Machine Learning-Based Active Bridge Deck Ice and Snow Melting Design Method for Energy Piles

Abstract

The present invention belongs to the technical fields of geothermal energy development and utilization, bridge engineering, and deicing and snow melting. A design method for active bridge deck deicing and snow melting of energy piles based on machine learning is proposed. Input sample attribute data to obtain the prediction results of the K-nearest neighbor algorithm for the snow melting efficiency of the buried pipe structure on the bridge deck and the prediction results of the improved support vector regression of the simulated annealing method for the thermally induced stress of the buried pipe structure on the bridge deck. Select the buried pipe structure parameters of the bridge deck corresponding to the snow melting efficiency and thermally induced stress, and use the federated learning energy pile active bridge deck deicing and snow melting design method including the bridge deck buried pipe design module, unit selection module, and energy pile buried pipe design module to determine the unit equipment and energy pile structure. The invention solves the problem of being unable to accurately evaluate and predict the thermal response of the buried pipes on the bridge deck. Federated learning takes into account the multi-node collaborative cooperation of the energy pile active bridge deck deicing and snow melting technology, calculates a scientific and reasonable scheme, and can improve the energy utilization rate of the system.

Description

Technical Field

[0001] The present invention relates to the technical fields of geothermal energy development and utilization, bridge engineering, and deicing and snow melting, and particularly relates to a design method for active bridge deck deicing and snow melting based on machine learning. Background Art

[0002] The phenomenon of snow accumulation and icing on bridge decks in winter seriously affects the transportation capacity of expressways and municipal roads in northern cities of China. Research shows that snow accumulation and icing in winter can reduce the road surface adhesion coefficient by 60-70%, resulting in vehicle braking failure and skidding, causing serious traffic accidents. Researchers have made a large number of research results in the technology of bridge deck deicing and snow melting, which can be mainly divided into two categories: passive technologies and active technologies. Among them, passive technologies include manual snow removal, mechanical snow removal, new snow melting agents, and unmanned ice removal vehicles. These methods are relatively mature and have been widely used. However, in practical applications, there are limitations such as corroding the bridge deck, polluting the environment, and being too costly. The emerging active technologies mainly include phase change material deicing, solar energy deicing and snow melting, electrothermal deicing and snow melting, and geothermal energy deicing and snow melting. The phase change material deicing technology has good effects, but the cost of phase change materials is too high. The solar energy technology has low energy consumption, but the deicing stability depends to a large extent on the lighting conditions and is poor in stability. The electrothermal method technology has good thermal stability, but consumes a large amount of electric energy and has a high operating cost. In comparison, shallow geothermal energy, as a widely distributed renewable clean energy, has broad prospects for popularization and application in deicing and snow melting.

[0003] The active bridge deck deicing and snow melting technology using energy piles extracts shallow geothermal energy through heat exchange pipes buried in bridge piles and the bridge deck. The heat exchange pipes buried in the energy piles can extract shallow geothermal energy, and then after being lifted by a heat pump unit, it is transported to the buried pipes on the bridge deck, finally achieving the purpose of bridge deck deicing and snow melting. It has the advantages of controllable construction cost, simple construction, high stability, and low operating cost.

[0004] The Chinese invention patent application number is CN202010660161.9, and the name is a road electric heating anti-icing and snow melting system and its laying method, which discloses a road electric heating anti-icing and snow melting system including a number of heating systems. The heating system includes a road section, a road surface wetness sensor, a temperature sensor, and a number of heating cables. The traditional spraying form is replaced by heating cables, and the road surface conditions can be monitored in real time for automatic heating and deicing.

[0005] The Chinese invention patent application number is CN 202110608504.1, and the name is an intelligent snow melting and ice melting system for urban sidewalks and its construction method, which discloses an intelligent snow melting and ice melting system for urban sidewalks including an intelligent control module, a data acquisition module, a sidewalk module, and a terminal system; the intelligent control module includes a server and an intelligent controller, and the data acquisition module includes a high-definition camera, an ice formation detector, and a temperature and humidity sensor.

[0006] The Chinese invention patent application number is CN 202210017291.X, and the name is an active road ice melting and snow melting system and its control method, which discloses a control method for obtaining road surface temperature, road ice and snow pictures and comparing with a database to control the start and stop of the system.

[0007] The Chinese invention patent application number is CN 202210226666.3, and the name is a buried pipe ground source heat pump road ice melting and snow melting system and method, which discloses a buried pipe ground source heat pump road ice melting and snow melting system including a ground source heat pump unit, a shallow geothermal heat exchange part structure and a heating pipeline part structure.

[0008] The above energy pile bridge deck ice melting and snow melting technology mainly focuses on the fields of system composition, equipment and construction methods. However, the research on evaluating the snow melting efficiency of the buried pipe structure on the bridge deck and the long-term impact of the repeated freezing and thawing process on the buried pipe structure of the bridge deck is relatively insufficient, and it is impossible to accurately describe the use effect and service life of the active energy pile bridge deck ice melting and snow melting technology. Accurately evaluating the snow melting efficiency and thermal characteristics of the buried pipes on the bridge deck is the basis for the design of the buried pipes on the bridge deck, which directly affects the practicability and durability of the overall system. In addition, the above research shows that the energy pile bridge deck ice melting and snow melting technology is a systematic technology composed of three sub-modules: the buried pipe end of the energy pile, the unit equipment end, and the buried pipe end of the bridge deck. The traditional design method based on the empirical safety factor requires sufficient professional knowledge and experience, and there are problems of unreasonable design and lack of scientificity. Therefore, a design calculation method that can comprehensively consider the mutual restraint relationship between the three modules is needed to provide scientific and reasonable design parameters for specific construction.

[0009] Machine learning is a science and technology that enables a computer to automatically analyze and obtain rules from a set of data through a suitable algorithm and use the rules to predict unknown data. A large number of data samples can be trained through machine learning to predict and evaluate the snow melting efficiency of the buried pipes on the bridge deck and the thermal characteristics generated by the repeated freezing and thawing process on the buried pipe structure of the bridge deck. On this basis, a design calculation model for the bridge deck ice melting and snow melting technology is established to design a buried pipe structure on the bridge deck that meets the snow melting and stress requirements, as well as the matching unit equipment and energy pile structure. Summary of the Invention

[0010] In order to overcome the following deficiencies and problems existing in the prior art: (1) It is difficult to quantify the snow melting efficiency of the buried pipe structure on the bridge deck and the long-term impact of the repeated freezing and thawing process on the buried pipe structure of the bridge deck; (2) There is a lack of a systematic design calculation method for the active energy pile bridge deck ice melting and snow melting technology. Therefore, this invention patent proposes a design method for the active energy pile bridge deck ice melting and snow melting based on machine learning. A complete and standardized design can improve the adaptability and stability of this technology and achieve the effect of all-weather efficient ice melting and snow melting.

[0011] Technical solution of the present invention:

[0012] A design method for active bridge deck de-icing and snow melting based on machine learning. The snow melting efficiency of the bridge deck buried pipe structure and the thermal stress of the bridge deck buried pipe structure caused by repeated freezing and thawing processes are mutually restricted. The energy pile bridge deck de-icing and snow melting technology design takes into account both the snow melting efficiency of the bridge deck buried pipe structure and the thermal stress of the bridge deck buried pipe structure, and calculates the matching unit equipment and energy pile structure; The design method for active bridge deck de-icing and snow melting based on machine learning first obtains the K-nearest neighbor prediction results of the snow melting efficiency of the bridge deck buried pipe structure and the support vector regression prediction results of the thermal stress of the bridge deck buried pipe through inputting the same sample attribute data, and selects the bridge deck buried pipe structure parameters corresponding to the snow melting efficiency and thermal stress. The energy pile active bridge deck de-icing and snow melting design method using the federated learning architecture determines the unit equipment and energy pile structure;

[0013] The K-nearest neighbor prediction method for the snow melting efficiency of the bridge deck buried pipe structure includes the following steps:

[0014] T1: Obtain the snow melting efficiency training sample data and the corresponding labels; The snow melting efficiency training sample attribute data includes the buried pipe geometric parameters of the bridge deck, the bridge deck material parameters, the meteorological parameters, and the heat exchange circulating liquid state parameters, and the label data includes the relevant information reflecting the snow melting efficiency of the bridge deck buried pipe structure, including the heat exchange efficiency, the heat exchange per unit length, the snow melting time, and the snow-free rate on the surface;

[0015] T2: Preprocess the snow melting efficiency training sample data, conduct quality analysis and cleaning processing of the snow melting efficiency training sample data, extract the distribution law, normalization processing, and sample data correlation evaluation between different labels of the cleaned snow melting efficiency training sample data, and divide it into a training set and a prediction set; Define the state vector, sort the sample attribute variables according to the correlation through sensitivity analysis, and take the sample attribute variables with higher ranking as the elements of the state parameters. The state vector at time t is expressed as X(t) = [x 1 (t),x 1 (t - 1),x 2 (t),x 2 (t - 1),x 3 (t),x 3 (t - 1),…,x n (t),x n (t - 1)];

[0016] T3: Select a distance metric criterion; the degree of correlation between each state vector X and the target vector Y in the state space can be represented by the distance in the state space. Therefore, selecting a distance metric criterion is the key to the K-nearest neighbor prediction model; the distance metric criterion is one of the Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, and cosine of the included angle, or a combined distance is designed according to the above distance metric criteria. The combined distance d xy = λd xy1 +(1 - λ)d xy2 represents the degree of correlation between each state vector X and the target vector Y in the state space;

[0017] In the formula, d xy1 is the selected distance metric criterion 1, d xy2 is the distance metric criterion 2, and λ is the proportion of the distance metric criterion 1;

[0018] The metric criteria are:

[0019] Cosine of the included angle:

[0020] Chebyshev distance: d xy = max i (x i - y i );

[0021] Manhattan distance:

[0022] Minkowski distance:

[0023] Euclidean distance:

[0024] Standardized Euclidean distance:

[0025] Mahalanobis distance:

[0026] T4: Form state vectors from the sample attributes of the training set, search for the K nearest neighbors of the current vector in the training set, and form a K-nearest neighbor prediction model; input the sample attribute data of the prediction set into the K-nearest neighbor prediction model, and the predicted mean value corresponding to the K nearest neighbors is the snow melting efficiency prediction result corresponding to the prediction set calculated by the K-nearest neighbor prediction model;

[0027] T5: Evaluate the snow melting efficiency prediction result of the bridge deck buried pipe structure, compare the prediction result of step T4 with the label corresponding to the prediction set sample to form a confusion matrix; further evaluate the model prediction error and model accuracy according to the confusion matrix; The closer its value is to 1, the higher the prediction accuracy of the model.

[0028] Sensitivity The closer its value is to 1, the better the fitting effect of the model.

[0029] Accuracy The closer its value is to 1, the stronger the prediction ability of the model.

[0030] Table 1 Confusion Matrix

[0031]

[0032] T6: Take the geometric parameters of the embedded pipes in the bridge deck, meteorological parameters, state parameters of the heat exchange circulating liquid, and bridge deck material parameters to be predicted as the attribute space of the prediction set, and perform the prediction calculation of the snow melting efficiency of the embedded pipe structure on the bridge deck;

[0033] The support vector regression prediction method for the thermally induced stress of the embedded pipes on the bridge deck includes the following steps:

[0034] S1: Obtain the training sample data of the thermally induced stress of the embedded pipe structure on the bridge deck and the corresponding labels; the training sample data of the thermally induced stress of the embedded pipe structure on the bridge deck includes the geometric parameters of the embedded pipes in the bridge deck, meteorological parameters, bridge deck material parameters, and state parameters of the heat exchange circulating liquid, and the label data includes the relevant information reflecting the thermally induced stress of the embedded pipe structure on the bridge deck;

[0035] S2: Preprocess the training sample data of the thermally induced stress of the embedded pipe structure on the bridge deck and analyze the data features; perform quality analysis, feature analysis, cleaning processing on the training sample data of the thermally induced stress of the embedded pipe structure on the bridge deck, and divide the training set and prediction set of the thermally induced stress of the embedded pipe structure on the bridge deck for supervised learning;

[0036] S3: Train the prediction model of the thermally induced stress of the embedded pipe structure on the bridge deck, and establish the prediction model of the thermally induced stress of the embedded pipe structure on the bridge deck according to the sample attribute information of the training set of the thermally induced stress of the embedded pipe structure on the bridge deck;

[0037] S4: Input the sample attribute information of the prediction set of the thermally induced stress of the embedded pipe structure on the bridge deck, and the sample attributes include: geometric parameters of the embedded pipes in the bridge deck, bridge deck material parameters, and state parameters of the heat exchange circulating liquid;

[0038] S5: Calculate the prediction results of the thermally induced stress corresponding to the prediction set of the thermally induced stress of the embedded pipe structure on the bridge deck according to the prediction model of the thermally induced stress of the embedded pipe structure on the bridge deck;

[0039] S6: Evaluate the prediction results of the thermally induced stress of the embedded pipe structure on the bridge deck, compare the prediction results obtained in S5 with the labels corresponding to the samples of the prediction set of the thermally induced stress of the embedded pipe structure on the bridge deck, calculate the prediction error of the prediction model of the thermally induced stress of the embedded pipe structure on the bridge deck, and evaluate its accuracy;

[0040] S7: Take the geometric parameters of the embedded pipes in the bridge deck, meteorological parameters, bridge deck material parameters, and the state parameters of the heat exchange circulating liquid as the attribute space of the prediction set for the thermally induced stress of the embedded pipe structure in the bridge deck, and perform the prediction calculation of the thermally induced stress of the embedded pipe structure in the bridge deck;

[0041] For the active bridge deck deicing and snow melting design method of energy piles based on the federated learning architecture, establish a federated learning algorithm framework for the active bridge deck deicing and snow melting design of energy piles based on the client-server architecture; implement the learning calculations of all sub-modules on the client side, and the client side includes 3 major modules: the embedded pipe design module for the bridge deck, the unit selection module, and the embedded pipe design module for the energy pile; after summarizing the desensitized parameters calculated by the client side to the central server for calculation, distribute them to each client side to update its local model until the global model is stable;

[0042] The specific steps are as follows:

[0043] D1: Obtain and organize bridge data, pile foundation data, and meteorological data, perform data preprocessing, analyze the characteristics of sample data, and distribute them to the corresponding calculation module clients;

[0044] D2: The 3 client sides of the embedded pipe design module for the bridge deck, the unit selection module, and the embedded pipe design module for the energy pile update their local models;

[0045] D3: The 3 client sides of the embedded pipe design module for the bridge deck, the unit selection module, and the embedded pipe design module for the energy pile upload the encrypted desensitized parameters to the central server in the form of public keys; the relevant data of client i is expressed as It can be encrypted as

[0046] D4: The central server decrypts the encrypted desensitized parameters uploaded by the 3 client sides using the private key, performs secure aggregation, and then updates the global shared model; the central server decodes the encrypted desensitized parameters through The federated learning on the central server side obtains the global shared model in one or more of the following ways: gradient averaging, federated averaging, and knowledge distillation; aggregate the federated learning of all client models based on knowledge distillation, and update the global model weights of the 3 client sides according to the formula where W n+1 is the global model parameter in the nth round, is the weight of the client sub-model uploaded by client i to the server in the nth round. After updating the model weights in each round, the central server calculates the error and accuracy of the global model. The central server can also control the transmission speed and the shutdown of model training.

[0047] D5: The central server generates a public key for encrypting and transmitting data using the global shared model and distributes it to each client; according to the global shared model, each local client updates the iteration results of other relevant clients as new sample data attribute parameters; for example, the client of the energy pile buried pipe scheme decision tree model updates the snow melting heat load calculated by the bridge deck buried pipe client and the circulating pump parameters and heat pump parameters calculated by the unit selection client according to the global shared model.

[0048] D6: Repeat steps D2 - D5 iteratively until the global shared model is stable. Finally, each client calculates the corresponding result according to the global shared model; calculate the matching design parameters in the active bridge deck deicing and snow melting system of the energy pile, including the bridge deck buried pipe structure, heat pump unit model, circulating pump model, and energy pile buried pipe geometric parameters.

[0049] The buried pipe geometric parameters of the bridge deck are the length of the heat exchange pipe, pipe spacing, pipe diameter, and pipe layout classification; the bridge deck material parameters are the concrete thermal conductivity, concrete temperature, concrete elastic modulus, and concrete expansion coefficient; the heat exchange circulating fluid state parameters are the heat exchange medium thermal conductivity, heat exchange medium temperature, heat exchange medium flow rate, circulating fluid flow rate, density, specific heat, and inlet temperature; the meteorological parameters are the ambient temperature, wind speed, humidity, and snowfall; the thermally induced stress of the bridge deck buried pipe structure includes thermally induced strain, thermally induced bending moment, and thermally induced warping deformation.

[0050] The objects of the cleaning process include incomplete data, incorrect data, duplicate data, and abnormal data; the cleaning process includes identifying invalid values, outliers, and missing values, and processing invalid values, outliers, and missing values. Optionally, the identification method of invalid values, outliers, and missing values can be implemented based on one of the following: 1. Calculate the maximum value, minimum value, median, and upper and lower quartiles of the data set and draw a box plot, and identify outliers or invalid values according to the upper and lower quartiles of the box plot; 2. Draw a scatter plot to visually identify outliers as outliers or invalid values by showing the positional relationship between two sets of data; 3. When the data follows a normal distribution, an outlier is defined as a value in a set of measured values whose deviation from the mean exceeds 3 times the standard deviation.

[0051] Optionally, the processing method of invalid values, outliers, and missing values can be implemented based on one of the following: 1. Direct deletion; 2. Fitting invalid values, outliers, and missing values according to the regression model or maximum likelihood estimation; 3. Filling invalid values, outliers, and missing values according to the statistical data characteristics of the mean, median, and mode.

[0052] The thermal stress prediction model of the bridge deck buried pipe structure is implemented based on one or more algorithms among the simulated annealing method, regression tree algorithm, random forest regression algorithm, support vector regression algorithm, multiple linear regression algorithm, improved support vector regression algorithm by simulated annealing method, and clustering regression algorithm.

[0053] When the thermal stress prediction model of the bridge deck buried pipe structure uses the improved support vector regression algorithm by simulated annealing method to train the thermal stress prediction model of the bridge deck buried pipe structure, the specific steps are as follows:

[0054] S3.1: To solve the relationship between the sample attribute variables and the sample label data, introduce the slack variable ξ i , and the penalty coefficient C to construct a non-linear segmentation support vector classifier considering soft margin; the prediction accuracy of the thermal stress prediction model of the bridge deck buried pipe structure and the self-stability of the model are expressed by the conditional extreme value function as:

[0055]

[0056] where, w is the normal vector of the hyperplane, C is the penalty coefficient, ξ, ξ * are the slack factors, ∈ is the hyperparameter determining the boundary width; y i is the measured result of the training sample; f(X i ) = w·Φ(X) + b is the classification hyperplane of the thermal stress prediction model of the bridge deck buried pipe structure; i is the training sample number; N is the number of training set samples; b is the intercept of the hyperplane; Φ(X) is the non-linear mapping function;

[0057] S3.2: Convert the above conditional extreme value function into a multi-variable function through the Lagrangian function for solution, let the partial derivatives of the Lagrangian function with respect to the optimization objectives w, b, ξ be 0 to obtain the Lagrange multipliers, and convert the original conditional extreme value function into a dual function, so as to find the minimum value within the prediction boundary;

[0058] S3.3: For the non-linear mapping function Φ(X) contained in the classification hyperplane in the thermal stress prediction model of the bridge deck buried pipe structure, the inner product φ(X i ) T φ(X j ) is processed by using a combination of one or more kernel functions among the following Gaussian kernel, linear kernel, polynomial kernel, and Sigmoid kernel;

[0059] Linear kernel: φ(X i ) T φ(X j ) = κ(X i , X j ) = X i T Xj ;

[0060] Polynomial kernel: φ(X i ) T φ(X j ) = κ(X i , X j ) = (αX i T X j + c) d ;

[0061] Gaussian kernel:

[0062] Sigmoid kernel: φ(X i ) T φ(X j ) = κ(X i , X j ) = tanh(αX i T X j + c);

[0063] Combined kernel: φ(X i ) T φ(X j ) = λκ 1 (X i , X j ) + (λ - 1)κ 2 (X i , X j );

[0064] S3.4: Improve the optimization of model parameters in the support vector regression algorithm using the segmented simulated annealing method: insensitive loss function μ, penalty coefficient C, hyperparameters γ, λ, α, c, d in the kernel function;

[0065] S3.4.1: Randomly generate an initial set of model parameters for cross - validation, and record the error value EEP as the current annealing system state E 0 , the initial temperature T 0 , the temperature T in the first annealing stage 1 , and the annealing end temperature is T 2 ;

[0066] S3.4.2: Perturb the model parameters according to the perturbation algorithm to form a new set of model parameters. After cross - validation, obtain the current annealing system state E n , and calculate ΔE = E n - E n-1 ;

[0067] where: m′ i is the perturbed variable, mi is the current variable, s is the perturbation ratio, is a random number in [0, 1], B i , A i is the current variable m i range;

[0068] S3.4.3: When ΔE < 0, accept the new model parameter set and jump to step S3.4.5; otherwise, accept the corresponding model parameter set according to the Metropolis criterion exp(ΔE / kT) - φ > 0 and jump to step S3.4.5; when the above conditions are not met, reject the critical state, execute step S3.4.2, re-perturb to generate a new model parameter set, and perform cross-validation until the parameter set acceptance condition in step S3.4.3 is satisfied;

[0069] S3.4.4: When obtaining the new state, cool down according to the cooling schedule When the set temperature T 1 is not reached, return to step S3.4.2; when the set temperature T 1 is reached, perform a new annealing plan;

[0070] S3.4.5: According to the new perturbation method and annealing plan, continue to perturb the parameter set of the first-stage annealing, and calculate the corresponding state parameter E n ;

[0071] S3.4.6: When ΔE < 0, accept the new model parameter set and jump to step S3.4.7; otherwise, accept the corresponding model parameter set according to the Metropolis criterion and jump to step S3.4.7; when the above conditions are not met, reject the parameter set, execute step S3.4.5, re-perturb to generate a new model parameter set, and perform cross-validation until the parameter set acceptance condition in step S3.4.6 is satisfied;

[0072] S3.4.7 Set the end temperature T 2 as the algorithm exit, and set the global maximum number of EEP calculations to N; when T 2 or N is reached, stop annealing. At this time, the cross-validation error E n of the accepted critical state is the lowest E n , and the corresponding parameters are the best prediction parameters, otherwise return to step S3.4.6.

[0073] The bridge deck buried pipe design module, based on the heat extraction capacity of the energy pile and the unit equipment parameters updated by the central server, establishes a local model of the decision tree for the buried pipe design of the bridge deck structure with meteorological parameters, bridge deck material parameters, and snow melting efficiency parameters as sample attributes on the client side, calculates the buried pipe parameters of the bridge deck as the optimal buried pipe scheme, and encrypts and uploads them to the central processor. The specific steps are as follows:

[0074] D2.1.1: Use the private key to decrypt the public key of the encrypted transmission data, combine the iterative results of the relevant clients of the energy pile buried pipe design module and the unit equipment selection module updated by the global shared model of the central server, and use the samples of the local client as the new sample data attribute parameters. Calculate the information gain of all attributes in the training samples according to the information entropy, and sort all sample attributes according to the information gain.

[0075] D2.1.2: Select the attribute with the largest information gain as the optimal attribute, and use the optimal attribute as the basis for sample division. Samples with the same value of the optimal attribute are used as the same sample set to form the root node.

[0076] D2.1.3: Then regard each root node as a complete data set, divide the samples according to the sub-optimal attribute as the basis, and use the samples with the same value of the sub-optimal attribute as the same sample set to form the leaf nodes, and iterate in turn to form a decision tree. Convert the buried pipe parameters of the bridge deck into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning.

[0077] The energy pile buried pipe design module, based on the heat required for snow melting of the bridge deck buried pipe and the unit equipment parameters updated by the central server, establishes a local model of the decision tree for the energy pile buried pipe design with geotechnical parameters, energy pile material parameters, and energy pile buried pipe geometric parameters as sample attributes on the client side, calculates the buried pipe parameters as the optimal buried pipe scheme, and encrypts and uploads them to the central processor. The energy pile material is the heat exchange pipe material, the concrete thermal conductivity, and the specific heat of the circulating liquid. The geotechnical parameters are the formation lithology, the geotechnical strength, and the geotechnical thermal conductivity.

[0078] The specific steps are as follows:

[0079] D2.2.1: Use the private key to decrypt the public key of the encrypted transmission data, and update the iterative results of the relevant clients of the bridge deck buried pipe design module and the unit equipment selection module according to the global shared model of the central server as the new sample data attribute parameters. Calculate the information gain of all attributes in the training samples according to the information entropy, and sort all attributes according to the information gain.

[0080] The calculation method of the information entropy is as follows;

[0081] The total information entropy of the given sample:

[0082] Total information entropy of the sample subset {s 1j , s 2j , …, s mj}: P ij is the probability of the sample with category C in the sample subset s j . i

[0083] Information entropy value of dividing samples according to attribute A: The corresponding information gain is: Gain(A) = I(s 1 , s 2 , …, s m ) - E(A).

[0084] D2.2.2: The larger the number of sample attribute values, the greater the split information, thus offsetting the influence of the number of sample attribute values; first find the information gain higher than the average value of sample attributes from the candidate sample attributes, and then select the sample attribute prediction with the highest gain rate as the branch attribute of the decision tree;

[0085] D2.2.3: Regard each root node as a complete data set, divide the samples according to the sub - optimal sample attribute, regard the samples with the same sub - optimal attribute value as the same sample set to form leaf nodes, solve the over - fitting problem of the decision tree model through pruning technology, and use the REP method, PEP method or MEP method for pruning the decision tree, and form the decision tree through iterative steps in turn; convert the energy pile buried pipe parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning;

[0086] The unit equipment selection and design module, combined with the updated bridge deck snow melting load and heat extraction of the energy pile from the central server, on the local client, establishes a random forest local model of unit equipment with unit equipment parameters as sample attributes, calculates the unit equipment parameters as the optimal unit plan and encrypts and uploads them to the central processor; the unit equipment parameters include: the refrigeration method of the heat pump unit, the refrigeration capacity of the heat pump unit, the flow rate of the heat pump unit, the flow rate of the chilled water pump, the head of the chilled water pump, the flow rate of the chilled water pump, the head information of the chilled water pump; the specific steps are as follows:

[0087] D2.3.1: Use the private key to decrypt the public key of the encrypted transmission data, update the iteration results of the relevant clients of the bridge deck buried pipe design module and the energy pile buried pipe module and the unit parameters of the local client as the new sample data attribute parameters according to the global shared model of the central server; use the Bootstrap sampling method to randomly generate sample subsets from the samples as the training samples of one of the decision tree models, and repeat the sampling k times to form k decision tree training samples;

[0088] ​D2.3.2: Train decision trees based on the subsets of attributes in k decision tree training samples to form k mutually independent random decision trees;

[0089] D2.3.3: The random selection classifier of the random forest votes on the unit equipment parameters predicted by the k decision trees, and the voting result is used as the optimal unit equipment of the random forest;

[0090] D2.3.4: Convert the unit equipment parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning.

[0091] The formation of the decision tree local model is generated by one or more of the following algorithms: ID3 algorithm, C4.5 algorithm, CART algorithm.

[0092] In step D1, the sample data can be obtained through at least one of the following: on-site monitoring technology, questionnaire survey technology, image intelligent recognition technology, and thermal imaging technology. Obtaining sample data includes the following information: bridge deck buried pipe parameters, environmental information parameters, geotechnical parameters, unit equipment parameters, energy pile buried pipe geometric parameters, and energy pile material parameters.

[0093] Advantages of the present invention: Compared with the existing energy pile bridge deck deicing and snow melting technology, the present invention has the following technical advantages:

[0094] (1) The present invention trains a snow melting efficiency model suitable for the bridge deck buried pipe structure through the K-nearest neighbor algorithm, which can accurately quantify the snow melting efficiency corresponding to different buried pipe forms, providing a basis for the reasonable design of the energy pile bridge deck deicing and snow melting system.

[0095] (2) The present invention obtains the thermal stress of the bridge deck buried pipe structure through the support vector regression algorithm improved by the simulated annealing method, which can correctly evaluate the influence of the repeated freezing and thawing process on the buried pipe bridge deck structure and improve the safety and stability of the design results.

[0096] (3) The energy pile active bridge deck deicing and snow melting federated learning design method proposed by the present invention can be used for the coordinated design of the three major modules of the bridge deck buried pipe, unit selection, and energy pile buried pipe in the energy pile bridge deck deicing and snow melting system. After fully considering the interaction, a reasonable and scientific overall design scheme can be obtained, improving the stability and reliability of the system. Description of the Drawings

[0097] Figure 1 is a schematic flow chart of an optional snow melting efficiency K-nearest neighbor prediction method for the bridge deck buried pipe structure in an embodiment of the present invention;

[0098] Figure 2 is a schematic diagram of the support vector regression structure of the thermal stress of the bridge deck buried pipe structure in an embodiment of the present invention;

[0099] Figure 3 It is a schematic diagram of the thermal stress support vector regression process of an optional bridge deck buried pipe structure in an embodiment of the present invention;

[0100] Figure 4 It is a schematic diagram of the support vector regression algorithm flow improved by the simulated annealing method in an embodiment of the present invention;

[0101] Figure 5 It is a schematic diagram of the design concept of the active bridge deck deicing and snow melting of the energy pile in an embodiment of the present invention;

[0102] Figure 6 It is a schematic diagram of the framework structure of the intelligent federated learning algorithm for the active bridge deck deicing and snow melting of the energy pile in an embodiment of the present invention;

[0103] Figure 7 It is a schematic diagram of the local model structure of the buried pipe design decision tree of an optional bridge deck buried pipe structure in an embodiment of the present invention;

[0104] Figure 8 It is a schematic diagram of the process of the local model of the buried pipe design decision tree of an optional bridge deck buried pipe structure in an embodiment of the present invention;

[0105] Figure 9 It is a schematic diagram of the local model structure of the buried pipe design decision tree of an optional energy pile in an embodiment of the present invention;

[0106] Figure 10 It is a schematic diagram of the process of the local model of the buried pipe design decision tree of an optional energy pile in an embodiment of the present invention;

[0107] Figure 11 It is a schematic diagram of the local model structure of the random forest of the unit equipment in an embodiment of the present invention;

[0108] Figure 12 It is a schematic diagram of the process of the local model of the random forest of the unit equipment in an embodiment of the present invention. Detailed implementation manners

[0109] The following describes in detail the specific implementation manners of the present invention patent. The protection scope of the present invention patent is not limited to the description of this implementation manner. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the protection scope of the present invention.

[0110] Figure 1 It is a schematic diagram of the process of the K-nearest neighbor prediction method for the snow melting efficiency of an optional bridge deck buried pipe structure in an embodiment of this patent, including the following steps:

[0111] Step T1: Snow melting efficiency training sample data and corresponding label collection. The acquired training sample data set includes geometric parameters of bridge deck buried pipes, bridge deck material parameters, meteorological parameters, and heat exchange circulation fluid state parameters. The label data includes relevant information reflecting the snow melting efficiency of the bridge deck buried pipe structure, such as heat exchange efficiency, heat exchange per meter, snow melting time, and surface snow-free rate. Among them, the geometric parameters of the bridge deck buried pipes include heat exchange pipe length, pipe spacing, pipe diameter, and pipe layout classification; the bridge deck material parameters include concrete thermal conductivity, concrete temperature, concrete elastic modulus, and concrete expansion coefficient; the heat exchange circulation fluid state parameters include heat exchange medium thermal conductivity, heat exchange medium temperature, and heat exchange medium flow rate.

[0112] Step T2: Preprocessing of snowmelt efficiency training sample data, quality analysis and cleaning of snowmelt efficiency training sample data, extracting data distribution law from the cleaned sample data, normalizing it, and evaluating the correlation of sample data between different labels. Define the state vector, and sort the sample attribute variables according to the correlation through sensitivity analysis, and take the attribute variables with the highest ranking as the elements of the state parameter. The state vector at time t can be expressed as X(t)=[x 1 (t),x 1 (t-1),x 2 (t),x 2 (t-1),x 3 (t),x 3 (t-1),…,x n (t),x n (t-1)]. Among them, data cleaning is carried out according to the following method: first, the maximum value, minimum value, median, and upper and lower quartiles of the data set are calculated and a box plot is drawn. The outliers or invalid values ​​are identified according to the upper and lower quartiles of the box plot. Secondly, the identified outliers or invalid values ​​are filled according to the mode. Table 1 shows the sample space and sample quality for the snow melting efficiency prediction of a certain energy pile bridge deck deicing and snow melting in Dalian.

[0113] Table 1 Sample space and sample quality for prediction of snow melting efficiency of an energy pile bridge deck for deicing and snow melting in Dalian

[0114]

[0115] Step T3: Select Chebyshev distance: d xy =max i (x i -y i ) The distance metric calculates the mutual distance between state vectors in the state space, reflecting the degree of correlation between each state vector X and the target vector Y in the state space.

[0116] Step T4: Input the sample attribute information of the prediction set, form a state vector of the training set sample attributes, search for the K nearest neighbors of the current vector in the training set, and form a K-nearest neighbor prediction model. The model calculates the snow melting efficiency prediction results corresponding to the prediction set.

[0117] Step T5: Evaluate the snow melting efficiency prediction results of the bridge deck buried pipe structure. Compare the prediction results of the prediction set with the labels corresponding to the prediction set samples to form a confusion matrix. According to the confusion matrix, the accuracy and sensitivity of the snow melting efficiency prediction model of a certain energy pile bridge deck deicing and snow melting system in Dalian can be further calculated to be 0.73 and 0.79 respectively.

[0118] Step T6: Input 30 groups of geometric parameters of the bridge deck buried pipes, meteorological parameters, heat exchange circulating liquid state parameters, and bridge deck material parameters to be predicted as the attribute space of the prediction set, and perform snow melting efficiency prediction calculations for the bridge deck buried pipe structure. The distribution characteristics of the prediction results are shown in Table 2.

[0119] Table 2 Distribution characteristics of the snow melting efficiency prediction results of 30 groups of bridge decks

[0120] Distribution Index Heat Exchange Power (kW) Heat Transfer per Linear Meter (W / m) Snow Melting Time (h) Snow-Free Surface Rate (%) Maximum Value 1.27 45 42 80 Minimum Value 0.74 15.4 5 0 Average Value 0.98 32.7 23 50 Standard Deviation 0.14 9.3 8.8 23.7

[0121] Figure 2 This is an optional flowchart for predicting the thermal stress of the bridge deck buried pipe structure in an embodiment of the patent, including the following steps:

[0122] Step S1: Collect the training sample data of the thermal stress of the bridge deck buried pipe structure and the corresponding labels. The obtained training sample data set includes the geometric parameters of the bridge deck buried pipes, meteorological parameters, bridge deck material parameters, and heat exchange circulating liquid state parameters, and the label data includes information related to the thermal stress of the bridge deck buried pipe structure. Among them, the buried pipe geometric parameters are the heat exchange pipe length, pipe spacing, pipe diameter, pipe layout classification, and buried pipe depth; the bridge deck material parameters are the heat transfer coefficient of the heat exchange medium, temperature of the heat exchange medium, flow rate of the heat exchange medium, heat transfer coefficient of the concrete, and temperature of the concrete; the meteorological parameters are the ambient temperature, wind speed, humidity, and snowfall; the heat exchange circulating liquid state parameters are the flow rate, density, specific heat, and inlet temperature of the circulating liquid; the thermal stress of the bridge deck buried pipe structure includes thermal strain, thermal bending moment, and thermal warping deformation.

[0123] Step S2: Preprocessing of the thermally induced stress training sample data of the bridge deck buried pipe structure. Clean and perform quality analysis on the thermally induced stress training sample data of the bridge deck buried pipe structure, extract the distribution law of the data, perform normalization processing on the data, evaluate the correlation of the sample data between different labels, and analyze the data characteristics of the training set and prediction set required for supervised learning. The specific operation process is as follows: First, draw a scatter plot of the sample attributes to visually identify outliers, that is, abnormal values or invalid values, by showing the positional relationship between two sets of data. Second, in order to reduce the amount of calculation, directly delete the sample data corresponding to the identified outliers or invalid values.

[0124] Table 3 Prediction sample space and sample quality of the thermally induced stress of the bridge deck buried pipe structure for deicing and snow melting of an energy pile in Dalian

[0125]

[0126]

[0127] Step S3: Training of the thermally induced stress model. Establish the training of the thermally induced stress prediction model for the bridge deck buried pipe structure according to the sample attribute information of the training set. Figure 3 and Figure 4 are respectively the flow chart and structure diagram of an optional simulated annealing method improved support vector regression for training the thermally induced stress prediction model of the bridge deck buried pipe structure in the patent embodiment. The specific steps are as follows:

[0128] Step S3.1: To solve the relationship between the sample attribute data and the sample label data, introduce a slack variable ξ i , and a penalty coefficient C to construct a non-linear segmentation support vector classifier considering soft margin. The prediction accuracy of the thermally induced stress prediction model of the bridge deck buried pipe structure and the self-stability of the model are represented by the conditional extreme value function.

[0129] Step S3.2: It is necessary to convert the above conditions into a multivariate function for solution through the Lagrangian function. Let the partial derivatives of the Lagrangian function with respect to the optimization objectives w, b, and ξ be 0 to obtain the Lagrange multipliers, and the original conditional extreme value function can be transformed into a dual function, so as to find the minimum value within the prediction boundary.

[0130] Step S3.3: Solve the inner product φ(X i ) T φ(X j ) = κ(X i , X j ) = tanh(αX i T X j +c) in the mapping function of the formula through the Sigmoid kernel: φ(X i ) T φ(Xj )。

[0131] Step S3.4: Improve the optimization of model parameters in the support vector regression model by using the segmented simulated annealing method, mainly including: the insensitive loss function μ, the penalty coefficient C, and the hyperparameters γ, λ, α, c, d in the kernel function.

[0132] Step S3.4.1: Randomly generate an initial parameter set for cross-validation, and record the error value EEP as the current annealing system state E 0 , the initial temperature T 0 , the temperature T in the first annealing stage 1 , and the annealing end temperature is T 2 。

[0133] Step S3.4.2: Perturb the parameters according to the perturbation algorithm to form a new parameter set. After cross-validation, obtain the current annealing system state E n , calculate ΔE = E n -E n-1 。

[0134] where: m′ i is the perturbed variable, m i is the current variable, s is the perturbation ratio, is a random number in [0,1], B i , A i are the ranges of the current variable m i ;

[0135] Step S3.4.3: If ΔE < 0, accept the new parameter set and jump to Step S3.4.5; otherwise, accept the corresponding parameter set according to the Metropolis criterion and jump to Step S3.4.5; if neither of the above conditions is satisfied, reject the critical state, then return to Step S3.4.2, re-perturb to generate a new parameter set, and perform cross-validation until the parameter set acceptance condition in Step S3.4.3 is satisfied.

[0136] Step S3.4.4: When a new state is obtained, cool down according to the cooling schedule . When the set temperature T 1 is not reached, return to Step S3.4.2; when the set temperature T 1 is reached, perform a new annealing plan.

[0137] Step S3.4.5: According to the new perturbation method and annealing plan, continue to perturb the parameter set of the first-stage annealing, and calculate the corresponding state parameter E n 。

[0138] Step S3.4.6: If ΔE < 0, accept the new parameter set and jump to Step S3.4.7; otherwise, accept the corresponding parameter set according to the Metropolis criterion and jump to Step S3.4.7; if neither of the above conditions is satisfied, reject the parameter set, execute Step S3.4.5, re-perturb to generate a new model parameter set, and perform cross-validation until the parameter set acceptance condition in Step S3.4.6 is met;

[0139] Step S3.4.7 Set the ending temperature to T 2 is the algorithm exit, and the global maximum number of EEP calculations is set to N. When reaching T 2 or N, stop annealing. At this time, the cross-validation error E n of the accepted critical state should be the lowest E n , and the corresponding parameters should be the best prediction parameters; otherwise, return to Step S3.4.6.

[0140] Step S4: Input the sample attribute information of the thermal stress prediction set of the bridge deck buried pipe structure. The sample attributes include: the buried pipe parameters of the bridge deck slab buried pipe structure, the bridge deck material parameters, and the heat exchange circulating liquid state parameters.

[0141] Step S5: Calculate the thermal stress prediction results corresponding to the thermal stress prediction set of the bridge deck buried pipe structure according to the thermal stress prediction model of the bridge deck buried pipe structure.

[0142] Step S6: Evaluate the thermal stress prediction results of the bridge deck buried pipe structure. Input the prediction set attribute information into the training model to calculate the corresponding prediction results. Compare the prediction results obtained in Step S5 with the labels corresponding to the prediction set samples, and calculate that the prediction error of the thermal stress prediction model of the bridge deck buried pipe structure is approximately 68.7%.

[0143] Step S7: Input 30 groups of geometric parameters of the bridge deck buried pipes, bridge deck material parameters, meteorological parameters, and heat exchange circulating liquid state parameters to be predicted as the prediction set attribute space, and perform thermal stress prediction calculations for the bridge deck buried pipe structure; Table 4 shows the distribution characteristics of the thermal stress prediction results of 30 groups of bridge deck buried pipe structures.

[0144] Table 4 Distribution characteristics of the thermal stress prediction results of 30 groups of bridge deck buried pipe structures

[0145]

[0146]

[0147] Another aspect of the embodiments of the present invention also provides an active bridge deck deicing and snow melting design method for energy piles, and its conceptual diagram is as Figure 5 shown. Establish a federated learning algorithm framework for active bridge deck deicing and snow melting design of energy piles based on the client-server architecture,Figure 6 It is a structural diagram of an active bridge deck deicing and snow melting design method for energy piles based on a federated learning architecture. All sub-module learning calculations are implemented on the client side, including three major modules: the bridge deck buried pipe design module, the unit selection module, and the energy pile buried pipe design module. Finally, the desensitized parameters calculated on the client side are aggregated to the central server for calculation, and then sent to each data holder (client) to update their local models until the global model is robust. The specific steps are as follows:

[0148] Step D1: Collect and organize bridge data, pile foundation data, and meteorological data through on-site investigation technology, questionnaire survey technology, and image intelligent recognition technology, perform simple data preprocessing, analyze the characteristics of sample data, and allocate them to the corresponding calculation module clients.

[0149] Step D2: Three clients, namely the bridge deck buried pipe design module, the unit selection module, and the energy pile buried pipe design module, update their local models. The local models of the three clients are as follows:

[0150] Local client 1 (bridge deck buried pipe design module): According to the heat extraction capacity of the energy pile and the unit equipment parameters updated by the central server, on the client side, a local model of the buried pipe design decision tree for the bridge deck buried pipe structure is established with meteorological parameters, bridge deck material parameters, and snow melting efficiency parameters as sample attributes, and the bridge deck buried pipe parameters are calculated as the optimal buried pipe scheme and encrypted and uploaded to the central processor. Figure 7 It is a structural diagram of an optional local model of the buried pipe design decision tree for the bridge deck buried pipe structure provided by an embodiment of the present invention. Taking the ID3 algorithm as an embodiment of the present invention, Figure 8 It is a flowchart of the local model of the buried pipe design decision tree for the bridge deck buried pipe structure of an embodiment. The specific steps of the s-th round of this client are as follows:

[0151] Step D2.1.1: Use the private key to decrypt the public key of the encrypted transmission data, combine the global shared model of the central server to update the iterative results of the relevant clients of the energy pile buried pipe design module and the unit equipment selection module, and use the samples of the local client as the basis for the new sample data attribute parameters. Calculate the information gain of all attributes in the training samples according to the information entropy, and sort all attributes according to the information gain.

[0152] Step D2.1.2: Select the attribute with the largest information gain as the optimal attribute, and perform sample partitioning based on the optimal attribute. Samples with the same value of the optimal attribute are used as the same sample set to form the root node.

[0153] Step D2.1.3: Then, regard each root node as a complete data set, divide the samples based on the sub-optimal attribute, form leaf nodes by taking samples with the same value of the sub-optimal attribute as the same sample set, and iteratively form a decision tree according to the PEP pruning method. Convert the bridge deck buried pipe parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model in federated learning.

[0154] Local client 2 (energy pile buried pipe design module): According to the updated heat demand for snow melting of the bridge deck buried pipe and the unit equipment parameters from the central server, on the client, establish a local model of the energy pile buried pipe design decision tree with geotechnical parameters, energy pile material parameters, and energy pile buried pipe geometric parameters as sample attributes, calculate the energy pile buried pipe parameters as the optimal buried pipe scheme, and encrypt and upload them to the central processor. Figure 9 This is a structural diagram of a local model of an energy pile buried pipe design decision tree provided by an embodiment of the present invention. The energy pile materials are the heat exchange pipe material, concrete thermal conductivity, and specific heat of the circulating fluid; the geotechnical parameters are formation lithology, geotechnical strength, and geotechnical thermal conductivity;. Taking the C4.5 algorithm as an embodiment of the present invention, Figure 10 This is a flowchart of a local model of an energy pile buried pipe design decision tree for an embodiment. The specific steps of the s-th round of this client are as follows:

[0155] Step D2.2.1: Use the private key to decrypt the public key of the encrypted transmission data, and update the iteration results of relevant clients of the bridge deck buried pipe design module and the unit equipment selection module according to the global shared model of the central server as the new sample data attribute parameters. Calculate the information gain of all attributes in the training samples according to the information entropy, and sort all attributes according to the information gain.

[0156] Step D2.2.2: The larger the number of sample attribute values, the greater the split information, thus offsetting the influence brought by the number of sample attribute values. First, find the sample attributes with information gain higher than the average level from the candidate sample attributes, and then select the sample attribute with the highest gain rate as the branch attribute of the decision tree for prediction.

[0157] Step D2.2.3: Then, regard each root node as a complete data set, divide the samples based on the sub-optimal attribute, form leaf nodes by taking samples with the same value of the sub-optimal attribute as the same sample set, and iteratively form a decision tree according to the REP pruning method. Convert the energy pile buried pipe parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model in federated learning.

[0158] Local client 3 (unit equipment selection design module): Figure 11A structure diagram of the random forest local model of the unit equipment provided by the present invention. Combining the snow melting load of the bridge deck and the heat extraction amount of the energy pile updated by the central server, on the local client, a random forest local model of the unit equipment is established with the unit equipment parameters as sample attributes, and the unit equipment parameters are calculated as the optimal unit plan and encrypted and uploaded to the central processor. The unit equipment parameters include: the refrigeration mode of the heat pump unit, the refrigeration capacity of the heat pump unit, the flow rate of the heat pump unit, the flow rate of the chilled water pump, the head of the chilled water pump, the flow rate of the chilled water pump, and the head information of the chilled water pump. Figure 12 The flowchart of the random forest local model of the unit equipment. The specific steps of the s-th round of the local model of the client are as follows:

[0159] Step D2.3.1: Use the private key to decrypt the public key of the encrypted transmission data, and update the iterative results of the relevant clients of the bridge deck buried pipe design module and the energy pile buried pipe module and the unit parameters of the local client based on the global shared model of the central server as the new sample data attribute parameters. Use the Bootstrap sampling method to randomly generate a sample subset from the samples as the training samples of one of the decision tree models, and repeat the sampling k times to form k decision tree training samples.

[0160] Step D2.3.2: Train the decision tree according to the attribute subsets in the k decision tree training samples to form k independent random decision trees.

[0161] Step D2.3.3: The random selection classifier of the random forest votes on the unit equipment parameters predicted by the k decision trees, and the voting result is used as the optimal unit equipment of the random forest.

[0162] Step D2.3.4: Convert the unit equipment parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning.

[0163] Step D3: The three clients of the bridge deck buried pipe design module, the unit selection module, and the energy pile buried pipe design module upload the encrypted desensitized parameters to the server in the form of public keys.

[0164] Step D4: The central server uses the private key to decrypt the encrypted desensitized parameters uploaded by the three clients, aggregates the decrypted parameters of all client models based on knowledge distillation for federated learning, and updates the global model weights of the M clients according to the following formula: Where W s+1 is the global model parameter of the s-th round, is the weight of the client sub-model uploaded by the i-th client to the server in the s-th round. After each round of model weight update, the central server calculates the error and accuracy of the global model. The central server can also control the transmission speed and the shutdown of model training.

[0165] Step D5: The central server generates a public key for encrypting and transmitting data using the global shared model and distributes it to each client. According to the global shared model, each local client updates the iteration results of other relevant clients as new sample data attribute parameters. For example, for the client of the energy pile buried pipe scheme decision tree model, it updates the snow melting heat load calculated by the bridge deck buried pipe client and the circulation pump parameters and heat pump parameters calculated by the unit selection client according to the global shared model.

[0166] Step D6: Repeat steps D2 - D5 continuously for iteration until the global model is robust. Finally, the client calculates the corresponding results according to the global model. The matching design parameters in a certain active bridge deck de - icing and snow melting system of energy piles in Dalian are calculated, including the bridge deck buried pipe structure, heat pump unit model, circulation pump model, energy pile structural characteristics, and buried pipe parameters, as shown in Table 5.

[0167] Table 5 Design Parameter Results of a Certain Energy Pile Green Low - Carbon Building in Dalian

[0168]

Claims

1. A design method for active bridge deck deicing and snow melting based on machine learning, characterized in that, the snow melting efficiency of the bridge deck buried pipe structure and the thermal stress of the bridge deck buried pipe structure caused by repeated freezing and thawing processes are mutually restricted. The energy pile bridge deck deicing and snow melting technology design takes into account both the snow melting efficiency of the bridge deck buried pipe structure and the thermal stress of the bridge deck buried pipe structure, and calculates the matching unit equipment and energy pile structure; The design method for active bridge deck deicing and snow melting based on machine learning obtains the K-nearest neighbor prediction results of the snow melting efficiency of the bridge deck buried pipe structure and the support vector regression prediction results of the thermal stress of the bridge deck buried pipe through inputting the same sample attribute data. The design method for active bridge deck deicing and snow melting using the federated learning architecture determines the unit equipment and energy pile structure by selecting the bridge deck buried pipe structure parameters corresponding to the snow melting efficiency and thermal stress; The K-nearest neighbor prediction method for the snow melting efficiency of the bridge deck buried pipe structure includes the following steps: T1: Obtain the snow melting efficiency training sample data and the corresponding labels; T2: Preprocessing of the training sample data for snow melting efficiency, including quality analysis and cleaning of the training sample data for snow melting efficiency. After cleaning, extract the distribution law, normalization processing, and evaluation of the sample data correlation between different labels of the training sample data for snow melting efficiency, and divide it into a training set and a prediction set; define the state vector, sort the sample attribute variables according to the correlation through sensitivity analysis, and take the sample attribute variables with higher ranking as the elements of the state parameters. The state vector at time t is expressed as X(t)=[x 1 (t), x 1 (t - 1), x 2 (t), x 2 (t - 1), x 3 (t), x 3 (t - 1), …, x n (t), x n (t - 1)]; T3: Select the distance metric criterion; T4: 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; input the sample attribute data of the prediction set into the K-nearest neighbor prediction model, and the predicted mean corresponding to the K nearest neighbors Calculate the predicted result of the snowmelt efficiency corresponding to the prediction set for the K-nearest neighbor prediction model; T5: Evaluate the prediction results of the snow melting efficiency of the bridge deck buried pipe structure. Compare the prediction results in step T4 with the labels corresponding to the prediction set samples to form a confusion matrix; further evaluate the model prediction error and model accuracy based on the confusion matrix; T6: Take the geometric parameters, meteorological parameters, heat exchange circulating liquid state parameters, and bridge deck material parameters of the bridge deck buried pipe to be predicted as the prediction set attribute space, and perform the prediction calculation of the snow melting efficiency of the bridge deck buried pipe structure; The support vector regression prediction method for the thermal stress of the bridge deck buried pipe includes the following steps: S1: Obtain the training sample data of the thermal stress of the bridge deck buried pipe structure and the corresponding labels; the training sample data of the thermal stress of the bridge deck buried pipe structure includes the geometric parameters of the bridge deck buried pipe, meteorological parameters, bridge deck material parameters, and heat exchange circulating liquid state parameters, and the label data includes the relevant information reflecting the thermal stress of the bridge deck buried pipe structure; S2: Preprocess the training sample data of the thermal stress of the bridge deck buried pipe structure and analyze the data features; perform quality analysis, feature analysis, cleaning processing on the training sample data of the thermal stress of the bridge deck buried pipe structure, and divide it into the training set of the thermal stress of the bridge deck buried pipe structure and the prediction set of the thermal stress of the bridge deck buried pipe structure for supervised learning; S3: Train the prediction model of the thermal stress of the bridge deck buried pipe structure, and establish the prediction model of the thermal stress of the bridge deck buried pipe structure according to the sample attribute information of the training set of the thermal stress of the bridge deck buried pipe structure; S4: Input the sample attribute information of the prediction set of the thermal stress of the bridge deck buried pipe structure, and the sample attributes include: the geometric parameters of the bridge deck buried pipe, the bridge deck material parameters, and the heat exchange circulating liquid state parameters; S5: Calculate the thermal stress prediction results corresponding to the prediction set of the thermal stress of the bridge deck buried pipe structure according to the prediction model of the thermal stress of the bridge deck buried pipe structure; S6: Evaluate the prediction results of the thermal stress of the bridge deck buried pipe structure. Compare the prediction results obtained in S5 with the labels corresponding to the samples of the prediction set of the thermal stress of the bridge deck buried pipe structure, calculate the prediction error of the prediction model of the thermal stress of the bridge deck buried pipe structure, and evaluate its accuracy; S7: Use the geometric parameters of the embedded pipes in the bridge deck, meteorological parameters, bridge deck material parameters, and heat exchange circulating fluid state parameters to form the attribute space of the prediction set for the thermal stress of the embedded pipe structure on the bridge deck, and perform the prediction calculation of the thermal stress of the embedded pipe structure on the bridge deck; For the energy pile active bridge deck deicing and snow melting design method based on the federated learning architecture, establish a federated learning algorithm framework for the energy pile active bridge deck deicing and snow melting design based on the client-server architecture; implement the learning calculations of all sub-modules on the client side, and the client includes three major modules: the embedded pipe design module for the bridge deck, the unit selection module, and the embedded pipe design module for the energy pile; after summarizing the desensitized parameters calculated by the client to the central server for calculation, distribute them to each client to update its local model until the global model is robust.

2. The energy pile active bridge deck deicing and snow melting design method based on machine learning according to claim 1, characterized in that the geometric parameters of the embedded pipes in the bridge deck are the length of the heat exchange pipes, the pipe spacing, the pipe diameter, and the pipe layout classification; the bridge deck material parameters are the concrete thermal conductivity, the concrete temperature, the concrete elastic modulus, and the concrete expansion coefficient; the heat exchange circulating fluid state parameters are the heat exchange medium thermal conductivity, the heat exchange medium temperature, the heat exchange medium flow rate, the circulating fluid flow rate, the density, the specific heat, and the inlet temperature; the meteorological parameters are the ambient temperature, the wind speed, the humidity, and the snowfall; the thermal stress of the embedded pipe structure on the bridge deck includes thermal strain, thermal bending moment, and thermal warping deformation.

3. The energy pile active bridge deck deicing and snow melting design method based on machine learning according to claim 1 or 2, characterized in that the objects of the cleaning process include incomplete data, incorrect data, duplicate data, and abnormal data; the content of the cleaning process includes identifying invalid values, outliers, and missing values, and processing invalid values, outliers, and missing values.

4. The energy pile active bridge deck deicing and snow melting design method based on machine learning according to claim 1, characterized in that the thermal stress prediction model of the embedded pipe structure on the bridge deck is implemented based on one or more algorithms among the simulated annealing method, the regression tree algorithm, the random forest regression algorithm, the support vector regression algorithm, the multiple linear regression algorithm, the improved support vector regression algorithm by the simulated annealing method, and the clustering regression algorithm.

5. The energy pile active bridge deck deicing and snow melting design method based on machine learning according to claim 4, characterized in that when the thermal stress prediction model of the embedded pipe structure on the bridge deck uses the improved support vector regression algorithm by the simulated annealing method to train the thermal stress prediction model of the embedded pipe structure on the bridge deck, the specific steps are as follows: S3.1: To solve the relationship between sample attribute variables and sample label data, slack variables are introduced , and a penalty coefficient C to construct a non-linear segmentation support vector classifier considering soft margins; the prediction accuracy and the self-stability of the prediction model for the thermal stress of the bridge deck buried pipe structure are expressed by the conditional extreme value function as follows: where w is the normal vector of the hyperplane, C is the penalty coefficient, ξ, ξ * is the slack factor, is the hyperparameter that determines the width of the decision boundary; is the measured result of the training sample; is the classification hyperplane of the prediction model for the thermal stress of the bridge deck buried pipe structure; i is the training sample number; N is the number of training set samples; b is the intercept of the hyperplane; is the non-linear mapping function; S3.2: Convert the above conditional extreme value function into a multivariate function for solution through the Lagrangian function. Let the partial derivatives of the Lagrangian function with respect to the optimization objective w , b , ξ be 0 to obtain the Lagrange multipliers, convert the original conditional extreme value function into a dual function, and thus find the minimum value within the prediction boundary; S3.3: Nonlinear mapping function contained in the classification hyperplane of the thermal stress prediction model for the bridge deck buried pipe structure , and the inner product is processed by using a combination of one or more of the following kernel functions: Gaussian kernel, linear kernel, polynomial kernel, and Sigmoid kernel; S3.4: Improve the optimization of model parameters in the support vector regression algorithm using the segmented simulated annealing method: insensitive loss function μ, Penalty coefficient C, hyperparameters in the kernel function γ , λ, α, c, d ; S3.4.1: Randomly generate an initial set of model parameters for cross-validation, and record the error value EEP as the current annealing system state E 0 , the initial temperature T 0 , the temperature T in the first annealing stage 1 , the annealing end temperature is T 2 ; S3.4.2: According to the perturbation algorithm perturb the model parameters to form a new set of model parameters, and obtain the current annealing system state E through cross-validation n , calculate ∆E = E n - E n-1 ; Wherein: m′ i is the variable after perturbation, m i is the current variable, s is the perturbation ratio, is a random number in [0, 1], B i and A i is the current variable m i range; S3.4.3: When ∆E < 0, accept the new model parameter set and jump to step S3.4.5; otherwise, accept the corresponding model parameter set according to the Metropolis criterion exp(∆E / kT) > 0, jump to step S3.4.5; when the above conditions are not met, reject the critical state, execute step S3.4.2, re-perturb to generate a new model parameter set, and perform cross-validation until the parameter set acceptance condition in step S3.4.3 is satisfied; S3.4.4: When obtaining the new state, cool down according to the cooling plan If the set temperature T 1 is not reached, return to step S3.4.2; if the set temperature T 1 is reached, perform a new annealing plan; S3.4.5: Continue to perturb the parameter set of the first-stage annealing according to the new perturbation method and annealing schedule, and calculate the corresponding state parameter E n ; S3.4.6: When ∆E < 0, accept the new model parameter set, and jump to step S3.4.7; otherwise, accept the corresponding model parameter set according to the Metropolis criterion, and jump to step S3.4.7; when the above conditions are not met, reject the parameter set, execute step S3.4.5, re-perturb to generate a new model parameter set, and perform cross-validation until the parameter set acceptance condition in step S3.4.6 is satisfied; S3.4.7 Set the ending temperature T 2 This is the algorithm exit. The global maximum number of EEP calculations is set to N; when reaching T 2 or N, stop annealing. At this time, the cross-validation error E n of the accepted critical state is the lowest E n , and the corresponding parameters are the best prediction parameters. Otherwise, return to step S3.4.

6.

6. The active bridge deck de-icing and snow melting design method based on machine learning according to claim 1, characterized in that, the bridge deck buried pipe design module, according to the heat extraction capacity of the energy pile and the unit equipment parameters updated by the central server, on the client, establishes a local model of the buried pipe design decision tree for the bridge deck buried pipe structure with meteorological parameters, bridge deck material parameters, and snow melting efficiency parameters as sample attributes, calculates the bridge deck buried pipe parameters as the optimal buried pipe scheme, and encrypts and uploads them to the central processor; the specific steps are as follows: D2.1.1: Use the private key to decrypt the public key of the encrypted transmission data, combine the global shared model of the central server to update the iterative results of the relevant clients of the energy pile buried pipe design module and the unit equipment selection module, and use the samples of the local client as the basis for the new sample data attribute parameters; calculate the information gain of all attributes in the training samples according to the information entropy, and sort all sample attributes according to the information gain; D2.1.2: Select the attribute with the largest information gain as the optimal attribute, and use the optimal attribute as the basis for sample division. Samples with the same value of the optimal attribute are used as the same sample set to form the root node; D2.1.3: Then regard each root node as a complete data set, and use the sub-optimal attribute as the basis for sample division. Samples with the same value of the sub-optimal attribute are used as the same sample set to form the leaf nodes, and iterate in turn to form a decision tree; convert the bridge deck buried pipe parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning; the energy pile buried pipe design module, according to the heat required for snow melting of the bridge deck buried pipe and the unit equipment parameters updated by the central server, on the client, establishes a local model of the energy pile buried pipe design decision tree with geotechnical parameters, energy pile material parameters, and energy pile buried pipe geometric parameters as sample attributes, calculates the buried pipe parameters as the optimal buried pipe scheme, and encrypts and uploads them to the central processor; the energy pile material is the heat exchange pipe material, the concrete thermal conductivity, and the specific heat of the circulating liquid; the geotechnical parameters are the formation lithology, the geotechnical strength, and the geotechnical thermal conductivity; the specific steps are as follows: D2.2.1: Use the private key to decrypt the public key of the encrypted transmission data, and update the iterative results of the relevant clients of the bridge deck buried pipe design module and the unit equipment selection module according to the global shared model of the central server as the new sample data attribute parameters; calculate the information gain of all attributes in the training samples according to the information entropy, and sort all attributes according to the information gain; D2.2.2: The larger the number of sample attribute values, the greater the split information, so as to offset the influence of the number of sample attribute values; first find the information gain higher than the sample attribute average value from the candidate sample attributes, and then select the sample attribute with the highest gain rate as the predicted branch attribute of the decision tree; D2.2.3: Treat each root node as a complete dataset, divide the samples based on the sub - optimal sample attribute, form leaf nodes by taking samples with the same value of the sub - optimal attribute as the same sample set, and use the REP method, PEP method or MEP method to prune the decision tree, and iteratively form the decision tree in turn; convert the energy pile buried pipe parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning. The unit equipment selection and design module, in combination with the updated bridge deck snow - melting load and heat extraction of the energy pile from the central server, on the local client, establishes a local random forest model of the unit equipment with the unit equipment parameters as sample attributes, calculates the unit equipment parameters as the optimal unit plan and encrypts and uploads them to the central processor; the unit equipment parameters include: the refrigeration mode of the heat pump unit, the refrigeration capacity of the heat pump unit, the flow rate of the heat pump unit, the flow rate of the chilled water pump, the head of the chilled water pump, the flow rate of the chilled water pump, the head information of the chilled water pump; the specific steps are as follows: D2.3.1: Use the private key to decrypt the public key of the encrypted transmission data, update the iteration results of the relevant clients of the bridge deck buried pipe design module and the energy pile buried pipe module and the unit parameters of the local client according to the global shared model of the central server as the new sample data attribute parameters; use the Bootstrap sampling method to randomly generate a sample subset from the samples as the training samples of one of the decision tree models, and repeat the sampling k times to form k decision tree training samples. D2.3.2: Train the decision tree according to the attribute subsets in the k decision tree training samples to form k independent random decision trees. D2.3.3: The random selection classifier of the random forest votes on the unit equipment parameters predicted by the k decision trees, and the voting result is used as the optimal unit equipment of the random forest; According to the method for designing active bridge deck de - icing and snow - melting of energy piles based on machine learning described in claim 1, in T3, the distance metric criterion is one of Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, cosine of the included angle, or a combined distance designed according to the above distance metric criteria, combined distance D2.3.4: Convert the unit equipment parameters into encrypted parameters using the public key and upload them to the central server for the iteration of the global shared model of federated learning.

7. According to the method for designing active bridge deck de - icing and snow - melting of energy piles based on machine learning described in claim 6, characterized in that the formation of the decision tree local model is generated by one or more of the following algorithms: ID3 algorithm, C4.5 algorithm, CART algorithm.

8. According to the method for designing active bridge deck de - icing and snow - melting of energy piles based on machine learning described in claim 1, characterized in that In T1, the snow - melting efficiency training sample attribute data includes bridge deck buried pipe geometric parameters, bridge deck material parameters, meteorological parameters, heat exchange circulating liquid state parameters, and the label data includes information related to the snow - melting efficiency of the bridge deck buried pipe structure, including heat exchange efficiency, heat transfer per unit length, snow - melting time, and surface snow - free rate.

9. The active bridge deck de-icing and snow melting design method based on machine learning according to claim 1, characterized in that, In T3, the distance metric criterion is one of Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, standardized Euclidean distance, Mahalanobis distance, cosine of the included angle, or a combined distance is designed according to the above distance metric criteria, and the combined distance represents the degree of correlation between each state vector X and the target vector Y in the state space; In the formula, is a selected distance metric criterion, is another selected distance metric criterion, λ is the ratio of a selected distance metric criterion.

10. The active bridge deck de-icing and snow melting design method based on machine learning according to claim 1, characterized in that, For the active bridge deck de-icing and snow melting design method based on the federated learning architecture, establish an active bridge deck de-icing and snow melting design federated learning algorithm framework based on the client-server architecture. The specific steps are as follows: D1: Obtain and organize bridge materials, pile foundation materials, and meteorological materials, perform data preprocessing, analyze the characteristics of sample data, and allocate them to the corresponding computing module clients; D2: The three clients of the bridge deck buried pipe design module, the unit selection module, and the energy pile buried pipe design module update their local models; D3: The three clients of the bridge deck buried pipe design module, the unit selection module, and the energy pile buried pipe design module upload the encrypted and desensitized parameters to the central server in the form of public keys; D4: The central server decrypts the encrypted and desensitized parameters uploaded by the three clients using the private key, performs secure aggregation, and then updates the global shared model; the central server decodes the encrypted and desensitized parameters; the federated learning on the central server side obtains the global shared model in one or more of the following ways: gradient averaging, federated averaging, and knowledge distillation; D5: The central server generates a public key for encrypting transmitted data from the global shared model and distributes it to each client; according to the global shared model, each local client updates the iteration results of other relevant clients as new sample data attribute parameters; D6: Repeat steps D2 to D5 continuously until the global shared model is robust. Finally, each client calculates the corresponding results according to the global shared model; calculate the matching design parameters in the active bridge deck de-icing and snow melting system of the energy pile, including the bridge deck buried pipe structure, the model of the heat pump unit, the model of the circulation pump, and the geometric parameters of the energy pile buried pipe.

Images