Tunnel ellipticity historical data collaborative analysis system and method based on federated learning

Abstract

The present invention relates to the technical field of data analysis systems, and in particular to a collaborative analysis system and method for tunnel ellipticity historical data based on federated learning, comprising a data center; a processing center, which is communicatively connected to the data center and is used to: receive an initial ellipticity processing model and an ellipticity detection model distributed by the data center; train the initial ellipticity processing model based on training data to generate a project-specific ellipticity processing model; upload the project-specific ellipticity processing model to the data center; receive an ellipticity detection model updated by the data center; a user interface module, which is communicatively connected to the data center and is used to: receive processing requirements sent by users; based on collaborative analysis results of the data center, provide users with processing solutions corresponding to the processing requirements in accordance with a predetermined mode, realize local processing of data and secure transmission of model parameters through a federated learning framework, effectively avoid direct sharing of original data, and greatly reduce the risk of data leakage.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis systems, and in particular to a system and method for collaboratively analyzing tunnel ellipticity historical data based on federated learning. Background Art

[0002] With the rapid advancement of modernization, tunnel engineering is increasingly used in transportation, water conservancy and other fields. During tunnel construction and operation, ovality is a crucial parameter that is directly related to the safety, stability and service life of the tunnel. Traditional tunnel ovality monitoring and analysis methods mainly rely on single-point measurement and simple data processing, which is difficult to fully and accurately reflect the actual condition of the tunnel.

[0003] In recent years, with the development of sensing technology and data analysis technology, some advanced tunnel ellipticity monitoring and analysis systems have gradually emerged. These systems usually use distributed sensor networks to collect a large amount of ellipticity data and use machine learning and other technologies for data analysis and prediction. However, these systems still face many challenges in practical applications.

[0004] First, tunnel ellipticity data is highly sensitive and involves engineering safety and commercial secrets. Traditional centralized data processing methods are prone to data leakage risks and are not conducive to multi-party data sharing and collaboration. Secondly, the geological conditions and construction methods of different tunnel projects vary greatly, and a single analysis model is difficult to adapt to the complex and changeable engineering reality. Furthermore, tunnel ellipticity data is usually scattered among different construction units and operating organizations. How to effectively integrate these scattered data resources and extract valuable knowledge is an urgent problem to be solved.

[0005] In addition, existing tunnel ellipticity analysis systems often lack a flexible model update mechanism, making it difficult to adapt to new data features and engineering requirements in a timely manner. At the same time, given the uneven quality of data, how to reasonably allocate the weights of different data sources and improve the accuracy of the overall analysis results is also a major challenge facing existing technologies.

[0006] In view of the above problems, there is an urgent need for a collaborative analysis system and method for tunnel ellipticity historical data that can effectively integrate scattered data resources, flexibly adapt to different engineering characteristics, and continuously optimize while protecting data privacy. Summary of the invention

[0007] The present invention aims to solve the problems existing in the existing tunnel ellipticity analysis technology, such as insufficient data privacy protection, difficulty in effective collaboration of multi-party data, poor model adaptability, and inflexible update mechanism. A tunnel ellipticity historical data collaborative analysis system and method based on federated learning are proposed.

[0008] The present invention proposes a collaborative analysis system for tunnel ellipticity historical data based on federated learning, including:

[0009] Data centers for:

[0010] Based on the ellipticity distribution data provided by all processing centers, an initial ellipticity processing model and an ellipticity detection model are established;

[0011] Distributing the initial ellipticity processing model and the ellipticity detection model to the processing center;

[0012] Receive ellipticity distribution data for tunnel projects;

[0013] Based on the ellipticity detection model, analyzing and classifying the ellipticity distribution data;

[0014] Select a processing center for the tunnel project based on the classification results and pre-set strategies;

[0015] receiving and updating a project-specific ellipticity processing model uploaded by the processing center;

[0016] updating the ellipticity detection model based on the project-specific ellipticity processing model;

[0017] A processing center, in communication with the data center, is used to:

[0018] Receiving an initial ellipticity processing model and an ellipticity detection model distributed by the data center;

[0019] Based on the training data, the initial ellipticity processing model is trained to generate a project-specific ellipticity processing model;

[0020] uploading the project-specific ellipticity processing model to the data center;

[0021] Receiving an ellipticity detection model updated by the data center;

[0022] A user interface module is connected to the data center for:

[0023] Receive processing requirements sent by users;

[0024] Based on the collaborative analysis results of the data center, a processing solution corresponding to the processing requirements is provided to the user in a predetermined mode.

[0025] Preferably, the data center is also used for:

[0026] Obtain the analysis results stored in each processing center as the target results;

[0027] According to the difficulty of all target results and the historical ellipticity data fed back by all processing centers, a preset weight algorithm is used to assign a collaborative processing weight value to each target result;

[0028] Based on the collaborative processing weight value of the target result and the preset algorithm corresponding to the corresponding collaborative processing weight value, a collaborative analysis result of each target result is obtained.

[0029] Preferably, the processing center further comprises:

[0030] The first level learning module is used to pre-process the historical data of ellipticity;

[0031] The second level learning module is used to perform federated learning using the preprocessed ellipticity history data.

[0032] Preferably, the ellipticity detection model is a decision tree model, which is used to analyze the ellipticity change trend and provide a basis for formulating a tunnel ellipticity treatment plan.

[0033] Preferably, the ellipticity processing model is a neural network model, which is used to analyze the relationship between the tunnel ellipticity change and each characteristic variable, and provide a basis for predicting the tunnel ellipticity change trend and formulating the ellipticity processing plan.

[0034] Preferably, the neural network model comprises:

[0035] Backbone network module, used to process multi-source data and generate multi-layer feature maps;

[0036] A feature extraction module, used to extract the multi-level feature graph generated by the backbone network module into multiple high-dimensional abstract features;

[0037] The feature mapping module is used to map the multiple high-dimensional abstract features output by the feature extraction module to the outputs corresponding to the ellipticity prediction task, the mileage prediction task and the grade prediction task respectively.

[0038] Preferably, the data center is also used for:

[0039] The connection relationship between each processing center is obtained based on the trust network between each processing center;

[0040] The objective function is established according to the relationship between the degree of trust of several processing centers in other processing centers, the score of the processing center on its own ellipticity historical data and the average score of the score of all processing centers on their own ellipticity historical data;

[0041] Based on the trained objective function and the connection relationship between the processing centers, a model aggregation solution based on trust network and federated learning is constructed.

[0042] Preferably, the data center is also used for:

[0043] According to the objective function, the ellipticity historical data stored in each processing center is collaboratively trained in a collaborative training manner to obtain an optimal decision for determining whether the aggregation strategy is met;

[0044] Determining the first-level learning module for performing data preprocessing on the ellipticity historical data determined to be in compliance with the aggregation strategy;

[0045] The second-level learning module is determined to perform federated learning on the ellipticity historical data after data preprocessing.

[0046] Preferably, the data center is also used for:

[0047] According to the difficulty of all target results and the historical ellipticity data fed back by all processing centers, a weight is set for each target result as the original weight of each target result;

[0048] According to the original weight of each target result, the average value of the original weight is calculated;

[0049] Based on the calculated average of the original weights, a co-processing weight value is assigned to each target result.

[0050] The collaborative analysis method of tunnel ellipticity historical data based on federated learning includes the following steps:

[0051] (1) The data center establishes the initial ellipticity processing model and ellipticity detection model based on the ellipticity distribution data provided by all processing centers;

[0052] (2) The data center distributes the initial ellipticity processing model and the ellipticity detection model to the processing center;

[0053] (3) After the tunnel project to be tested is started, the ellipticity distribution data of the tunnel project is obtained and uploaded to the data center;

[0054] (4) The data center uses the ellipticity detection model to analyze and classify the uploaded data, selects a processing center for the tunnel project based on the classification and preset strategy, and sends the initial ellipticity processing model to the processing center;

[0055] (5) The processing center trains the initial ellipticity processing model using the training data to obtain a project-specific ellipticity processing model, and uploads the model to the data center;

[0056] (6) the data center updates the ellipticity detection model according to the project-specific ellipticity processing model uploaded by the processing center, and sends the updated ellipticity detection model to the processing center;

[0057] (7) If the tunnel project has completed the formulation of the ellipticity processing plan, the data center receives the ellipticity processing model uploaded by the processing center and continues to update the ellipticity detection model, otherwise it returns to step (3);

[0058] (8) Repeat steps (3) to (7) until the ellipticity treatment plan for all tunnel projects is completed.

[0059] The system of the present invention innovatively introduces a federated learning framework to achieve collaborative analysis of data without leaving the local area, effectively solving the problem of data privacy protection. At the same time, the system adopts a flexible model design and update strategy, which can adapt to the characteristics of different tunnel projects, and continuously learn from new data to continuously improve the analysis accuracy. In addition, the present invention also proposes a model aggregation scheme based on a trust network and an innovative weight distribution mechanism, which effectively improves the robustness of the system and the reliability of the analysis results.

[0060] Specifically, the beneficial effects of the present invention are mainly reflected in the following aspects:

[0061] First, in terms of data security, the present invention realizes local data processing and secure transmission of model parameters through the federated learning framework, effectively avoiding direct sharing of original data and greatly reducing the risk of data leakage. This not only protects the commercial secrets of all parties involved, but also creates conditions for collaborative analysis of multi-party data, which is conducive to the accumulation and sharing of industry knowledge.

[0062] Secondly, in terms of model performance, the system of the present invention realizes collaborative learning and sharing of knowledge by integrating data and models from multiple processing centers. This approach not only significantly improves the generalization ability of the model, enabling it to adapt to different tunnel engineering characteristics, but also makes full use of data resources from all parties to improve the overall analysis accuracy. Especially in new tunnel projects with scarce data, the advantages of this collaborative learning are more obvious.

[0063] Furthermore, the system of the present invention has strong adaptability and scalability. By designing a flexible model update mechanism, the system can continuously learn from new project data and continuously optimize model performance. This dynamic learning capability enables the system to timely adapt to new situations and new problems in engineering practice, providing continuous technical support for the long-term safe operation of tunnels.

[0064] In terms of data processing, the model aggregation scheme based on the trust network and the innovative weight distribution mechanism proposed in this invention effectively improve the system's ability to process data of different qualities. This not only improves the reliability of the analysis results, but also enhances the system's resistance to potential malicious data and improves the robustness of the entire system.

[0065] Finally, from the application perspective, the system of the present invention provides powerful technical support for the full life cycle management of tunnel projects. From real-time monitoring and early warning in the construction phase, to maintenance decision support in the operation phase, to the planning and design of future projects, the system can provide accurate and reliable data analysis results. This not only helps to improve the safety and economy of tunnel projects, but also contributes to the technological progress of the entire industry.

[0066] In general, this invention effectively solves several key problems faced by existing tunnel ellipticity analysis technologies through innovative technical solutions, and provides new ideas and tools for intelligent and refined management of tunnel engineering. Its advantages in data security, model performance, system adaptability, etc. will make important contributions to the safe and efficient development of tunnel engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a logic block diagram of the whole system of the present invention.

[0068] Figure 2 The present invention is a logic block diagram of the model processing and model updating.

[0069] Figure 3 The logical block diagram of the model training and iteration of the present invention. DETAILED DESCRIPTION

[0070] See also Figure 1-3 The present invention provides a system and method for collaborative analysis of tunnel ellipticity historical data based on federated learning. The system includes a data center 1, a processing center 2 and a user interface module 3, and realizes safe and efficient collaborative analysis of tunnel ellipticity historical data through federated learning.

[0071] Specifically, the system of the present invention includes a data center 1, a processing center 2 and a user interface module 3. The data center 1, as the core of the system, is responsible for the initialization, distribution and update of the model. The processing center 2 is responsible for the storage of local data and model training. The user interface module 3 provides an interface for interacting with the user, realizes the reception of processing requirements and the provision of processing solutions.

[0072] The functions of data center 1 mainly include the following aspects: First, based on the ellipticity distribution data provided by all processing centers 2, the initial ellipticity processing model T0 and the ellipticity detection model A0 are established. The ellipticity distribution data here usually includes information in multiple dimensions such as geometric parameters, geological conditions, and construction methods of the tunnel. The establishment of the initial model adopts the transfer learning technology in machine learning, which makes full use of existing knowledge and improves the initial performance of the model.

[0073] Secondly, the data center 1 distributes the initial ellipticity processing model T0 and the ellipticity detection model A0 to each processing center 2. This step adopts a secure data transmission protocol to ensure the security of the model during transmission. Preferably, the present invention adopts a secure transmission method based on homomorphic encryption, so that even if the model is intercepted, sensitive information therein cannot be obtained.

[0074] Again, when a new tunnel project is launched, Data Center 1 receives the ellipticity distribution data of the project. These data usually come from the real-time monitoring system at the tunnel construction site, including the real-time ellipticity data of the tunnel, changes in geological conditions and other information. Data Center 1 then uses the ellipticity detection model A0 to analyze and classify these data. During the analysis process, the present invention uses a deep learning algorithm based on the attention mechanism, which can automatically identify the key factors that have a greater impact on the ellipticity change.

[0075] After the classification is completed, the data center 1 selects the most suitable processing center 2 for the tunnel project according to the classification results and the preset strategy. The selection strategy takes into account multiple factors such as the computing power, historical processing experience, and current load of the processing center 2. Preferably, the present invention adopts a selection algorithm based on multi-objective optimization, and its mathematical expression is as follows:

[0076] S = argmax i (w1C i +w2E i -w3L i )

[0077] Among them, S is the selected processing center, C i represents the computing power of the ith processing center, E i Indicates its historical processing experience, L i represents its current load, w1, w2 and w3 are weight coefficients. By adjusting these weight coefficients, different selection strategies can be flexibly adapted.

[0078] Data Center 1 is also responsible for receiving and updating the project-specific ellipticity processing model Ti uploaded by Processing Center 2. This process uses the FedAvg algorithm, the core idea of ​​which is to perform weighted averaging of the model parameters of each processing center 2. The specific update formula is as follows:

[0079]

[0080] Among them, T global is the updated global model, N is the number of processing centers involved in the update, n i is the data volume of the ith processing center, n is the total data volume, T i is the local model of the i-th processing center.

[0081] Finally, Data Center 1 processes the model T based on the updated project-specific ellipticity i , update the ellipticity detection model Ai+1. This update process adopts the incremental learning method, and only fine-tunes some parameters of the model, thereby maintaining the stability of the model while improving its adaptability to new data.

[0082] The main functions of processing center 2 include receiving the initial model distributed by data center 1, training the model based on local training data, generating a project-specific ellipticity processing model Ti, and uploading it to data center 1. During the model training process, processing center 2 uses differential privacy technology to protect the privacy of the original data by adding carefully designed noise to the training data. The specific noise adding method is as follows:

[0083]

[0084] in, is the data after adding noise, x is the original data, Indicates that the mean is 0 and the variance is σ 2 The selection of σ requires a trade-off between privacy protection and data usefulness. This paper recommends setting σ to 0.1 times the standard deviation of the original data.

[0085] The user interface module 3 is responsible for receiving the processing requirements sent by the user, and providing the user with a processing solution corresponding to the processing requirements according to a predetermined mode based on the collaborative analysis results of the data center 1. This module uses natural language processing technology, which can understand the user's natural language input and convert complex analysis results into easy-to-understand processing suggestions.

[0086] The system of the present invention also includes some optimization designs. For example, the data center 1 is also used to obtain the analysis results stored in each processing center 2 as the target results, and according to the difficulty of all target results and the historical ellipticity data fed back by all processing centers 2, a preset weight algorithm is used to assign a collaborative processing weight value to each target result. This design takes into account the differences in data quality and processing capabilities of different processing centers 2, and improves the accuracy of the overall analysis results by assigning different weights.

[0087] The weight allocation algorithm adopts an improved algorithm based on the entropy weight method, and its mathematical expression is as follows:

[0088]

[0089] Among them, w i is the weight of the ith processing center, H i is the information entropy of its data, and N is the total number of processing centers. Information entropy H i The calculation formula is:

[0090]

[0091] Among them, K is the number of categories of data, p ik is the proportion of the k-th category data in the i-th processing center.

[0092] Based on the assigned collaborative processing weight values ​​and the corresponding preset algorithms, data center 1 can obtain collaborative analysis results for each target result. This process adopts the idea of ​​ensemble learning to weightedly fuse the analysis results of multiple processing centers 2, thereby obtaining a more accurate and stable final result.

[0093] In addition, the processing center 2 of the present invention also includes a first-level learning module and a second-level learning module. The first-level learning module is used to perform data preprocessing on the ellipticity historical data, including steps such as data cleaning, outlier detection, and feature engineering. The second-level learning module is used to perform federated learning using the preprocessed ellipticity historical data. This two-level learning structure effectively improves learning efficiency and model performance.

[0094] In general, the system of the present invention realizes safe and efficient collaborative analysis of tunnel ellipticity historical data through federated learning, which not only protects data privacy but also improves the accuracy and efficiency of analysis. The system has broad application prospects in tunnel construction and maintenance, and can effectively improve the construction quality and operational safety of tunnels. The system of the present invention further adopts advanced ellipticity detection models and processing models to improve the accuracy and efficiency of tunnel ellipticity analysis. Specifically, the ellipticity detection model of the present invention adopts a decision tree model, while the ellipticity processing model is constructed based on a neural network model.

[0095] Preferably, the ellipticity detection model of the present invention is a decision tree model. The model is mainly used to analyze the trend of ellipticity changes and provide a basis for the formulation of tunnel ellipticity treatment plans. The decision tree model is selected because it has good interpretability and high computational efficiency. In an embodiment of the present invention, the decision tree model uses the CART (Classification and Regression Trees) algorithm, which can handle classification and regression problems at the same time and is very suitable for tunnel ellipticity data that has both discrete attributes (such as geological type) and continuous attributes (such as ellipticity values).

[0096] The construction process of the decision tree model can be expressed as the following mathematical formula:

[0097]

[0098] The Gain function represents information gain, which can be calculated using the Gini index or entropy. For example, when using the Gini index, the Gain function can be expressed as:

[0099]

[0100] Where K is the number of categories, p i is the proportion of the i-th category, n left and n right are the number of samples of the left and right child nodes after segmentation, and n is the total number of samples.

[0101] In the system of the present invention, the ellipticity processing model adopts a neural network model. The model is used to analyze the relationship between the tunnel ellipticity change and each characteristic variable, and provides a basis for predicting the tunnel ellipticity change trend and formulating the ellipticity processing plan. The neural network model is selected because it has a strong nonlinear fitting ability and can capture the complex pattern in the tunnel ellipticity change.

[0102] In a preferred embodiment of the present invention, the neural network model adopts a multi-layer perceptron (MLP) structure, including an input layer, multiple hidden layers and an output layer. Each hidden layer uses a ReLU activation function, and the output layer selects a suitable activation function according to the task type. For example, for regression tasks (such as predicting a specific ellipticity value), an activation function may not be used; for classification tasks (such as predicting the level of ellipticity change), a softmax function may be used.

[0103] The forward propagation process of the neural network model can be expressed by the following mathematical formula:

[0104] h (l) =f(W (l) Jh (l-1)+b (l) )

[0105] Among them, h (l) represents the output of the lth layer, W (l) and b (l) are the weight matrix and bias vector of the lth layer, respectively, and f is the activation function. For the ReLU activation function, it is defined as:

[0106] f(x)=max(0,x)

[0107] In order to further improve the performance of the model, the neural network model of the present invention also includes a backbone network module, a feature extraction module and a feature mapping module. The backbone network module is used to process multi-source data and generate multi-layer feature maps. In one embodiment of the present invention, the backbone network adopts the ResNet50 structure, which effectively solves the gradient vanishing problem of the deep network by introducing residual connections.

[0108] The feature extraction module is used to extract the multi-level feature maps generated by the backbone network module into multiple high-dimensional abstract features. This process can be achieved through global average pooling or self-attention mechanism. For example, when using the self-attention mechanism, it can be expressed as:

[0109]

[0110] Among them, Q, K and are query, key, and value matrices, respectively, and d k is the dimension of the key vector.

[0111] The feature mapping module is responsible for mapping the multiple high-dimensional abstract features output by the feature extraction module to the outputs corresponding to the ellipticity prediction task, mileage prediction task, and grade prediction task. This multi-task learning design can make full use of the correlation between tasks and improve the generalization ability of the model.

[0112] The system of the present invention also introduces a model aggregation scheme based on a trust network, which further improves the efficiency and robustness of federated learning. Specifically, the data center obtains the connection relationship between each processing center based on the trust network between each processing center. This connection relationship not only considers the direct trust between the processing centers, but also the indirect trust relationship, thereby forming a complete trust network.

[0113] When building a trust network, the present invention adopts a method based on a graph neural network (GNN). First, each processing center is regarded as a node in the graph, and the trust relationship between processing centers is regarded as an edge. Then, a graph convolutional network (GCN) is used to learn the representation of the node:

[0114]

[0115] in, is the adjacency matrix with self-loops added, yes The degree matrix, H (l) is the node feature matrix of the lth layer, W (l) is a learnable weight matrix and σ is a nonlinear activation function.

[0116] Based on the learned node representation, the system of the present invention further establishes an objective function. The function comprehensively considers the relationship between the degree of trust of the processing center in other processing centers, the score of the processing center on its own ellipticity historical data, and the average score of all processing centers on their own ellipticity historical data. Preferably, the objective function can be expressed as:

[0117]

[0118] Among them, w ij represents the degree of trust that processing center i has in processing center j, f(h i ) is the characteristic representation of processing center i, s i is the score of the processing center i on its own ellipticity historical data, is the average of all treatment center scores, and λ is the balance parameter.

[0119] By optimizing this objective function, the system of the present invention can learn a model aggregation scheme that takes into account both the trust relationship between processing centers and the data quality. This scheme improves the efficiency of federated learning while also enhancing the system's robustness to malicious or low-quality data.

[0120] In general, the present invention greatly improves the analysis efficiency and accuracy of tunnel ellipticity historical data by introducing advanced ellipticity detection and processing models, combined with a model aggregation scheme based on a trust network. These innovations not only provide more reliable decision-making support for tunnel construction and maintenance, but also provide new ideas for the application of federated learning in the engineering field. In the further optimization of the present invention, the data center also adopts an innovative collaborative training mechanism. Specifically, the data center uses a collaborative training method to collaboratively train the ellipticity historical data stored in each processing center according to the aforementioned objective function, so as to obtain the optimal decision for determining whether it complies with the aggregation strategy. This collaborative training mechanism not only improves the overall performance of the model, but also effectively prevents abnormal data from individual processing centers from having a negative impact on the global model.

[0121] In a preferred embodiment of the present invention, collaborative training uses a method based on Federated dDistillation. This method first trains the model locally in each processing center, and then transmits the model output (not the model parameters) to the data center. The data center aggregates these outputs, generates a global "soft label", and then returns these soft labels to each processing center for further local training. This process can be expressed by the following mathematical formula:

[0122]

[0123] Among them, y soft is the generated soft label, N is the number of processing centers, and f i (x) is the prediction of the i-th processing center model for input x.

[0124] Based on this collaborative training mechanism, the system of the present invention can better utilize data dispersed in various processing centers while avoiding the privacy risks that may be caused by directly transmitting raw data or model parameters.

[0125] In addition, the system of the present invention also optimizes the design of the learning module. Specifically, the system determines a first-level learning module for preprocessing the historical ellipticity data that is judged to meet the aggregation strategy. This preprocessing process includes steps such as data cleaning, feature extraction, and standardization, aiming to improve the efficiency and effect of the subsequent learning process.

[0126] In the data cleaning stage, the system uses an anomaly detection algorithm based on the Local Outlier Factor (LOF). The core idea of ​​the LOF algorithm is to compare the local density of a sample with its neighboring samples to identify outliers. Its mathematical expression is as follows:

[0127]

[0128] Among them, LOF k (A) is the local anomaly factor of point A, N k (A) is the set of k nearest neighbors of A, lrd k (A) is the local reachable density of A.

[0129] In the feature extraction stage, the system uses a combination of principal component analysis (PCA) and autoencoder. PCA is used for linear feature extraction, while autoencoder is used to capture nonlinear features in the data. The loss function of the autoencoder can be expressed as:

[0130]

[0131] Here, f and g are the encoder and decoder functions respectively, and λ is the weight of the regularization term.

[0132] The system of the present invention also defines a second-level learning module for performing federated learning on the historical ellipticity data after data preprocessing. At this stage, the system adopts an improved federated average (FedAvg) algorithm. Unlike the traditional FedAvg algorithm, the algorithm of the present invention introduces a weighted mechanism based on trust, so that more trusted processing centers have greater influence in model updates. The specific update formula is as follows:

[0133]

[0134] Among them, w t+1 is the updated global model parameter, n i is the data volume of the i-th processing center, trust i is the trustworthiness of the ith processing center, is the model parameter after local update of the i-th processing center in the t+1th round.

[0135] In order to further improve the performance and reliability of the system, the present invention also designs an innovative weight allocation mechanism. Specifically, the data center sets a weight for each target result based on the difficulty of all target results and the historical ellipticity data fed back by all processing centers, which serves as the original weight of each target result. This design takes into account the quality differences of different data sources and helps to improve the accuracy of the final analysis results.

[0136] In one embodiment of the present invention, the calculation of the original weight adopts an improved algorithm based on the entropy weight method. First, the information entropy of each processing center data is calculated:

[0137]

[0138] Among them, H i is the information entropy of the data of the ith processing center, p ij is the proportion of the jth category of data in the ith processing center, and m is the number of data categories.

[0139] Then, the original weights are calculated based on the information entropy:

[0140]

[0141] Among them, w i is the original weight of the ith processing center, and N is the total number of processing centers.

[0142] Next, the system calculates the average of the original weights based on the original weights of each target result. This step is intended to obtain a global reference standard for subsequent weight adjustments. The formula for calculating the average is as follows:

[0143]

[0144] Among them, w i is the original weight of the ith processing center, and N is the total number of processing centers.

[0145] Finally, based on the calculated average value of the original weights, the system assigns a co-processing weight value to each target result. This process uses an adaptive weight adjustment algorithm, the core idea of ​​which is to dynamically adjust the weight according to the difference between the original weight and the average value. The specific adjustment formula is as follows:

[0146]

[0147] in, is the adjusted collaborative processing weight value, and α is an adjustable parameter used to control the intensity of the adjustment. In a preferred embodiment of the present invention, the value range of α is 0.1 to 0.5.

[0148] This weight distribution mechanism not only takes into account the quality and difficulty of the data, but also introduces a dynamic adjustment mechanism to better adapt to the characteristics of different data distributions and processing centers.

[0149] Finally, the present invention also proposes a collaborative analysis method for tunnel ellipticity historical data based on federated learning, which includes the following steps:

[0150] (1) The data center establishes the initial ellipticity processing model and ellipticity detection model based on the ellipticity distribution data provided by all processing centers;

[0151] (2) The data center distributes the initial ellipticity processing model and the ellipticity detection model to the processing center;

[0152] (3) After the tunnel project to be tested is started, the ellipticity distribution data of the tunnel project is obtained and uploaded to the data center;

[0153] (4) The data center uses the ellipticity detection model to analyze and classify the uploaded data, selects a processing center for the tunnel project based on the classification and preset strategy, and sends the initial ellipticity processing model to the processing center;

[0154] (5) The processing center trains the initial ellipticity processing model using the training data to obtain a project-specific ellipticity processing model, and uploads the model to the data center;

[0155] (6) the data center updates the ellipticity detection model according to the project-specific ellipticity processing model uploaded by the processing center, and sends the updated ellipticity detection model to the processing center;

[0156] (7) If the tunnel project has completed the formulation of the ellipticity processing plan, the data center receives the ellipticity processing model uploaded by the processing center and continues to update the ellipticity detection model, otherwise it returns to step (3);

[0157] (8) Repeat steps (3) to (7) until the ellipticity processing scheme for all tunnel projects is formulated. This method includes a series of steps, from establishing the initial model in the data center, to local training in the processing center, and then to updating and iterating the model, forming a complete federated learning process.

[0158] In this method, the model update strategy is particularly noteworthy. When the tunnel project completes the formulation of the ellipticity treatment plan, the system will not stop the learning process immediately, but continue to update the ellipticity detection model. This design enables the system to continuously learn from new projects and continuously improve the performance and adaptability of the model.

[0159] In general, the present invention constructs an efficient, reliable and secure collaborative analysis system for tunnel ellipticity historical data through an innovative collaborative training mechanism, optimized learning module design, flexible weight distribution strategy and a complete federated learning method. This system can not only effectively protect data privacy, but also make full use of data resources scattered in various processing centers to provide accurate decision support for tunnel construction and maintenance. At the same time, the system's adaptive learning ability also enables it to continuously evolve and adapt to changing engineering needs and data characteristics.

[0160] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A collaborative analysis system of tunnel ellipticity historical data based on federated learning, characterized in that: include: Data centers for: Based on the ellipticity distribution data provided by all processing centers, an initial ellipticity processing model and an ellipticity detection model are established; Distributing the initial ellipticity processing model and the ellipticity detection model to the processing center; Receive ellipticity distribution data for tunnel projects; Based on the ellipticity detection model, analyzing and classifying the ellipticity distribution data; Select a processing center for the tunnel project based on the classification results and pre-set strategies; receiving and updating a project-specific ellipticity processing model uploaded by the processing center; updating the ellipticity detection model based on the project-specific ellipticity processing model; A processing center, in communication with the data center, is used to: Receiving an initial ellipticity processing model and an ellipticity detection model distributed by the data center; Based on the training data, the initial ellipticity processing model is trained to generate a project-specific ellipticity processing model; uploading the project-specific ellipticity processing model to the data center; Receiving an ellipticity detection model updated by the data center; A user interface module is connected to the data center for: Receive processing requirements sent by users; Based on the collaborative analysis results of the data center, a processing solution corresponding to the processing requirements is provided to the user in a predetermined mode.

2. The system according to claim 1, characterized in that The data center is also used for: Obtain the analysis results stored in each processing center as the target results; According to the difficulty of all target results and the historical ellipticity data fed back by all processing centers, a preset weight algorithm is used to assign a collaborative processing weight value to each target result; Based on the collaborative processing weight value of the target result and the preset algorithm corresponding to the corresponding collaborative processing weight value, a collaborative analysis result of each target result is obtained.

3. The system according to claim 1, characterized in that The processing center also includes: The first level learning module is used to pre-process the historical data of ellipticity; The second level learning module is used to perform federated learning using the preprocessed ellipticity history data.

4. The system according to claim 1, characterized in that The ellipticity detection model is a decision tree model, which is used to analyze the ellipticity change trend and provide a basis for formulating a tunnel ellipticity treatment plan.

5. The system according to claim 1, characterized in that The ellipticity processing model is a neural network model, which is used to analyze the relationship between the tunnel ellipticity change and each characteristic variable, and provide a basis for predicting the tunnel ellipticity change trend and formulating the ellipticity processing plan.

6. The system according to claim 5, characterized in that The neural network model includes: Backbone network module, used to process multi-source data and generate multi-layer feature maps; A feature extraction module, used to extract the multi-level feature graph generated by the backbone network module into multiple high-dimensional abstract features; The feature mapping module is used to map the multiple high-dimensional abstract features output by the feature extraction module to the outputs corresponding to the ellipticity prediction task, the mileage prediction task and the grade prediction task respectively.

7. The system according to claim 1, characterized in that The data center is also used for: The connection relationship between each processing center is obtained based on the trust network between each processing center; The objective function is established according to the relationship between the degree of trust of several processing centers in other processing centers, the score of the processing center on its own ellipticity historical data and the average score of the score of all processing centers on their own ellipticity historical data; Based on the trained objective function and the connection relationship between the processing centers, a model aggregation solution based on trust network and federated learning is constructed.

8. The system according to claim 7, characterized in that The data center is also used for: According to the objective function, the ellipticity historical data stored in each processing center is collaboratively trained in a collaborative training manner to obtain an optimal decision for determining whether the aggregation strategy is met; Determining the first-level learning module for performing data preprocessing on the ellipticity historical data determined to be in compliance with the aggregation strategy; The second-level learning module is determined to perform federated learning on the ellipticity historical data after data preprocessing.

9. The system according to claim 1, characterized in that The data center is also used for: According to the difficulty of all target results and the historical ellipticity data fed back by all processing centers, a weight is set for each target result as the original weight of each target result; According to the original weight of each target result, the average value of the original weight is calculated; Based on the calculated average of the original weights, a co-processing weight value is assigned to each target result.

10. A method for collaboratively analyzing tunnel ellipticity historical data based on federated learning based on the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: (1) The data center establishes the initial ellipticity processing model and ellipticity detection model based on the ellipticity distribution data provided by all processing centers; (2) The data center distributes the initial ellipticity processing model and the ellipticity detection model to the processing center; (3) After the tunnel project to be tested is started, the ellipticity distribution data of the tunnel project is obtained and uploaded to the data center; (4) The data center uses the ellipticity detection model to analyze and classify the uploaded data, selects a processing center for the tunnel project based on the classification and preset strategy, and sends the initial ellipticity processing model to the processing center; (5) The processing center trains the initial ellipticity processing model using the training data to obtain a project-specific ellipticity processing model, and uploads the model to the data center; (6) the data center updates the ellipticity detection model according to the project-specific ellipticity processing model uploaded by the processing center, and sends the updated ellipticity detection model to the processing center; (7) If the tunnel project has completed the formulation of the ellipticity processing plan, the data center receives the ellipticity processing model uploaded by the processing center and continues to update the ellipticity detection model, otherwise it returns to step (3); (8) Repeat steps (3) to (7) until the ellipticity treatment plan for all tunnel projects is completed.

Images