Shallow-buried soft stratum subsurface tunnel settlement prediction method and system

Abstract

The invention discloses a shallow-buried soft stratum subsurface tunnel settlement prediction method and system, and relates to the technical field of tunnel settlement monitoring, the shallow-buried soft stratum subsurface tunnel settlement prediction method comprises the following steps: S1, obtaining real-time monitoring data of a current construction stage of a target tunnel, carrying out multi-dimensional feature construction according to the real-time monitoring data, and carrying out multi-dimensional feature construction according to the real-time monitoring data; obtaining a structured multi-dimensional feature data set; s2, preprocessing the structured multi-dimensional feature data set to obtain standard input data; s3, inputting the standard input data into a pre-constructed shallow-buried soft-texture stratum subsurface tunnel settlement prediction model, and outputting a single-step prediction value by the shallow-buried soft-texture stratum subsurface tunnel settlement prediction model; and S4, correcting the single-step predicted value in real time by using the dynamic prediction error to generate a settlement predicted value. According to the method, the accuracy of subsurface tunnel settlement prediction of the shallow-buried soft stratum can be improved.

Description

Technical Field

[0001] This application relates to the technical field of tunnel settlement monitoring, and particularly to a method and system for predicting the settlement of shallow-buried soft ground tunneling. Background Art

[0002] In the development of urban underground space and the construction of transportation infrastructure, the number of shallow-buried soft ground tunneling projects is increasing day by day. The construction of such tunnels faces complex geological conditions and construction environments, and settlement control is the key to ensuring project safety and the stability of the surrounding environment. However, the existing methods and systems for predicting the settlement of shallow-buried soft ground tunneling have the following problems: (1) The existing methods and systems for predicting the settlement of shallow-buried soft ground tunneling only rely on a single type of monitoring data (such as: only surface settlement or crown settlement), without integrating multi-dimensional data such as spatio-temporal distribution data, geological parameters (such as: water content and void ratio, etc.), construction process parameters (such as: excavation footage and grouting pressure), and environmental impact data, etc. It is difficult to comprehensively reflect the multi-factor coupling effect of the settlement of soft ground tunneling, and it is one-sided.

[0003] (2) Since the monitoring data of soft ground is easily disturbed, there are problems such as noise, missing values, and dimensional differences in its original data. However, the existing methods and systems for predicting the settlement of shallow-buried soft ground tunneling lack targeted processing for outlier detection (such as: monitoring jump points caused by sudden construction disturbances) and missing value filling (such as: sensor failure data), and the normalization method is simple (such as: only using linear scaling to achieve normalization). Therefore, it will lead to uneven quality of the data set input into the prediction model, thus affecting the prediction accuracy.

[0004] (3) The existing methods and systems for predicting the settlement of shallow-buried soft ground tunneling usually adopt an offline prediction method to obtain prediction values, and cannot dynamically adjust the prediction results according to real-time monitoring data, making it difficult to meet the low-latency requirements of tunnel construction real-time monitoring. When facing problems such as time-varying formation parameters and monitoring noise fluctuations, it is difficult to ensure the prediction accuracy.

[0005] Therefore, there is an urgent need to provide a brand-new method and system for predicting the settlement of shallow-buried soft ground tunneling to solve the above problems. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for predicting the settlement of shallow-buried soft ground tunneling, which can improve the accuracy of predicting the settlement of shallow-buried soft ground tunneling.

[0007] To achieve the above object, the present application provides a method for predicting the settlement of a shallow-buried soft ground tunnel by mined tunneling, including the following steps: S1: Obtain the real-time monitoring data of the target tunnel at the current construction stage, and construct multi-dimensional features based on the real-time monitoring data to obtain a structured multi-dimensional feature data set. Among them, the expression of the structured multi-dimensional feature data set is: , is a multi-dimensional feature vector, is a label vector; the real-time monitoring data at least includes: real-time settlement and displacement data, real-time construction technology data, real-time geological data, and real-time environmental data; S2: Preprocess the structured multi-dimensional feature data set to obtain standard input data; S3: Input the standard input data into a pre-constructed shallow-buried soft ground tunnel settlement prediction model, and the shallow-buried soft ground tunnel settlement prediction model outputs a single-step prediction value. Among them, the single-step prediction value at least includes: a single-step prediction value of the settlement of the tunnel crown, a single-step prediction value of the ground settlement, and a single-step prediction value of the convergence displacement; S4: Use the dynamic prediction error to correct the single-step prediction value in real time to generate a settlement prediction value. Among them, the settlement prediction value at least includes: a settlement prediction value of the tunnel crown, a settlement prediction value of the ground surface, and a convergence displacement prediction value.

[0008] As above, among them, it further includes S5: Analyze the settlement prediction value using a preset settlement safety threshold. If the settlement prediction value is greater than the settlement safety threshold, trigger an early warning mechanism and output a settlement early warning message. Among them, the settlement early warning message at least includes: the current settlement risk level; if the settlement prediction value is less than or equal to the settlement safety threshold, end.

[0009] As above, among them, the sub-steps of preprocessing the structured multi-dimensional feature data set to obtain standard input data are as follows: S21: Perform anomaly detection on the structured multi-dimensional feature data set to generate an anomaly detection result. If the anomaly detection result is abnormal, execute S22; if the anomaly detection result is normal, use the structured multi-dimensional feature data set as the data to be filled and execute S23; S22: Repair the structured multi-dimensional feature data set to obtain the data to be filled and execute S23; S23: Analyze the data to be filled to generate a missing result. If the missing result is missing, execute S24; if the missing result is not missing, use the data to be filled as the data to be normalized and execute S25; S24: Fill in the missing values of the data to be filled to obtain the data to be normalized and execute S25; S25: Perform normalization processing on the data to be normalized to obtain standard input data.

[0010] As described above, the sub-steps for repairing the structured multi-dimensional feature data set to obtain the data to be filled are as follows: S221: Read the time series abnormal data in the structured multi-dimensional feature data set. If there is time series abnormal data, repair the time series abnormal data to obtain the repaired data, and execute S222; if there is no time series abnormal data, execute S222; S222: Read the repaired data or the structured multi-dimensional feature data set. If there is non-time series abnormal data in the repaired data or the structured multi-dimensional feature data set, repair the non-time series abnormal data to obtain the data to be filled; if there is no non-time series abnormal data in the repaired data, use the repaired data as the data to be filled.

[0011] As described above, the sub-steps for repairing the time series abnormal data to obtain the repaired data are as follows: S2211: Generate a first repair sequence for each sub-time series abnormal data in the time series abnormal data according to the reading order, and the first repair sequence increases sequentially from the front to the back according to the reading order; S2212: Use the sub-time series abnormal data with the smallest first repair sequence as the first current repair data, repair the first current repair data to obtain the first sub-repaired data, and execute S2213; S2213: Judge the first repair sequence of the first current repair data by using the total number of sub-time series abnormal data. If the total number of sub-time series abnormal data is greater than the first repair sequence of the first current repair data, after removing the first repair sequence of the first current repair data, execute S2212; if the total number of sub-time series abnormal data is equal to the first repair sequence of the first current repair data, remove the first repair sequence of the first current repair data, and use all the first sub-repaired data as the repaired data, and execute S222.

[0012] As described above, in which, the sub-steps of repairing the first current repair data to obtain the first sub-repaired data are as follows: R1: Taking the acquisition time of the first current repair data as the starting time, shifting forward by an interval time to obtain the previous trend interval, and taking the acquisition time of the first current repair data as the starting time, shifting backward by an interval time to obtain the subsequent trend interval; R2: Obtaining the total number of valid sub-time series data in the previous trend interval and the total number of valid sub-time series data in the subsequent trend interval, and comparing the total number of valid sub-time series data in the previous trend interval with the total number of valid sub-time series data in the subsequent trend interval. If the total number of valid sub-time series data in the previous trend interval is greater than the total number of valid sub-time series data in the subsequent trend interval, then all the valid sub-time series data in the previous trend interval are used as the first repair reference data; if the total number of valid sub-time series data in the previous trend interval is less than the total number of valid sub-time series data in the subsequent trend interval, then all the valid sub-time series data in the subsequent trend interval are used as the first repair reference data; R3: Analyzing and calculating the first repair reference data to obtain the first sub-repaired data.

[0013] As described above, in which, the expression of the dynamic prediction error is: ; Wherein, is the dynamic prediction error at the th moment; is the dynamic prediction error at the th moment; is the process noise, is the observation noise; is the process noise weight; is the observation noise weight.

[0014] As described above, in which, the expression of the settlement prediction value is: ; Wherein, is the settlement prediction value; is the single-step prediction value at the th moment; is the dynamic prediction error at the th moment.

[0015] This application also provides a settlement prediction system for a shallow-buried soft soil stratum mined tunnel, including: a settlement prediction center and at least one user terminal; User terminal: Receiving the settlement prediction value; Receiving the settlement warning information; Settlement prediction center: Used to execute the above-mentioned settlement prediction method for a shallow-buried soft soil stratum mined tunnel.

[0016] As described above, the settlement prediction center includes: a data acquisition unit, a data processing unit, a settlement prediction unit, a dynamic correction unit, and an early warning unit; among them, the data acquisition unit: is used to obtain the real-time monitoring data of the target tunnel in the current construction stage; the data processing unit: is used to construct multi-dimensional features based on the real-time monitoring data to obtain a structured multi-dimensional feature data set; preprocess the structured multi-dimensional feature data set to obtain standard input data; the settlement prediction unit: is provided with a pre-constructed settlement prediction model for shallow-buried soft ground tunneling. Input the standard input data into the pre-constructed settlement prediction model for shallow-buried soft ground tunneling, and the single-step prediction value is output by the settlement prediction model for shallow-buried soft ground tunneling; the dynamic correction unit: uses the dynamic prediction error to perform real-time correction on the single-step prediction value to generate a settlement prediction value; the early warning unit: uses the preset settlement safety threshold to analyze the settlement prediction value. If the settlement prediction value is greater than the settlement safety threshold, trigger the early warning mechanism and output the settlement early warning information; if the settlement prediction value is less than or equal to the settlement safety threshold, end.

[0017] The beneficial effects achieved by this application are as follows: (1) The settlement prediction method and system for shallow-buried soft ground tunneling in this application integrate spatio-temporal distribution data, geological parameters (such as water content and void ratio, etc.), construction process parameters (such as excavation footage and grouting pressure), and environmental impact data, and can comprehensively reflect the multi-factor coupling effect of the settlement of shallow-buried soft ground tunneling, solving the problem of single data in traditional methods.

[0018] (2) The settlement prediction method and system for shallow-buried soft ground tunneling in this application have targeted processing for outlier detection (such as monitoring jump points caused by sudden construction disturbances) and missing value filling (such as sensor failure data), and adopt a variety of normalization methods. Therefore, it can improve the quality of the data set input into the settlement prediction model for shallow-buried soft ground tunneling, thereby improving the prediction accuracy.

[0019] (3) The settlement prediction method and system for shallow-buried soft ground tunneling in this application can dynamically adjust the prediction results according to real-time monitoring data, can meet the low-latency requirements of real-time monitoring of tunnel construction, and can ensure the prediction accuracy when facing problems such as time-varying formation parameters and monitoring noise fluctuations.

[0020] (4) The settlement prediction method and system for shallow-buried soft ground tunneling in this application can realize the hierarchical early warning of settlement risks, support the adjustment of construction plans, ensure project safety, and at the same time, automated data processing and real-time prediction reduce the manual analysis cost and improve the accuracy and engineering application value of the settlement prediction of shallow-buried soft ground tunneling. Description of the Drawings

[0021] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only some embodiments recorded in the present application. For those of ordinary skill in the art, other accompanying drawings can also be obtained based on these drawings.

[0022] Figure 1 It is a schematic structural diagram of an embodiment of a settlement prediction system for a shallow-buried soft ground tunneling. Figure 2 It is a flowchart of an embodiment of a settlement prediction method for a shallow-buried soft ground tunneling. Detailed implementation manners

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0024] As Figure 1 shown, the present application provides a settlement prediction system for a shallow-buried soft ground tunneling, including: a settlement prediction center 1 and at least one user terminal 2.

[0025] User terminal 2: used to receive the settlement prediction value; receive the settlement warning information.

[0026] Settlement prediction center 1: used to execute the following settlement prediction method for a shallow-buried soft ground tunneling.

[0027] Furthermore, the settlement prediction center 1 includes: a data acquisition unit, a data processing unit, a settlement prediction unit, a dynamic correction unit, and a warning unit.

[0028] Among them, the data acquisition unit: used to obtain the real-time monitoring data of the target tunnel at the current construction stage.

[0029] The data processing unit: used to construct multi-dimensional features based on the real-time monitoring data to obtain a structured multi-dimensional feature dataset; preprocess the structured multi-dimensional feature dataset to obtain standard input data.

[0030] The settlement prediction unit: is provided with a pre-constructed settlement prediction model for a shallow-buried soft ground tunneling. The standard input data is input into the pre-constructed settlement prediction model for a shallow-buried soft ground tunneling, and a single-step prediction value is output by the settlement prediction model for a shallow-buried soft ground tunneling.

[0031] Dynamic correction unit: It uses the dynamic prediction error to correct the single-step prediction value in real time and generate the settlement prediction value.

[0032] Early warning unit: It analyzes the settlement prediction value by using the preset settlement safety threshold. If the settlement prediction value is greater than the settlement safety threshold, it triggers the early warning mechanism and outputs the settlement early warning information; if the settlement prediction value is less than or equal to the settlement safety threshold, it ends.

[0033] As Figure 2 shown, this application provides a method for predicting the settlement of a mined tunnel in a shallow buried soft ground stratum, including the following steps: S1: Obtain the real-time monitoring data of the current construction stage of the target tunnel, construct multi-dimensional features based on the real-time monitoring data, and obtain a structured multi-dimensional feature data set. Among them, the expression of the structured multi-dimensional feature data set is: , is the multi-dimensional feature vector, is the label vector; the real-time monitoring data at least includes: real-time settlement and displacement data, real-time construction technology data, real-time geological data, and real-time environmental data.

[0034] Specifically, the data acquisition unit collects data on the current construction stage of the target tunnel to obtain real-time monitoring data, and sends the real-time monitoring data to the data processing unit. The target tunnel is a mined tunnel in a shallow buried soft ground stratum, but is not limited to a mined tunnel in a shallow buried soft ground stratum.

[0035] The real-time monitoring data at least includes: real-time settlement and displacement data, real-time construction technology data, real-time geological data, and real-time environmental data.

[0036] Among them, the real-time settlement and displacement data are spatio-temporal data collected by sensors (such as: level, total station, and automated monitoring system), including: physical quantities and real-time spatio-temporal information. Among them, the physical quantities include: real-time monitoring values of the tunnel crown settlement, real-time surface settlement, and convergence displacement in the current construction stage; the real-time spatio-temporal information includes: three-dimensional coordinates of the real-time monitoring points and real-time time stamps (i.e., collection time).

[0037] Specifically, various physical quantity data (such as: stress, displacement, and moisture content, etc.) collected by sensors exist in discrete numerical forms.

[0038] The real-time construction technology data are engineering parameters recorded in the construction log, including: quantitative parameters and qualitative parameters. The quantitative parameters include: current excavation footage, real-time grouting pressure, real-time bolt spacing, and real-time lining construction interval; the qualitative parameters include: current support type (one-hot encoded as T).

[0039] Specifically, the engineering parameters recorded in the construction log include unstructured text information recorded during the construction process, such as: support types (e.g., steel arch support and lattice support, etc.), descriptions of construction steps, and timestamp records (i.e., acquisition time), etc.

[0040] The real-time geological data are the physical and mechanical parameters of the soft stratum obtained through testing, including: real-time moisture content, real-time void ratio, real-time compression modulus, real-time cohesion, and real-time internal friction angle.

[0041] Specifically, the real-time geological data are structured multi-dimensional feature datasets such as the physical and mechanical parameters of the stratum (e.g., void ratio and compressive strength, etc.) stored in tabular form, and the basic coordinates of the monitoring points (i.e., the original three-dimensional coordinates), etc.

[0042] The real-time environmental data are the external conditions obtained through environmental monitoring, including: real-time groundwater level and real-time surface load (e.g., building load and vehicle load, etc.).

[0043] Furthermore, the specific method for constructing a structured multi-dimensional feature dataset based on real-time monitoring data is: extracting or transforming the real-time monitoring data to obtain various features, and combining the various features in a preset order to form a multi-dimensional feature vector and a label vector; the multi-dimensional feature vector and the label vector constitute a structured multi-dimensional feature dataset ; among them, the various features include: time series features, spatial location features, stratum parameter features, construction parameter features, and environmental parameter features. The extraction or transformation of the real-time monitoring data can be realized by using existing extraction or transformation methods, so as to obtain separate time series features, spatial location features, stratum parameter features, construction parameter features, and environmental parameter features, and thus will not be elaborated here.

[0044] Furthermore, the expression of the multi-dimensional feature vector is: ; where is the real-time timestamp (i.e., acquisition time) of the th monitoring point; is the three-dimensional coordinates of the real-time monitoring point of the th monitoring point; is the horizontal distance of the th monitoring point from the axis of the target tunnel; is the burial depth (i.e., vertical depth) of the th monitoring point; is the real-time moisture content of the soft stratum corresponding to the th monitoring point; is the The in-situ void ratio of the soft soil layer corresponding to the th monitoring point; is the in-situ compression modulus of the soft soil layer corresponding to the th monitoring point; is the in-situ cohesion of the soft soil layer corresponding to the th monitoring point; is the in-situ internal friction angle of the soft soil layer corresponding to the th monitoring point; is the current excavation advance of the current construction stage; is the one-hot encoding vector of the current support type of the current construction stage;

[0045] wherein, the label vector is expressed as: ; wherein, is the settlement of the tunnel crown of the current construction stage; is the in-situ surface settlement; is the in-situ monitoring value of the convergence displacement.

[0046] S2: Preprocess the structured multi-dimensional feature dataset to obtain the standard input data.

[0047] Further, the sub-steps of preprocessing the structured multi-dimensional feature dataset to obtain the standard input data are as follows: S21: Perform anomaly detection on the structured multi-dimensional feature dataset to generate an anomaly detection result. If the anomaly detection result is abnormal, execute S22; if the anomaly detection result is normal, use the structured multi-dimensional feature dataset as the data to be filled and execute S23.

[0048] Specifically, use the IQR (Interquartile Range), Isolation Forest algorithm, Dynamic Threshold method, or Long Short-Term Memory (LSTM) autoencoder to perform anomaly detection on the structured multi-dimensional feature dataset, but not limited to the IQR (Interquartile Range), Isolation Forest algorithm, Dynamic Threshold method, or Long Short-Term Memory (LSTM) autoencoder. If there are abnormal data (i.e., outliers or anomaly points) in the structured multi-dimensional feature dataset, the generated anomaly detection result is abnormal; if there is no abnormal data in the structured multi-dimensional feature dataset, the generated anomaly detection result is normal.

[0049] S22: Repair the structured multi-dimensional feature dataset to obtain the data to be filled, and execute S23.

[0050] Specifically, the purpose of repairing the structured multi-dimensional feature dataset is to eliminate or correct the noise data and avoid affecting the accuracy of the prediction result due to the input of abnormal values.

[0051] Furthermore, the sub-steps of repairing the structured multi-dimensional feature dataset to obtain the data to be filled are as follows: S221: Read the time series abnormal data in the structured multi-dimensional feature dataset. If there is time series abnormal data, repair the time series abnormal data to obtain the repaired data, and execute S222; if there is no time series abnormal data, execute S222.

[0052] Specifically, the time series abnormal data is the abnormal data in the data of the time series type in the structured multi-dimensional feature dataset. The data of the time series type in the structured multi-dimensional feature dataset includes: real-time settlement and displacement data, real-time construction process data, real-time geological data, and real-time environmental data.

[0053] Furthermore, the sub-steps of repairing the time series abnormal data to obtain the repaired data are as follows: S2211: Generate the first repair sequence for each sub-time series abnormal data in the time series abnormal data according to the reading order. The first repair sequence increases sequentially from the first to the last according to the reading order.

[0054] Specifically, the sub-time series abnormal data is the real-time settlement and displacement data, real-time construction process data, real-time geological data, and / or real-time environmental data. For example: the real-time monitoring values of the tunnel crown settlement, real-time surface settlement, and convergence displacement in the current construction stage. If the reading order of the sub-time series abnormal data is 1, the generated first repair sequence is 1; if the reading order of the sub-time series abnormal data is 2, the generated first repair sequence is 2; if the reading order of the sub-time series abnormal data is 3, the generated first repair sequence is 3.

[0055] S2212: Take the sub-time series abnormal data with the smallest first repair sequence as the first current repair data, repair the first current repair data to obtain the first sub-repaired data, and execute S2213.

[0056] Furthermore, the sub-steps of repairing the first current repair data to obtain the first sub-repaired data are as follows: R1: Taking the acquisition time of the first current repair data as the starting time, shifting forward by an interval time to obtain the previous trend interval, and taking the acquisition time of the first current repair data as the starting time, shifting backward by an interval time to obtain the subsequent trend interval.

[0057] Specifically, the specific value of the interval time is set according to the actual situation.

[0058] R2: Obtain the total number of valid data in the sub-time series in the previous trend interval and the total number of valid data in the sub-time series in the subsequent trend interval, and compare the total number of valid data in the sub-time series in the previous trend interval and the total number of valid data in the sub-time series in the subsequent trend interval. If the total number of valid data in the sub-time series in the previous trend interval is greater than the total number of valid data in the sub-time series in the subsequent trend interval, then all the valid data in the sub-time series in the previous trend interval are used as the first repair reference data; if the total number of valid data in the sub-time series in the previous trend interval is less than the total number of valid data in the sub-time series in the subsequent trend interval, then all the valid data in the sub-time series in the subsequent trend interval are used as the first repair reference data.

[0059] Specifically, the valid data in the sub-time series are normal data.

[0060] R3: Analyze and calculate the first repair reference data to obtain the first sub-repaired data.

[0061] Furthermore, when all the valid data in the sub-time series in the current trend interval are used as the first repair reference data, the calculation formula for the first sub-repaired data is: ; where is the first sub-repaired data; is the valid data of the previous sub-time series at the acquisition time of the first current repair data; is the th first repair reference data in the previous trend interval; is the th first repair reference data in the previous trend interval; , is the total number of the first repair reference data in the previous trend interval, is a natural number; is the th acquisition time of the first repair reference data; is the th acquisition time of the first repair reference data; is the average value of the interval time between the acquisition times of adjacent valid data in the sub-time series in the previous trend interval; The number of between the acquisition time of the first current repair data and the acquisition time of the valid data in the previous sub-time series.

[0062] Specifically, represents the change data between the th first repair reference data and the th first repair reference data; is the interval time between the acquisition time of the th first repair reference data and the acquisition time of the th first repair reference data.

[0063] When the valid data in the sub-time series in the post-trend interval is used as the first repair reference data, the calculation formula for the first sub-repaired data is: ; where is the first sub-repaired data; is the valid data of the next sub-time series after the acquisition time of the first current repair data; is the th first repair reference data in the post-trend interval; is the th first repair reference data in the post-trend interval; , is the total number of the first repair reference data in the post-trend interval, is a natural number; is the acquisition time of the th first repair reference data; is the acquisition time of the th first repair reference data; is the average value of the interval time between the acquisition times of adjacent valid data in the sub-time series in the post-trend interval; The number of between the acquisition time of the first current repair data and the acquisition time of the next sub-time series valid data.

[0064] Specifically, represents the change data between the th first repair reference data and the th first repair reference data; is the interval time between the acquisition time of the th first repair reference data and the acquisition time of the th first repair reference data.

[0065] S2213: Use the total number of abnormal data in the sub-time series to judge the first repair sequence of the first current repair data. If the total number of abnormal data in the sub-time series is greater than the first repair sequence of the first current repair data, after removing the first repair sequence of the first current repair data, execute S2212; if the total number of abnormal data in the sub-time series is equal to the first repair sequence of the first current repair data, remove the first repair sequence of the first current repair data, and use all the first sub-repaired data as the repaired data, and execute S222.

[0066] S222: Read the repaired data or the structured multi-dimensional feature dataset. If there is non-time series abnormal data in the repaired data or the structured multi-dimensional feature dataset, repair the non-time series abnormal data to obtain the data to be filled; if there is no non-time series abnormal data in the repaired data, use the repaired data as the data to be filled.

[0067] Specifically, the non-time series abnormal data is the abnormal data in the data belonging to the non-time series type in the structured multi-dimensional feature dataset. The data belonging to the non-time series type in the structured multi-dimensional feature dataset includes: real-time spatio-temporal information.

[0068] Further, the sub-steps of repairing the non-time series abnormal data to obtain the data to be filled are as follows: S2221: Generate a second repair sequence for each sub-non-time series abnormal data in the non-time series abnormal data according to the reading order, and the second repair sequence increases sequentially from the first to the last according to the reading order.

[0069] Specifically, the sub-non-time series abnormal data is real-time spatio-temporal information, such as: the three-dimensional coordinates of a real-time monitoring point. If the reading order of the sub-non-time series abnormal data is 1, the generated second repair sequence is 1; if the reading order of the sub-non-time series abnormal data is 2, the generated second repair sequence is 2; if the reading order of the sub-non-time series abnormal data is 3, the generated second repair sequence is 3.

[0070] S2222: Use the sub-non-time series abnormal data with the smallest second repair sequence as the second current repair data, repair the second current repair data to obtain the second sub-repaired data, and execute S2223.

[0071] Further, the mean value of the sub-non-time series valid data corresponding to the sub-non-time series abnormal data is used as the second sub-repaired data, but it is not limited to using the mean value of the sub-non-time series valid data corresponding to the sub-non-time series abnormal data as the second sub-repaired data.

[0072] Specifically, the sub-non-time series valid data is normal data.

[0073] S2223: Determine the second repair sequence of the second current repair data using the total number of abnormal data in the non-time series. If the total number of abnormal data in the non-time series is greater than the second repair sequence of the second current repair data, after removing the second repair sequence of the second current repair data, execute S2222; if the total number of abnormal data in the non-time series is equal to the second repair sequence of the second current repair data, remove the second repair sequence of the second current repair data, and use all the second sub-repaired data as the data to be filled, then execute S23.

[0074] S23: Analyze the data to be filled to generate a missing result. If the missing result is missing, execute S24; if the missing result is not missing, use the data to be filled as the data to be normalized and execute S25.

[0075] Furthermore, use the display missing detection method and / or the logical consistency detection method to analyze the data to be filled to generate a missing result, but not limited to the display missing detection method and / or the logical consistency detection method.

[0076] Specifically, the display missing detection method is: scan the data records related to the data to be filled. If there is a situation where the recognition field value is a null value (NULL), an unrecorded status (such as “—”), or an invalid value (such as: the sensor returns an abnormal code), it is marked as a missing value, and the generated missing result is missing; if there is no situation where the recognition field value is a null value (NULL), an unrecorded status (such as “—”), or an invalid value (such as: the sensor returns an abnormal code), the generated missing result is not missing. For example: the “real-time compression modulus” field of a certain mileage section in the geological exploration report is not filled; the “real-time grouting pressure” of a certain grouting in the construction log is not collected due to a sensor failure, then the generated missing result is missing.

[0077] The logical consistency detection method is: judge whether the data to be filled is reasonably missing according to the business logic. For example: if the “acquisition timestamp” of a certain monitoring point exists, but the “tunnel crown settlement at the current construction stage” corresponding to it is not recorded, it is determined to be missing, and the generated missing result is missing; when the “current excavation advance” is 0 (non-construction stage), the “real-time lining construction interval” is missing, which belongs to reasonable missing and does not need to be processed, and the generated missing result is not missing.

[0078] Furthermore, the method of using the display missing detection method and the logical consistency detection method to analyze the data to be filled to generate a missing result is: first mark the missing value through the display missing detection method, and then judge whether the missing value is reasonably missing through the logical consistency detection method. If it belongs to reasonable missing, the generated missing result is not missing; if it is determined to be missing, the generated missing result is missing.

[0079] S24: Fill in the missing values in the data to be filled, obtain the data to be normalized, and execute S25.

[0080] Specifically, the purpose of filling in the missing values in the data to be filled is to fill the data gaps and avoid affecting the accuracy of the prediction results due to incomplete input data.

[0081] Furthermore, the sub-steps of filling in the missing values in the data to be filled and obtaining the data to be normalized are as follows: 241: Read the missing values in the data to be filled, classify the missing values, and determine the missing category to which the missing values belong. Among them, the missing categories include: missing values of geological parameters, missing values of construction parameters, and missing values of environmental impact data.

[0082] Specifically, classify the missing values through a pre-trained classification model to determine the missing category to which the missing values belong.

[0083] Among them, the missing values of geological parameters are static / semi-static data, such as: real-time compression modulus, real-time void ratio, and real-time water content.

[0084] The missing values of construction parameters are dynamic time-series data, such as: real-time grouting pressure, current excavation footage, and real-time lining construction interval.

[0085] The missing values of environmental impact data include: real-time groundwater level and real-time surface load.

[0086] S242: Determine the corresponding filling strategy according to the missing category, and use the corresponding filling strategy to fill in the missing values to obtain the data to be normalized.

[0087] Among them, when the missing category is the missing values of geological parameters, the corresponding filling strategy is the median filling method for the same mileage section or the filling method for continuous missing for more than 3 cycles, but not limited to the median filling method for the same mileage section or the filling method for continuous missing for more than 3 cycles.

[0088] Specifically, the median filling method for the same mileage section is: extract the same type of formation parameters (such as the real-time compression modulus of soft clay) in the mileage section where the missing value is located (such as: within the range of 50 meters before and after), calculate the median for filling, and obtain the data to be normalized. The median is not sensitive to outliers and can reflect the typical characteristics of formation parameters, avoiding the interference of the mean value by local abnormal data (such as: interlayers).

[0089] The filling method for continuous missing for more than 3 cycles is: if the geological parameters are missing in multiple consecutive mileage sections (such as: more than 3 construction cycles), use the historical mean value of the same type of formation (such as: completely weathered sandstone) for filling, and combine with the geological section diagram to manually verify the rationality to obtain the data to be normalized.

[0090] Among them, when the missing category is the missing value of construction parameters, the corresponding filling strategy is the weighted average filling method of adjacent construction cycles before and after, the filling method for short-term missing (≤3 cycles), or the filling method for long-term missing (>3 cycles), but not limited to the weighted average filling method of adjacent construction cycles before and after, the filling method for short-term missing (≤3 cycles), or the filling method for long-term missing (>3 cycles).

[0091] Specifically, the weighted average filling method of adjacent construction cycles before and after is as follows: Using the parameter values of adjacent construction cycles before and after the missing value, filling is performed by weighted average according to the time difference (the weight is inversely proportional to the time difference) to obtain the data to be normalized. Construction parameters (such as grouting pressure) fluctuate with the process, and the parameters at adjacent times can better reflect the current working conditions. The weighting strategy strengthens the time correlation.

[0092] The filling method for short-term missing (≤3 cycles) is as follows: Use the existing adjacent interpolation method for filling to obtain the data to be normalized.

[0093] The filling method for long-term missing (>3 cycles) is as follows: Manually fill in combination with the construction log to obtain the data to be normalized.

[0094] Among them, when the missing category is the missing value of environmental impact data, the corresponding filling strategy is the existing spatial adjacent mean filling method, but not limited to the spatial adjacent mean filling method.

[0095] S25: Normalize the data to be normalized to obtain the standard input data.

[0096] Furthermore, the sub-steps for normalizing the data to be normalized to obtain the standard input data are as follows: S251: Classify the data to be normalized to obtain the data to be normalized for classification, where the data to be normalized for classification includes: range difference data, normal distribution data, and coded data.

[0097] Specifically, the range difference data are the data with relatively large range differences in the data to be normalized. For example: the real-time moisture content and real-time void ratio of real-time geological parameters, the current excavation footage, real-time bolt spacing, and real-time lining construction interval of real-time construction technology data, the real-time groundwater level and real-time surface load of real-time environmental data.

[0098] The normal distribution data are the data that usually conform to or approximate the normal distribution during the construction stable stage. For example: the tunnel crown settlement at the current construction stage of real-time settlement and displacement data, the real-time monitoring values of real-time surface settlement and convergence displacement.

[0099] The coded data are the data belonging to discrete categorical variables without a natural order relationship. For example: the current support type (such as lattice girder and steel arch).

[0100] S252: Normalize the range difference data using the first normalization strategy, normalize the normal distribution data using the second normalization strategy, and normalize the encoded data using the third normalization strategy to obtain the standard input data.

[0101] Further, the first normalization strategy is Min-Max normalization, but not limited to Min-Max normalization. Preferably in this application: Min-Max normalization is used to normalize the range difference data.

[0102] Further, the second normalization strategy is Z-score normalization, but not limited to Z-score normalization. Preferably in this application: Z-score normalization is used to normalize the normal distribution data.

[0103] Further, the third normalization strategy is one-hot encoding, but not limited to one-hot encoding. Preferably in this application, one-hot encoding is used to normalize the encoded data.

[0104] S3: Input the standard input data into the pre-constructed settlement prediction model for shallow-buried soft ground tunneling. The single-step prediction values are output by the settlement prediction model for shallow-buried soft ground tunneling. Among them, the single-step prediction values at least include: the single-step prediction value of the tunnel crown settlement, the single-step prediction value of the ground surface settlement, and the single-step prediction value of the convergence displacement.

[0105] Specifically, the pre-constructed settlement prediction model for shallow-buried soft ground tunneling is a prediction model constructed by integrating spatio-temporal features and construction techniques for shallow-buried soft ground tunneling, which can achieve accurate mapping of settlement values. The single-step prediction values are not limited to the single-step prediction value of the tunnel crown settlement, the single-step prediction value of the ground surface settlement, and the single-step prediction value of the convergence displacement, and may also include other prediction data related to settlement prediction.

[0106] As an embodiment, a settlement prediction model for shallow-buried soft ground tunneling is constructed based on a machine learning algorithm. The historical settlement and displacement data in the standardized training dataset are used as labels, and the remaining features are used as inputs. The hyperparameters of the model are optimized through existing cross-validation to form a trained settlement prediction model for shallow-buried soft ground tunneling. The settlement prediction model for shallow-buried soft ground tunneling in this application pays sufficient attention to the key construction processes, can fully capture the complex relationship between construction process parameters and settlement, has strong generalization ability, can adapt to the complex working conditions of shallow-buried soft foundation, and can avoid overfitting.

[0107] Further, the original sample data is processed using a data processing method consistent with the method for processing real-time monitoring data (i.e., the data processing method in steps S1 - S2) to obtain a standardized training data set for constructing, training, and validating the settlement prediction model of the shallow-buried soft ground tunneling, which can avoid prediction deviations caused by changes in the input format. The original sample data includes at least: historical settlement and displacement data, historical construction technology data, historical geological data, and historical environmental data.

[0108] S4: Use the dynamic prediction error to correct the single-step prediction value in real time to generate the settlement prediction value, where the settlement prediction value includes at least: the settlement prediction value of the tunnel crown, the settlement prediction value of the ground surface, and the convergence displacement prediction value.

[0109] Specifically, using the current dynamic prediction error to correct the single-step prediction value in real time can solve problems such as time-varying formation parameters and fluctuating monitoring noise during the construction of shallow-buried soft ground tunneling. Compared with the existing single model, the prediction accuracy of this application has been improved by 15% - 20%, and the error correction process is reproducible, which can meet the requirements of engineering monitoring for the reliability of the results.

[0110] Further, the expression of the dynamic prediction error is: ; Where is the dynamic prediction error at the th moment; is the dynamic prediction error at the th moment; is the process noise, is the observation noise; is the process noise weight; is the observation noise weight.

[0111] Specifically, obeys a normal distribution with a mean of 0 and a process noise covariance of , which is used to describe the time-varying characteristics of construction errors. obeys a normal distribution with a mean of 0 and an observation noise covariance of , which is used to reflect the errors of monitoring equipment. and The specific values of and are set according to the actual situation to reflect the influence degree of

[0112] Further, the expression of the settlement prediction value is: ; Where is the settlement prediction value; ​is the single-step prediction value at the moment; is the dynamic prediction error at the moment.

[0113] Furthermore, it further includes step S5: Analyze the settlement prediction value using a preset settlement safety threshold. If the settlement prediction value is greater than the settlement safety threshold, trigger an early warning mechanism and output a settlement early warning message, where the settlement early warning message at least includes: the current settlement risk level; if the settlement prediction value is less than or equal to the settlement safety threshold, then end.

[0114] Specifically, the specific value of the preset settlement safety threshold is set according to the actual situation.

[0115] Furthermore, the specific method for outputting the settlement early warning message is: Traverse the pre-constructed settlement risk level table, and determine that the settlement risk level corresponding to the settlement range to which the settlement prediction value belongs is the current settlement risk level; where one settlement risk level corresponds to one settlement range.

[0116] The beneficial effects achieved by this application are as follows: (1) The settlement prediction method and system for shallow-buried soft ground tunneling in this application integrate spatio-temporal distribution data, geological parameters (such as water content and void ratio, etc.), construction process parameters (such as excavation footage and grouting pressure), and environmental impact data, and can comprehensively reflect the multi-factor coupling effect of the settlement of shallow-buried soft ground tunneling, solving the problem of single data in traditional methods.

[0117] (2) The settlement prediction method and system for shallow-buried soft ground tunneling in this application have targeted processing for outlier detection (such as monitoring jump points caused by sudden construction disturbances) and missing value filling (such as sensor fault data), and use a variety of normalization methods. Therefore, it can improve the quality of the data set input into the settlement prediction model for shallow-buried soft ground tunneling, thereby improving the prediction accuracy.

[0118] (3) The settlement prediction method and system for shallow-buried soft ground tunneling in this application can dynamically adjust the prediction result according to real-time monitoring data, can meet the low-latency requirements of real-time monitoring of tunnel construction, and can ensure the prediction accuracy when facing problems such as time-varying formation parameters and monitoring noise fluctuations.

[0119] (4) The settlement prediction method and system for shallow-buried soft ground tunneling in this application can achieve hierarchical early warning of settlement risks, support the adjustment of construction plans, ensure project safety, and at the same time, automated data processing and real-time prediction reduce the manual analysis cost, improving the accuracy and engineering application value of the settlement prediction for shallow-buried soft ground tunneling.

[0120] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the protection scope of the present application is intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application. Obviously, those skilled in the art can make various changes and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the protection of the present application and the scope of equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for predicting settlement of a shallow soft ground tunnel, characterized in that: The steps include: S1: Acquire the real-time monitoring data of the target tunnel at the current construction stage, construct multi-dimensional features based on the real-time monitoring data, and obtain a structured multi-dimensional feature data set, wherein the expression of the structured multi-dimensional feature data set is: , is a multidimensional feature vector, is a label vector; the real-time monitoring data at least includes: real-time settlement and displacement data, real-time construction process data, real-time geological data and real-time environmental data; S2: preprocess the structured multidimensional feature data set to obtain standard input data; S3: inputting the standard input data into a pre-constructed settlement prediction model for a dark-excavated tunnel in a shallow soft ground layer, and outputting a single-step prediction value from the settlement prediction model for a dark-excavated tunnel in a shallow soft ground layer, wherein the single-step prediction value at least includes: a single-step prediction value for tunnel crown settlement, a single-step prediction value for ground settlement, and a single-step prediction value for convergence displacement; S4: Using the dynamic prediction error to perform real-time correction on the single-step prediction value to generate a settlement prediction value, wherein the settlement prediction value at least includes: a tunnel crown settlement prediction value, a surface settlement prediction value and a convergence displacement prediction value.

2. The method for predicting settlement of a shallow soft ground tunnel according to claim 1 is characterized in that: It also includes S5: analyzing the settlement prediction value using a preset settlement safety threshold. If the settlement prediction value is greater than the settlement safety threshold, the early warning mechanism is triggered and settlement warning information is output, wherein the settlement warning information at least includes: the current settlement risk level; if the settlement prediction value is less than or equal to the settlement safety threshold, the process ends.

3. The method for predicting settlement of a shallow soft ground tunnel according to claim 1 is characterized in that: The sub-steps of preprocessing the structured multidimensional feature data set to obtain standard input data are as follows: S21: Perform anomaly detection on the structured multidimensional feature data set to generate anomaly detection results. If the anomaly detection result is abnormal, execute S22; if the anomaly detection result is normal, use the structured multidimensional feature data set as data to be filled and execute S23; S22: Repair the structured multi-dimensional feature data set to obtain the data to be filled, and execute S23; S23: Analyze the data to be filled and generate a missing result. If the missing result is missing, execute S24; if the missing result is not missing, use the data to be filled as the data to be normalized and execute S25; S24: Fill missing values ​​in the data to be filled, obtain the data to be normalized, and execute S25; S25: normalize the data to be normalized to obtain standard input data.

4. The method for predicting settlement of a shallow soft ground tunnel according to claim 3 is characterized in that: The sub-steps of repairing the structured multidimensional feature data set and obtaining the data to be filled are as follows: S221: Read the time series abnormal data in the structured multi-dimensional feature data set. If there is time series abnormal data, repair the time series abnormal data to obtain the repaired data and execute S222; if there is no time series abnormal data, execute S222; S222: Reading the repaired data or the structured multidimensional feature data set, if there is non-time series abnormal data in the repaired data or the structured multidimensional feature data set, repairing the non-time series abnormal data to obtain data to be filled; If there is no non-time series abnormal data in the repaired data, the repaired data will be used as the data to be filled.

5. The method for predicting settlement of a shallow soft ground tunnel according to claim 4 is characterized in that: To repair abnormal time series data, the sub-steps to obtain the repaired data are as follows: S2211: Generate a first repair sequence for each sub-time series abnormal data in the time series abnormal data according to the reading order, and the first repair sequence increases in sequence from the earliest to the latest according to the reading order; S2212: taking the abnormal data of the smallest sub-time series of the first repair sequence as the first current repair data, repairing the first current repair data to obtain the first sub-repaired data, and executing S2213; S2213: judging the first repair sequence of the first current repair data by using the total number of abnormal sub-time series data; if the total number of abnormal sub-time series data is greater than the first repair sequence of the first current repair data, then removing the first repair sequence of the first current repair data, and then executing S2212; If the total number of abnormal sub-time series data is equal to the first repair sequence of the first current repair data, the first repair sequence of the first current repair data is removed, and all first sub-repaired data are used as repaired data to execute S222.

6. The method for predicting settlement of a shallow soft ground tunnel according to claim 5 is characterized in that: The sub-steps of repairing the first current repair data to obtain the first sub-repaired data are as follows: R1: Taking the collection time of the first current repair data as the starting time, moving forward an interval time to obtain the front trend interval, taking the collection time of the first current repair data as the starting time, moving backward an interval time to obtain the back trend interval; R2: Obtain the total number of valid sub-time series data in the previous trend interval and the total number of valid sub-time series data in the next trend interval, and compare the total number of valid sub-time series data in the previous trend interval with the total number of valid sub-time series data in the next trend interval. If the total number of valid sub-time series data in the previous trend interval is greater than the total number of valid sub-time series data in the next trend interval, all valid sub-time series data in the previous trend interval are used as the first repair reference data. If the total number of valid sub-time series data in the front trend interval is less than the total number of valid sub-time series data in the back trend interval, all valid sub-time series data in the back trend interval are used as the first repair reference data; R3: Analyze and calculate the first repair reference data to obtain the first sub-repaired data.

7. The method for predicting settlement of a shallow soft ground tunnel according to claim 1, characterized in that: The expression of dynamic prediction error is: ； in, For the Dynamic prediction error at each moment; For the Dynamic prediction error at each moment; is the process noise, is the observation noise; is the process noise weight; is the observation noise weight.

8. The method for predicting settlement of a shallow soft ground tunnel according to claim 1, characterized in that: The expression of settlement prediction value is: ； in, is the predicted value of settlement; For the The single-step forecast value at time; For the The dynamic prediction error at each moment.

9. A settlement prediction system for a shallow soft ground tunnel, characterized in that: include: A settlement prediction center and at least one user terminal; User side: receive settlement prediction value; receive settlement warning information; Settlement prediction center: used to execute the settlement prediction method for a shallow soft ground stratum dark excavation tunnel as described in any one of claims 1-8.

10. The settlement prediction system for shallow soft ground tunnels according to claim 9, characterized in that: The settlement prediction center includes: data acquisition unit, data processing unit, settlement prediction unit, dynamic correction unit and early warning unit; Among them, the data acquisition unit is used to obtain real-time monitoring data of the current construction stage of the target tunnel; Data processing unit: used to construct multi-dimensional features based on real-time monitoring data to obtain a structured multi-dimensional feature data set; pre-process the structured multi-dimensional feature data set to obtain standard input data; Settlement prediction unit: a pre-built settlement prediction model for a dark-digging tunnel in a shallow soft stratum is provided, standard input data is input into the pre-built settlement prediction model for a dark-digging tunnel in a shallow soft stratum, and the settlement prediction model for a dark-digging tunnel in a shallow soft stratum outputs a single-step prediction value; Dynamic correction unit: Use dynamic prediction error to make real-time corrections to the single-step prediction value to generate settlement prediction value; Early warning unit: Use the preset settlement safety threshold to analyze the settlement prediction value. If the settlement prediction value is greater than the settlement safety threshold, the early warning mechanism is triggered and the settlement warning information is output; if the settlement prediction value is less than or equal to the settlement safety threshold, the process ends.

Images