Space-time evolution data processing method and system applied to tunnel deformation analysis

Abstract

The application provides a kind of space-time evolution data processing method and system applied to tunnel deformation analysis, by obtaining the time series monitoring sequence of different monitoring dimensions in tunnel construction process, based on this, the space-time evolution correlation model of tunnel deformation is constructed, by analyzing the dynamic coupling relationship of structure response data and geological environment action data, generate a space-time correlation feature matrix, and based on this, generate the feature evolution sequence of tunnel deformation, nonlinear mutation characteristics extraction is carried out on the feature evolution sequence, and the mutation feature index set is obtained;According to the mutation feature index set and the preset cumulative deformation rate threshold, the dynamic division of tunnel deformation safety level is carried out, and the safety level division result is generated, based on the safety level division result and the spatial distribution data of tunnel construction section, the spatial positioning and boundary contour extraction of potential risk damage area are carried out, and the risk damage area distribution map is obtained.The application can accurately evaluate and effectively control the safety risk of tunnel construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a spatio-temporal evolution data processing method and system applied to tunnel deformation analysis. BACKGROUND

[0002] In the process of tunnel engineering construction and operation, the deformation law and development trend of the tunnel structure under the action of external environment can be revealed through the processing and analysis of monitoring data. At present, the processing of tunnel deformation data is mostly difficult to fully reflect the dynamic coupling relationship between the response of the tunnel structure and the action of the geological environment, and the identification ability of the nonlinear mutation characteristics in the deformation process is limited, which leads to the situation that the early warning is lagged or misjudged in the risk level division, and it is difficult to accurately capture the spatial distribution characteristics of the potential damage area in the risk area positioning, affecting the accurate assessment and effective control of the safety risk of tunnel construction. SUMMARY

[0003] The present application provides a spatio-temporal evolution data processing method and system applied to tunnel deformation analysis.

[0004] In a first aspect, the present application provides a spatio-temporal evolution data processing method applied to tunnel deformation analysis, which comprises: acquiring time series monitoring sequences of different monitoring dimensions in the process of tunnel construction, wherein the time series monitoring sequences contain tunnel structure response data and geological environment action data recorded continuously at a preset sampling interval; constructing a spatio-temporal evolution correlation model of tunnel deformation based on the time series monitoring sequences, generating a spatio-temporal correlation feature matrix containing spatio-temporal coupling effect parameters by analyzing the dynamic coupling relationship between the structure response data and the geological environment action data; generating a feature evolution sequence of tunnel deformation based on the spatio-temporal correlation feature matrix, extracting nonlinear mutation characteristics of the feature evolution sequence to obtain a mutation feature index set representing the deformation trend; performing dynamic division of the safety level of tunnel deformation according to the mutation feature index set and a preset cumulative deformation rate threshold, and generating a safety level division result containing multiple early warning threshold intervals; based on the safety level division result and the spatial distribution data of the tunnel construction section, performing spatial positioning and boundary contour extraction of the potential risk damage area to obtain a risk damage area distribution map of the tunnel construction section.

[0005] In a second aspect, the present application provides a computer system, comprising: a memory having a computer program stored therein; a processor for loading the computer program to realize the spatio-temporal evolution data processing method applied to tunnel deformation analysis as described above.

[0006] The application can accurately capture the complex interaction between structure and environmental factors in the tunnel deformation process, and improve the comprehensiveness of deformation analysis; by generating a feature evolution sequence of tunnel deformation and extracting nonlinear mutation characteristics, a mutation feature index set representing the deformation trend is obtained, which can effectively identify key turning points in the deformation process and improve the early warning ability of potential risks; according to the mutation feature index set and the preset cumulative deformation rate threshold, the dynamic division of the safety level of tunnel deformation is carried out, and the safety level division result containing multiple early warning threshold intervals is generated, which can realize the fine evaluation of risk level and avoid the limitations of traditional fixed threshold evaluation method; based on the safety level division result and the spatial distribution data of the tunnel construction section, the spatial positioning and boundary contour extraction processing of the potential risk damage area are carried out, and the risk damage area distribution atlas of the tunnel construction section is obtained, which can accurately determine the spatial position and morphological characteristics of the risk area. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 is a flowchart of a spatio-temporal evolution data processing method applied to tunnel deformation analysis provided by an embodiment of the application.

[0008] Figure 2 is a composition schematic diagram of a computer system provided by an embodiment of the application. DETAILED DESCRIPTION

[0009] Please refer to Figure 1 is a flowchart of a spatio-temporal evolution data processing method applied to tunnel deformation analysis provided by an embodiment of the application, which can be executed by a computer system, and the method can include the following steps:

[0010] Step S100: acquiring time sequence monitoring sequences of different monitoring dimensions in the tunnel construction process, the time sequence monitoring sequences containing tunnel structure response data and geological environmental action data recorded continuously at a preset sampling interval.

[0011] The tunnel structure response data refers to response information generated by the tunnel structure when subjected to various forces and environmental effects, such as displacement, stress, strain and other data of the tunnel lining, which reflect the changes in the mechanical state of the tunnel structure itself. The geological environment effect data are data related to the geological environment in which the tunnel is located, including underground water level changes, physical and mechanical properties of rock and soil mass, seismic activity and other information, which reflect the effects and influences of the geological environment on the tunnel structure. These data can be obtained by using various monitoring equipment. For tunnel structure response data, displacement sensors can be used to collect displacement data of the tunnel lining, and strain gauges can be used to measure strain data of the tunnel structure. For geological environment effect data, underground water level monitoring instruments can be used to obtain underground water level change data, and geological radars can be used to detect the physical properties of rock and soil mass. For example, the preset sampling interval is 1 hour, and every 1 hour, the displacement sensor records the displacement data of the tunnel lining at a specific position, and the underground water level monitoring instrument records the height of the underground water level, so as to obtain the tunnel structure response data and the geological environment effect data arranged in time sequence, and obtain the time series monitoring sequence.

[0012] Step S200: constructing a spatio-temporal evolution correlation model of tunnel deformation based on the time series monitoring sequence, generating a spatio-temporal correlation feature matrix containing spatio-temporal coupling effect parameters by analyzing the dynamic coupling relationship between the structure response data and the geological environment effect data.

[0013] The spatio-temporal evolution correlation model of tunnel deformation is used to describe the evolution law of tunnel deformation in time and space, and through the analysis of the dynamic coupling relationship between the structure response data and the geological environment effect data, the interaction mechanism between the two can be revealed. The spatio-temporal coupling effect parameters reflect the interaction intensity of different monitoring dimensions in time and space, and the spatio-temporal correlation feature matrix is a matrix formed by integrating these parameters, which is used for tunnel deformation analysis.

[0014] As an implementation manner, step S200 can specifically include steps S210-S260 as follows:

[0015] Step S210: time axis coordination is performed on the structure response data and the geological environment effect data in the time series monitoring sequence, the data of different sampling intervals are adjusted to a unified time reference, and a time-synchronized monitoring data pair is obtained, each time point in the monitoring data pair contains corresponding structure response items and environmental effect items.

[0016] Since the structural response data and the geological environment action data can be collected by different monitoring devices at different sampling intervals, time axis coordination is needed. Unified time reference refers to aligning all data according to a common time standard. Time-synchronized monitoring data pair refers to a pair of data that simultaneously contains structural response data and geological environment action data at the same time point. For example, the sampling interval of the structural response data is 30 minutes, and the sampling interval of the geological environment action data is 1 hour. Through time axis coordination, the structural response data is filtered or interpolated according to an interval of 1 hour to make it consistent with the time point of the geological environment action data, thereby obtaining a time-synchronized monitoring data pair. For example, the linear interpolation algorithm is used to process the structural response data, and the approximate value of the intermediate time point is calculated according to the data values of the adjacent time points to realize the time synchronization with the geological environment action data.

[0017] Step S220: dividing the monitoring data pair into training samples and validation samples according to data distribution characteristics, the training samples are used for model parameter learning, and the validation samples are used for model performance evaluation, and the division ratio is dynamically determined according to the data time span.

[0018] Data distribution characteristics refer to the distribution law of data in time and space, such as the concentration degree and fluctuation range of data. Training samples are a data set used to train the spatiotemporal evolution correlation model of tunnel deformation. The parameters of the model are learned through these data, so that the model can better fit the data. Validation samples are used to evaluate the performance of the trained model and check the prediction accuracy and generalization ability of the model. The division ratio is dynamically determined according to the data time span, which means adjusting the proportion of training samples and validation samples according to the length of time covered by the data. For example, when the data time span is short, more data can be divided into training samples to make full use of limited data for model training; when the data time span is long, the proportion of validation samples can be appropriately increased to more accurately evaluate the model performance.

[0019] Step S230: constructing a combined network architecture containing a time-dependent feature extraction network and a spatial correlation feature extraction network, the time-dependent feature extraction network is used to capture the time evolution law of the structural response data, and the spatial correlation feature extraction network is used to model the spatial distribution characteristics of the geological environment action data, and the network input layer dimension matches the feature dimension of the monitoring data pair.

[0020] The combined network architecture is a network structure that combines a time-dependent feature extraction network and a spatial correlation feature extraction network. Through this architecture, the time and spatial features of the data can be processed simultaneously. The time-dependent feature extraction network can use a long short-term memory network (LSTM), which can effectively process sequence data and capture the time evolution law in the structural response data, such as the trend of tunnel lining displacement over time. The spatial correlation feature extraction network can use a convolutional neural network (CNN), which is good at processing data with spatial structure and can model the spatial distribution characteristics of geological environmental action data, such as the distribution of groundwater level at different spatial locations. The input layer dimension of the network matches the feature dimension of the monitoring data pair, which means that the number of neurons in the input layer is consistent with the number of features contained in the monitoring data pair, to ensure that the data can be correctly input into the network.

[0021] Step S240: input the structural response sub-item in the training sample into the time-dependent feature extraction network, generate the structural response time series feature vector through forward and backward time feature fusion, and input the environmental action sub-item into the spatial correlation feature extraction network, generate the environmental action spatial feature vector through spatial topological relationship modeling.

[0022] The structural response sub-item refers to the part of the monitoring data pair that belongs to the structural response data, and the environmental action sub-item refers to the part that belongs to the geological environmental action data. Forward and backward time feature fusion refers to the integration of forward and backward features of structural response data in time to more comprehensively capture the time evolution information of the data. In the LSTM network, the hidden states at different time steps are fused through the processes of forward propagation and backward propagation to generate the structural response time series feature vector. Spatial topological relationship modeling refers to the analysis and representation of the connection and distribution relationship of geological environmental action data in space. In the CNN network, the spatial features of geological environmental action data are extracted through the operations of convolutional and pooling layers to generate the environmental action spatial feature vector. For example, the tunnel lining displacement data in the training sample is input into the LSTM network, and the structural response time series feature vector is generated after network processing; the groundwater level distribution data is input into the CNN network, and the relationship between groundwater level at different spatial locations is modeled through the network to generate the environmental action spatial feature vector.

[0023] Step S250: model the correlation between the structural response time series feature vector and the environmental action spatial feature vector through the feature interaction layer of the network, calculate the interaction strength of the two at different time scales, and generate the coupling feature tensor containing the spatio-temporal correlation.

[0024] The feature interaction layer is the part of the combined network architecture that is used to realize the interaction between the structural response time series feature vector and the environmental action spatial feature vector. Correlation modeling refers to analyzing the relationship between the two vectors and finding their interaction patterns at different time scales. The interaction intensity reflects the degree of correlation between the structural response data and the geological environmental action data at different time scales. The coupled feature tensor is a tensor that contains three dimensions of time, space, and features, which integrates the spatiotemporal correlation between the structural response data and the geological environmental action data at different time scales.

[0025] As an implementation, step S250 can specifically include steps S251-S256:

[0026] Step S251: Divide the structural response time series feature vector and the environmental action spatial feature vector into feature segment groups according to the time window, each segment group containing structural feature segments and environmental feature segments at consecutive time points, the window length being determined according to the periodic characteristics of the monitoring data, and generating multiple groups of time series feature segments.

[0027] The feature segment group is a grouping obtained by dividing the structural response time series feature vector and the environmental action spatial feature vector according to the time window, each grouping containing structural feature segments and environmental feature segments within the time window. The periodic characteristics of the monitoring data refer to the repeated regularity exhibited by the data over time, for example, some geological environmental factors may have a seasonal variation period. According to this periodic characteristic, the length of the time window can be determined to better capture the variation of the data. For example, if the monitoring data shows that the groundwater level has a periodic variation of once a year, the length of the time window can be set to one year.

[0028] Step S252: Perform feature discretization on each group of feature segments, converting continuous feature values into interval identifiers, adjusting the number of interval divisions according to the feature value distribution range, and generating a discretized feature matrix, wherein each element represents the interval belonging of the corresponding feature dimension.

[0029] Feature discretization is the process of converting continuous feature values into discrete interval identifiers. The number of interval divisions is adjusted according to the feature value distribution range, which means that the feature values are divided into how many intervals according to the maximum and minimum values of the feature values and the density of their distribution. The discretized feature matrix is a matrix, and each element in the matrix represents the interval to which the feature value of the corresponding feature dimension belongs. For example, for the tunnel lining displacement feature, the feature value range is between 0-100mm, and according to its distribution, this range is divided into 10 intervals, each interval corresponding to an interval identifier. For the tunnel lining displacement feature value in each feature segment, it is converted into the corresponding interval identifier, thereby generating a discretized feature matrix.

[0030] Step S253: Construct a feature combination transaction set based on the discretized feature matrix, wherein each transaction corresponds to a feature combination relationship within a time window, and the transaction item is the joint identification of the feature dimension and the interval identifier, and a feature transaction dataset is generated.

[0031] The feature combination transaction set is a set composed of multiple transactions, and each transaction represents the combination relationship of the interval to which the feature value of different feature dimensions belongs within a time window. The transaction item is the joint identification of the feature dimension and the interval identifier, which is used to uniquely identify the state of a feature in a certain interval. For example, for the two feature dimensions of tunnel lining displacement and groundwater level, within a certain time window, the feature value of tunnel lining displacement falls in interval 3, and the feature value of groundwater level falls in interval 5. The corresponding transaction items can be represented as "tunnel lining displacement_interval 3" and "groundwater level_interval 5". Combining these two transaction items forms a transaction. By processing each feature segment in the discretized feature matrix in this way, a feature combination transaction set is constructed, and a feature transaction dataset is generated.

[0032] Step S254: Perform association rule mining on the feature transaction dataset through co-occurrence pattern recognition, filter the feature combinations that repeatedly appear in different time windows, calculate the appearance frequency and conditional appearance probability of the combination, and generate a candidate association rule set.

[0033] In the feature transaction dataset, the association rules between the structural response data and the geological environment action data are mined through co-occurrence pattern recognition. Filtering the feature combinations that repeatedly appear in different time windows refers to finding the combinations of feature dimensions and interval identifiers that appear simultaneously in multiple time windows. The appearance frequency refers to the proportion of the number of times a certain feature combination appears in all time windows to the total number of time windows. The conditional appearance probability is the probability of the appearance of another feature combination under the condition that a certain feature combination has appeared. By calculating these values, a candidate association rule set is generated. For example, in the tunnel monitoring data, it is found that the feature combination "tunnel lining displacement_interval 3" and "groundwater level_interval 5" appears simultaneously in multiple time windows, and its appearance frequency and conditional appearance probability are calculated, which are used as candidate association rules. In actual operation, the Apriori algorithm can be used for co-occurrence pattern recognition, and through layer-by-layer search, the frequent item set is found, and then the candidate association rules are generated.

[0034] Step S255: Screen the candidate association rule set, retain the rules whose appearance frequency and conditional appearance probability both meet the preset conditions, remove false association rules through association strength evaluation, and generate a preliminary association rule set.

[0035] The preset conditions are pre-defined thresholds for occurrence frequency and condition occurrence probability. A rule is retained only if both its occurrence frequency and condition occurrence probability exceed these thresholds. Association strength assessment evaluates the reliability of association rules, removing potentially spurious association rules. For example, with a preset occurrence frequency threshold of 0.3 and a condition occurrence probability threshold of 0.5, for each rule in the candidate association rule set, its occurrence frequency and condition occurrence probability are checked against these thresholds; if they are, the rule is retained, otherwise it is removed. Simultaneously, association strength is assessed by calculating metrics such as lift; rules with a lift less than 1 are considered spurious association rules and removed.

[0036] Step S256: Arrange the initial association rule set in chronological order, merge rules that are sequential in time and have consistent feature combinations, and generate an association rule set containing time evolution characteristics. The association rule set is used to guide the association modeling process of the feature interaction layer.

[0037] Arranging the initial association rule set chronologically is to better reflect the evolution of rules over time. Merging rules that are temporally consecutive and have consistent feature combinations means combining rules with the same feature combinations appearing in adjacent time windows to form a rule with a longer time span. This generated association rule set incorporates temporal evolution characteristics and can more accurately reflect the temporal correlation changes between structural response data and geological environmental impact data. For example, if association rules for "tunnel lining displacement_interval 3" and "groundwater level_interval 5" appear in three consecutive time windows, merging these three rules into one rule will cover all three time windows. This association rule set, incorporating temporal evolution characteristics, will guide the association modeling process in the feature interaction layer, enabling the model to better capture the interaction relationships between structural response data and geological environmental impact data at different time scales.

[0038] In one implementation, step S250 may further include the following steps S257 to S2512:

[0039] Step S257: Determine the association dimensions between structural response features and environmental action features based on the association rule set, generate a feature dimension mapping relationship table, record the association types and influence directions between different feature dimensions, and generate a feature association dimension index.

[0040] The association rule set reflects the association relationship between the structure response data and the geological environmental action data. According to the association rule set, the association dimension between the structure response features and the environmental action features is determined, that is, it is found out which feature dimensions are associated. The feature dimension mapping relationship table is a table used to record the association type (such as positive correlation, negative correlation, etc.) and the influence direction (that is, how the change of one feature influences another feature) between different feature dimensions. The feature association dimension index is an index used to quickly find and locate the feature dimension association information. For example, it is found through the association rule set that there is an association between the tunnel lining displacement and the underground water level, and the association type between the two feature dimensions is recorded in the feature dimension mapping relationship table as positive correlation, and the influence direction is that the rise of the underground water level may lead to the increase of the tunnel lining displacement. According to these information, the feature association dimension index is generated, which is convenient for subsequent calculation and analysis.

[0041] Step S258: The occurrence frequency of the association rule in the association rule set and the conditional occurrence probability are combined by a preset proportion to obtain a rule influence strength value of each association rule, and the rule influence strength values of all the association rules are arranged in order to obtain a rule importance sequence.

[0042] The preset proportion is a preset weight proportion of the occurrence frequency and the conditional occurrence probability. The rule influence strength value is a value obtained by combining the occurrence frequency and the conditional occurrence probability by weighting, which reflects the importance of the association rule in the entire association relationship. The weight of the occurrence frequency is adjusted according to the uniformity of the distribution of the association rule in different time windows. If the occurrence frequency of an association rule in each time window is relatively uniform, the weight of the occurrence frequency can be set to be relatively high; otherwise, the weight is appropriately reduced. The weight of the conditional occurrence probability is set according to the difference of the field source types contained in the antecedent of the association rule. Different field source types of the association rule may have different importance, so the weight of the conditional occurrence probability needs to be adjusted according to the field source type. For example, for the association rule "tunnel lining displacement_interval 3->underground water level_interval 5", assuming that the preset occurrence frequency weight is 0.6 and the conditional occurrence probability weight is 0.4, the rule influence strength value of the rule is calculated according to the occurrence frequency and the conditional occurrence probability of the rule in different time windows. The rule influence strength values of all the association rules are arranged in order from large to small to obtain the rule importance sequence.

[0043] Step S259: According to the rule importance sequence, different association dimensions are assigned with interaction weights, the feature dimensions corresponding to the rules with high importance are given with higher weights, and the weight values are mapped to a unified numerical range through normalization processing to generate a feature interaction weight matrix.

[0044] The rule importance sequence reflects the importance of each association rule, and according to the sequence, the interaction weight is assigned to different association dimensions, that is, for the feature dimensions involved in the rule with high importance, a higher weight is given to represent that these feature dimensions have greater influence in the association modeling process. Normalization is the process of mapping the weight value to a unified numerical range (such as [0, 1]), which can ensure the comparability of the weights of different feature dimensions. The feature interaction weight matrix is a matrix, and the elements in the matrix represent the interaction weights between different feature dimensions. For example, in the rule importance sequence, the importance of the rule "tunnel lining displacement_interval 3 -> groundwater level_interval 5" is high, and in the feature interaction weight matrix, the element value corresponding to the two feature dimensions of tunnel lining displacement and groundwater level will be relatively high.

[0045] Step S2510: The structural response time series feature vector and the environmental action space feature vector are weighted and combined according to the feature interaction weight matrix to calculate the feature interaction values at different time scales, and a multi-scale interaction feature matrix is generated, the dimension of which matches the number of time scales of the feature vector.

[0046] Weighted combination means that each element in the structural response time series feature vector and the environmental action space feature vector is multiplied by the corresponding weight in the feature interaction weight matrix, and then the results are added to obtain the feature interaction value. The feature interaction value at different time scales reflects the association degree of the structural response data and the geological environmental action data at different time lengths. The multi-scale interaction feature matrix is a matrix, the dimension of which matches the number of time scales of the feature vector, and each element in the matrix represents the feature interaction value at a certain time scale. For example, if the feature vector contains information of 10 time scales, then the multi-scale interaction feature matrix is a 10-dimensional matrix.

[0047] Step S2511: Feature fusion is performed on the multi-scale interaction feature matrix to integrate the interaction features at different time scales through cross-scale feature association analysis, and a coupled feature tensor with spatiotemporal consistency is generated, and the tensor dimension contains time, space and feature three dimensions.

[0048] Feature fusion is the process of integrating interactive features of different time scales in the multi-scale interactive feature matrix. Cross-scale feature correlation analysis refers to analyzing the correlation between interactive features at different time scales and finding their common features and variation rules. Through this analysis, interactive features at different time scales are integrated together to generate a coupling feature tensor with spatiotemporal consistency. The dimension of the coupling feature tensor includes time, space, and feature, which integrates the information of structural response data and geological environmental action data in terms of time, space, and feature. For example, the principal component analysis (PCA) method is used to fuse the features of the multi-scale interactive feature matrix, project the interactive features at different time scales into a low-dimensional space, and then construct a coupling feature tensor with spatiotemporal consistency based on the projection results.

[0049] Step S2512: Calculate the spatiotemporal coupling effect parameters based on the coupling feature tensor. The parameter values reflect the interaction intensity of different monitoring dimensions at a given spatiotemporal location. Arrange the parameter values according to the time and space dimensions to generate the core parameter items of the spatiotemporal correlation feature matrix.

[0050] The spatiotemporal coupling effect parameters are calculated based on the coupling feature tensor and reflect the interaction intensity of different monitoring dimensions (such as tunnel structural response data and geological environmental action data) at a given spatiotemporal location. Arrange these parameter values according to the time and space dimensions to form the core parameter items of the spatiotemporal correlation feature matrix. For example, for a certain time point and spatial location, calculate the spatiotemporal coupling effect parameter between the tunnel lining displacement and groundwater level at that location based on the coupling feature tensor, and then arrange and arrange all the parameter values of the time points and spatial locations to generate the core parameter items of the spatiotemporal correlation feature matrix. For example, use the tensor decomposition method to process the coupling feature tensor to calculate the spatiotemporal coupling effect parameters and arrange them according to the time and space dimensions.

[0051] Step S260: Evaluate the prediction accuracy of the coupling feature tensor based on the validation sample, adjust the network weight parameters through the error feedback mechanism, and iteratively optimize until the prediction accuracy meets the preset requirements to generate a spatiotemporal correlation feature matrix containing spatiotemporal coupling effect parameters. The spatiotemporal coupling effect parameters are used to quantify the influence relationship between different monitoring dimensions.

[0052] The verification sample is used to evaluate the prediction accuracy of the coupling feature tensor. The error feedback mechanism refers to a mechanism for adjusting the weight parameters of the network according to the error between the predicted value and the actual value of the verification sample. Iterative optimization refers to the process of repeatedly adjusting the weight parameters and evaluating the prediction accuracy until the prediction accuracy meets the preset requirements. The spatiotemporal correlation feature matrix contains spatiotemporal coupling effect parameters, which are used to quantify the influence relationship between different monitoring dimensions. For example, using mean square error (MSE) as an evaluation index, the mean square error between the predicted value and the actual value of the coupling feature tensor for the verification sample is calculated. If the error is large, the weight parameters of the network are adjusted through the back propagation algorithm, and prediction and evaluation are performed again until the mean square error is less than the preset threshold. Through multiple iterations of optimization, the spatiotemporal correlation feature matrix containing the spatiotemporal coupling effect parameters is finally generated.

[0053] Step S300: generating a feature evolution sequence of tunnel deformation based on the spatiotemporal correlation feature matrix, performing nonlinear mutation characteristic extraction on the feature evolution sequence to obtain a mutation feature index set representing the deformation trend.

[0054] The feature evolution sequence of tunnel deformation is a sequence reflecting the change of tunnel deformation with time generated according to the spatiotemporal correlation feature matrix. Nonlinear mutation characteristic extraction refers to finding out those parts that suddenly change from the feature evolution sequence, which may indicate that the trend of tunnel deformation has changed. The mutation feature index set is a set of indexes for representing the trend of tunnel deformation, containing relevant information of the mutation interval. For example, by processing the spatiotemporal correlation feature matrix in chronological order, the feature values corresponding to each time point are extracted to form the feature evolution sequence of tunnel deformation. Then the sequence is analyzed to find out the positions of sudden change of feature values, and the relevant features of these positions are extracted to form the mutation feature index set. For example, using the difference algorithm to process the feature evolution sequence, the difference value of the feature values of adjacent time points is calculated, and when the difference value exceeds a certain threshold, it is considered that a mutation has occurred, and the relevant mutation feature index is extracted.

[0055] As an implementation, step S300 can specifically include steps S310-S360:

[0056] Step S310: performing dominant evolution feature extraction on the spatiotemporal correlation feature matrix to obtain a low-dimensional dominant feature matrix, the row dimension of the matrix corresponds to the time axis, and the column dimension corresponds to the dominant feature dimension.

[0057] The principal evolution feature extraction is a process of extracting those features that play a major role in the evolution of tunnel deformation from the spatiotemporal correlation feature matrix. The low-dimensional principal feature matrix is a matrix with a lower dimension, which only contains the principal evolution features, and the row dimension corresponds to the time axis and the column dimension corresponds to the principal feature dimension. For example, the principal component analysis (PCA) method is used to process the spatiotemporal correlation feature matrix, and the principal components with the largest variance are found. The spatiotemporal correlation feature matrix is projected onto these principal components to obtain the low-dimensional principal feature matrix.

[0058] Step S320: The low-dimensional principal feature matrix is arranged in time sequence to construct a tunnel deformation feature evolution sequence indexed by time, and each time point in the sequence contains multi-dimensional deformation feature values at the corresponding time, and the feature dimension is consistent with the column dimension of the low-dimensional principal feature matrix.

[0059] The low-dimensional principal feature matrix is arranged in time sequence to enable the data to be displayed and analyzed in chronological order. The tunnel deformation feature evolution sequence indexed by time is a sequence identified by time, and each time point corresponds to a set of multi-dimensional deformation feature values, and the dimensions of these feature values are the same as the column dimensions of the low-dimensional principal feature matrix. For example, the low-dimensional principal feature matrix has 3 columns, representing three principal evolution features. Therefore, each time point in the tunnel deformation feature evolution sequence contains the values of the three features. The low-dimensional principal feature matrix is sorted in time sequence to construct the tunnel deformation feature evolution sequence.

[0060] Step S330: The feature evolution sequence is processed by a sliding window to obtain a plurality of sequence segments with time overlap, each sequence segment containing deformation feature values at consecutive time points, and the window size is determined according to the time resolution of the sequence.

[0061] Sliding window processing is a method of local analysis on sequence data, which slides a fixed-size window on the sequence and forms a sequence segment by cutting the data within the window each time. The time resolution refers to the time interval between adjacent time points in the sequence, and the window size is determined according to the time resolution to ensure that the change information in the sequence can be captured. The sequence segments with time overlap refer to the fact that there are some time points that are the same between adjacent sequence segments, which ensures the continuity of the sequence. For example, assuming that the time resolution of the feature evolution sequence is 1 day and the window size is set to 7 days, then each time the window is slid to cut 7 days of deformation feature values to form a sequence segment, and adjacent sequence segments may have a 3-day time overlap.

[0062] As an implementation, step S330 can specifically include steps S331-S335:

[0063] Step S331: Analyze the time resolution characteristic of the feature evolution sequence, determine the time interval of adjacent time points in the sequence, the time interval reflects the intensive degree of data collection, and generate a sequence time interval parameter.

[0064] The time resolution characteristic refers to the distribution and interval of time points in the feature evolution sequence. By analyzing this characteristic, the time interval between adjacent time points is determined, which reflects the intensive degree of data collection. The sequence time interval parameter is a quantitative representation of this time interval.

[0065] Step S332: Set the initial size of the sliding window based on the sequence time interval parameter, the initial size is an integer multiple of the time interval, and generate an initial window parameter.

[0066] According to the sequence time interval parameter, the initial size of the sliding window is set, in order to be able to capture the change information in the sequence completely, the initial size is set to be an integer multiple of the time interval. The initial window parameter is a record of the initial size of the sliding window. For example, the sequence time interval parameter is 1 hour, the initial size of the sliding window is set to 7 hours, and the initial window parameter is generated.

[0067] Step S333: Adjust the initial window parameter, correct the window size according to the change frequency of the feature value in the feature evolution sequence, reduce the window size in the region where the feature value changes frequently to improve the local detail capture ability, and increase the window size in the region where the feature value changes slowly to reduce the calculation amount, and generate a dynamic window parameter.

[0068] The change frequency of the feature value refers to the change speed of the feature value between different time points. According to this change frequency, the initial window parameter is adjusted, in the region where the feature value changes frequently, the window size is reduced to capture the local change details more accurately; in the region where the feature value changes slowly, the window size is increased to reduce unnecessary calculation amount. The dynamic window parameter is the adjusted window size parameter. For example, by calculating the difference of the feature value between adjacent time points in the feature evolution sequence, the change frequency of the feature value is determined. In the time period where the feature value changes frequently, the window size is reduced from 7 hours to 3 hours; in the time period where the feature value changes slowly, the window size is increased to 10 hours, and the dynamic window parameter is generated.

[0069] Step S334: Slide and segment the feature evolution sequence according to the dynamic window parameter, set the window moving step to be a fixed proportion of the window size, generate a sequence segment set, so that the adjacent windows maintain time overlap, and the overlapping part is used to maintain the time continuity of the sequence segment.

[0070] Sliding segmentation refers to sliding on the feature evolution sequence according to the dynamic window parameter, and each time a sequence segment is formed by intercepting part of the data in the window. The window moving step refers to the distance of the window sliding each time, which is set as a fixed proportion of the window size, so that there is a certain time overlap between adjacent windows. The sequence segment set is a set composed of all intercepted sequence segments. For example, the dynamic window parameter is 5 hours, and the window moving step is set to 50% of the window size, i.e. 2.5 hours, so that each time the window is slid by 2.5 hours to intercept the 5-hour deformation feature value to form a sequence segment, and there is a 2.5-hour time overlap between adjacent sequence segments.

[0071] Step S335: boundary smoothing is performed on each sequence segment, and feature mutations at the window boundaries are eliminated through edge feature value transition processing to keep the feature value changes within the segment continuous, and a smoothed sequence segment is generated.

[0072] Boundary smoothing is a process of processing the boundaries of the sequence segment, and feature mutations that may occur at the window boundaries are eliminated through edge feature value transition processing. The edge feature value transition processing can use linear interpolation and other methods to calculate the transition value at the boundary according to the feature values near the boundary, so that the feature value changes within the segment remain continuous. The smoothed sequence segment is the sequence segment after boundary smoothing processing. For example, for each sequence segment, the linear interpolation method is used to process the feature values at the boundaries, the transition values are calculated, and the feature values at the boundaries are replaced with the transition values, thereby generating a smoothed sequence segment.

[0073] Step S340: non-linear dynamic characteristic parameters of each sequence segment are calculated, including trajectory divergence characteristic parameters and trajectory complexity characteristic parameters, and a segment dynamic parameter set is generated, wherein the trajectory divergence characteristic parameters reflect the divergence degree of the sequence segment evolution over time, and the trajectory complexity characteristic parameters reflect the spatial distribution complexity of the sequence segment.

[0074] The non-linear dynamic characteristic parameters are used to describe the dynamic change characteristics of the sequence segment. The trajectory divergence characteristic parameters reflect the separation degree between the feature values of the sequence segment in the time evolution process. If the parameter value is large, it means that the sequence segment changes more and more dispersed over time; the trajectory complexity characteristic parameters reflect the complexity of the spatial distribution of the sequence segment. The larger the parameter value, the more complex the spatial distribution of the sequence segment. The segment dynamic parameter set is a set composed of the non-linear dynamic characteristic parameters of each sequence segment. For example, the maximum Lyapunov exponent is used to calculate the trajectory divergence characteristic parameters, and the correlation dimension is used to calculate the trajectory complexity characteristic parameters.

[0075] Step S350: identifying nonlinear mutation positions in the feature evolution sequence based on the fragment dynamics parameter set, calculating the parameter variation degree of adjacent sequence fragments, and marking potential mutation points when the parameter variation degree exceeds the preset determination condition, and merging time-adjacent potential mutation points into a mutation interval through mutation point clustering processing.

[0076] The nonlinear mutation position refers to the position where the feature value suddenly changes in the feature evolution sequence. These mutation positions are identified by calculating the parameter variation degree of adjacent sequence fragments, which can be obtained by calculating the difference of nonlinear dynamics characteristic parameters of adjacent sequence fragments. The preset determination condition is a threshold value set in advance. When the parameter variation degree exceeds this threshold value, the position is marked as a potential mutation point. The mutation point clustering processing is a process of merging time-adjacent potential mutation points into a mutation interval, which can more accurately describe the range of mutation. For example, the preset determination condition is that the parameter variation degree is greater than 0.5, and when the difference of trajectory divergence characteristic parameters of adjacent sequence fragments is greater than 0.5, the position is marked as a potential mutation point. Then, time-adjacent analysis is performed on all potential mutation points, and potential mutation points with a time interval less than a certain value are merged into a mutation interval.

[0077] As an implementation, step S350 can specifically include steps S351-S356 as follows:

[0078] Step S351: calculating the change rate of each parameter value in the fragment dynamics parameter set, the change rate being the ratio of the difference between the current fragment parameter value and the previous fragment parameter value to the time interval, and the time interval being the difference between the center time points of the two fragments, and generating a parameter change rate sequence.

[0079] The change rate of the parameter value reflects the change speed of the parameter value between adjacent sequence fragments. By calculating the difference between the current fragment parameter value and the previous fragment parameter value, and dividing it by the difference between the center time points of the two fragments, the parameter change rate is obtained. The parameter change rate sequence is a sequence composed of the change rates of all parameter values. For example, for the trajectory divergence characteristic parameter, the difference between the trajectory divergence characteristic parameter value of the current sequence fragment and the trajectory divergence characteristic parameter value of the previous sequence fragment is calculated, and then divided by the difference between the center time points of the two fragments to obtain the change rate of the parameter.

[0080] Step S352: obtaining a preset parameter change rate determination threshold, the determination threshold being dynamically determined according to the overall distribution characteristics of the parameter change rate sequence, so that the parameter variation within the normal fluctuation range will not be misjudged as a mutation, and generating a mutation determination threshold.

[0081] The preset parameter change rate determination threshold is a criterion for determining whether the parameter change is a mutation. The threshold is dynamically determined according to the overall distribution characteristics of the parameter change rate sequence, for example, the mean and standard deviation of the parameter change rate sequence can be used to determine the threshold, so that the parameter change within the normal fluctuation range will not be misjudged as a mutation. The mutation determination threshold is the final threshold for determining the mutation. For example, the mean and standard deviation of the parameter change rate sequence are calculated, and the mean plus a certain multiple of the standard deviation is taken as the mutation determination threshold, so that the parameter change within the normal fluctuation range will not be marked as a mutation.

[0082] Step S353: Mark the parameter change points in the parameter change rate sequence that exceed the mutation determination threshold as potential mutation points, record the time stamp and corresponding parameter change rate value of each potential mutation point, and generate a potential mutation point set.

[0083] When a parameter change rate value in the parameter change rate sequence exceeds the mutation determination threshold, the parameter change point is marked as a potential mutation point. The time stamp and corresponding parameter change rate value of each potential mutation point are recorded to form a potential mutation point set. For example, the parameter change rate sequence is traversed, the parameter change points that exceed the mutation determination threshold are marked as potential mutation points, their time stamps and parameter change rate values are recorded, and a potential mutation point set is generated.

[0084] Step S354: Time proximity analysis is performed on the potential mutation point set, the time interval between adjacent potential mutation points is calculated, and when the time interval is less than a preset merging condition, the adjacent potential mutation points are determined as time-proximal mutation points, and a mutation point proximity group is generated.

[0085] Time proximity analysis is a process of analyzing the time interval between adjacent potential mutation points in the potential mutation point set. The preset merging condition is a pre-set time interval threshold. When the time interval between adjacent potential mutation points is less than the threshold, they are determined as time-proximal mutation points. The mutation point proximity group is a group composed of time-proximal mutation points. For example, the preset merging condition is that the time interval is less than 3 days, and when the time interval between adjacent potential mutation points is less than 3 days, they are grouped into a mutation point proximity group.

[0086] Step S355: The potential mutation points in the mutation point proximity group are merged into a mutation interval through mutation point clustering processing, the start time of the interval is the time stamp of the earliest mutation point in the group, and the end time of the interval is the time stamp of the latest mutation point in the group, and a preliminary mutation interval is generated.

[0087] Mutation point clustering processing is a process of merging potential mutation points in the mutation point proximity group into a mutation interval. The start time of the mutation interval is the time stamp of the earliest mutation point in the group, and the end time of the interval is the time stamp of the latest mutation point in the group. The preliminary mutation interval is an interval obtained by such merging processing.

[0088] Step S356: Interval verification is performed on the preliminary mutation interval, the characteristic value fluctuation range of the characteristic evolution sequence in the interval is calculated, the interval is marked as a noise interval and removed when the fluctuation range does not reach a preset significant condition, and a final mutation interval set is generated.

[0089] Interval verification is a process of checking the effectiveness of the preliminary mutation interval. By calculating the characteristic value fluctuation range of the characteristic evolution sequence in the interval, it is determined whether the interval truly represents a mutation. The preset significant condition is a fluctuation range threshold value set in advance. When the characteristic value fluctuation range in the interval does not reach this threshold value, the interval is marked as a noise interval and removed. The final mutation interval set is a set composed of the mutation intervals remaining after interval verification. For example, the preset significant condition is that the characteristic value fluctuation range is greater than 10. When the characteristic value fluctuation range of the characteristic evolution sequence in a preliminary mutation interval is less than 10, the interval is marked as a noise interval and removed.

[0090] Step S360: Interval features of the mutation interval are extracted, including interval start time, interval end time, parameter variation amplitude, and statistical features of the kinetic parameters in the interval. The characteristic values are arranged in time sequence after standardization processing, a mutation feature index set representing the deformation trend is generated, and each index item of the mutation feature index set contains the time attribute and kinetic characteristic parameters of the mutation interval.

[0091] The interval features of the mutation interval are some characteristic information describing the mutation interval, including the start and end time of the interval, the parameter variation amplitude, and the statistical features of the kinetic parameters in the interval (such as mean value, standard deviation, etc.). Standardization processing is a process of mapping these characteristic values to a unified numerical range (such as [0, 1]), so that different characteristic values have comparability. Arranging in time sequence is to sort the extracted interval features according to the start time of the mutation interval. The mutation feature index set is a set composed of these standardized interval features, each index item contains the time attribute and kinetic characteristic parameters of the mutation interval, and is used to represent the trend of tunnel deformation.

[0092] Step S400: According to the mutation feature index set and the preset cumulative deformation rate threshold, dynamic division of the tunnel deformation safety level is performed, and a safety level division result containing multiple early warning threshold intervals is generated.

[0093] The mutation feature index set reflects the relevant feature conditions of the tunnel deformation in different mutation intervals, and the preset cumulative deformation rate threshold is a standard for defining different safety levels. By combining the mutation feature index set with the threshold, the safety level of the tunnel deformation can be dynamically divided. The multi-level early warning threshold interval refers to the set interval range of the cumulative deformation rate corresponding to each safety level, and the safety level division result comprehensively considers the multi-level early warning threshold interval and the safety level information corresponding to each interval.

[0094] As an implementation manner, the step S400 can specifically include steps S410-S460:

[0095] Step S410: Extracting a deformation rate feature from the feature evolution sequence, and generating a cumulative deformation rate sequence in time sequence, wherein each element corresponds to a deformation rate value of a time point, and the value reflects the development speed of the deformation at the time point.

[0096] The feature evolution sequence is generated based on the space-time correlation feature matrix and reflects the change of the tunnel deformation over time. The deformation rate feature is feature information that can reflect the development speed of the tunnel deformation at each time point in the feature evolution sequence. The deformation rate feature can be extracted from the feature evolution sequence by calculating the ratio of the change amount of the feature value of adjacent time points to the time interval. For example, for the tunnel lining displacement feature, the displacement difference between two adjacent time points is calculated, and then divided by the time interval of the two time points to obtain the average deformation rate in the time period, which is taken as the deformation rate feature of the corresponding time point.

[0097] The extracted deformation rate feature is arranged in time sequence to obtain a cumulative deformation rate sequence. Each element in the sequence corresponds to a specific time point, and the value of the element is the deformation rate value of the time point, which reflects the development speed of the tunnel deformation at the time point. For example, if the deformation rate value of a time point is large, it indicates that the tunnel deformation develops rapidly at that time; otherwise, if the deformation rate value is small, it indicates that the tunnel deformation develops relatively slowly.

[0098] Step S420: Comparing the cumulative deformation rate sequence with a preset initial early warning threshold, the initial early warning threshold includes a plurality of continuous threshold intervals, each interval corresponds to a different safety level, and generating a preliminary safety level sequence, wherein each time point is marked with a corresponding safety level label.

[0099] The preset initial warning threshold is a series of preset continuous threshold intervals, each of which corresponds to a set safety level. By comparing each element in the cumulative deformation rate sequence with these threshold intervals one by one, the corresponding safety level is determined according to the interval in which the element is located. In this way, each time point in the cumulative deformation rate sequence can be labeled with the corresponding safety level tag, and a preliminary safety level sequence is generated. For example, the initial warning threshold is set to several continuous intervals, which correspond to different safety levels such as "safe", "mild danger", "moderate danger", "high danger", etc. For each deformation rate feature value in the cumulative deformation rate sequence, determine the threshold interval it belongs to, and label it with the corresponding safety level. If the deformation rate feature value corresponding to a certain time point is in the threshold interval corresponding to "mild danger", then the time point is labeled as "mild danger". By comparing and labeling all elements in the cumulative deformation rate sequence, a preliminary safety level sequence is finally generated.

[0100] Step S430: Based on the preliminary safety level sequence, analyze the actual coverage of each threshold interval, calculate the density of data distribution in the threshold interval, and generate interval coverage statistical features. The density reflects the frequency of the safety level corresponding to the threshold interval.

[0101] The preliminary safety level sequence contains safety level information corresponding to each time point. By analyzing this information, the actual coverage of each threshold interval can be understood. The density of data distribution in the threshold interval is calculated, that is, the number of time points contained in each threshold interval is counted, so as to measure the frequency of the safety level corresponding to the threshold interval appearing in the entire monitoring time period. The interval coverage statistical features are the quantitative representation of the coverage and data density of each threshold interval. For example, in the analysis of the preliminary safety level sequence, the number of time points belonging to the threshold interval corresponding to the "safe" level is counted, as well as the number of time points belonging to the threshold interval corresponding to the "mild danger" level, etc. The number of time points in each threshold interval is compared with the total number of time points to obtain the data distribution proportion of each threshold interval, which constitutes the interval coverage statistical features.

[0102] Step S440: Adjust the initial warning threshold according to the interval coverage statistical features, expand the coverage range of the interval with a data distribution frequency higher than the average level, and narrow the coverage range of the interval with a data distribution frequency lower than the average level, to generate the adjusted warning threshold interval.

[0103] The interval coverage statistical features reflect the actual situation of data distribution in each threshold interval. According to these features, the initial early warning threshold is adjusted, and the purpose is to make the early warning threshold more accurately reflect the actual safety situation of tunnel deformation. For the threshold interval with data distribution frequency higher than the average level, it means that the safety level corresponding to the interval appears more frequently in actual monitoring, and expanding its coverage range can more reasonably cover more situations conforming to the safety level. For the interval with data distribution frequency lower than the average level, narrowing its coverage range can avoid unnecessary misjudgment and make the early warning more accurate. The adjusted early warning threshold interval is the new threshold interval obtained after such adjustment.

[0104] As an implementation, step S440 can specifically include steps S441-S446:

[0105] Step S441: Extract the interval data density in the interval coverage statistical features, the interval data density is the ratio of the number of mutation intervals contained in the interval to the interval width, and generate the interval density sequence.

[0106] The interval coverage statistical features contain detailed information of each threshold interval, and the interval data density is an important indicator. By calculating the ratio of the number of mutation intervals contained in the interval to the interval width, the data density of the interval can be obtained. Arranging the interval data density of each threshold interval in a certain order generates the interval density sequence. For example, in the analysis of tunnel deformation monitoring data, for each threshold interval, the number of mutation intervals contained therein is counted, and the corresponding interval data density is calculated in combination with the width of the interval. Then, the interval data densities of all threshold intervals are arranged in sequence to form the interval density sequence, which can intuitively reflect the difference in data density in different threshold intervals.

[0107] Step S442: Calculate the density deviation of each interval, and the density deviation is the ratio of the current interval density to the average density of all intervals. The ratio greater than 1 indicates that the interval data distribution is dense, and the ratio less than 1 indicates that the interval data distribution is sparse, and generate the density deviation sequence.

[0108] The average density of all intervals is a value obtained by summing the interval data densities of all threshold intervals and then dividing by the number of intervals. For each threshold interval, the ratio of its interval data density to the average density is calculated to obtain the density deviation of the interval. This ratio can intuitively reflect whether the data distribution of the interval is dense or sparse. The density deviation of each interval is arranged in a certain order to generate a density deviation sequence. For example, during the calculation process, if the density deviation of a certain threshold interval is greater than 1, it means that the data distribution of the interval is relatively dense, that is, the security level corresponding to the interval appears more frequently in the monitoring data; if the density deviation is less than 1, it means that the data distribution of the interval is relatively sparse, and the corresponding security level appears less frequently.

[0109] Step S443: Determine the adjustment direction and adjustment amplitude of the threshold interval based on the density deviation sequence. The interval with dense data distribution expands the boundary to both sides, and the interval with sparse data distribution shrinks the boundary to the center. The adjustment amplitude is positively correlated with the density deviation, and the boundary adjustment amount is generated.

[0110] The density deviation sequence reflects the density of the data distribution of each threshold interval. According to this sequence, the adjustment direction and amplitude of each threshold interval can be determined. For the interval with dense data distribution, it means that the security level corresponding to the interval appears more frequently in actual monitoring. In order to more comprehensively cover related situations, the boundary of the interval is expanded to both sides; for the interval with sparse data distribution, in order to avoid unnecessary misjudgment, its boundary is shrunk to the center. The adjustment amplitude is positively correlated with the density deviation, that is, the greater the density deviation, the greater the adjustment amplitude. The adjustment amount of each threshold interval boundary is calculated by such a rule to generate the boundary adjustment amount. For example, for the interval with dense data distribution and large density deviation, the amplitude of boundary expansion is relatively large; and for the interval with sparse data distribution and small density deviation, the amplitude of boundary shrinkage is relatively small.

[0111] Step S444: Modify the interval boundary of the initial early warning threshold according to the boundary adjustment amount, calculate the new upper and lower boundary values of the interval, so that the adjusted interval boundary does not exceed the value range of the cumulative deformation rate sequence.

[0112] The boundary adjustment amount determines the adjustment magnitude and direction of each threshold interval boundary. The interval boundaries of the initial warning threshold are modified according to these adjustment amounts, and the new upper and lower boundary values are obtained through corresponding calculations. During the modification process, it is necessary to ensure that the adjusted interval boundaries do not exceed the value range of the cumulative deformation rate sequence, so as to ensure that the adjusted warning threshold interval has practical significance. For example, for a certain threshold interval, the upper and lower boundaries are subjected to corresponding addition and subtraction operations according to the boundary adjustment amount, and new boundary values are obtained. At the same time, it is checked whether the new boundary values are between the minimum and maximum values of the cumulative deformation rate sequence. If they exceed the range, appropriate adjustments are made to make them meet the requirements.

[0113] Step S445: Conflict detection is performed on the modified interval boundaries. When the boundaries of adjacent intervals overlap or have gaps, boundary coordination processing is performed to make the interval boundaries continuous and non-overlapping, and the coordination principle is to preferentially retain the interval range with high data density.

[0114] After the interval boundaries of the initial warning threshold are modified, the boundaries of adjacent intervals may overlap or have gaps. Conflict detection on the modified interval boundaries is to find out these situations that do not meet the requirements. When it is found that the boundaries of adjacent intervals overlap or have gaps, boundary coordination processing is needed. The coordination principle is to preferentially retain the interval range with high data density, because the safety level corresponding to the interval with high data density appears more frequently in actual monitoring and can better reflect the actual situation of tunnel deformation. For example, if the boundaries of two adjacent threshold intervals overlap, the data densities of the two intervals are compared, the interval range with high data density is retained, and the boundary of the other interval is adjusted accordingly to make the interval boundaries continuous and non-overlapping.

[0115] Step S446: The coordinated interval boundaries are smoothed to eliminate the sudden jumps of the boundary values, make the boundary transition of adjacent intervals comply with the natural evolution law, and generate the adjusted warning threshold interval.

[0116] The interval boundaries after conflict detection and coordination processing may have sudden jumps, which do not comply with the natural evolution law and may lead to inaccurate warning. The smoothing processing of the coordinated interval boundaries is to eliminate these sudden jumps and make the boundary transition of adjacent intervals more natural and continuous. Smoothing algorithms such as moving average method and spline interpolation method can be used to process the boundary values. For example, using the moving average method, several data points near the boundary are averaged, and the average value is used to replace the original boundary value, so that the boundary transition is smoother. Through such smoothing processing, the adjusted warning threshold interval is generated, which can more accurately reflect the safety condition of tunnel deformation and provide reliable basis for subsequent safety evaluation and warning.

[0117] Step S450: match the accumulated deformation rate sequence with the adjusted early warning threshold intervals again to determine the final safety level corresponding to each mutation interval, and generate a time sequence safety level sequence.

[0118] After obtaining the adjusted early warning threshold intervals, each deformation rate feature value in the accumulated deformation rate sequence is matched with these new threshold intervals again. According to the matching result, the final safety level corresponding to each mutation interval is determined. These final safety levels are arranged in chronological order to generate a time sequence safety level sequence. For example, for each element in the accumulated deformation rate sequence, it is determined which adjusted early warning threshold interval it is in, thereby determining the final safety level of the mutation interval corresponding to the element. If a deformation rate feature value is in the threshold interval corresponding to "moderate danger" after adjustment, the final safety level of the mutation interval is marked as "moderate danger". Through the one-by-one matching and marking of all elements in the accumulated deformation rate sequence, a time sequence safety level sequence reflecting the change of the tunnel deformation safety level over time is generated.

[0119] Step S460: calculate the transition rule and persistence characteristic of different safety levels in the time sequence safety level sequence, the transition rule reflects the evolution path between levels, and the persistence characteristic reflects the stability of each level, and generate a safety level division result containing multiple early warning threshold intervals, level transition rules and persistence characteristics.

[0120] The time sequence safety level sequence records the change of the tunnel deformation safety level over time. Through analysis of the sequence, the transition rule between different safety levels and the persistence characteristic of each safety level can be calculated. The transition rule reflects the evolution path from one safety level to another, reflecting the dynamic change trend of the tunnel deformation safety situation; the persistence characteristic reflects the stability of each safety level in a period of time. The multi-level early warning threshold interval, level transition rule and persistence characteristic are integrated to generate the final safety level division result.

[0121] As an implementation manner, step S460 can specifically include steps S461-S466 as follows:

[0122] Step S461: perform level transition event extraction on the time sequence safety level sequence, identify the positions where the safety level changes at adjacent time points, record the level identifiers before and after the transition, generate a level transition event set, and each event in the event set contains a transition start time, an original level and a target level.

[0123] The time-series safety level sequence contains information on how the tunnel deformation safety level changes over time. Extracting level transition events from this sequence involves identifying the points where the safety level changes between adjacent time points. When a difference in safety level is found between adjacent time points, the start time of the transition, the original level before the transition, and the target level after the transition are recorded. This information is then organized into a level transition event set, where each event clearly records the specific details of a safety level transition. For example, in the time-series safety level sequence, if the safety level at a certain time point is "safe," and it changes to "slightly dangerous" at the next time point, the start time of the transition, the original level "safe," and the target level "slightly dangerous" are recorded, forming a level transition event. By checking and recording all adjacent time points in the time-series safety level sequence, a complete level transition event set is generated.

[0124] Step S462: Construct a level conversion probability matrix based on the level conversion event set. The matrix elements represent the probability of converting from one level to another. Calculate the conversion probability by the frequency of the conversion events and generate a level conversion probability distribution.

[0125] The safety level transition event set records the specific details of all safety level transitions. A safety level transition probability matrix is ​​constructed based on this event set, where each element represents the probability of transitioning from one safety level to another. The transition probability is calculated by statistically analyzing the frequency of each transition event, and these probabilities are then organized into a safety level transition probability distribution. For example, the probability of transitioning from "Safe" to "Slightly Hazardous" is obtained by counting the number of events that transition from "Safe" to "Slightly Hazardous" and then dividing by the total number of events that transition from "Safe". This calculation is performed for all possible safety level transitions to obtain the elements in the safety level transition probability matrix, thus generating the safety level transition probability distribution. This probability distribution intuitively reflects the likelihood of transitions between different safety levels, providing an important basis for predicting changes in tunnel deformation safety conditions.

[0126] Step S463: Calculate the duration sequence of each security level. The duration is the length of time during which the level appears consecutively. Divide the temporal security level sequence into multiple level duration segments through sequence segmentation. Each duration segment contains a level identifier and a duration value, generating a set of level duration segments.

[0127] The time sequence of security levels records the change of security levels over time. The duration sequence of each security level is calculated, i.e. the length of time that each security level continuously appears is counted. The time sequence of security levels is segmented into multiple level duration segments through sequence segmentation processing, each duration segment containing the identification of the level and the duration value. These level duration segments are sorted into a level duration segment set. For example, in the time sequence of security levels, if the "safe" level continuously appears for a period of time, the length of this period of time and the level identification "safe" are recorded to form a level duration segment. Through analysis and segmentation of the entire time sequence of security levels, a level duration segment set containing all security level duration conditions is generated.

[0128] Step S464: statistical analysis is performed on the level duration segment set to calculate the average duration and duration fluctuation range of each security level, the average duration reflecting the stability of the level and the fluctuation range reflecting the dispersion of the duration of the level, and a level duration characteristic parameter is generated.

[0129] The level duration segment set records the duration of each security level, and statistical analysis is performed on the set to calculate the average duration and duration fluctuation range of each security level. The average duration is the average duration of the security level in all duration segments, which can reflect the stability of the level, and the longer the average duration, the more stable the level; the duration fluctuation range reflects the dispersion of the duration of the security level, and the larger the fluctuation range, the more unstable the duration of the level. Through calculation and analysis of the duration data of each security level in the level duration segment set, a level duration characteristic parameter is generated.

[0130] Step S465: the level transition probability distribution and the level duration characteristic parameter are fused to construct a security level evolution model, the model containing the transition paths between levels and the stable time characteristics of each level, and a level evolution law description is generated.

[0131] The level transition probability distribution reflects the possibility of transition between different security levels, and the level duration characteristic parameter reflects the stability of each security level. The two are fused to construct a security level evolution model. The model comprehensively considers the transition paths between levels and the stable time characteristics of each level, and can comprehensively describe the evolution law of the security level of the tunnel deformation. Through integration and analysis of the level transition probability distribution and the level duration characteristic parameter, a level evolution law description is generated.

[0132] Step S466: the level evolution law description is combined with the adjusted early warning threshold interval to generate a security level division result containing multiple early warning threshold intervals, level transition probability distribution and level duration characteristic parameter.

[0133] The grade evolution law describes the dynamic change trend and conversion relationship of the deformation safety grade of the tunnel, and the adjusted early warning threshold interval clearly defines the cumulative deformation rate range corresponding to different safety grades. By combining the two, plus the previously calculated grade conversion probability distribution and grade duration characteristic parameters, the final safety grade classification result is generated. This result comprehensively contains the multi-level early warning threshold interval, the possibility of grade conversion, the stability of each safety grade, and other information, which can provide more accurate and detailed basis for the safety evaluation and early warning of the tunnel.

[0134] Step S500: Based on the safety grade classification result and the spatial distribution data of the tunnel construction section, the spatial positioning and boundary contour extraction of the potential risk damage zone are performed to obtain the risk damage zone distribution atlas of the tunnel construction section.

[0135] The safety grade classification result provides the safety grade information of different areas of the tunnel, and the spatial distribution data of the tunnel construction section describes the specific location and range of the tunnel construction section in space. Based on the two, the spatial positioning of the potential risk damage zone can be performed to determine its specific location in the tunnel construction section. At the same time, the boundary contour of the potential risk damage zone is extracted by a certain method to clearly define its range in space. Finally, these information are integrated to obtain the risk damage zone distribution atlas of the tunnel construction section. This atlas can intuitively show the location, range and corresponding safety grade of the potential risk damage zone in the tunnel construction section, and provide an important reference for the risk management and decision-making of the tunnel construction.

[0136] As an implementation manner, step S500 can specifically include steps S510-S560:

[0137] Step S510: Time and space correlation is performed between the multi-level early warning threshold interval in the safety grade classification result and the spatial distribution data of the tunnel construction section. Each spatial position unit matches the corresponding safety grade information according to the time stamp to generate a spatial data set with safety grade attribute. The data set contains spatial coordinates, safety grade identifier and early warning level.

[0138] The multi-level early warning threshold interval in the safety level division result defines the cumulative deformation rate range corresponding to different safety levels, and the spatial distribution data of the tunnel construction section records the specific coordinate information of each spatial location unit of the tunnel construction section. By associating the two in space and time, the corresponding safety level information is matched for each spatial location unit according to the time stamp. In this way, a spatial data set with safety level attributes is generated, which contains spatial coordinates, safety level identifiers, and early warning levels and other information. For example, at a certain time point, for a spatial location unit in the tunnel construction section, according to the matching of its cumulative deformation rate and the multi-level early warning threshold interval, the corresponding safety level identifier and early warning level are determined, and these information are associated with the spatial coordinates of the location unit. By performing such association processing on all spatial location units in the tunnel construction section, a complete spatial data set with safety level attributes is generated.

[0139] Step S520: Classify the spatial data set according to safety level identifiers, aggregate units with the same safety level and adjacent spatial positions into preliminary risk areas, and generate multiple groups of preliminary risk areas through area aggregation processing.

[0140] The spatial data set with safety level attributes contains safety level information of each spatial location unit. The data set is classified according to safety level identifiers, and spatial location units with the same safety level are classified into one category. Then, in the same category, units with adjacent spatial positions are aggregated to form preliminary risk areas. Through such area aggregation processing, multiple groups of preliminary risk areas are generated. These preliminary risk areas can preliminarily reflect the risk distribution in the tunnel construction section under different safety levels. For example, for spatial location units with a safety level of “high risk”, adjacent units are aggregated to form a preliminary risk area. All spatial location units of different safety levels are processed in this way to obtain multiple groups of preliminary risk areas corresponding to different safety levels.

[0141] As an implementation manner, step S520 can specifically include steps S521-S526:

[0142] Step S521: Extract safety level identifiers from the spatial data set, divide the data set into multiple level sub-data sets according to the identifier values, each level sub-data set contains spatial location units of the corresponding safety level, and generate level grouping data.

[0143] The spatial data set with security level attribute contains security level identification information of each spatial location unit. The security level identification is extracted from the data set, and the data set is divided into multiple level sub-data sets according to different identification values. Each level sub-data set only contains spatial location units of the corresponding security level, and the level grouping data is generated through such division. For example, the spatial location units with security level identification of "safe" in the spatial data set are extracted to form a level sub-data set, and the units with security level identification of "mild danger" form another level sub-data set.

[0144] Step S522: Spatial adjacency analysis is performed on each level grouping data to determine the spatial connection relationship between spatial location units, and a unit adjacency relationship table is generated. Adjacent units refer to units having common boundaries or common vertices in the spatial coordinate system.

[0145] Each level grouping data contains spatial location units of the same security level. Spatial adjacency analysis of these level grouping data is to determine the spatial connection relationship between each spatial location unit in them. In the spatial coordinate system, units having common boundaries or common vertices are defined as adjacent units. By analyzing and comparing the position information of spatial location units in each level grouping data, the adjacency relationship between them is determined, and these relationships are recorded to generate a unit adjacency relationship table. This table can clearly show the adjacency between spatial location units of the same security level, providing a basis for subsequent regional aggregation. For example, for a spatial location unit in a level grouping data, check whether it has common boundaries or common vertices with other units in space, and if so, record it as an adjacent relationship to finally form a unit adjacency relationship table.

[0146] Step S523: Based on the unit adjacency relationship table, the spatial location units in the level grouping data are regionally aggregated. Starting from any unmarked unit, units adjacent to it and having the same security level identification are merged into a region, and the merging process continues until no new unit can be added, generating an initial aggregated region.

[0147] The unit adjacency relation table records the adjacency relation between each space location unit under the same security level. Based on this table, the space location units in the level grouping data are regionally aggregated. Starting from any unmarked unit, the units adjacent to it and having the same security level identifier are found according to the unit adjacency relation table, and they are merged into a region. This process is repeatedly repeated, and the adjacent units of the newly merged units are also added to the region, until there is no new unit that meets the conditions to be added. In this way, the initial aggregated region is generated. For example, in the level grouping data, starting from an unmarked space location unit, the units adjacent to it and having the same security level identifier are found according to the unit adjacency relation table, and they are merged into a region. Then continue to find the units that meet the conditions in the periphery of the region, and continuously expand the range of the region, until no new unit can be added, forming an initial aggregated region. Each level grouping data is processed in this way to obtain multiple initial aggregated regions.

[0148] Step S524: Scale screening is performed on the initial aggregated region, the number of space location units contained in the region is calculated, and the region whose number does not reach the preset minimum scale is marked as an isolated region, and an isolated region marking set is generated.

[0149] The initial aggregated region may vary in size, and some regions may contain only a few space location units. These regions may be formed due to data errors or local special circumstances, and have less impact on the overall risk assessment. The scale screening is performed on the initial aggregated region, that is, the number of space location units contained in each region is calculated. The region whose number does not reach the preset minimum scale is marked as an isolated region, and the marking information of these isolated regions is sorted into an isolated region marking set. For example, the preset minimum scale is 5 space location units. For a certain initial aggregated region, if the number of units it contains is less than 5, it is marked as an isolated region. By performing such scale screening on all initial aggregated regions, a complete isolated region marking set is generated.

[0150] Step S525: The space location units corresponding to the isolated region marking set are removed from the level grouping data, and the remaining units are regionally aggregated again to generate a preliminary risk region containing a sufficient scale, and the number of preliminary risk regions is related to the number of levels of the level grouping data.

[0151] The isolated region marker set records the spatial location unit information corresponding to the initial aggregation region that does not reach the preset minimum size. The spatial location units corresponding to the isolated region marker set are removed from the hierarchical grouping data, and small regions that may affect the accuracy of the overall risk assessment are removed. Then, the remaining spatial location units are re-aggregated into regions according to the previous region aggregation method, and units adjacent to each other and having the same security level identifier are combined into a region. Through such re-aggregation, a preliminary risk region containing a sufficient size is generated. The number of preliminary risk regions is related to the number of levels of the hierarchical grouping data, and each security level may correspond to one or more preliminary risk regions. For example, for the hierarchical grouping data of each security level, after removing the units corresponding to the isolated regions, the region aggregation is re-performed to obtain the preliminary risk region under the security level. Finally, a plurality of preliminary risk regions corresponding to different security levels are obtained, which can more accurately reflect the risk distribution in the tunnel construction section.

[0152] Step S526: The generated preliminary risk region is uniquely identified, each region is assigned a unique region number, the security level identifier corresponding to the region and the spatial location unit coordinates contained are recorded, and a plurality of preliminary risk regions are generated, each group of regions contains complete spatial attribute information.

[0153] In order to facilitate the management and analysis of the preliminary risk region, it is necessary to uniquely identify the generated preliminary risk region. A unique region number is assigned to each preliminary risk region, and the security level identifier corresponding to the region and the spatial location unit coordinates contained are recorded. These information is sorted to generate a plurality of preliminary risk regions, each group of regions contains complete spatial attribute information. For example, a preliminary risk region is assigned a number "R1", and its security level identifier is recorded as "moderate danger", and the coordinates of all spatial location units contained in the region are recorded. All preliminary risk regions are identified and information recorded in this way to form a plurality of preliminary risk regions with complete spatial attribute information, which provides a basis for subsequent risk analysis and processing.

[0154] Step S530: Calculate the spatial form features of each group of preliminary risk regions, including the region extension range, the shape compactness degree and the spatial distribution center of gravity, the region extension range reflects the coverage size of the region in space, the shape compactness degree reflects the concentration degree of the region, and the spatial distribution center of gravity reflects the center position of the region, and generate a region form feature set.

[0155] Each initial risk area group possesses a specific spatial morphology. The spatial morphological characteristics of these areas are calculated, including their extent, compactness, and center of gravity. The extent is determined by measuring the maximum span of the area, reflecting its spatial coverage. Compactness is measured by calculating the relationship between the area's area and perimeter, reflecting its concentration; higher compactness indicates a more concentrated area. The center of gravity is calculated by weighted averaging of the coordinates of all spatial units within the area, reflecting its central location. These spatial morphological characteristics of each initial risk area group are then compiled to generate a set of regional morphological characteristics. For example, for a given initial risk area, its extent is determined by measuring its maximum distance in all directions, and its compactness is calculated (e.g., using the circularity ratio formula, which is the ratio of the area's area to the area of ​​a circle with the same perimeter. A circularity ratio closer to 1 indicates a more circular shape, meaning a higher concentration and a more compact form; a circularity ratio much less than 1 indicates a more dispersed shape). The center of gravity is then calculated using a weighted average of the coordinates. This calculation and organization is performed on all preliminary risk areas to obtain a complete set of regional morphological features.

[0156] Step S540: Based on the regional morphological feature set, select target risk areas with actual risk significance. When the area's extension range exceeds the preset scale condition and the compactness of its shape meets the preset concentration condition, retain the area and remove areas that do not meet the extension standard or the compactness standard.

[0157] The regional morphological feature set records the spatial morphological characteristics of each group of preliminary risk areas. Based on this feature set, target risk areas with actual risk significance are selected. Preset size and preset concentration conditions are pre-defined standards used to determine whether a preliminary risk area has actual risk significance. When the regional extension of a preliminary risk area exceeds the preset size condition and its morphological compactness meets the preset concentration condition, it indicates that the area has a certain spatial scale and is relatively concentrated, potentially posing a significant risk; therefore, the area is retained. Areas that do not meet the extension or compactness criteria are considered to have relatively low risk and are eliminated. For example, the preset size condition is a regional extension exceeding a certain set value, and the preset concentration condition is a morphological compactness exceeding a certain threshold. For a given preliminary risk area, if its regional extension is less than the preset size condition or its morphological compactness is less than the preset concentration condition, it is eliminated; if both conditions are met, it is retained as a target risk area. Through this selection process, target risk areas with actual risk significance are chosen from the preliminary risk areas.

[0158] Step S550: Boundary contour extraction is performed on the target risk area. Boundary point coordinates of the area are obtained through edge recognition processing. The boundary points are arranged in order of spatial position to form a closed boundary line, and a boundary contour coordinate sequence is generated.

[0159] After the target risk area is determined, its boundary contour needs to be extracted. The boundary point coordinates of the area are obtained through edge recognition processing. The edge recognition processing can use some image processing or spatial analysis methods, such as edge detection algorithms, to find the boundary points of the target risk area and the surrounding area. These boundary points are arranged in order of spatial position to form a closed boundary line. The coordinates of the points on the boundary line are recorded to generate a boundary contour coordinate sequence. This sequence can accurately describe the boundary shape and position of the target risk area. For example, for a target risk area, the edge detection algorithm is used to find its boundary points, which are then arranged in order of clockwise or counterclockwise direction to form a closed boundary line. The coordinates of the points on the boundary line are recorded to obtain the boundary contour coordinate sequence.

[0160] As an implementation, step S550 can specifically include steps S551-S556:

[0161] Step S551: Map the spatial position units of the target risk area to a two-dimensional plane grid. Each spatial position unit corresponds to a grid point in the grid, and the coordinates of the grid point correspond to the actual coordinates of the spatial position unit. A region grid mapping diagram is generated.

[0162] The target risk area is composed of multiple spatial position units. These spatial position units are mapped to a two-dimensional plane grid, and each spatial position unit is assigned a corresponding grid point in the grid. The coordinates of the grid point correspond to the actual coordinates of the spatial position unit, and a region grid mapping diagram is generated through such mapping. This mapping diagram can represent the spatial information of the target risk area in the form of a two-dimensional grid, facilitating subsequent edge recognition processing. For example, for each spatial position unit in the target risk area, find the corresponding position in the two-dimensional plane grid according to its actual coordinates, and mark it as a grid point. All spatial position units are mapped in this way to form a region grid mapping diagram.

[0163] Step S552: Edge enhancement is performed on the region grid mapping diagram to highlight the difference between the region boundary units and the internal units in the grid. The enhancement processing is achieved through comparison of the properties of adjacent units, and an edge-enhanced grid diagram is generated.

[0164] The regional grid map shows the distribution of the target risk area on a two-dimensional grid. In order to more clearly identify the boundaries of the region, edge enhancement processing is performed on the regional grid map. Edge enhancement processing is achieved by comparing the properties of adjacent grid cells, highlighting the differences between boundary cells and internal cells. For example, the security level identifier, spatial position relationship and other attributes of adjacent grid cells can be compared, and boundary cells can be specially marked or enhanced. Through such processing, an edge-enhanced grid map is generated, in which the boundaries of the region are more obvious, facilitating subsequent accurate identification of boundary points.

[0165] Step S553: Identify the boundary cells in the edge-enhanced grid map, the boundary cells are cells that have at least one adjacent cell that does not belong to the target risk area, the adjacent relationship is determined based on the adjacency rules of the spatial grid, and a set of boundary cells is generated.

[0166] In the edge-enhanced grid map, the boundary cells are cells that have at least one adjacent cell that does not belong to the target risk area, and the adjacent relationship is determined based on the adjacency rules of the spatial grid, such as in a two-dimensional grid, cells adjacent to the top, bottom, left and right are considered adjacent cells. By checking each grid cell in the edge-enhanced grid map, it is determined whether it meets the conditions of the boundary cell, and the cells that meet the conditions are filtered out to generate a set of boundary cells. For example, for a grid cell in the edge-enhanced grid map, check whether all its adjacent cells above, below, left and right belong to the target risk area, if there is one adjacent cell that does not belong to the target risk area, then the cell is a boundary cell, and it is added to the set of boundary cells. Through the checking and screening of all grid cells, a complete set of boundary cells is obtained.

[0167] Step S554: Extract the center coordinates of the boundary cell set, the center coordinates are the geometric center positions of the boundary cells, and generate an initial coordinate set of boundary points.

[0168] The boundary cell set records the boundary cell information of the target risk area. The center coordinates of these boundary cells are extracted, which refer to the geometric center positions of the boundary cells. By calculating the coordinate information of each boundary cell in the boundary cell set, the center coordinates are obtained, and these center coordinates are arranged into an initial coordinate set of boundary points. For example, for a boundary cell, the coordinates of the geometric center are calculated according to its coordinate range in the two-dimensional grid, and the coordinates are added to the initial coordinate set of boundary points.

[0169] Step S555: Spatially sort the initial coordinate set of boundary points, connect the boundary points in a clockwise or counterclockwise direction to form a closed polygon, the sorting is determined according to the polar angle relationship of the boundary points, the polar angle is calculated with the region barycenter as the pole, and an ordered boundary point sequence is generated.

[0170] The initial coordinate set of the boundary points contains the initial coordinate information of the boundary points of the target risk region. The coordinates are spatially sorted, and the boundary points are connected in a clockwise or counterclockwise direction to form a closed polygon. The sorting is determined according to the polar angle relationship of the boundary points, which is calculated with the region center as the pole. First, the region center of the target risk region is calculated, and then the polar angle of each boundary point is calculated with the center as the pole. The boundary points are sorted according to the size of the polar angle, and the sorted boundary points are connected in sequence to form an ordered boundary point sequence. For example, the polar angle of each boundary point relative to the pole is calculated with the region center as the pole, the boundary points are sorted in order from small to large according to the polar angle, and then the boundary points are connected in sequence to form a closed polygon, obtaining an ordered boundary point sequence.

[0171] Step S556: Redundant point removal is performed on the ordered boundary point sequence, key turning points are retained through a boundary smoothing algorithm, redundant points that do not significantly contribute to the description of the region shape are deleted, and a simplified boundary contour coordinate sequence is generated, with the distance between adjacent points in the sequence being kept within a preset range to ensure contour continuity.

[0172] The ordered boundary point sequence records the ordered coordinate information of the boundary points of the target risk region. Redundant point removal processing is performed on this sequence, and a boundary smoothing algorithm is used to determine which points are key turning points and which points are redundant points that do not significantly contribute to the description of the region shape. The key turning points are retained, the redundant points are deleted, and a simplified boundary contour coordinate sequence is generated. During processing, the distance between adjacent points in the sequence is kept within a preset range to ensure the continuity of the boundary contour. For example, some boundary smoothing algorithms such as the Douglas-Peucker algorithm are used to determine whether each point is a key turning point according to a preset threshold. Points that have less impact on the description of the region shape are deleted. At the same time, the distance between adjacent points is checked, and if the distance is too large, appropriate interpolation processing is performed to keep the distance between adjacent points within a preset range.

[0173] Step S560: The boundary contour coordinate sequence is fitted to a regular geometric contour line, the contour line is simplified through key feature point retention processing, the key turning points reflecting the region shape are retained, and a risk damage area distribution map containing the spatial position, boundary contour and safety level information of the risk region is generated. The spatial coordinate system of the distribution map is consistent with the spatial distribution data of the tunnel construction section.

[0174] The boundary contour coordinate sequence records the boundary point coordinate information of the target risk area. The coordinate sequence is fitted into a regular geometric contour line, for example, a circle, an ellipse, a polygon or the like. In the fitting process, the contour line is simplified through key feature point reservation processing, and only key turning points that can reflect the region shape are reserved, and some unnecessary details are removed. Based on this, the contour line can be made more concise and clear, while accurately describing the general shape of the region. The spatial position information of the risk area, the simplified boundary contour information and the corresponding safety level information are integrated to generate a risk damage area distribution map. The spatial coordinate system of the distribution map is consistent with the spatial distribution data of the tunnel construction section, which is convenient for association and analysis with the overall information of the tunnel construction section. For example, the boundary contour coordinate sequence is fitted into a polygon using a least square method or the like, key turning points are reserved by analyzing the curvature and other characteristics of the boundary points, and the contour line is simplified. Then, the spatial position of the risk area, the simplified boundary contour and the safety level information are sorted into a map, which can intuitively show the distribution of the risk damage area in the tunnel construction section, and provide a visual basis for risk management of the tunnel construction.

[0175] Please refer to Figure 2 , Figure 2 A structural schematic diagram of a computer system provided by an embodiment of the present application is provided, which at least includes a processor 101, a communication interface 102 and a memory 103. The processor 101, the communication interface 102 and the memory 103 can be connected through a bus or other means. The processor 101 (or a central processing unit (CPU)) is the calculation core and control core of the computer system, which can parse various instructions in the computer system and process various data of the computer system. The communication interface 102 can optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and can be used for transmitting and receiving data under the control of the processor 101; the communication interface 102 can also be used for transmitting and interacting data within the computer system. The memory 103 is a memory device in the computer system, which is used for storing programs and data. It can be understood that the memory 103 herein can include a built-in memory of the computer system, and of course can also include an extended memory supported by the computer system. The memory 103 provides a storage space, which stores an operating system of the computer system, and the present application does not limit this.

[0176] In one embodiment, the processor 101 executes the computer program in the memory 103 to perform the time-space evolution data processing method applied to tunnel deformation analysis provided by the above embodiments of the present application.

Claims

1. A spatiotemporal evolution data processing method applied to tunnel deformation analysis, characterized in that, The method includes: The time-series monitoring sequence of different monitoring dimensions during tunnel construction is obtained. The time-series monitoring sequence includes tunnel structure response data and geological environment effect data continuously recorded at preset sampling intervals. Among them, the tunnel structure response data refers to the response information of the tunnel structure under various forces and environmental effects, including displacement data, stress data, and strain data of the tunnel lining, which reflects the changes in the mechanical state of the tunnel structure itself. The geological environment effect data is data related to the geological environment in which the tunnel is located, including groundwater level change information, physical and mechanical property information of rock and soil, and seismic activity information, which reflects the role and influence of the geological environment on the tunnel structure. Based on the time-series monitoring sequence, a spatiotemporal evolution correlation model of tunnel deformation is constructed. By analyzing the dynamic coupling relationship between structural response data and geological environmental impact data, a spatiotemporal correlation feature matrix containing spatiotemporal coupling effect parameters is generated. Specifically, this includes: coordinating the structural response data and geological environmental impact data in the time-series monitoring sequence along the time axis, adjusting data from different sampling intervals to a unified time reference to obtain time-synchronized monitoring data pairs. Each time point in the monitoring data pair contains corresponding structural response and environmental impact components. The monitoring data pairs are divided into training samples and validation samples according to data distribution characteristics. Training samples are used for model parameter learning, and validation samples are used for model performance evaluation. A combined network architecture is constructed, including a time-dependent feature extraction network and a spatial correlation feature extraction network. The time-dependent feature extraction network is used to capture the temporal evolution law of structural response data, and the spatial correlation feature extraction network is used for... The modeling method identifies the spatial distribution characteristics of geological environmental impact data, matching the network input layer dimension with the feature dimensions of the monitoring data pairs. Structural response components from the training samples are input into a time-dependent feature extraction network, generating a structural response time-series feature vector through forward and backward temporal feature fusion. Environmental impact components are input into a spatial correlation feature extraction network, generating an environmental impact spatial feature vector through spatial topological modeling. The network's feature interaction layer models the correlation between the structural response time-series feature vector and the environmental impact spatial feature vector, calculating their interaction strength at different time scales to generate a coupled feature tensor containing spatiotemporal correlation. The prediction accuracy of the coupled feature tensor is evaluated based on validation samples, and the network weight parameters are adjusted iteratively until the prediction accuracy meets preset requirements, generating a spatiotemporal correlation feature matrix containing spatiotemporal coupling effect parameters. These spatiotemporal coupling effect parameters are used to quantify the influence relationships between different monitoring dimensions. Based on the spatiotemporal correlation feature matrix, a feature evolution sequence of tunnel deformation is generated, and nonlinear mutation characteristics are extracted from the feature evolution sequence to obtain a set of mutation feature indicators characterizing the deformation trend. Based on the set of mutation feature indicators and the preset cumulative deformation rate threshold, the tunnel deformation safety level is dynamically divided, and a safety level division result containing multiple warning threshold intervals is generated. Based on the safety level classification results and the spatial distribution data of the tunnel construction section, the spatial positioning and boundary contour extraction of the potential risk damage area are performed to obtain the risk damage area distribution map of the tunnel construction section.

2. The method according to claim 1, characterized in that, The network's feature interaction layer models the correlation between the structural response temporal feature vector and the environmental action spatial feature vector, calculates the interaction strength between the two at different time scales, and generates a coupled feature tensor containing spatiotemporal correlation, including: The structural response time-series feature vector and the environmental action space feature vector are divided into feature segment groups according to time windows. Each segment group contains structural feature segments and environmental feature segments at consecutive time points. The window length is determined according to the periodicity of the monitoring data, and multiple sets of time-series feature segments are generated. For each set of feature segments, feature discretization is performed, converting continuous feature values ​​into interval identifiers. The number of interval divisions is adjusted according to the distribution range of feature values, generating a discretized feature matrix, where each element represents the interval belonging to the corresponding feature dimension. Based on the discretized feature matrix, a feature combination transaction set is constructed, where each transaction corresponds to a feature combination relationship within a time window, and the transaction item is a joint identifier of feature dimension and interval identifier, thus generating a feature transaction dataset. By recognizing co-occurrence patterns, the feature transaction dataset is analyzed to mine association patterns, and feature combinations that appear repeatedly in different time windows are filtered out. The frequency of occurrence of the combination and the probability of occurrence of the condition are calculated to generate a set of candidate association rules. The candidate association rule set is filtered, and rules whose occurrence frequency and condition occurrence probability both meet preset conditions are retained. False association rules are removed through association strength evaluation, and a preliminary association rule set is generated. The initial association rule set is arranged in chronological order, and rules that are sequential in time and have consistent feature combinations are merged to generate an association rule set containing time evolution characteristics. The association rule set is used to guide the association modeling process of the feature interaction layer.

3. The method according to claim 2, characterized in that, The step of modeling the correlation between the temporal feature vector of the structural response and the spatial feature vector of the environmental interaction through the feature interaction layer of the network, calculating the interaction strength between the two at different time scales, and generating a coupled feature tensor containing spatiotemporal correlation, further includes: Based on the association rule set, the association dimensions between structural response features and environmental action features are determined, a feature dimension mapping relationship table is generated, the association types and influence directions between different feature dimensions are recorded, and a feature association dimension index is generated. The frequency of occurrence and the probability of occurrence of conditions in the association rule set are weighted and combined according to a preset ratio to obtain the rule influence strength value of each association rule. The rule influence strength values ​​of all association rules are arranged in order to obtain the rule importance sequence. The weight of the frequency of occurrence is adjusted according to the uniformity of the distribution of association rules in different time windows, and the weight of the probability of occurrence of conditions is set according to the differences in the field source types contained in the antecedent of the association rule. Based on the importance sequence of the rules, interaction weights are assigned to different association dimensions. The weight values ​​are normalized and mapped to a uniform numerical range to generate a feature interaction weight matrix. The structural response time-series feature vector and the environmental action space feature vector are weighted and combined according to the feature interaction weight matrix to calculate the feature interaction values ​​at different time scales and generate a multi-scale interaction feature matrix. The matrix dimension matches the number of time scales of the feature vector. The multi-scale interaction feature matrix is ​​fused, and the interaction features at different time scales are integrated through cross-scale feature correlation analysis to generate a coupled feature tensor. Based on the aforementioned coupling feature tensor, spatiotemporal coupling effect parameters are calculated. The parameter values ​​reflect the interaction strength of different monitoring dimensions at a set spatiotemporal location. The parameter values ​​are arranged according to time and space dimensions to generate the core parameter items of the spatiotemporal correlation feature matrix.

4. The method according to claim 1, characterized in that, The process involves generating a feature evolution sequence of tunnel deformation based on the spatiotemporal correlation feature matrix, extracting nonlinear mutation characteristics from the feature evolution sequence, and obtaining a set of mutation feature indicators characterizing the deformation trend, including: The spatiotemporal correlation feature matrix is ​​subjected to dominant evolution feature extraction to obtain a low-dimensional dominant feature matrix, where the row dimension of the matrix corresponds to the time axis and the column dimension corresponds to the dominant feature dimension. The low-dimensional dominant feature matrix is ​​arranged in chronological order to construct a tunnel deformation feature evolution sequence indexed by time. Each time point in the sequence contains multi-dimensional deformation feature values ​​corresponding to that time, and the feature dimension is consistent with the column dimension of the low-dimensional dominant feature matrix. The feature evolution sequence is processed by a sliding window to obtain multiple sequence segments with temporal overlap. Each sequence segment contains deformation feature values ​​at consecutive time points. The window size is determined according to the temporal resolution of the sequence. Calculate the nonlinear dynamic characteristic parameters of each sequence segment, including trajectory divergence characteristic parameters and trajectory complexity characteristic parameters, and generate a segment dynamic parameter set. The trajectory divergence characteristic parameters reflect the degree of divergence of the sequence segment over time, and the trajectory complexity characteristic parameters reflect the spatial distribution complexity of the sequence segment. Based on the fragment dynamic parameter set, nonlinear mutation locations in the feature evolution sequence are identified, the degree of parameter change of adjacent sequence fragments is calculated, and when the degree of parameter change exceeds the preset judgment condition, it is marked as a potential mutation point. Through mutation point clustering processing, potential mutation points that are close in time are merged into mutation intervals. The interval features of the mutation interval are extracted, including the interval start time, interval end time, parameter change amplitude, and statistical features of dynamic parameters within the interval. After standardization, the feature values ​​are arranged in chronological order to generate a set of mutation feature indicators that characterize the deformation trend. Each indicator in the set of mutation feature indicators contains the time attribute and dynamic characteristic parameters of the mutation interval.

5. The method according to claim 4, characterized in that, The sliding window processing of the feature evolution sequence yields multiple temporally overlapping sequence segments, each containing deformation feature values ​​at consecutive time points. The window size is determined based on the temporal resolution of the sequence, including: Analyze the temporal resolution characteristics of the feature evolution sequence, determine the time interval between adjacent time points in the sequence, the time interval reflects the density of data collection, and generate sequence time interval parameters; The initial size of the sliding window is set based on the sequence time interval parameter, and the initial size is an integer multiple of the time interval, and the initial window parameter is generated. The initial window parameters are adjusted, and the window size is corrected according to the frequency of feature value changes in the feature evolution sequence to generate dynamic window parameters; The feature evolution sequence is slide-segmented according to the dynamic window parameters, and the window movement step size is set to a fixed proportion of the window size to generate a set of sequence segments so that the temporal overlap between adjacent windows is maintained. The overlapping part is used to maintain the temporal continuity of the sequence segments. Each sequence segment is smoothed at its boundaries. By processing the transition of edge feature values, feature abrupt changes at the window boundaries are eliminated, so that the feature value changes within the segment remain continuous, thus generating a smooth sequence segment.

6. The method according to claim 4, characterized in that, The process involves identifying nonlinear mutation locations in the feature evolution sequence based on the fragment dynamic parameter set, calculating the degree of parameter change in adjacent sequence fragments, and marking potential mutation points when the degree of parameter change exceeds a preset judgment condition. Then, a mutation point clustering process is used to merge temporally adjacent potential mutation points into mutation intervals, including: Calculate the rate of change of each parameter value in the fragment dynamics parameter set to generate a parameter rate of change sequence; Obtain the preset threshold for the rate of change of parameters. The threshold is dynamically determined based on the overall distribution characteristics of the parameter rate of change sequence, and a mutation threshold is generated. Mark the parameter change points in the parameter change rate sequence that exceed the mutation determination threshold as potential mutation points, record the timestamp of each potential mutation point and the corresponding parameter change rate value, and generate a set of potential mutation points. Temporal proximity analysis is performed on the set of potential mutation points to calculate the time interval between adjacent potential mutation points. When the time interval is less than a preset merging condition, the mutation points are determined to be temporally adjacent mutation points, and a mutation point neighbor group is generated. By clustering mutation points, potential mutation points in neighboring groups of the mutation point are merged into mutation intervals. The start time of the interval is the timestamp of the earliest mutation point in the group, and the end time of the interval is the timestamp of the latest mutation point in the group, thus generating preliminary mutation intervals. The initial mutation intervals are validated by calculating the eigenvalue fluctuation range of the feature evolution sequence within the interval. Intervals whose fluctuation range does not meet the preset significance condition are marked as noise intervals and removed, thus generating the final mutation interval set.

7. The method according to claim 1, characterized in that, The dynamic classification of tunnel deformation safety levels based on the set of mutation feature indicators and a preset cumulative deformation rate threshold generates a safety level classification result containing multiple warning threshold ranges, including: The parameter change amplitude of each mutation interval is extracted from the set of mutation feature indicators and arranged in chronological order to generate a cumulative deformation rate sequence, wherein each element corresponds to the deformation rate feature value of the mutation interval, and the feature value reflects the development speed of deformation within the interval. The cumulative deformation rate sequence is compared with a preset initial warning threshold, which contains multiple consecutive threshold intervals, each interval corresponding to a different safety level, to generate a preliminary safety level sequence, wherein each time point is marked with a corresponding safety level label. Based on the preliminary security level sequence, the actual coverage of each threshold interval is analyzed, the density of data distribution within the threshold interval is calculated, and interval coverage statistical characteristics are generated. The density reflects the frequency of occurrence of the security level corresponding to the threshold interval. The initial warning threshold is adjusted based on the interval coverage statistical characteristics. The coverage range is expanded for intervals with data distribution frequency higher than the average level, and the coverage range is reduced for intervals with data distribution frequency lower than the average level, thus generating an adjusted warning threshold interval. The cumulative deformation rate sequence is rematched with the adjusted warning threshold interval to determine the final safety level corresponding to each mutation interval, and a time-series safety level sequence is generated. The transition patterns and persistence characteristics of different security levels in the time-series security level sequence are calculated. The transition patterns reflect the evolution path between levels, and the persistence characteristics reflect the stability of each level. A security level classification result containing multi-level warning threshold ranges, level transition patterns, and persistence characteristics is generated.

8. The method according to claim 7, characterized in that, The step of adjusting the initial warning threshold based on the interval coverage statistical characteristics, expanding the coverage range for intervals with dense data distribution and narrowing the coverage range for intervals with sparse data distribution, and generating the adjusted warning threshold interval through a threshold boundary adaptive correction algorithm includes: Extract the interval data density from the interval coverage statistical features. The interval data density is the ratio of the number of abrupt intervals contained within an interval to the interval width, and generate an interval density sequence. Calculate the density deviation for each interval. The density deviation is the ratio of the density of the current interval to the average density of all intervals. A ratio greater than 1 indicates that the data distribution in the interval is dense, and a ratio less than 1 indicates that the data distribution in the interval is sparse. Generate a density deviation sequence. Based on the density deviation sequence, the adjustment direction and adjustment magnitude of the threshold interval are determined. The interval with dense data distribution expands its boundaries to both sides, and the interval with sparse data distribution shrinks its boundaries towards the center. The adjustment magnitude is positively correlated with the density deviation, and the boundary adjustment amount is generated. The interval boundary of the initial warning threshold is corrected according to the boundary adjustment amount, and new upper and lower boundary values ​​of the interval are calculated so that the adjusted interval boundary does not exceed the value range of the cumulative deformation rate sequence. Conflict detection is performed on the corrected interval boundaries. When the boundaries of adjacent intervals overlap or have gaps, boundary coordination is used to make the interval boundaries continuous and non-overlapping. The boundaries of the coordinated intervals are smoothed to eliminate abrupt changes in boundary values, ensuring that the transition between adjacent intervals conforms to the laws of natural evolution, thus generating the adjusted warning threshold intervals.

9. A computer system, characterized in that, include: A memory, wherein a computer program is stored; A processor is configured to load the computer program to implement the spatiotemporal evolution data processing method for tunnel deformation analysis as described in any one of claims 1-8.

Images