A loess roadbed deformation early warning method and system

Abstract

The present application provides a loess subgrade deformation early warning method and system for predicting the deformation amount of a loess subgrade through the water content of the subgrade, and then early warning based on the deformation amount, so as to improve the loess subgrade deformation early warning effect. The loess subgrade deformation early warning method comprises time series processing of loess subgrade rainfall data and drainage ditch flow data obtained by real-time monitoring, construction of a set of time lag features, selection of features from the time lag features according to a selection principle of a first data set, so as to select a feature subset for predicting the water content at a future time point; inputting the feature subset into a first prediction model to obtain a water content prediction value of the loess subgrade; and inputting the water content prediction value and real-time loess subgrade soil pressure data into a second prediction model to obtain a deformation amount prediction value of the loess subgrade. The present application improves the safety and sustainability of the loess subgrade.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering, specifically to a method and system for early warning of deformation of loess roadbed. Background Technology

[0002] Loess has large pores and well-developed vertical joints, with a permeability in the longitudinal direction far greater than in the lateral direction. Due to rainfall, the loess particle skeleton collapses rapidly after being soaked in water, resulting in a significant reduction in strength. Furthermore, the deformation of loess often exhibits abrupt changes in the longitudinal direction. This property easily leads to uneven settlement of buildings, causing significant damage to the safety and economic efficiency of structures.

[0003] Currently, most monitoring equipment for loess subgrade deformation is vertical deformation monitoring devices, which are typically installed inside the subgrade and on the pavement. This monitoring method transmits deformation data to a computer system in real time according to a predetermined monitoring frequency, providing continuous and long-term monitoring capabilities. However, because loess deformation often occurs abruptly, this monitoring method is less effective in providing early warnings of loess subgrade deformation, making it difficult to achieve timely warnings.

[0004] During the research process, the applicant discovered that loess deformation is significantly affected by moisture content and earth pressure, while moisture content is influenced by rainfall in the loess subgrade and the flow rate of its drainage ditches. As highway and railway engineering projects increasingly demand higher standards for subgrade deformation and settlement, existing monitoring methods are insufficient to predict deformation, especially when dealing with unforeseen circumstances.

[0005] Therefore, improving the early warning effect of loess roadbed deformation has become an urgent need in the fields of land management and engineering. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a loess subgrade deformation early warning method and system that predicts the deformation of loess subgrade by subgrade moisture content and then uses the deformation to provide early warning, so as to improve the accuracy of loess subgrade deformation early warning.

[0007] The technical solution adopted by this invention to solve its technical problem is: a method for early warning of deformation of loess roadbed, comprising the following steps:

[0008] A time series dataset is constructed by measuring the rainfall, drainage ditch flow, and moisture content of the loess roadbed over a certain period of time.

[0009] A moisture content feature set is constructed by progressively moving the rainfall and drainage ditch flow data of the loess roadbed in the time series dataset forward. The correlation coefficient between the moisture content feature set and the moisture content is then used to determine the number of lag features of rainfall and drainage ditch flow. The correlation coefficient is then analyzed to understand the degree of influence of rainfall and drainage ditch flow at different lag times on the moisture content and to determine the importance of the features. Features are then selected based on their importance, and these selected features are finally used as the first dataset.

[0010] The first dataset is used to train a soil moisture content prediction model for loess roadbed until the preset conditions are met, so as to obtain the first prediction model.

[0011] Based on geotechnical tests, data on the deformation of loess under corresponding earth pressure and water content were obtained, and a second dataset including earth pressure, water content and deformation was established.

[0012] The second dataset was used to train a loess subgrade deformation prediction model until the requirements were met, in order to obtain the second prediction model.

[0013] Time series processing is performed on the real-time monitoring data of rainfall and drainage ditch flow on the loess roadbed to construct a set of time lag feature sets. The time lag feature sets cover data including at least the current time and one or more previous time points. Then, features are selected from the time lag feature sets according to the feature selection principle of the first dataset to select a feature subset for predicting the moisture content at a future time point.

[0014] The feature subset is input into the first prediction model to obtain the predicted moisture content of the loess roadbed;

[0015] The predicted moisture content and the real-time earth pressure data of the loess subgrade are input into the second prediction model to obtain the predicted deformation value of the loess subgrade.

[0016] Warnings are issued based on the predicted deformation value, and when the predicted deformation value exceeds a set value, a warning message is issued.

[0017] Furthermore, both the first and second prediction models employ the XGboost model.

[0018] Furthermore, the specific steps for constructing a moisture content feature set by progressively moving forward the rainfall and drainage ditch flow data of the loess roadbed are as follows:

[0019] Keeping the moisture content sequence unchanged, the rainfall sequence R(1) and the drainage ditch flow sequence D(1) are shifted forward by one time unit to obtain the R(2) and D(2) sequences. This process is repeated until the R(N) and D(N) sequences are obtained. These sequences R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) constitute the lag feature set of rainfall and drainage ditch flow.

[0020] Furthermore, the rainfall R1 and drainage ditch flow D1 at the initial time t1 will have a lag effect on the loess moisture content at subsequent times t2, t3, ..., tm. In this way, the features R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) are constructed.

[0021] Furthermore, as the rainfall and drainage flow series are shifted forward as a whole, we assume there are three sets of data: Rt, Dt, and St, where Rt represents the rainfall series, Dt represents the drainage flow series, and St represents the moisture content series. The rainfall and drainage flow time series are shifted to the right by 1 to N units, and the missing data at the beginning of the lag feature time series is filled with the first data, while the missing data at the end is filled with the last data.

[0022] Furthermore, the selection of features based on their importance involves selecting the features with the highest correlation coefficient with the moisture content.

[0023] A loess subgrade deformation early warning system for the above-mentioned loess subgrade deformation early warning method includes a data acquisition module, a data processing module, a moisture content prediction module, a deformation prediction module, and an early warning module.

[0024] The data acquisition module is used to collect rainfall, drainage ditch flow, and earth pressure data of the loess roadbed.

[0025] The data processing module is connected to the data acquisition module to perform time series processing on the collected loess roadbed rainfall data and drainage ditch flow data, construct a set of time lag feature sets, and then select features from the time lag feature sets based on the determined importance conclusions of the features, so as to select a feature subset for predicting the moisture content at a future time point.

[0026] The moisture content prediction module is connected to the data processing module to predict the moisture content of the loess subgrade based on the feature subset.

[0027] The deformation prediction module is connected to the moisture content prediction module and the data acquisition module to predict the deformation of the loess subgrade based on the predicted moisture content and the current earth pressure data.

[0028] The early warning module is connected to the deformation prediction module to issue an early warning based on the magnitude of the predicted deformation value. When the predicted deformation value exceeds a set value, an early warning message is issued.

[0029] The beneficial effects of the present invention are as follows: The loess subgrade deformation early warning method and system of the present invention can predict the deformation of the loess subgrade at a future moment or a period of time based on the monitored rainfall, drainage ditch flow and earth pressure data, and can issue early warnings in advance, with better early warning effect, improving the timeliness and efficiency of loess subgrade monitoring, and making it easier to respond in advance, thereby improving the safety and sustainability of loess subgrade. Attached Figure Description

[0030] Figure 1 This is a flowchart of the early warning method of the present invention;

[0031] Figure 2 This is a graph showing the processing of characteristic data after moisture content analysis;

[0032] Figure 3 This is a structural diagram of the early warning system of the present invention; Detailed Implementation

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] During the research process, the applicant discovered that loess deformation is significantly affected by moisture content and earth pressure. Moisture content is influenced by rainfall in the loess subgrade and the flow rate of its drainage ditches, but the effects of rainfall and drainage ditches on the subgrade moisture content have a lag. To accurately predict the deformation of the loess subgrade using moisture content and earth pressure, such as... Figure 1 As shown, a method for early warning of loess roadbed deformation according to the present invention includes the following steps:

[0035] Step 1: Measure the rainfall, drainage ditch flow, and moisture content of the loess roadbed over a certain period of time to form a time series dataset. A tubular moisture meter is preferred for measuring the moisture content.

[0036] Step 2: Construct a moisture content feature set by progressively moving the rainfall and drainage ditch flow data of the loess roadbed in the time series dataset forward. Then, use the correlation coefficient between the moisture content feature set and the moisture content to determine the number of lag features for rainfall and drainage ditch flow. Analyze the correlation coefficients to understand the impact of rainfall and drainage ditch flow at different lag times on the moisture content and determine the importance of the features. Select features based on their importance, such as selecting features with a correlation coefficient greater than 0. Finally, use these selected features as the first dataset.

[0037] Step 3: Divide the first dataset into a first training set and a first validation set. Use the first training set to train the loess subgrade soil moisture content prediction model and use the first validation set to validate the prediction model until the preset conditions are met to obtain the first prediction model. The root mean square error can be used to evaluate whether the first prediction model meets the requirements.

[0038] Step 4: Based on the geotechnical test, obtain the deformation data of loess under the corresponding earth pressure and water content, and establish a second dataset including earth pressure, water content and deformation.

[0039] Step 5: Divide the second dataset into a second training set and a second validation set. Use the second training set to train the loess subgrade deformation prediction model and use the second validation set to validate the prediction model until the preset conditions are met to obtain the second prediction model. The root mean square error can be used to evaluate whether the second prediction model meets the requirements.

[0040] Step 6: Perform time series processing on the real-time monitoring data of rainfall and drainage ditch flow of the loess roadbed to construct a set of time lag feature sets. The time lag feature sets cover data including at least the current time and one or more previous time nodes. Then, select features from the time lag feature sets according to the feature selection principle of the first dataset determined in Step 2 to select or optimize the feature subset used for predicting the moisture content at a future time point.

[0041] Step 7: Input the feature subset into the first prediction model to obtain the predicted moisture content of the loess roadbed;

[0042] Step 8: Input the predicted moisture content and the real-time earth pressure monitoring data of the loess subgrade into the second prediction model to obtain the predicted deformation value of the loess subgrade.

[0043] Step 9: Issue an early warning based on the predicted deformation value. When the predicted deformation value exceeds the set value, an early warning message is issued. The set value is generally the deformation value when the loess undergoes a longitudinal abrupt change, which can be obtained from experiments.

[0044] Rainfall can be monitored using rainfall monitoring sensors, drainage ditch flow can be monitored using flow meters, and earth pressure can be monitored using earth pressure sensors.

[0045] like Figure 2As shown, in this embodiment of the invention, the rainfall sequence R(1) and the drainage ditch flow sequence D(1) are shifted forward by one time unit while keeping the moisture content sequence unchanged, to obtain the R(2) and D(2) sequences. This process is repeated until the R(N) and D(N) sequences are obtained. These sequences R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) constitute the lag feature set of rainfall and drainage ditch flow. In this embodiment of the invention, a day is used as a time unit. In some embodiments, an hour or two hours are also used as a time unit.

[0046] In this invention, the rainfall R1 and drainage ditch flow rate D1 at the initial time t1 will have a lag effect on the loess moisture content at subsequent times t2, t3, ..., tm. In this way, features R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) are constructed, that is, the number of lag features for rainfall and drainage ditch flow rate is determined. In this embodiment of the invention, m is 4.

[0047] In this embodiment of the invention, during the process of shifting the rainfall sequence and the drainage ditch flow sequence as a whole, it is assumed that there are three sets of data: Rt, Dt, and St, where Rt represents the rainfall sequence, Dt represents the drainage ditch flow sequence, and St represents the moisture content sequence. The rainfall and drainage ditch flow time series are shifted to the right by 1 to N units, and the missing data at the beginning of the lag feature time series is filled with the first data, while the missing data at the end is filled with the last data.

[0048] Correlation coefficient calculation between moisture content feature set and moisture content: The linear cross-correlation coefficients between loess moisture content data (St) and rainfall data (Rt), and between loess moisture content data and drainage ditch flow data (Dt), are calculated. The correlation operations are represented as RLS = (St, Rt) for rainfall and RLD = (St, Dt) for drainage ditch flow, respectively. The results will be two sequences: RLS and RLD, representing the correlation coefficients between loess moisture content and rainfall, and drainage ditch flow, at different lag times. The specific calculation of the correlation coefficients uses existing correlation coefficient calculation methods and will not be described in detail here.

[0049] In embodiments of the present invention, the correlation coefficient between rainfall and moisture content is calculated as follows:

[0050] The correlation coefficient between the rainfall and moisture content on that day was 0.2634;

[0051] The correlation coefficient between 1-day delayed rainfall and moisture content was 0.7344;

[0052] The correlation coefficient between 2-day delayed rainfall and moisture content was 0.2467;

[0053] The correlation coefficient between 3-day delayed rainfall and moisture content was -0.1916;

[0054] Analysis of the correlation coefficients reveals that the current rainfall is most relevant to predicting the moisture content of the next day. In other words, the rainfall data for the current day is the most important for predicting the moisture content of the next day. Therefore, if the highest correlation coefficient is used as the selection criterion, the rainfall data for the current day can be selected as the input for predicting the moisture content of the next day. The same method can be applied to predict the moisture content at other times. The analysis and selection of drainage ditch flow rate follows the same principle. This constructs the first dataset. Thus, in step six, when predicting the moisture content of the next day, only the current rainfall data needs to be selected based on the above selection principle.

[0055] The loess subgrade deformation early warning method of this invention considers the lagging effects of rainfall and drainage ditch flow on the moisture content of the loess subgrade. It constructs a moisture content feature set by progressively moving forward, and then, based on the correlation coefficient between the feature set and the moisture content (i.e., its importance), identifies features that will have a significant impact on a future time point and trains the first prediction model accordingly. This allows for more accurate and effective prediction of the loess subgrade moisture content. When predicting loess subgrade deformation for early warning, based on real-time monitoring data of rainfall and drainage ditch flow, as well as the correlation between the previous feature set and the moisture content and the importance of the features, it analyzes and selects a subset of features that will have a significant impact on a future time point or a period of time as input to the first prediction model. Then, based on real-time monitoring data of loess subgrade earth pressure, a second prediction model is used to predict the deformation of the loess subgrade at a future time point or a period of time. Finally, an early warning is issued based on the predicted deformation value. This allows for earlier warnings, resulting in better warning effectiveness, improved timeliness and efficiency of loess subgrade monitoring, more targeted road maintenance, reduced risk of sudden disasters, and enhanced safety and sustainability of highway and railway projects.

[0056] The first and second prediction models can be either machine learning models or artificial neural network models; in this invention, both adopt the XGboost model.

[0057] The present invention also provides a loess roadbed deformation early warning system, including a data acquisition module, a data processing module, a moisture content prediction module, a deformation prediction module and an early warning module;

[0058] The data acquisition module is used to collect rainfall, drainage ditch flow, and earth pressure data of the loess roadbed.

[0059] The data processing module is connected to the data acquisition module to perform time series processing on the collected loess roadbed rainfall data and drainage ditch flow data, construct a set of time lag feature sets, and then select features from the time lag feature sets based on the determined importance conclusions of the features, so as to select a feature subset for predicting the moisture content at a future time point.

[0060] The moisture content prediction module is connected to the data processing module to predict the moisture content of the loess subgrade based on the feature subset.

[0061] The deformation prediction module is connected to the moisture content prediction module and the data acquisition module to predict the deformation of the loess subgrade based on the predicted moisture content and the current earth pressure data.

[0062] The early warning module is connected to the deformation prediction module to issue an early warning based on the magnitude of the predicted deformation value. When the predicted deformation value exceeds a set value, an early warning message is issued.

[0063] This system can predict the deformation of loess subgrade at a future moment or over a period of time based on monitored rainfall, drainage ditch flow, and earth pressure data. Finally, it issues an early warning based on the magnitude of the predicted deformation. This allows for earlier warnings, resulting in better warning effectiveness and improving the timeliness and efficiency of loess subgrade monitoring, thereby enhancing the safety and sustainability of highway and railway projects.

Claims

1. A loess subgrade deformation early warning method, characterized in that, Includes the following steps: A time series dataset is constructed by measuring the rainfall, drainage ditch flow, and moisture content of the loess roadbed over a certain period of time. A moisture content feature set is constructed by progressively moving the rainfall and drainage ditch flow data of the loess roadbed in the time series dataset forward. The correlation coefficient between the moisture content feature set and the moisture content is then used to determine the number of lag features of rainfall and drainage ditch flow. The correlation coefficient is then analyzed to understand the degree of influence of rainfall and drainage ditch flow at different lag times on the moisture content and to determine the importance of the features. Features are then selected based on their importance, and these selected features are finally used as the first dataset. The first dataset is used to train a soil moisture content prediction model for loess roadbed until the preset conditions are met, so as to obtain the first prediction model. Based on geotechnical tests, data on the deformation of loess under corresponding earth pressure and water content were obtained, and a second dataset including earth pressure, water content and deformation was established. The second dataset was used to train a loess subgrade deformation prediction model until the requirements were met, in order to obtain the second prediction model. Time series processing is performed on the real-time monitoring data of rainfall and drainage ditch flow on the loess roadbed to construct a set of time lag feature sets. The time lag feature sets cover data including at least the current time and one or more previous time points. Then, features are selected from the time lag feature sets according to the selection principle of the first dataset to select a feature subset for predicting the moisture content at a future time point. The feature subset is input into the first prediction model to obtain the predicted moisture content of the loess roadbed; The predicted moisture content and the real-time earth pressure data of the loess subgrade are input into the second prediction model to obtain the predicted deformation value of the loess subgrade. Warnings are issued based on the predicted deformation value, and when the predicted deformation value exceeds a set value, a warning message is issued.

2. The method for early warning of loess roadbed deformation as described in claim 1, characterized in that: Both the first and second prediction models use the XGboost model.

3. The method for early warning of loess roadbed deformation as described in claim 1, characterized in that: The specific steps for constructing a moisture content feature set by progressively moving forward the rainfall and drainage ditch flow data of the loess roadbed are as follows: Keeping the moisture content sequence unchanged, the rainfall sequence R(1) and the drainage ditch flow sequence D(1) are shifted forward by one time unit to obtain the R(2) and D(2) sequences. This process is repeated until the R(N) and D(N) sequences are obtained. These sequences R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) constitute the lag feature set of rainfall and drainage ditch flow.

4. The method for early warning of loess roadbed deformation as described in claim 1, characterized in that: The rainfall R1 and drainage ditch flow D1 at the initial time t1 will have a lag effect on the loess moisture content at subsequent times t2, t3, ..., tm. In this way, the features R(1), R(2), ..., R(N) and D(1), D(2), ..., D(N) are constructed.

5. The method for early warning of loess roadbed deformation as described in claim 1, characterized in that: Assuming there are three sets of data when shifting the rainfall and drainage flow time series forward, Rt, Dt, and St, where Rt represents the rainfall series, Dt represents the drainage flow series, and St represents the moisture content series, the rainfall and drainage flow time series are shifted to the right by 1 to N units. The first data point fills the gaps in the lagging feature time series, and the last data point fills the gaps in the lagging feature time series.

6. The method for early warning of loess roadbed deformation as described in claim 1, characterized in that: The selection of features based on their importance involves choosing the feature with the highest correlation coefficient with the moisture content.

7. A loess subgrade deformation early warning system for use in the loess subgrade deformation early warning method as described in any one of claims 1 to 6, characterized in that, It includes a data acquisition module, a data processing module, a moisture content prediction module, a deformation prediction module, and an early warning module; The data acquisition module is used to collect rainfall, drainage ditch flow, and earth pressure data of the loess roadbed. The data processing module is connected to the data acquisition module to perform time series processing on the collected loess roadbed rainfall data and drainage ditch flow data, construct a set of time lag feature sets, and then select features from the time lag feature sets based on the determined importance conclusions of the features, so as to select a feature subset for predicting the moisture content at a future time point. The moisture content prediction module is connected to the data processing module to predict the moisture content of the loess subgrade based on the feature subset. The deformation prediction module is connected to the moisture content prediction module and the data acquisition module to predict the deformation of the loess subgrade based on the predicted moisture content and the current earth pressure data. The early warning module is connected to the deformation prediction module to issue an early warning based on the magnitude of the predicted deformation value. When the predicted deformation value exceeds a set value, an early warning message is issued.

Images