Urban underground space settlement prediction method based on transfer learning

Abstract

The invention relates to the technical field of building informatization, in particular to an urban underground space settlement prediction method and system based on transfer learning, and the method comprises the steps: obtaining geological feature data, construction process data and historical settlement monitoring data of a target city; obtaining geological feature data, construction process data and historical settlement monitoring data of the source domain city with rich settlement data; constructing and training a source domain neural network model based on the source domain city data; constructing a target domain neural network model based on the target city data and the source domain neural network model; according to the target domain neural network model and the source domain neural network model, executing a transfer learning process to obtain a fusion model; and based on the fusion model, generating an underground space settlement prediction result of the target city, and by introducing a transfer learning technology, the method can effectively utilize data to enrich settlement knowledge of the city and improve the prediction precision of the data-deficient city.

Description

Technical Field

[0001] The present invention relates to the field of building information technology, and more specifically, to a method and system for predicting urban underground space settlement based on transfer learning. Background Art

[0002] With the continuous advancement of urbanization, underground space development has become an important means to alleviate urban land shortage and improve land use efficiency. However, underground space development will inevitably cause surface subsidence, posing a potential threat to the safety of urban buildings and infrastructure. Therefore, accurate prediction of urban underground space subsidence is crucial to ensure urban safety and sustainable development.

[0003] Traditional settlement prediction methods mainly include empirical formula method, numerical simulation method and artificial intelligence method. The empirical formula method is simple and intuitive, but it is difficult to consider complex geological conditions and construction factors, and the prediction accuracy is limited. Although the numerical simulation method can simulate the settlement mechanism well, the calculation is complex, time-consuming, and requires high accuracy of parameters. In recent years, with the development of machine learning technology, settlement prediction methods based on artificial intelligence have been widely used. This type of method can automatically learn complex nonlinear relationships and has strong prediction capabilities.

[0004] However, there are still some problems with existing artificial intelligence prediction methods. First, these methods usually require a large amount of historical data to train the model, and in newly developed cities or regions, there is often a lack of sufficient settlement monitoring data, which makes model training difficult and the prediction accuracy is not high. Secondly, the geological conditions and engineering characteristics of different cities vary greatly. When the model trained in one city is directly applied to another city, there will often be a large prediction deviation. In addition, existing methods focus more on single-point settlement prediction and lack comprehensive analysis and risk assessment of the settlement trend of the entire region. Summary of the invention

[0005] The present invention aims to solve the above technical problems and proposes a method for predicting urban underground space settlement based on transfer learning. This method innovatively integrates multi-source data and transfer learning technology, effectively solves problems such as insufficient data and poor model generalization ability, and realizes regional settlement trend analysis and risk warning.

[0006] The present invention provides a method for predicting urban underground space settlement based on transfer learning, comprising:

[0007] The acquisition steps include:

[0008] Obtain geological characteristic data, construction process data and historical settlement monitoring data of the target city;

[0009] Obtain geological characteristic data, construction technology data and historical settlement monitoring data of source cities with rich settlement data;

[0010] Processing steps include:

[0011] Based on the source domain city data, construct and train a source domain neural network model;

[0012] Based on the target city data and the source domain neural network model, construct a target domain neural network model;

[0013] According to the target domain neural network model and the source domain neural network model, a transfer learning process is performed to obtain a fusion model;

[0014] Output steps include:

[0015] Based on the fusion model, the underground space settlement prediction result of the target city is generated.

[0016] Preferably, the obtaining step specifically includes:

[0017] Obtain the geological characteristic data through geological survey, including soil stratification, thickness and settlement trend;

[0018] Obtaining the construction process data through engineering records, including support structure type, excavation method and support type;

[0019] The historical settlement monitoring data are obtained through on-site monitoring equipment, including the relative settlement between the measuring point and the horizontal reference, the cumulative settlement of the monitoring point and the settlement rate.

[0020] Preferably, constructing the source domain neural network model in the processing step specifically includes:

[0021] Preprocessing the source domain city data, including data cleaning, standardization and outlier processing;

[0022] Construct a deep neural network structure including multiple hidden layers, each layer is equipped with a nonlinear activation function; use a back propagation algorithm to train the source domain neural network model until the loss function converges. Preferably, constructing the target domain neural network model in the processing step specifically includes: copying the structure of the source domain neural network model as the initial target domain neural network model; fine-tuning the initial target domain neural network model using the target city data; and updating the parameters of the target domain neural network model based on the gradient of the loss function.

[0023] Preferably, performing the transfer learning process in the processing step specifically includes:

[0024] Calculate the difference between the source domain and the target domain data distribution;

[0025] Based on the differences, dynamically adjust the degree of source domain knowledge transfer;

[0026] The loss functions of the source domain and the target domain are combined to optimize the parameters of the fusion model.

[0027] Preferably, the dynamically adjusting the degree of source domain knowledge migration specifically includes:

[0028] Introduce the weight coefficient λ, where 0≤λ≤1;

[0029] When λ=0, the target domain data is fully used; when λ=1, the source domain data is fully used; the optimal λ value is determined through iterative optimization.

[0030] Preferably, the model evaluation step is also included:

[0031] Using validation set data to evaluate the prediction accuracy of the fusion model;

[0032] Calculate evaluation metrics such as mean absolute percentage error;

[0033] Based on the evaluation results, it is determined whether the fusion model needs to be further optimized.

[0034] Preferably, the method further includes the step of inversion parameters:

[0035] Based on the historical settlement monitoring data, key parameters affecting settlement are obtained by inversion;

[0036] Input the key parameters into the fusion model as additional features;

[0037] The fusion model is updated to improve its ability to capture sedimentation mechanisms.

[0038] Preferably, the output step further comprises:

[0039] Generate regional subsidence trend map of target city;

[0040] Identify areas of higher risk of subsidence;

[0041] Output settlement warning information.

[0042] As a preferred embodiment, the model updating step is also included:

[0043] Regularly obtain the latest settlement monitoring data of target cities;

[0044] Based on the latest data, updating the parameters of the fusion model;

[0045] The output step is re-executed to generate updated settlement prediction results.

[0046] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0047] First, by introducing transfer learning technology, the present invention can effectively utilize the settlement knowledge of data-rich cities and improve the prediction accuracy of data-deficient cities. This knowledge transfer mechanism not only solves the problem of insufficient data, but also greatly improves the generalization ability of the model, enabling it to adapt to the geological conditions and engineering characteristics of different cities.

[0048] Secondly, the present invention adopts a multimodal data fusion strategy, which comprehensively considers multiple factors such as geological characteristics, construction technology and historical settlement data. This comprehensive data utilization method enables the model to more accurately capture the various factors that affect settlement, thereby improving the accuracy and reliability of the prediction.

[0049] Furthermore, the present invention achieves accurate adaptation of the model to the characteristics of the target city by dynamically adjusting the degree of source domain knowledge transfer. This adaptive mechanism ensures that the model can fully learn the specific characteristics of the target city while retaining useful prior knowledge, thereby achieving optimal knowledge transfer between different cities.

[0050] In addition, the present invention not only provides single-point settlement prediction, but also generates regional settlement trend graphs and risk warning information. This comprehensive output provides comprehensive decision support for urban planning, engineering design and risk management, greatly improving the practicality and value of the prediction results.

[0051] Finally, the present invention introduces a continuous model update mechanism to ensure that the prediction model can maintain its accuracy and applicability over time. This dynamic update method can better adapt to changes in geological conditions, engineering activities and other factors, and provide long-term and reliable settlement prediction support.

[0052] In summary, the urban underground space settlement prediction method based on transfer learning proposed in this invention effectively solves many problems existing in the prior art through the organic combination of multiple innovative technologies, and significantly improves the accuracy, reliability and practicality of settlement prediction. The application of this method will provide strong technical support for the safe development and management of urban underground space, and is of great significance to promoting the sustainable development of cities. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The figure is a flow chart of the method of the present invention.

[0054] Figure 2 It is a logic block diagram of the data acquisition step of the present invention.

[0055] Figure 3 The figure is a logic block diagram for processing data and building models according to the present invention.

[0056] Figure 4 A logical block diagram for performing transfer learning of the present invention.

[0057] Figure 5 A logical block diagram for model evaluation and updating of the present invention. DETAILED DESCRIPTION

[0058] Please refer to Figure 1-5 The present invention provides a method for predicting urban underground space settlement based on transfer learning. The method innovatively integrates multi-source data and transfer learning technology to effectively solve the problems of insufficient data and low prediction accuracy in urban underground space settlement prediction. The specific implementation methods of the present invention will be described in detail below.

[0059] The method of the present invention comprises an acquisition step, a processing step and an output step.

[0060] In the acquisition step, the method first obtains the geological characteristic data, construction process data and historical settlement monitoring data of the target city. These data are the basic inputs of the prediction model. At the same time, the method also obtains the corresponding data of the source city with rich settlement data. The source city is usually a city that has carried out a large amount of underground space development and has a complete monitoring system. Its data can provide valuable prior knowledge for the prediction of the target city.

[0061] In the processing step, the method first constructs and trains a source domain neural network model based on the source domain city data. This model captures the general law of urban underground space settlement in the source domain. Then, based on the target city data and the source domain neural network model, a target domain neural network model is constructed. This step achieves the initial transfer of knowledge. Then, the method performs the transfer learning process to obtain a fusion model. This process achieves the deep transfer and adaptation of source domain knowledge to the target domain.

[0062] In the output step, this method generates the underground space settlement prediction results of the target city based on the fusion model. This result comprehensively considers the source domain knowledge and target domain characteristics, and can provide reliable decision support for urban underground space development.

[0063] Furthermore, the acquisition step of the present invention includes a more specific data acquisition method.

[0064] For geological characteristic data, this method obtains them through geological surveys. These data include soil stratification, thickness, and settlement trend. For example, soil stratification may include multiple layers such as top fill, silty clay, and sandy clay, and the thickness of each layer may vary from a few meters to tens of meters. The settlement trend may be linear, nonlinear, or periodic.

[0065] For construction process data, this method obtains it through engineering records. These data include support structure type, excavation method and support type. For example, the support structure may be soil nail wall, pile anchor support or underground continuous wall; the excavation method may be open excavation, covered excavation or dark excavation; the support type may be internal support, anchor cable or no support, etc.

[0066] For historical settlement monitoring data, this method obtains them through on-site monitoring equipment. These data include the relative settlement between the measuring point and the horizontal benchmark, the cumulative settlement of the monitoring point, and the settlement rate. For example, the relative settlement may be in the millimeter to centimeter level, the cumulative settlement may reach tens of centimeters, and the settlement rate may vary from a few millimeters to a few centimeters per day.

[0067] This detailed data acquisition method ensures the comprehensiveness and accuracy of the input data, laying a solid foundation for subsequent model training and prediction.

[0068] The present invention adopts a series of innovative processing methods when constructing the source domain neural network model.

[0069] First, this method preprocesses the source domain urban data. This includes data cleaning, standardization, and outlier processing. Data cleaning can remove erroneous or irrelevant data points, such as abnormal readings caused by equipment failure. Standardization can map data of different scales to the same scale, usually using the z-score standardization method:

[0070]

[0071] Among them, z is the standardized value, x is the original value, μ is the mean, and σ is the standard deviation.

[0072] Outliers can be processed using the 3σ rule, that is, data that exceeds 3 standard deviations from the mean value are considered outliers and deleted or corrected.

[0073] Next, this method constructs a deep neural network structure containing multiple hidden layers, each layer is equipped with a nonlinear activation function. A typical network structure may contain 4-6 hidden layers, and the number of neurons in each layer may range from dozens to hundreds. Commonly used nonlinear activation functions include ReLU, Sigmoid, and Tanh. For example, the expression of the ReLU function is:

[0074] f(x)=max(0,x)

[0075] This depth structure can effectively capture the nonlinear relationship in the complex process of underground space settlement.

[0076] Finally, this method uses the back propagation algorithm to train the source domain neural network model until the loss function converges. The loss function is usually the mean square error (MSE):

[0077]

[0078] Where n is the number of samples, y i is the true value, is the predicted value.

[0079] Preferably, when the rate of change of the loss function is less than 1e -6 When , the model can be considered to have converged. This threshold is selected based on experience and can be fine-tuned according to specific circumstances in practical applications.

[0080] Through the above steps, this method can build a source domain neural network model with strong prediction ability, laying the foundation for the subsequent transfer learning process. Compared with the traditional single city prediction model, this method can better utilize existing knowledge and improve the accuracy and generalization ability of prediction.

[0081] The present invention adopts an efficient migration strategy when constructing the target domain neural network model.

[0082] First, this method copies the structure of the source domain neural network model as the initial target domain neural network model. This approach is based on the assumption that although the sedimentation laws of the source domain and the target domain are different, there are certain similarities. By copying the source domain model structure, an initial model with certain predictive capabilities can be quickly obtained, which greatly speeds up the convergence of the model.

[0083] Next, the method fine-tunes the initial target domain neural network model using the target city data. This step is the core of transfer learning, which allows the model to adapt to the specific conditions of the target city. During fine-tuning, usually only the parameters of the last few layers of the network are updated, while the parameters of the first few layers remain unchanged. This is because the first few layers of the neural network usually capture more general features, while the later layers are more related to specific tasks.

[0084] Preferably, in one embodiment of the present invention, only the parameters of the last two layers of the network can be updated. This approach can fully adapt to the characteristics of the target domain while retaining the source domain knowledge. Of course, the specific number of layers of parameters to be updated can be adjusted according to the similarity between the source domain and the target domain.

[0085] Finally, this method updates the parameters of the target domain neural network model based on the gradient of the loss function. The gradient descent method is used here, and its update rule can be expressed as:

[0086]

[0087] Among them, θ t represents the parameter at the tth iteration, η represents the learning rate, L(θt ) represents the loss function, Represents the gradient operator.

[0088] Preferably, the learning rate η can be set to a smaller value, such as 0.001, to ensure that the model can converge stably. In practical applications, an adaptive learning rate method, such as the Adam optimizer, can also be used to further improve the training effect of the model.

[0089] Through the above steps, this method can effectively transfer the source domain knowledge to the target domain while maintaining sensitivity to the characteristics of the target domain, thereby constructing a neural network model that can both utilize prior knowledge and adapt to the characteristics of the target city.

[0090] The present invention adopts a dynamic adjustment strategy when executing the transfer learning process to achieve optimal transfer of source domain knowledge.

[0091] First, this method calculates the difference in data distribution between the source domain and the target domain. This step is crucial because it determines to what extent the source domain knowledge can be directly applied to the target domain. There are many methods for calculating distribution differences. The present invention preferably uses the Maximum Mean Discrepancy (MMD) method. The calculation formula of MMD is as follows:

[0092]

[0093] Among them, X and Y represent the data of the source domain and the target domain respectively, φ is a feature mapping function, is the reproducing kernel Hilbert space.

[0094] Next, this method dynamically adjusts the degree of source domain knowledge migration based on the calculated distribution difference. Specifically, when the distribution difference is small, more reliance can be placed on source domain knowledge; when the distribution difference is large, more reliance needs to be placed on target domain data. This dynamic adjustment strategy can be achieved through a weight coefficient λ, which ranges from [0,1].

[0095] Finally, this method combines the loss functions of the source domain and the target domain to optimize the parameters of the fusion model. The fused loss function can be expressed as:

[0096] L total =λL source +(1-λ)L target +γMMD 2 (X,Y)

[0097] Among them, L source and L target are the loss functions of the source domain and the target domain respectively, and γ is a balance parameter used to control the influence of the distribution difference term.

[0098] Preferably, γ can be set to a value between 0.1 and 1, and the specific value can be determined by cross-validation. The design of this fusion loss function allows the model to fully utilize the source domain knowledge while learning the target domain features, thereby achieving effective knowledge transfer.

[0099] Through the above steps, this method can adaptively adjust the degree of knowledge transfer according to the specific conditions of the source domain and the target domain, thereby constructing a fusion model that can fully utilize prior knowledge and adapt well to the characteristics of the target city. Compared with simple transfer or complete retraining, this method can better balance knowledge transfer and target adaptation, thereby improving the accuracy and reliability of settlement prediction.

[0100] The present invention introduces a key weight coefficient λ when dynamically adjusting the degree of source domain knowledge transfer. The reasonable setting of this coefficient is crucial to the effect of transfer learning.

[0101] First, this method defines the range of lambda as [0,1]. When lambda = 0, the model uses the target domain data completely, which is equivalent to training a new model from scratch on the target domain; when lambda = 1, the model uses the source domain data completely, which is equivalent to directly applying the source domain model to the target domain. In practical applications, the optimal lambda value is usually between these two extremes.

[0102] Next, the method determines the optimal λ value through iterative optimization. Preferably, a grid search method can be used, that is, a series of λ values ​​are uniformly selected in the interval [0,1], such as {0, 0.1, 0.2, ..., 0.9, 1}, and then a model is trained for each λ value, and finally the λ value with the best performance on the validation set is selected.

[0103] Another more efficient method is to use Bayesian Optimization. This method can intelligently select the next λ value to try by constructing a probabilistic model between λ and model performance, thereby finding the optimal solution faster. The objective function of Bayesian optimization can be defined as:

[0104] f(λ)=-MSE val (λ)

[0105] Among them, MSE val (λ) represents the mean square error of the model trained using the λ value on the validation set.

[0106] Preferably, in one embodiment of the present invention, the number of iterations of the Bayesian optimization can be set to 50, which is usually sufficient to find a better lambda value. Of course, if computing resources permit, the number of iterations can be increased to obtain more accurate results.

[0107] Through the above steps, this method can find an optimal λ value and achieve precise control of the degree of source domain knowledge migration. Compared with the fixed knowledge migration strategy, this method can better adapt to the differences in geological conditions and engineering characteristics between different cities, thereby improving the generalization ability and prediction accuracy of the settlement prediction model.

[0108] The present invention also includes an important model evaluation step for verifying the predictive performance of the fusion model.

[0109] First, this method uses the validation set data to evaluate the prediction accuracy of the fusion model. The validation set is a data set independent of the training set, which is used to simulate the performance of the model in actual applications. Preferably, the total data set can be divided into a training set, a validation set, and a test set in a ratio of 7:2:1. This division method can provide enough data for model verification and final testing while ensuring sufficient training data.

[0110] Next, this method calculates evaluation indicators such as mean absolute percentage error (MAPE). The calculation formula of MAPE is:

[0111]

[0112] Among them, A i is the actual value, is the predicted value and n is the number of samples.

[0113] Preferably, in one embodiment of the present invention, a MAPE of less than 10% can be used as a standard for good model performance. This standard is set based on the actual needs of underground space engineering, because in most cases, an error within 10% can meet the needs of engineering design and risk assessment.

[0114] In addition to MAPE, this method can also calculate indicators such as root mean square error (RMSE) and coefficient of determination (R^2) to comprehensively evaluate the prediction performance of the model.

[0115] Finally, based on the evaluation results, this method determines whether the fusion model needs to be further optimized. If the model performance does not meet the preset standards, the following optimization strategies can be adopted:

[0116] 1. Adjust the network structure, such as increasing or decreasing the number of hidden layers, changing the number of neurons in each layer, etc.

[0117] 2. Try different activation functions, such as switching from ReLU to Leaky ReLU or ELU.

[0118] 3. Adjust the learning rate or adopt a learning rate decay strategy.

[0119] 4. Add regularization terms, such as L1 / L2 regularization or Dropout, to prevent overfitting.

[0120] 5. Increase training data or use data enhancement techniques.

[0121] Through this comprehensive model evaluation and optimization process, this method can ensure that the final fusion model has good prediction performance and generalization ability, providing reliable technical support for urban underground space settlement prediction.

[0122] The present invention also includes an innovative inversion parameter step, which aims to further improve the model's ability to capture the sedimentation mechanism.

[0123] First, this method is based on historical settlement monitoring data to invert the key parameters that affect settlement. These key parameters usually include soil compression index, secondary compression index, preload, etc. The inversion process can use optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms. Taking genetic algorithms as an example, its objective function can be defined as:

[0124]

[0125] Where θ represents the parameter set to be inverted, is the measured settlement value at the i-th time point, is the theoretical sedimentation value calculated using the parameter θ, and n is the total number of time points.

[0126] Preferably, in one embodiment of the present invention, the population size of the genetic algorithm can be set to 100, and the number of iterations can be set to 500. These parameter settings can usually achieve a good balance between computational efficiency and optimization effect.

[0127] Next, this method uses the key parameters obtained from the inversion as additional features and inputs them into the fusion model. This step realizes the combination of physical model and data-driven model, which can significantly improve the interpretability and prediction accuracy of the model. Specifically, these parameters can be concatenated with the original feature vector to form a new input vector:

[0128] X new =[X original ,θ inverted ],

[0129] Among them, X original is the original eigenvector, θ inverted is the parameter vector obtained by inversion.

[0130] Finally, this method updates the fusion model to improve its ability to capture the sedimentation mechanism. This step can be achieved by training the model again, but it should be noted that in order to maintain the sensitivity of the model to the original features, a smaller learning rate, such as 0.0001, can be used.

[0131] Through the above steps, this method realizes the organic combination of physical model and data-driven model, which not only utilizes the flexibility and adaptability of data-driven method, but also incorporates the prior knowledge of physical model, thus constructing a more reliable and interpretable settlement prediction model.

[0132] The output step of the present invention includes not only generating single-point settlement prediction results, but also generating more comprehensive settlement information.

[0133] First, this method generates a regional subsidence trend map of the target city. This trend map can intuitively show the subsidence distribution of the entire city or a specific area. Preferably, Kriging interpolation (Kriging) can be used to generate a continuous subsidence distribution map. The basic formula of Kriging interpolation is as follows:

[0134]

[0135] Among them, Z * (x0) is the estimated value Z(x i ) is the observed value of the known point, λ i is the weight coefficient, satisfying

[0136] Preferably, in one embodiment of the present invention, ordinary kriging can be selected and a spherical semivariogram model can be used. This setting performs well in most geostatistical applications.

[0137] Next, the method identifies areas with a higher risk of subsidence. This can be achieved by setting a subsidence threshold. For example, areas with an annual subsidence rate of more than 20 mm / year can be marked as high-risk areas. This threshold is set based on engineering experience, but can be adjusted according to specific circumstances in actual applications.

[0138] The algorithm for identifying high-risk areas can be expressed as:

[0139]

[0140] Where R(x,y) represents the risk status of the location (x,y), 1 represents high risk, and 0 represents low risk. S(x,y) represents the annual sedimentation rate of the location (x,y), S threshold is the set threshold.

[0141] Finally, the method outputs settlement warning information. The warning information may include the location, area, maximum settlement rate, etc. of the high-risk area. Preferably, the warning information may be divided into three levels: yellow warning (annual settlement rate 20-30 mm), orange warning (30-50 mm) and red warning (>50 mm).

[0142] Through the above steps, this method not only provides detailed settlement prediction results, but also generates intuitive settlement trend diagrams and targeted risk warning information. These output results can provide comprehensive decision support for urban planning, engineering design and risk management.

[0143] The present invention also includes an important model updating step to ensure that the predictive model maintains its accuracy and applicability over time.

[0144] First, this method obtains the latest settlement monitoring data of the target city on a regular basis. The specific meaning of "regularly" may vary from project to project, but it can usually be set to once a month or once a quarter.

[0145] Preferably, in one embodiment of the present invention, an automated data acquisition system, such as a distributed optical fiber sensor network, may be used to achieve real-time or quasi-real-time data updating.

[0146] Next, this method updates the parameters of the fusion model based on the latest data. This step adopts the idea of ​​online learning, allowing the model to continuously adjust its parameters based on new data. Specifically, the mini-batch gradient descent method can be used to update the parameters. The update formula is as follows:

[0147]

[0148] Among them, θ t is the current model parameter, η is the learning rate, L is the loss function, X t:t+n and Y t:t+n are the features and labels of the newly acquired n samples respectively.

[0149] Preferably, the learning rate η can be set to a smaller value, such as 0.001, to ensure that the model can be updated stably without drastic fluctuations. At the same time, a learning rate decay strategy can be adopted to gradually reduce the learning rate over time, such as:

[0150] The learning rate decay strategy can be expressed as:

[0151]

[0152] Among them, η0 is the initial learning rate, decay_rate is the decay rate, and t is the number of updates.

[0153] The performance degradation calculation formula in the performance monitoring mechanism is:

[0154]

[0155] If Performance_Drop exceeds a certain threshold, such as 10%, the model retraining process is triggered.

[0156] Through this continuous updating and monitoring mechanism, this method can ensure that the settlement prediction model always maintains its accuracy and applicability, providing reliable technical support for the long-term safety management of urban underground space. Compared with static models, this dynamic update method can better adapt to changes in geological conditions, engineering activities and other factors, thereby providing more accurate and timely settlement prediction results.

[0157] It should be noted that the above description 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. The urban underground space settlement prediction method based on transfer learning is characterized by: include: The acquisition steps include: Obtain geological characteristic data, construction process data, and historical settlement monitoring data for the target city; Obtain geological characteristic data, construction technology data, and historical settlement monitoring data of source cities with rich settlement data; Processing steps include: Based on the source domain city data, construct and train a source domain neural network model; Constructing a target domain neural network model based on the target city data and the source domain neural network model; Performing a transfer learning process based on the target domain neural network model and the source domain neural network model to obtain a fusion model; Output steps include: Based on the fusion model, underground space settlement prediction results of the target city are generated.

2. The method according to claim 1, characterized in that The obtaining step specifically includes: Obtaining the geological characteristic data through geological survey, including soil stratification, thickness and settlement trend; Obtaining the construction process data from engineering records, including support structure type, excavation method, and support type; The historical settlement monitoring data are obtained through on-site monitoring equipment, including the relative settlement between the measuring point and the horizontal benchmark, the cumulative settlement of the monitoring point and the settlement rate.

3. The method according to claim 1, characterized in that The processing steps of constructing the source domain neural network model specifically include: Preprocessing the source domain city data, including data cleaning, standardization and outlier processing; Construct a deep neural network structure with multiple hidden layers, each equipped with a nonlinear activation function; The source domain neural network model is trained using a back-propagation algorithm until the loss function converges.

4. The method according to claim 1, wherein The processing steps of constructing the target domain neural network model specifically include: Copying the structure of the source domain neural network model as an initial target domain neural network model; Fine-tuning the initial target domain neural network model using the target city data; Based on the gradient of the loss function, the parameters of the target domain neural network model are updated.

5. The method according to claim 1, wherein The transfer learning process in the processing steps specifically includes: Calculate the difference in data distribution between the source domain and the target domain; Based on the differences, dynamically adjust the degree of source domain knowledge transfer; The loss functions of the source domain and the target domain are combined to optimize the parameters of the fusion model.

6. The method according to claim 5, characterized in that The dynamic adjustment of the degree of source domain knowledge transfer specifically includes: Introduce the weight coefficient λ, where 0≤λ≤1; When λ=0, the target domain data is fully used; when λ=1, the source domain data is fully used; The optimal λ value is determined through iterative optimization.

7. The method according to claim 1, characterized in that A model evaluation step is also included: Use validation set data to evaluate the prediction accuracy of the fusion model; Calculate evaluation metrics such as mean absolute percentage error; Based on the evaluation results, it is determined whether the fusion model needs to be further optimized.

8. The method according to claim 1, characterized in that It also includes the inversion parameter step: Based on the historical settlement monitoring data, key parameters affecting settlement are obtained by inversion; Inputting the key parameters into the fusion model as additional features; The fusion model is updated to improve its ability to capture sedimentation mechanisms.

9. The method according to claim 1, characterized in that The output step further includes: Generate regional subsidence trend map of target city; Identify areas of higher subsidence risk; Output settlement warning information.

10. The method according to claim 1, characterized in that It also includes a model update step: Regularly obtain the latest settlement monitoring data of target cities; Based on the latest data, updating the parameters of the fusion model; The output step is re-executed to generate updated settlement prediction results.

Images