Shield settlement deformation monitoring method and system based on gray neural network
Through a gray neural network-based method, multi-source data is fused and model parameters are dynamically adjusted, the problem of settlement deformation prediction in shield construction is solved, and real-time monitoring and prediction of high accuracy and adaptability is achieved.
Patent Information
- Application Number
- CN202411131873.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-11-22
AI Technical Summary
During the shield construction process, geological conditions, construction parameters and surrounding environmental factors affect the settlement deformation, making prediction difficult. Traditional methods have limited ability to process uncertain data, making it difficult to meet the needs of real-time monitoring and prediction.
A gray neural network-based method is adopted to integrate multi-source data, optimize the original time series, build a gray prediction model, and dynamically adjust the model parameters based on feature selection and real-time data feedback to achieve real-time prediction and adjustment.
It improves the prediction accuracy and adaptability of settlement deformation in shield construction, meets the needs of real-time monitoring and prediction, and ensures the timeliness and reliability of the prediction results.
Smart Images

Figure CN119005435A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent prediction technology, and in particular to a shield settlement deformation monitoring method and system based on grey neural network. Background Art
[0002] During the shield construction process, factors such as geological conditions, construction parameters and surrounding environment will affect settlement and deformation, making prediction difficult. Data sources are diverse and of varying quality. Traditional prediction methods have limited processing capabilities for uncertain data. Actual projects require real-time monitoring and prediction, and traditional batch processing methods are difficult to meet the needs of real-time prediction. Summary of the invention
[0003] The present invention provides a shield settlement deformation monitoring method and system based on grey neural network, which integrates multi-source data, optimizes the original time series, constructs a grey prediction model using the cumulative generation sequence, constructs a grey neural network model based on feature selection and grey prediction results, and dynamically adjusts model parameters based on real-time data feedback and environment-impact tables, so that the model can adapt to changes in the construction environment in real time, realize real-time prediction and adjustment, improve the accuracy and adaptability of the prediction, meet the needs of real-time monitoring and prediction during shield construction, and ensure the timeliness and reliability of the prediction results.
[0004] The present invention provides a shield settlement deformation monitoring method based on grey neural network, comprising: Step 1: Obtain relevant data that affect settlement and deformation during shield construction around the construction area, and merge relevant data from different sources to form a relevant data set; Step 2: According to the relevant data sets, the cumulative generation sequence and the original time series are formed, the original time series is optimized, and the relationship model between the cumulative generation sequence and the optimized time series is established, and then the grey prediction model parameters are determined and the grey prediction model is constructed; Step 3: Determine the feature dimensions in the relevant data set through a feature selection process, build a grey neural network model based on the feature dimensions combined with the prediction results of the grey prediction model, and predict the settlement deformation during the shield tunneling process based on the grey neural network model; Step 4: Adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the predicted results of settlement deformation.
[0005] The present invention provides a shield settlement deformation monitoring method based on grey neural network, which is characterized in that relevant data affecting settlement deformation during shield construction are obtained around the construction area, and relevant data from different sources are merged to form a relevant data set, including: Divide the construction area vertically according to its geological layers, and determine the data collection strategy for each vertical area based on key project factors; According to the data acquisition strategy, relevant data affecting the settlement deformation of the corresponding longitudinal area during the shield construction process are obtained, and the relevant data are fused to form a relevant data set based on the fusion result.
[0006] The present invention provides a shield settlement deformation monitoring method based on a grey neural network, which obtains relevant data affecting the settlement deformation of a corresponding longitudinal area during shield construction according to the data acquisition strategy, and fuses the relevant data to form a relevant data set based on the fusion result, including: Check the existing data sets in the project to determine the data characteristics of the initial data corresponding to each longitudinal region, and select the data fusion method based on the data characteristics of the corresponding longitudinal region and the project requirements; Based on the data characteristics of different longitudinal regions, the initial data of the corresponding longitudinal region is featured, sparse points are judged for the initial data of each longitudinal region, and the sparse points are interpolated according to the judgment results, so as to obtain the relevant data of the corresponding longitudinal region; Perform spatial superposition of the longitudinal regions, check the consistency between different longitudinal regions, and perform spatial alignment of the relevant data of different longitudinal regions based on the consistency check results; Perform event alignment on relevant data in different longitudinal regions according to event stamps; The data fusion method is used to fuse the relevant data of different longitudinal regions according to the spatial alignment result and the event alignment result, thereby forming a relevant data set.
[0007] The present invention provides a shield settlement deformation monitoring method based on grey neural network, which forms a cumulative generation sequence and an original time series according to relevant data sets, optimizes the original time series, and establishes a relationship model between the cumulative generation sequence and the optimized time series, including: Form the original time series from the relevant dataset: ;in, Indicated in k The relevant data value at a time point; Decompose the original time series into several scale components, perform denoising and reconstruction, and obtain the reconstructed time signal: ;in, represents the reconstructed time signal; Indicated in the scale factor Time Shift Factor The wavelet function at ; represents the scale factor; jRepresents the discretization index of the scale factor; J Indicates the maximum scale; k Represents the discrete time axis k time point; t Represents the index of the time series; represents the time shift factor; Indicates time t Related data signals; represents the mother wavelet function; Represents discrete scale factor j and discrete translation factor k Wavelet function under the condition of ; According to the reconstructed time signal, an optimized time series after denoising is obtained, and a cumulative generation sequence is constructed according to the optimized time series, and then a relationship model is determined based on the cumulative generation sequence and the optimized time series.
[0008] The present invention provides a shield settlement deformation monitoring method and system based on grey neural network, determines grey prediction model parameters, and constructs a grey prediction model, including: Grey prediction model basic formula: ;in, Indicates that the cumulative generated sequence is k The value at a point in time; Indicates that the cumulative generated sequence is k-1 The value at a point in time, ; Represents the relevant data value at the kth time point; a Represents the first parameter of the grey prediction model; b Represents the second parameter of the grey prediction model; Indicates c1 The autoregressive coefficient of the first lag term; Indicates c2 The moving average coefficient of the second lag term; represents independent and identically distributed random terms; c1 It represents the number corresponding to the first lag of the autoregressive term; c2 Represents the corresponding number of the second lag term of the moving average term; p represents the total number of first lag items; q represents the total number of second lag terms; Represents the first c1 The first lag value of the first lag term; Indicates c2The lag noise value of the second lag term; A represents the system dynamics corresponding to the grey prediction model and the internal structure of the optimized time series; B represents the system noise and external influence of the grey prediction model; A grey prediction model is constructed according to the basic formula and the relationship model.
[0009] The present invention provides a shield settlement deformation monitoring method based on grey neural network, which determines the feature dimensions in the relevant data set through a feature selection process, constructs a grey neural network model according to the feature dimensions combined with the prediction results of the grey prediction model, and predicts the settlement deformation amount in the shield process based on the grey neural network model, including: Select feature selection methods based on the data type of the relevant dataset and determine the feature dimensions and important features in the relevant dataset; Combine the important features with the prediction results of the grey prediction model to form the input features of the grey neural network model; Conduct term analysis on shield settlement deformation to determine the task type and model requirements needed to predict settlement deformation values; Determine the input layer according to the input features, set the hidden layer based on the task type, and then build the hierarchical structure of the grey neural network model; select the corresponding loss function, optimizer and evaluation index according to the task type, data characteristics and model requirements, and then compile the grey neural network model; and determine the grey neural network model by combining the hierarchical structure and the compilation results; The settlement deformation during shield tunneling is predicted based on the grey neural network model.
[0010] The present invention provides a shield settlement deformation monitoring method based on grey neural network, which adjusts the grey neural network model based on the environment-impact table and real-time data feedback to optimize the prediction result of the settlement deformation, including: Determine the environment-impact table based on relevant data sets; Use real-time data to update the environment-impact table to obtain the impact of current environmental factors; Select new important features according to the updated environment-influence table and update the input features of the grey neural network; According to real-time data, the incremental learning technology is used to adjust the parameters of the grey neural network model corresponding to the training model, and then the parameters of the grey neural network model are adjusted to achieve feedback adjustment of the grey neural network model.
[0011] The present invention provides a shield settlement deformation monitoring system based on grey neural network, comprising: Fusion module: It obtains relevant data that affects settlement and deformation during shield construction around the construction area, and fuses relevant data from different sources to form a relevant data set; Prediction model module: It forms cumulative generation sequence and original time series according to relevant data sets, optimizes the original time series, and establishes the relationship model between the cumulative generation sequence and the optimized time series, thereby determining the parameters of the grey prediction model and constructing the grey prediction model; Neural network model module: determine the feature dimensions in the relevant data set through the feature selection process, build a gray neural network model based on the feature dimensions combined with the prediction results of the gray prediction model, and predict the settlement deformation during the shield process based on the gray neural network model; Feedback adjustment module: adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the prediction results of settlement deformation.
[0012] Compared with the prior art, the beneficial effects of the present application are as follows: by fusing multi-source data, optimizing the original time series, using the cumulative generation sequence to build a gray prediction model, feature selection combined with the gray prediction results to build a gray neural network model, based on real-time data feedback and environment-impact table, dynamically adjust the model parameters, so that the model can adapt to changes in the construction environment in real time, realize real-time prediction and adjustment, improve the accuracy and adaptability of the prediction, meet the needs of real-time monitoring and prediction during shield construction, and ensure the timeliness and reliability of the prediction results.
[0013] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.
[0014] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 It is a flow chart of a shield settlement deformation monitoring method based on grey neural network provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a shield settlement deformation monitoring system based on a grey neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0016] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.
[0017] Embodiment 1: The embodiment of the present invention provides a shield settlement deformation monitoring method based on grey neural network, such as Figure 1 As shown, including: Step 1: Obtain relevant data that affect settlement and deformation during shield construction around the construction area, and merge relevant data from different sources to form a relevant data set; Step 2: According to the relevant data sets, the cumulative generation sequence and the original time series are formed, the original time series is optimized, and the relationship model between the cumulative generation sequence and the optimized time series is established, and then the grey prediction model parameters are determined and the grey prediction model is constructed; Step 3: Determine the feature dimensions in the relevant data set through a feature selection process, build a grey neural network model based on the feature dimensions combined with the prediction results of the grey prediction model, and predict the settlement deformation during the shield tunneling process based on the grey neural network model; Step 4: Adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the predicted results of settlement deformation.
[0018] In this embodiment, the relevant data includes geological condition data, construction parameter data and surrounding environment data.
[0019] In this embodiment, the process of data fusion is to integrate multi-source data to form a unified data set, which specifically includes: preprocessing different data sources, including denoising, standardization, etc.; matching data in different longitudinal regions based on event alignment and spatial alignment results; and performing data fusion according to the selected fusion method. For example, Kalman filtering is used for dynamic data fusion, or weighted average method is used to integrate multi-source data.
[0020] In this embodiment, the process of optimizing the original time series is to obtain an optimized time series by decomposing the original time series into multiple scale components and performing denoising and reconstruction.
[0021] In this embodiment, the original time series is time series data directly obtained from the data set, which are usually observations arranged in chronological order. For the settlement deformation in shield construction, it may include stratum settlement, advancement speed, climate conditions, etc.
[0022] In this embodiment, the accumulated sequence is the result of the accumulation operation on the original time series, which is used to reduce the volatility and randomness of the data and make the sequence smoother, thereby being more suitable for establishing a prediction model.
[0023] In this embodiment, the relational model is used to describe the relationship between the cumulatively generated sequence and the original time series (or its optimized version), and the parameters in the grey prediction model can be determined through this model.
[0024] In this embodiment, the grey prediction model is a data prediction model specifically used to process incomplete and uncertain information. It performs accumulation generation operations on a small amount of sample data to construct a differential model, thereby performing prediction.
[0025] In this embodiment, the grey neural network model combines the advantages of the grey prediction model and the neural network model, and improves the prediction accuracy and the ability to handle nonlinear relationships by taking the output of the grey prediction model as one of the input features of the neural network model.
[0026] In this embodiment, the environmental-impact table is a table used to record the degree of influence of various environmental factors on specific targets (such as shield construction settlement deformation). This table helps identify and analyze the main influencing factors by listing different environmental factors and their specific impact on the predicted target. For example, if the environmental factor is geological conditions and the corresponding variable is soil density, the impact degree is high and the note is geological exploration data record.
[0027] In this embodiment, the feature selection method is mainly used to select the most representative and predictive features from the data set. Common methods include: filtering method: screening features through statistical methods, including Pearson correlation coefficient, variance analysis, chi-square test, etc. For example, the Pearson correlation coefficient is used to screen features that are highly correlated with settlement deformation; packaging method: using a machine learning algorithm (such as recursive feature elimination, RFE) to evaluate the feature set, for example, using RFE to gradually eliminate unimportant features and keep the most effective set of features; embedded method: embedding feature selection in a learning algorithm, such as Lasso regression and random forest importance evaluation, for example, obtaining the coefficient of the feature through Lasso regression, and using features with non-zero coefficients for selection.
[0028] In this embodiment, feedback adjustment refers to the process of online adjustment of model parameters based on model prediction results and real-time input data. The specific steps are: obtaining real-time construction monitoring data and environmental monitoring data, updating the environment-impact table based on real-time data and analysis results, determining new impact features and their importance, reselecting and extracting current important features based on the updated environment-impact table, updating the input features of the gray neural network model, ensuring the timeliness and relevance of the model input data, using incremental learning technology, real-time training of the model based on new data batches, adjusting model parameters, using new input features and prediction results, calculating prediction errors, and adjusting model parameters through optimization algorithms.
[0029] In this embodiment, the specific process of optimizing the settlement deformation is to obtain multi-source data during the construction process and perform preprocessing, such as missing value filling, normalization, etc.; extract important features and construct feature vectors; construct cumulative generation sequences and optimized time series based on historical data, determine the parameters of the grey prediction model and make predictions; comprehensively consider important features and grey prediction results, design a neural network structure, and conduct joint training; obtain and update the environment-impact table in real time, and re-evaluate the impact of environmental factors; apply incremental learning technology to adjust the parameters of the grey neural network model in real time to optimize model performance; adjust the model structure and parameters based on model prediction errors and feedback data to ensure that the model responds accurately in real time and provide high-precision settlement deformation predictions.
[0030] The working principle and beneficial effects of the above technical solution are: by fusing multi-source data, optimizing the original time series, using the cumulative generation sequence to build a gray prediction model, feature selection combined with the gray prediction results to build a gray neural network model, based on real-time data feedback and environment-impact table, dynamically adjust the model parameters, so that the model can adapt to changes in the construction environment in real time, realize real-time prediction and adjustment, improve the accuracy and adaptability of the prediction, meet the needs of real-time monitoring and prediction during shield construction, and ensure the timeliness and reliability of the prediction results.
[0031] Embodiment 2: The embodiment of the present invention provides a shield settlement deformation monitoring method based on a grey neural network, which obtains relevant data affecting settlement deformation during shield construction around the construction area, and fuses relevant data from different sources to form a relevant data set, including: Divide the construction area vertically according to its geological layers, and determine the data collection strategy for each vertical area based on key project factors; According to the data acquisition strategy, relevant data affecting the settlement deformation of the corresponding longitudinal area during the shield construction process are obtained, and the relevant data are fused to form a relevant data set based on the fusion result.
[0032] In this embodiment, the specific process of vertical division is: obtaining a comprehensive geological exploration report to determine the geological conditions of the entire construction area, and then obtaining detailed geological profile data; identifying the geological layers in the construction area, such as topsoil, clay layer, sand layer, rock layer, etc., based on the geological report and on-site survey results; further refining the geological layers and performing vertical segmentation according to depth. For example, one layer is divided every 10 meters, or divided according to the specific geological features identified.
[0033] In this embodiment, the data collection strategy is to first determine the collection target of each longitudinal area, including but not limited to the following categories: geological attribute data, mechanical property data, hydrological data and environmental data. The collection methods include drilling sampling, field monitoring, geological radar and groundwater monitoring.
[0034] The working principle and beneficial effects of the above technical solution are: through the vertical division of the geological levels of the construction area, determine the data collection strategy based on the key factors of the project, collect and integrate regional relevant data, form a comprehensive data set to improve data consistency and accuracy, improve the prediction accuracy of the model under different geological conditions, reflect the changes in geological and construction conditions in real time, and meet real-time monitoring needs.
[0035] Embodiment 3: The embodiment of the present invention provides a shield settlement deformation monitoring method based on a grey neural network, which obtains relevant data affecting the settlement deformation of a corresponding longitudinal area during shield construction according to the data acquisition strategy, and fuses the relevant data to form a relevant data set based on the fusion result, including: Check the existing data sets in the project to determine the data characteristics of the initial data corresponding to each longitudinal region, and select the data fusion method based on the data characteristics of the corresponding longitudinal region and the project requirements; Based on the data characteristics of different longitudinal regions, the initial data of the corresponding longitudinal region is featured, sparse points are judged for the initial data of each longitudinal region, and the sparse points are interpolated according to the judgment results, so as to obtain the relevant data of the corresponding longitudinal region; Perform spatial superposition of the longitudinal regions, check the consistency between different longitudinal regions, and perform spatial alignment of the relevant data of different longitudinal regions based on the consistency check results; Perform event alignment on relevant data in different longitudinal regions according to event stamps; The data fusion method is used to fuse the relevant data of different longitudinal regions according to the spatial alignment result and the event alignment result, thereby forming a relevant data set.
[0036] In this embodiment, the process of determining project requirements is to understand the main objectives of the shield construction process, such as ensuring that surface settlement is controlled within the specified range and that the construction progress meets expectations; assessing the geological risks and environmental risks that may be faced during the construction process, such as stratum deformation and groundwater leakage; referring to the construction data and experience of similar projects in the past to understand which data are most critical for predicting settlement and deformation; clarifying the monitoring and data recording requirements that need to be met based on relevant engineering regulations and standards; considering the prediction models, algorithms, etc. that the project needs to use, and understanding the specific data requirements of these technologies.
[0037] In this embodiment, the existing data sets in the project generally cover basic data and real-time monitoring data in multiple aspects, including but not limited to: geological data, construction data, environmental data, and monitoring data.
[0038] In this embodiment, data characteristics refer to the data attributes and characteristics involved in each longitudinal area, including: data distribution: such as the mean, variance, and distribution shape of the data; data integrity: whether there are missing values or outliers; time series: whether the data has time continuity and dependence; spatial correlation: the correlation between data in different geological layers and spatial locations.
[0039] In this embodiment, the data fusion method is selected according to data characteristics and project requirements, including: simple averaging method, weighted averaging method, Kalman filtering and Bayesian fusion.
[0040] In this embodiment, the specific process of feature highlighting is to extract features that better represent the essential characteristics of the data from the initial data, including: converting the data to a unified scale to eliminate data differences in different dimensions; using principal component analysis or other methods to extract main features; identifying and eliminating outliers or noise data to improve data quality; performing sparse point judgment on key features and filling in missing points based on interpolation methods (such as linear interpolation and spline interpolation).
[0041] In this embodiment, the process of spatial superposition is to check and align the data of different longitudinal regions to ensure spatial consistency; map the data of different longitudinal regions to a unified spatial coordinate system; check the data consistency between regions through statistical analysis or visualization tools; and adjust the inconsistent data to align them in space. Common methods include interpolation, extrapolation, Kriging and other spatial alignment techniques.
[0042] In this embodiment, the process of event alignment is to ensure the consistency of different data sources in the time dimension. The process is: converting the timestamps of different data sources into a unified standard time format; synchronously sampling the data according to the timestamp to ensure the time consistency between the data; marking key event points (such as construction nodes, emergencies), and aligning the corresponding data.
[0043] In this embodiment, the formation of relevant data sets is that the fused data is integrated and processed, and finally forms the following data sets, which can be used for different purposes: comprehensive data set: contains all initial data and processed related data; feature data set: extracts a subset of key features for model training and prediction; historical data set: historical records before and after fusion, which are used for subsequent analysis and trend prediction; real-time monitoring data set: used for dynamic monitoring and real-time prediction.
[0044] The working principle and beneficial effects of the above technical solution are: by determining the characteristics of the initial data, selecting a suitable data fusion method, highlighting the features and interpolating sparse points of the initial data, and finally using a fusion method to fuse data from different regions through spatial overlay inspection and event alignment, a comprehensive data set is formed to improve prediction accuracy, effectively deal with uncertainty and diversity of data, improve adaptability to complex construction environments, meet the requirements of real-time monitoring and prediction of projects, and improve the timeliness and reliability of prediction results.
[0045] Embodiment 4: The embodiment of the present invention provides a shield settlement deformation monitoring method based on a grey neural network, which forms a cumulative generation sequence and an original time series according to relevant data sets, optimizes the original time series, and establishes a relationship model between the cumulative generation sequence and the optimized time series, including: Form the original time series from the relevant dataset: ;in, Indicated in k The relevant data value at a time point; Decompose the original time series into several scale components, perform denoising and reconstruction, and obtain the reconstructed time signal: ;in, represents the reconstructed time signal; Indicated in the scale factor Time Shift Factor The wavelet function at ; represents the scale factor; j Represents the discretization index of the scale factor; J Indicates the maximum scale; k Represents the discrete time axis k time point; t Represents the index of the time series; represents the time shift factor; Indicates time t Related data signals; represents the mother wavelet function; Represents discrete scale factor j and discrete translation factor k Wavelet function under the condition of ; According to the reconstructed time signal, an optimized time series after denoising is obtained, and a cumulative generation sequence is constructed according to the optimized time series, and then a relationship model is determined based on the cumulative generation sequence and the optimized time series.
[0046] In this embodiment, the scale component is the manifestation of the original time series at different time scales. The original time series can be decomposed into multiple scale components through wavelet transform, and each scale component represents the characteristics of the signal at different frequencies, including low-frequency components: usually containing the overall trend or slowly changing part of the signal; high-frequency components: containing the rapidly changing details or noise in the signal. In the wavelet decomposition process, the original time series is decomposed into components of different scales through wavelet transform, and each component can be regarded as a sub-signal within a specific frequency range.
[0047] In this embodiment, the cumulatively generated sequence represents the cumulative value of the optimized time series, reflecting the changing trend of the time series under the cumulative effect.
[0048] The working principle and beneficial effects of the above technical solution are: by decomposing the original time series into multiple scale components and denoising and reconstructing, an optimized time series is obtained, and a relationship model is constructed using the cumulative generated sequence and the optimized time series, thereby improving the prediction accuracy and real-time performance of settlement deformation in shield construction, improving the quality and accuracy of input data of the prediction model, enhancing the robustness of the model in dealing with changes in geological conditions, construction parameters and environmental factors, and improving the efficiency of overall data processing and real-time prediction.
[0049] Embodiment 5: The embodiment of the present invention provides a shield settlement deformation monitoring method and system based on grey neural network, determines grey prediction model parameters, and constructs a grey prediction model, including: Grey prediction model basic formula: ;in, Indicates that the cumulative generated sequence is k The value at a point in time; Indicates that the cumulative generated sequence is k-1 The value at a point in time, ; Represents the relevant data value at the kth time point; a Represents the first parameter of the grey prediction model; b Represents the second parameter of the grey prediction model; Indicates c1 The autoregressive coefficient of the first lag term; Indicates c2 The moving average coefficient of the second lag term; represents independent and identically distributed random terms; c1 It represents the number corresponding to the first lag of the autoregressive term; c2 Represents the corresponding number of the second lag term of the moving average term; p represents the total number of first lag items; q represents the total number of second lag terms; Represents the first c1 The first lag value of the first lag term; Indicates c2 The lag noise value of the second lag term; A represents the system dynamics corresponding to the grey prediction model and the internal structure of the optimized time series; B represents the system noise and external influence of the grey prediction model; A grey prediction model is constructed according to the basic formula and the relationship model.
[0050] In this embodiment, the autoregressive term refers to a component in a time series model that uses past values to predict current or future values. Generally speaking, the autoregressive term is represented as a linear combination of the past data of the time series itself.
[0051] In this embodiment, the first lag term refers to the values of the first p time points used to predict the current value in the time series.
[0052] In this embodiment, the second lag term is related to the noise in the prediction model and is used in the moving average model. The lag noise term used is the error term at the previous q time points.
[0053] In this embodiment, the moving average term is a linear combination used to describe noise or error terms in the time series.
[0054] In this embodiment, the working principle of constructing the gray prediction model is based on the provided basic formula and relationship model, by defining the cumulative generation sequence, applying autoregression and moving average, establishing the difference equation, using historical data to estimate the model parameters through the least squares method, and establishing the final prediction model based on the optimized time series and the cumulative generation sequence.
[0055] The working principle and beneficial effect of the above technical solution are: constructing a grey prediction model by accumulating the lag term and noise term of the generated sequence and the optimized time series. The model comprehensively considers the system dynamics, internal structure, system noise and external influences, improves the accuracy and real-time performance of predicting the settlement deformation of shield construction, improves the prediction accuracy of uncertain data, meets the real-time monitoring and prediction needs of the project, and protects construction safety.
[0056] Embodiment 6: The embodiment of the present invention provides a shield settlement deformation monitoring method based on a grey neural network, which determines the feature dimensions in the relevant data set through a feature selection process, constructs a grey neural network model according to the feature dimensions combined with the prediction results of the grey prediction model, and predicts the settlement deformation amount in the shield process based on the grey neural network model, including: Select feature selection methods based on the data type of the relevant dataset and determine the feature dimensions and important features in the relevant dataset; Combine the important features with the prediction results of the grey prediction model to form the input features of the grey neural network model; Conduct term analysis on shield settlement deformation to determine the task type and model requirements needed to predict settlement deformation values; Determine the input layer according to the input features, set the hidden layer based on the task type, and then build the hierarchical structure of the grey neural network model; select the corresponding loss function, optimizer and evaluation index according to the task type, data characteristics and model requirements, and then compile the grey neural network model; and determine the grey neural network model by combining the hierarchical structure and the compilation results; The settlement deformation during shield tunneling is predicted based on the grey neural network model.
[0057] In this embodiment, data types include numerical types: such as soil density, moisture content, shield machine propulsion force, etc.; time series types: such as real-time monitored surface settlement data, propulsion speed, etc.; classification or label types: such as soil category, environment type, etc.; geographic space types: such as the geographical location of the stratum, groundwater level distribution, etc.
[0058] In this embodiment, the process of determining the task type and model requirements through noun analysis is to determine the category of the task and the requirements of the model by analyzing the key terms of the problem. In the shield settlement deformation prediction problem, the typical task type is a regression problem, whose goal is to predict a continuous variable (i.e., the settlement deformation value); according to the task type and engineering requirements, determine the functions and requirements that the model needs to have: Accuracy requirements: The model must have high accuracy to ensure that the prediction results are safe and reliable for the project, real-time: The model should be able to process real-time data and quickly generate prediction results, robustness: The model should be able to process multi-source heterogeneous data and have noise resistance and fault tolerance.
[0059] In this embodiment, the process of constructing the hierarchical structure determines the dimension and structure of the input layer based on the selected important features and the prediction results of the grey prediction model. For example, if 10 important features are selected, the number of nodes in the input layer is 10, plus the grey prediction results, a total of 11 input nodes; the number of layers and nodes of the hidden layer are set according to the task type and data characteristics, and are usually tuned through experiments and cross-validation. For example, you can try a single hidden layer or multiple hidden layers: single hidden layer: 5-10 nodes (suitable for simple problems), multiple hidden layers: such as 2 layers, 10-20 nodes per layer (suitable for complex problems); the number of nodes in the output layer should match the prediction target. For single value prediction, the output layer usually has only 1 node.
[0060] In this embodiment, compilation refers to the process of configuring the loss function, optimizer and evaluation index of the constructed neural network model so that it can be trained and predicted: select the loss function that matches the task type, such as the mean square error (MSE) or root mean square error (RMSE) commonly used for regression problems; select the appropriate optimizer according to the data characteristics and model requirements, such as SGD, Adam, RMSprop, etc. For example, the Adam optimizer is suitable for processing large-scale and high-noise data; select the appropriate evaluation index to measure the model performance, such as mean square error, mean absolute error, etc.
[0061] The working principle and beneficial effects of the above technical solution are as follows: after determining the important features through the feature selection method, combine them with the output of the grey prediction model to form the input features of the grey neural network model. According to the task type and data characteristics, the neural network structure is optimized, and the appropriate loss function and optimizer are selected to accurately predict the settlement deformation of the shield, improve the real-time prediction capability of the settlement deformation during the shield construction process, fully tap the data value, and improve the overall performance and efficiency of the prediction model.
[0062] Embodiment 7: The embodiment of the present invention provides a shield settlement deformation monitoring method based on a grey neural network, which adjusts the grey neural network model based on an environment-impact table and real-time data feedback to optimize the prediction result of the settlement deformation, including: Determine the environment-impact table based on relevant data sets; Use real-time data to update the environment-impact table to obtain the impact of current environmental factors; Select new important features according to the updated environment-influence table and update the input features of the grey neural network; According to real-time data, the incremental learning technology is used to adjust the parameters of the grey neural network model corresponding to the training model, and then the parameters of the grey neural network model are adjusted to achieve feedback adjustment of the grey neural network model.
[0063] In this embodiment, the degree of influence can be determined by data analysis and statistical methods, such as calculating the correlation coefficient between each environmental factor and the settlement deformation, and a higher correlation coefficient indicates a greater influence.
[0064] In this embodiment, the input features are feature variables used to construct the grey neural network model, which can be divided into two categories: original features: important features directly selected from the data set, such as geological conditions, construction parameters and monitoring data; derived features: new features generated by data transformation or combination, such as cumulatively generated sequences, trend items, etc.
[0065] In this embodiment, the process corresponding to updating the input features is to select new important features according to the real-time update of the environment-impact table and use them as the input of the gray neural network, for example: current important features: soil density, advancement speed, groundwater level change, cumulative generation sequence.
[0066] The working principle and beneficial effects of the above technical solution are: by updating the environment-impact table in real time, reselecting important features and adjusting the input features of the gray neural network. Using incremental learning technology, the model parameters are adjusted in real time to achieve dynamic optimization of the gray neural network model. Ultimately, the accuracy and adaptability of settlement deformation prediction during shield construction are improved, the uncertainty of geological conditions, construction parameters and environmental factors on settlement deformation prediction is reduced, and the model better adapts to real-time data changes on site and achieves real-time prediction.
[0067] Embodiment 8: The embodiment of the present invention provides a shield settlement deformation monitoring system based on grey neural network, such as Figure 2 As shown, including: Fusion module: It obtains relevant data that affects settlement and deformation during shield construction around the construction area, and fuses relevant data from different sources to form a relevant data set; Prediction model module: It forms cumulative generation sequence and original time series according to relevant data sets, optimizes the original time series, and establishes the relationship model between the cumulative generation sequence and the optimized time series, thereby determining the parameters of the grey prediction model and constructing the grey prediction model; Neural network model module: determine the feature dimensions in the relevant data set through the feature selection process, build a gray neural network model based on the feature dimensions combined with the prediction results of the gray prediction model, and predict the settlement deformation during the shield process based on the gray neural network model; Feedback adjustment module: adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the prediction results of settlement deformation.
[0068] The working principle and beneficial effects of the above technical solution are: by fusing multi-source data, optimizing the original time series, using the cumulative generation sequence to build a gray prediction model, feature selection combined with the gray prediction results to build a gray neural network model, based on real-time data feedback and environment-impact table, dynamically adjust the model parameters, so that the model can adapt to changes in the construction environment in real time, realize real-time prediction and adjustment, improve the accuracy and adaptability of the prediction, meet the needs of real-time monitoring and prediction during shield construction, and ensure the timeliness and reliability of the prediction results.
[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A shield settlement deformation monitoring method based on grey neural network, characterized in that: include: Step 1: Obtain relevant data that affect settlement and deformation during shield construction around the construction area, and merge relevant data from different sources to form a relevant data set; Step 2: According to the relevant data sets, the cumulative generation sequence and the original time series are formed, the original time series is optimized, and the relationship model between the cumulative generation sequence and the optimized time series is established, and then the grey prediction model parameters are determined and the grey prediction model is constructed; Step 3: Determine the feature dimensions in the relevant data set through a feature selection process, build a grey neural network model based on the feature dimensions combined with the prediction results of the grey prediction model, and predict the settlement deformation during the shield tunneling process based on the grey neural network model; Step 4: Adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the predicted results of settlement deformation.
2. The shield settlement deformation monitoring method based on grey neural network according to claim 1 is characterized in that: Obtain relevant data on the settlement and deformation that affect the shield construction process around the construction area, and merge relevant data from different sources to form a relevant data set, including: Divide the construction area vertically according to its geological layers, and determine the data collection strategy for each vertical area based on key project factors; According to the data acquisition strategy, relevant data affecting the settlement deformation of the corresponding longitudinal area during the shield construction process are obtained, and the relevant data are fused to form a relevant data set based on the fusion result.
3. The shield settlement deformation monitoring method based on grey neural network according to claim 2 is characterized in that: According to the data acquisition strategy, relevant data affecting the settlement deformation of the corresponding longitudinal area during the shield construction process are obtained, and the relevant data are fused to form a relevant data set based on the fusion result, including: Check the existing data sets in the project to determine the data characteristics of the initial data corresponding to each longitudinal region, and select the data fusion method based on the data characteristics of the corresponding longitudinal region and the project requirements; Based on the data characteristics of different longitudinal regions, the initial data of the corresponding longitudinal region is featured, sparse points are judged for the initial data of each longitudinal region, and the sparse points are interpolated according to the judgment results, so as to obtain the relevant data of the corresponding longitudinal region; Perform spatial superposition of the longitudinal regions, check the consistency between different longitudinal regions, and perform spatial alignment of the relevant data of different longitudinal regions based on the consistency check results; Perform event alignment on relevant data in different longitudinal regions according to event stamps; The data fusion method is used to fuse the relevant data of different longitudinal regions according to the spatial alignment result and the event alignment result, thereby forming a relevant data set.
4. The shield settlement deformation monitoring method based on grey neural network according to claim 1 is characterized in that: According to the relevant data sets, the cumulative generation sequence and the original time series are formed, the original time series is optimized, and the relationship model between the cumulative generation sequence and the optimized time series is established, including: Form the original time series from the relevant dataset: ;in, Indicated in k The relevant data value at a time point; Decompose the original time series into several scale components, perform denoising and reconstruction, and obtain the reconstructed time signal: ;in, represents the reconstructed time signal; Indicated in the scale factor Time Shift Factor The wavelet function at ; represents the scale factor; j Represents the discretization index of the scale factor; J Indicates the maximum scale; k Represents the discrete time axis k time point; t Represents the index of the time series; represents the time shift factor; Indicates time t Related data signals; represents the mother wavelet function; Represents discrete scale factor j and discrete translation factor k Wavelet function under the condition of ; According to the reconstructed time signal, an optimized time series after denoising is obtained, and a cumulative generation sequence is constructed according to the optimized time series, and then a relationship model is determined based on the cumulative generation sequence and the optimized time series.
5. A shield settlement deformation monitoring method and system based on grey neural network according to claim 4, characterized in that: Determine the parameters of the grey prediction model and build the grey prediction model, including: Grey prediction model basic formula: ;in, Indicates that the cumulative generated sequence is k The value at a point in time; Indicates that the cumulative generated sequence is k-1 The value at a point in time, ; Represents the relevant data value at the kth time point; a Represents the first parameter of the grey prediction model; b Represents the second parameter of the grey prediction model; Indicates c1 The autoregressive coefficient of the first lag term; Indicates c2 The moving average coefficient of the second lag term; represents independent and identically distributed random terms; c1 It represents the number corresponding to the first lag of the autoregressive term; c2 Represents the corresponding number of the second lag term of the moving average term; p represents the total number of first lag items; q represents the total number of second lag terms; Represents the first c1 The first lag value of the first lag term; Indicates c2 The lag noise value of the second lag term; A represents the system dynamics corresponding to the grey prediction model and the internal structure of the optimized time series; B represents the system noise and external influence of the grey prediction model; A grey prediction model is constructed according to the basic formula and the relationship model.
6. The shield settlement deformation monitoring method based on grey neural network according to claim 3 is characterized in that: The feature dimensions in the relevant data set are determined through a feature selection process, a grey neural network model is constructed according to the feature dimensions combined with the prediction results of the grey prediction model, and the settlement deformation amount in the shield process is predicted based on the grey neural network model, including: Select feature selection methods based on the data type of the relevant dataset and determine the feature dimensions and important features in the relevant dataset; Combine the important features with the prediction results of the grey prediction model to form the input features of the grey neural network model; Conduct term analysis on shield settlement deformation to determine the task type and model requirements needed to predict settlement deformation values; Determine the input layer according to the input features, set the hidden layer based on the task type, and then build the hierarchical structure of the grey neural network model; select the corresponding loss function, optimizer and evaluation index according to the task type, data characteristics and model requirements, and then compile the grey neural network model; and determine the grey neural network model by combining the hierarchical structure and the compilation results; The settlement deformation during shield tunneling is predicted based on the grey neural network model.
7. The shield settlement deformation monitoring method based on grey neural network according to claim 1 is characterized in that: Adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the predicted results of settlement deformation, including: Determine the environment-impact table based on relevant data sets; Use real-time data to update the environment-impact table to obtain the impact of current environmental factors; Select new important features according to the updated environment-influence table and update the input features of the grey neural network; According to real-time data, the incremental learning technology is used to adjust the parameters of the grey neural network model corresponding to the training model, and then the parameters of the grey neural network model are adjusted to achieve feedback adjustment of the grey neural network model.
8. A shield settlement deformation monitoring system based on grey neural network, characterized in that: include: Fusion module: It obtains relevant data that affects settlement and deformation during shield construction around the construction area, and fuses relevant data from different sources to form a relevant data set; Prediction model module: It forms cumulative generation sequence and original time series according to relevant data sets, optimizes the original time series, and establishes the relationship model between the cumulative generation sequence and the optimized time series, thereby determining the parameters of the grey prediction model and constructing the grey prediction model; Neural network model module: determine the feature dimensions in the relevant data set through the feature selection process, build a gray neural network model based on the feature dimensions combined with the prediction results of the gray prediction model, and predict the settlement deformation during the shield process based on the gray neural network model; Feedback adjustment module: adjust the grey neural network model based on the environment-impact table and real-time data feedback to optimize the prediction results of settlement deformation.
Citation Information
Cited By
Shallow-buried soft stratum subsurface tunnel settlement prediction method and system
CN120086779A
A method and system for predicting settlement of a tunnel dug in shallow soft strata
CN120086779B
Shield construction earthwork volume prediction method based on Kriging interpolation and classification regression tree
CN120317460A
Foundation pit excavation deformation prediction method, device, equipment and medium
CN120821958A
Cement-based material permeation corrosion degradation analysis method and system based on deep learning
CN120954574A
