corresponding conditional extremum function is:

[0107]

[0108] S3.2: Transform the above conditional extremum function into a multivariate function using the Lagrangian function for solution, and let the Lagrangian function be the optimum. , , The partial derivatives are 0, yielding the Lagrange multipliers. This transforms the original conditional extremum function into a dual function, thus finding the minimum value within the constraint region. The Lagrange function is as follows:

[0109] S3.3: Using polynomial kernel functions to process the inner product of mapping functions in multivariate functions polynomial kernel Inner products can also be processed using Gaussian kernels, linear kernels, Sigmoid kernels, and combined kernel functions;

[0110] S3.4: Optimization of model parameters in support vector regression model using piecewise simulated annealing: insensitive loss function Penalty coefficient C, hyperparameters in the kernel function , , , , The specific steps are as follows:

[0111] S3.4.1: Randomly generate an initial parameter set for interactive verification, and denote the error value EEP as the current annealing system state. initial temperature The temperature of annealing stage one The annealing end temperature is .

[0112] S3.4.2: According to the perturbation algorithm The parameters are perturbed to form a new parameter set, and the current annealing system state is obtained through interactive verification. ,calculate .

[0113] S3.4.3: If If the new parameter set is accepted, proceed to step S3.4.5; otherwise, follow the Metropolis criterion. If the condition is met, accept the corresponding parameter set and jump to step S3.4.5; if none of the above conditions are met, reject the critical state and execute step S3.4.4.

[0114] S3.4.4: If the parameter set is rejected, return to step S3.4.2, re-perturb to generate a new parameter set, and perform interactive verification until the parameter set acceptance condition in step S3.4.3 is met.

[0115] S3.4.5: Upon obtaining a new state, proceed according to the cooling plan. Cooling will occur if the set temperature is not reached. Return to step S3.4.2 when the set temperature is reached. At that time, a new annealing plan will be carried out. .

[0116] S3.4.6: Following the new perturbation method and annealing plan, continue to perturb the parameter set of the first-stage annealing and calculate the corresponding state parameters. .

[0117] S3.4.7: If If the new parameter set is accepted, proceed to step S3.4.9; otherwise, follow the Metropolis criterion. If the condition is met, accept the corresponding parameter set and proceed to step S3.4.9; if none of the above conditions are met, reject the parameter set and proceed to step S3.4.8.

[0118] S3.4.8: If the parameter set is rejected, return to step S3.4.6, re-perturb to generate a new parameter set, and perform interactive verification until the parameter set acceptance condition in step S3.4.9 is met.

[0119] S3.4.9: Set the end temperature to... As the algorithm's exit point, the global maximum number of EEP calculations is set to... When it reaches or Stop annealing at this point; the critical state cross-validation error accepted at this time... It should be the lowest The corresponding parameters should be the optimal prediction parameters; otherwise, return to step S3.4.6.

[0120] S4: Input the sample attribute information of the prediction set. The sample attributes include: geometric parameters of the buried pipes of the energy station structure, thermal properties of the soil and rock mass, and state parameters of the heat exchange circulating fluid. Calculate the heat exchange efficiency prediction result corresponding to the prediction set based on the heat exchange efficiency prediction model of the energy station structure. The buried pipe parameters of the energy station structure mainly include: buried pipe type, buried pipe spacing, buried pipe length, buried pipe depth, and thermal properties of the heat exchange pipe. The specific buried pipe parameters of the energy station structure mainly include: buried pipe type, buried pipe spacing, buried pipe length, buried pipe depth, and thermal properties of the heat exchange pipe. The specific soil and rock mass parameters mainly include: groundwater depth, soil and rock mass temperature, and soil and rock mass thermal conductivity. The specific state parameters of the heat exchange circulating fluid are: circulating fluid velocity, density, specific heat, and inlet temperature.

[0121] S5: Compare the prediction results with the labels corresponding to the prediction set samples, calculate the model prediction error, and evaluate the model accuracy;

[0122] S6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform prediction calculation of the heat exchange efficiency of the station structure.

[0123] like Figure 15 As shown, it also includes the prediction of thermal stress in the nearest neighbor structure K of the designed energy underground subway station, including the following steps;

[0124] T1: Collection of sample data and corresponding label data. The training sample dataset includes the geometric parameters of the buried pipes in the energy station structure, the material parameters of the energy station structure, the geotechnical mechanics parameters, and the state parameters of the heat exchange circulating fluid. Label data includes information reflecting the thermal stress of the energy station structure: thermal stress, thermal strain, thermal bending moment, and thermal additional settlement. The sample data and corresponding label data are classified according to energy base plate, energy tunnel, and energy support piles. The geometric parameters of the buried pipes in the energy station structure are heat exchange pipe length, pipe spacing, pipe diameter, pipe layout classification, and pipe depth. The material parameters of the energy station structure are concrete thermal conductivity, concrete temperature, concrete elastic modulus, concrete expansion coefficient, heat exchange medium thermal conductivity, heat exchange medium temperature, and heat exchange medium flow velocity. The geotechnical mechanics parameters are geotechnical elastic modulus, groundwater depth, groundwater flow velocity, and geotechnical thermal expansion coefficient. The state parameters of the heat exchange circulating fluid are heat exchange medium thermal conductivity, specific heat, temperature, and flow velocity.

[0125] T2: Sample data preprocessing, including extracting data distribution patterns, data normalization, evaluating the correlation between sample data of different labels, performing quality analysis and data cleaning on the sample data. The data cleaning targets erroneous data, outlier data, incomplete data, and duplicate data. The data cleaning process includes identifying invalid values, outliers, and missing values; processing outliers, invalid values, and missing values; defining a state vector X and a target vector Y; and sorting the sample attribute variables according to their correlation through sensitivity analysis, taking the top-ranked attributes as elements of the state parameters. The state vector at time t can be expressed as: The specific data cleaning process is as follows: 1. Use scatter plots to show the positional relationship between the two sets of data and visually identify outliers as abnormal or invalid values; 2. The identified invalid, outlier, and missing values ​​can be directly deleted; The handling methods for invalid, outlier, and missing values ​​can be based on one of the following: 1. Direct deletion; 2. Fill in invalid, outlier, and missing values ​​according to the statistical data characteristics of mean, median, and mode; 3. Fit invalid, outlier, and missing values ​​according to regression models or maximum likelihood estimation.

[0126] T3: Measured using the Euclidean distance metric. Calculate the distance between each state vector in the state space to reflect the correlation between each state vector X and the target vector Y in the state space. Manhattan distance, Chebyshev distance, Minkowski distance, normalized Euclidean distance, Mahalanobis distance, cosine angle, and combined distance can also be used.

[0127] T4: Input the sample attribute information of the prediction set. The sample attributes include: buried pipe parameters and soil and rock mechanical parameters of the energy station structure; form a state vector from the sample attributes of the training set, search for the K nearest neighbors of the current vector in the training set to form a K nearest neighbor prediction model, and use the K nearest neighbor prediction model to predict the corresponding thermal stress prediction result of the set. The expression of the K nearest neighbor prediction model is: ;

[0128] T5: Evaluate the prediction results of thermal stress in the energy station structure, compare the prediction results of the prediction set with the labels corresponding to the samples in the prediction set to form a confusion matrix, calculate the sensitivity, accuracy, and precision based on the confusion matrix, and further evaluate the model prediction error and model precision;

[0129] Sensitivity: The closer the value is to 1, the better the model fit.

[0130] Accuracy: The closer the value is to 1, the stronger the model's predictive ability.

[0131] Accuracy The closer the value is to 1, the higher the model's prediction accuracy.

[0132] Table 1 Confusion Matrix

[0133]

[0134] T6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform thermal stress prediction calculation of the energy station structure.

[0135] The energy-efficient underground subway station design method proposed in this invention can fully consider the coordination between the station's indoor user end, unit equipment selection, and the buried pipe ends of the three types of energy station structures. The resulting design scheme can greatly conform to the energy-saving and emission-reduction design concept. The support vector regression algorithm predicts the heat exchange efficiency of the designed energy station. Compared with traditional numerical simulation methods, it has a lower technical threshold and lower computational cost. By using the k-nearest neighbor regression training model to predict the thermal stress of the designed energy station, the impact of the heat exchange process on the station structure can be accurately assessed. Compared with traditional experimental methods and on-site monitoring methods, it can significantly reduce the input of human and material resources.

[0136] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A federated learning based energy underground subway station design method, characterized in that, Includes the following steps: D1: Collect and organize station data, including structural information parameters, environmental information parameters, foundation information parameters, system equipment parameters, and energy station structural parameters. Perform necessary data preprocessing, establish local client-side local models, analyze sample data characteristics, and allocate them to the corresponding local client-side local models. The local client-side local models include station load calculation modules, unit equipment selection modules, energy support pile buried pipe design modules, energy tunnel buried pipe design modules, and energy base plate buried pipe design modules. D2: Each local client updates its local model and uploads the calculation results to the central server; D3: The central server uses its private key to decrypt and securely aggregate the encrypted and de-identified parameters uploaded by each local client's local model, and then updates the global shared model. D4: The central server generates a public key for encrypting data transmission from the global shared model and distributes it to each local client model. Each local client model updates the iteration results of other related client models based on the global shared model as the new sample data attribute parameters. D5: Repeat steps D2-D4 iteratively until the global shared model is robust. The local models of each local client calculate the matching design parameters in the energy underground subway station based on the global shared model. The design parameters include: station indoor load end, heat pump unit model, circulating pump model, energy support pile geometry and buried pipe parameters, energy tunnel geometry and buried pipe parameters, and energy base plate geometry and buried pipe parameters.

2. The energy underground subway station design method based on federated learning according to claim 1, characterized in that, The station load calculation module updates the local model, including the following steps: D2.1.1: Decrypt the public key of the encrypted data using the private key, update the design parameters of the relevant clients of the unit equipment end, energy support pile buried pipe end, energy tunnel buried pipe end, and energy base plate buried pipe end according to the global shared model of the central server, and use them as the new sample data attribute parameters. Based on the sample data, use the ambient temperature, ambient humidity, passenger flow behavior, indoor equipment behavior, lighting behavior, ventilation behavior, station equipment heat, enclosure structure material, enclosure structure size, and enclosure structure thermal property parameters as the sample attribute space set, and use the station's heating and cooling load as the learning target. D2.1.2: Introducing slack variables , and penalty coefficient A nonlinear segmentation support vector classifier considering soft margins is constructed to represent the relationship between the station load prediction sample attribute space and the station load. The loss function represents the model's prediction accuracy and its inherent stability; the specific expression is as follows: , and obtain its corresponding conditional extremum function; D2.1.3: The above conditional extremum function is transformed into a multivariate function for solution using the Lagrangian function, allowing the Lagrangian function to be applied to the optimization objective. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which can transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region; D2.1.4: Processing the inner product of mapping functions in the computation model using kernel functions; D2.1.5: Optimize the parameters of the support vector regression model using grid search, particle swarm optimization, PSO algorithm, and genetic algorithm: insensitive loss function. Penalty coefficient Hyperparameters in the kernel function: , , , , ; D2.1.6: Input the target basic feature parameters of the station, use the trained support vector regression model to predict and calculate the annual hourly cooling and heating load of the target station, and upload the station cooling and heating load results to the server in the form of encrypted and de-identified parameters in the form of public key, so as to prepare for the global shared model iteration of federated learning on the central server.

3. The federated learning based energy underground subway station design method of claim 1, wherein, The unit equipment selection module updates the local model, including the following steps: D2.2.1: Decrypt the public key of the encrypted data using the private key, update the iteration results of the relevant clients of the station load calculation module, energy support pile buried pipe design module, energy tunnel buried pipe design module, and energy base plate buried pipe design module according to the global shared model of the central server, and use them as the attribute parameters of the new sample data. Calculate the information gain of all attributes of the training samples according to the information entropy, and sort all attributes according to the information gain. D2.2.2: Find the attributes with information gain higher than the average level from the candidate attributes, and select the attribute with the highest gain rate to predict as the branch attribute of the decision tree; D2.2.3: Treat each root node as a complete dataset, divide the samples according to the suboptimal attribute, and form leaf nodes by taking samples with the same attribute value as the same sample set. Use the REP method to prune the decision tree, and iterate to form a decision tree. The calculated unit parameters are converted into encrypted parameters using the public key and uploaded to the central server for global shared model iteration of federated learning on the central server.

4. The federated learning based energy underground subway station design method of claim 1, wherein, The energy support pile buried pipe design module updates the local model, including the following steps; D2.3.1: Use the private key to decrypt the encrypted iteration results of the station load calculation module and the unit equipment selection module related clients of the global shared model update of the central server, and combine the energy support pile buried pipe parameters of the local client as new sample data attribute parameters. Use the Bootstrap sampling method to randomly generate a sample subset from the sample as a training sample of one of the decision tree models. Repeat the sampling k times to form k decision tree training samples. D2.3.2: Train decision trees based on a subset of attributes from k training samples to form k independent random decision trees; D2.3.3: Vote on the pipe laying schemes predicted by the k decision trees, and use the voting result as the optimal pipe laying scheme of the random forest; D2.3.4: Convert the optimal buried pipe parameters of the energy support pile into encrypted parameters using the public key and upload them to the central server for global shared model iteration in federated learning on the central server side.

5. The federated learning based energy underground subway station design method of claim 1, wherein, Both the energy tunnel buried pipe design module and the energy base plate buried pipe design module are based on the random forest local model to calculate the energy tunnel buried pipe design parameters and energy base plate buried pipe design parameters that meet the load requirements of the station equipment area, the unit equipment parameters and thermal characteristics as the optimal solution and then encrypted and uploaded to the central processing unit.

6. The federated learning based energy underground subway station design method of claim 1, wherein, It also includes predicting the structural heat exchange efficiency of the designed energy-efficient underground subway stations, including the following steps: S1: Collection of training sample data and corresponding tag data on the heat exchange efficiency of the energy station structure. The training sample data includes geometric and material parameters of the energy station structure, geometric parameters of buried pipes, thermal properties of soil and rock, state parameters of heat exchange circulating fluid, ambient temperature, and wind speed. The tag data includes relevant information reflecting the heat exchange performance of the energy tunnel, energy base plate, and energy support piles. S2: Perform quality analysis and data cleaning on the training sample data, extract the distribution pattern of the data, normalize the data, evaluate the correlation between training sample data of different labels, divide the training set and prediction set required for supervised learning, and distinguish between the three types of energy station structures: energy support piles, energy tunnels, and energy base plates. S3: Establish a predictive model for the structural heat exchange efficiency of the energy station based on the sample attribute information of the training set, and train it. S4: Input the sample attribute information of the prediction set. The sample attributes include: geometric parameters of the buried pipes of the energy station structure, thermal properties of the soil and rock, and state parameters of the heat exchange circulating fluid. Calculate the heat exchange efficiency prediction result corresponding to the prediction set based on the heat exchange efficiency prediction model of the energy station structure. S5: Compare the prediction results with the labels corresponding to the prediction set samples, calculate the model prediction error, and evaluate the model accuracy; S6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform prediction calculation of the heat exchange efficiency of the station structure.

7. The federated learning based energy underground subway station design method of claim 6, wherein, The geometric parameters of the energy station structure include heat exchange tube length, tube spacing, tube diameter, and tube layout classification; the material parameters of the energy station structure include heat exchange medium thermal conductivity, heat exchange medium temperature, heat exchange medium flow velocity, concrete thermal conductivity, and concrete temperature; the geotechnical thermal property parameters include geotechnical thermal conductivity, groundwater depth, groundwater flow velocity, and geotechnical temperature; the state parameters of the heat exchange circulating fluid include circulating fluid flow velocity, density, specific heat, and inlet temperature; the heat exchange efficiency of the energy station structure includes comprehensive thermal conductivity, heat exchange per meter, heat extraction, and heat exchange power; the data cleaning objects include incomplete data, erroneous data, duplicate data, and abnormal data; the data cleaning content includes identifying invalid values, outliers, and missing values, and processing invalid values, outliers, and missing values.

8. The federated learning based energy underground subway station design method of claim 6, wherein, In step S3, the simulated annealing method is used to improve the support vector regression algorithm to train the prediction model for the heat transfer efficiency of the energy station structure. The specific steps are as follows: S3.1: To resolve the relationship between sample attribute data and sample label data, slack variables are introduced. , and penalty coefficient Constructing a nonlinear segmentation support vector classifier that considers soft margins allows the model's prediction accuracy and stability to be improved through a loss function. This indicates that the corresponding conditional extremum function can be obtained; S3.2: Transform the above conditional extremum function into a multivariate function using the Lagrangian function for solution, and let the Lagrangian function be the optimum. , , The partial derivatives are 0, so we obtain the Lagrange multipliers, which transform the original conditional extremum function into the dual function, thereby finding the minimum value within the constraint region; S3.3: Using kernel functions to process and calculate the inner product of mapping functions in multivariate functions; S3.4: Adopting piecewise simulated annealing method to optimize model parameters in support vector regression model: insensitive loss function , penalty coefficient C, hyperparameters in kernel function , , , , .

9. The federated learning based energy underground subway station design method of claim 1, wherein, This also includes predicting the thermally induced stress in the K-nearest neighbor structure of the designed underground subway station. Includes the following steps; T1: Collection of sample data and corresponding label data. The training sample dataset includes the geometric parameters of the buried pipes of the energy station structure, the material parameters of the energy station structure, the geotechnical parameters, and the state parameters of the heat exchange circulating fluid. The label data includes relevant information reflecting the thermal stress of the energy station structure: thermally induced warping deformation, thermally induced strain, thermally induced bending moment, and thermally induced additional settlement. The sample data and corresponding label data are classified according to energy base plate, energy tunnel, and energy support pile. T2: Sample data preprocessing, extraction of data distribution patterns, data normalization, evaluation of correlation between sample data of different labels, quality analysis and data cleaning of sample data, definition of state vector X and target vector Y, and sorting of sample attribute variables according to correlation through sensitivity analysis, taking the top-ranked attributes as elements of state parameters. The state vector at time t can be expressed as: ； T3: Use the Euclidean distance metric to measure the distance between each state vector in the state space, reflecting the degree of correlation between each state vector X and the target vector Y in the state space; T4: Input the sample attribute information of the prediction set. The sample attributes include: buried pipe parameters of the energy station structure and soil and rock mechanical parameters. The attributes of the training set samples are used to form a state vector. The K nearest neighbors of the current vector are searched within the training set to form a K-nearest neighbor prediction model. This model is then used to predict the corresponding thermal stress. The expression for the K-nearest neighbor prediction model is: ; T5: Evaluate the prediction results of thermal stress in the energy station structure, compare the prediction results of the prediction set with the labels corresponding to the prediction set samples to form a confusion matrix, calculate the sensitivity, accuracy, and precision based on the confusion matrix, and further evaluate the model prediction error and model precision; T6: Input the basic characteristic parameters of the energy station structure to be predicted as the prediction set attribute space, and perform thermal stress prediction calculation of the energy station structure.

10. A method for designing underground subway stations based on federated learning according to claim 9, characterized in that, The geometric parameters of the buried pipes in the energy station structure include heat exchange pipe length, pipe spacing, pipe diameter, pipe layout classification, and burial depth. The material parameters of the energy station structure include concrete thermal conductivity, concrete temperature, concrete elastic modulus, and concrete expansion coefficient. The geotechnical parameters include soil elastic modulus, groundwater depth, groundwater flow velocity, and soil thermal expansion coefficient. The state parameters of the heat exchange circulating fluid include heat exchange medium thermal conductivity, specific heat, temperature, and flow velocity. The data cleaning objects include erroneous data, abnormal data, incomplete data, and duplicate data. The data cleaning content includes identifying invalid values, outliers, and missing values; and processing outliers, invalid values, and missing values.

Images