Dynamic prediction method for deformation joint of underground comprehensive pipe gallery

Abstract

The invention discloses a dynamic prediction method for a deformation joint of an underground comprehensive pipe gallery. The method comprises the following steps: S1, collecting and fusing multi-source monitoring data; s2, data preprocessing and feature engineering; s3, constructing a dynamic prediction model; s4, dynamic prediction and performance evaluation; s5, early warning and decision support are carried out, in the step S1, temperature field data including the air temperature in the pipe gallery, the concrete temperature (inner and outer surfaces) of the pipe gallery structure and the soil temperature outside the pipe gallery are acquired, and the measuring range ranges from-40 DEG C to 80 DEG C; according to the method, a multi-source monitoring data acquisition and dynamic prediction model system is established, continuous monitoring and trend prediction of the deformation joint of the underground comprehensive pipe gallery are achieved, multi-dimensional data such as the temperature field, the soil pressure and the underground water level are integrated, time sequence analysis is combined, and the deformation joint of the underground comprehensive pipe gallery is obtained. The development rule of the deformation joint can be accurately captured, and a prospective evaluation basis is provided for the safety of a pipe gallery structure.

Description

Technical Field

[0001] This invention relates to the technical field of dynamic prediction methods for expansion joints in utility tunnels, and particularly to a dynamic prediction method for expansion joints in underground integrated utility tunnels. Background Technology

[0002] Underground utility tunnels refer to structures and ancillary facilities located below the urban surface to accommodate two or more types of public utility pipelines or specialized pipelines. Public utility pipelines include power, communication (including monitoring lines), radio and television, water supply, drainage, heating, gas, fire protection pipelines, traffic signals, emergency optical cables, etc. Ancillary facilities include drainage, ventilation, lighting, electrical, communication, fire protection, safety monitoring systems, and monitoring and management rooms used to maintain the normal operation of the utility tunnel. Incorporating public utility pipelines into utility tunnels avoids the hassle of repeated road excavation due to the laying or maintenance of pipelines. At the same time, since the pipelines do not come into contact with soil and groundwater, soil corrosion of the pipelines is avoided, thus extending the service life of the pipelines. Currently, monitoring of expansion joints in underground utility tunnels mainly relies on regular manual inspections and static measurements. This method suffers from significant timeliness issues, failing to capture instantaneous changes and continuous development trends of expansion joints in a timely manner. Furthermore, existing monitoring methods often employ single-point data collection, making it difficult to comprehensively reflect the combined effects of multiple factors such as temperature changes, soil pressure fluctuations, and groundwater level rises and falls on expansion joints. In addition, traditional methods lack effective predictive analysis capabilities, failing to make forward-looking judgments on the future state of expansion joints. This results in a lag in the safety assessment of the utility tunnel structure, often leading to irreversible structural damage when abnormal situations occur. Therefore, this invention proposes a dynamic prediction method for expansion joints in underground utility tunnels to address the aforementioned problems. Summary of the Invention

[0003] Given the current technology, monitoring of expansion joints in underground utility tunnels mainly relies on regular manual inspections and static measurements. This method suffers from significant timeliness issues, failing to capture instantaneous changes and continuous development trends of expansion joints in a timely manner. Furthermore, existing monitoring methods often employ single-point data acquisition, making it difficult to comprehensively reflect the combined effects of multiple factors such as temperature changes, soil pressure fluctuations, and groundwater level rises and falls on expansion joints. In addition, traditional methods lack effective predictive analysis capabilities, failing to make forward-looking judgments on the future state of expansion joints. This results in a lag in the safety assessment of utility tunnel structures, often leading to irreversible structural damage when abnormal situations occur. Therefore, this invention proposes a dynamic prediction method for expansion joints in underground utility tunnels.

[0004] The present invention proposes a method for dynamic prediction of expansion joints in underground utility tunnels, comprising the following steps: S1: Multi-source monitoring data acquisition and fusion; S2: Data preprocessing and feature engineering; S3: Construct a dynamic prediction model; S4: Dynamic Prediction and Performance Evaluation; S5: Early warning and decision support.

[0005] Preferably, in step S1, the temperature field data are first measured: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃. Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 1-5 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 1-5 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 50Hz to 200Hz. Time factors: year, month, day, hour, etc., are used to capture long-term trends and cyclical patterns.

[0006] Furthermore, by collecting multi-source environmental parameters such as temperature field, soil pressure, groundwater level, and vibration data, a comprehensive monitoring data system is established. These parameters can reflect the comprehensive impact of the environment in which the utility tunnel structure is located, providing a sufficient data foundation for subsequent analysis and ensuring the integrity of the prediction model input.

[0007] Preferably, in step S1, the expansion joint response data (target variable) is: Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 1-5 times / hour. Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored. The measurement range is 10-30mm, and the sampling frequency is 1-5 times / hour.

[0008] Furthermore, by monitoring key indicators such as the change in expansion joint width and the amount of three-dimensional misalignment in real time, the actual response data of the utility tunnel structure can be obtained. These data directly reflect the working state of the expansion joint and provide reliable target variables for establishing an accurate prediction model.

[0009] Preferably, in step S2, missing and abnormal values ​​caused by sensor malfunctions are cleaned up; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 24-48 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns.

[0010] Furthermore, through data cleaning and standardization, outliers and dimensional differences in the monitoring data are eliminated. Then, by constructing lag features, sliding statistical features, and periodic features, the temporal patterns in the data are fully explored, providing high-quality feature inputs for the prediction model.

[0011] Preferably, in step S3, a combined model of long short-term memory network and attention mechanism is used. The model selection is LSTM+Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions.

[0012] Furthermore, a model architecture combining long short-term memory networks and attention mechanisms is adopted, which can capture long-term dependencies in time series and dynamically focus on data features at key time points, thereby improving the model's ability to model complex deformation patterns.

[0013] Preferably, in step S3, the model input and output are: Input: A time window containing all multi-source feature data from the past 30-60 days.

[0014] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. Furthermore, by setting reasonable input and output time windows, a mapping relationship is established between historical monitoring data and future prediction periods. Training sets, validation sets, and test sets are used for model training and validation respectively to ensure the model's generalization ability and prediction accuracy.

[0015] Preferably, in step S4, dynamic rolling prediction: after the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data, and then predict the state of the next step or the next few steps, so as to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination.

[0016] Furthermore, a dynamic rolling forecasting approach is adopted, which updates the forecast input in real time as new monitoring data is continuously acquired, so that the forecast results can reflect the latest structural state, and the forecast performance is comprehensively measured through multi-dimensional evaluation indicators.

[0017] Preferably, in step S5, the prediction result is compared with a preset safety threshold: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0018] Furthermore, by setting safety thresholds and a tiered early warning mechanism, the prediction results are transformed into specific operation and maintenance guidance. When the predicted value approaches the threshold, an early warning signal is issued in a timely manner, providing decision support for the safety management of the utility tunnel.

[0019] The beneficial effects of this invention are: 1. By establishing a multi-source monitoring data acquisition and dynamic prediction model system, continuous monitoring and trend prediction of deformation joints in underground integrated utility tunnels can be achieved. This method integrates multi-dimensional data such as temperature field, earth pressure, and groundwater level, and combines time series analysis to accurately capture the development law of deformation joints, providing a forward-looking assessment basis for the safety of utility tunnel structures. By collecting multi-source environmental parameters such as temperature field, earth pressure, groundwater level, and vibration data, a comprehensive monitoring data system is established. These parameters can reflect the comprehensive influence of the environment in which the utility tunnel structure is located, providing a sufficient data foundation for subsequent analysis and ensuring the integrity of the prediction model input. 2. By real-time monitoring of key indicators such as the change in expansion joint width and the amount of three-dimensional misalignment, the actual response data of the pipe gallery structure is obtained. These data directly reflect the working state of the expansion joint and provide reliable target variables for establishing an accurate prediction model. 3. Through data cleaning and standardization, outliers and dimensional differences in the monitoring data are eliminated. Then, by constructing lag features, sliding statistical features, and periodic features, the temporal patterns in the data are fully explored, providing high-quality feature inputs for the prediction model. The model architecture, which combines long short-term memory networks with attention mechanisms, can capture long-term dependencies in time series and dynamically focus on data features at key time points, thereby improving the model's ability to model complex deformation patterns. This invention establishes a multi-source monitoring data acquisition and dynamic prediction model system to achieve continuous monitoring and trend prediction of deformation joints in underground integrated utility tunnels. This method integrates multi-dimensional data such as temperature field, earth pressure, and groundwater level, and combines time series analysis to accurately capture the development law of deformation joints, providing a forward-looking assessment basis for the safety of utility tunnel structures. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the workflow proposed in this invention. Detailed Implementation

[0021] The present invention will be further explained below with reference to specific embodiments.

[0022] Reference Figure 1 Example 1 This embodiment proposes a dynamic prediction method for expansion joints in underground utility tunnels, including the following steps: S1: Multi-source monitoring data acquisition and fusion, firstly, temperature field data: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃; Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 3 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 4 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 150Hz; Time factors: year, month, day, hour, etc., used to capture long-term trends and cyclical patterns; Expansion joint response data (target variable): Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 2 times / hour; Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored, with a range of 10-30mm and a sampling frequency of 3 times / hour; S2: Data preprocessing and feature engineering, cleaning up missing and outlier values ​​caused by sensor malfunctions; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 24 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns; S3: Build a dynamic prediction model using a combination of long short-term memory network and attention mechanism. Model selection: LSTM + Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The Attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions. Model input and output: Input: A time window containing all multi-source feature data from the past 30 days.

[0023] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. S4: Dynamic prediction and performance evaluation, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data and then predict the state of the next step or the next few steps to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; S5: Early warning and decision support, comparing the prediction results with preset safety thresholds: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0024] Reference Figure 1 Example 2 This embodiment proposes a dynamic prediction method for expansion joints in underground utility tunnels, including the following steps: S1: Multi-source monitoring data acquisition and fusion, firstly, temperature field data: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃; Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 2 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 3 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 200Hz. Time factors: year, month, day, hour, etc., used to capture long-term trends and cyclical patterns; Expansion joint response data (target variable): Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 5 times / hour; Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored, with a range of 10-30mm and a sampling frequency of 3 times / hour; S2: Data preprocessing and feature engineering, cleaning up missing and outlier values ​​caused by sensor malfunctions; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 24-48 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns; S3: Build a dynamic prediction model using a combination of long short-term memory network and attention mechanism. Model selection: LSTM + Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The Attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions. Model input and output: Input: A time window containing all multi-source feature data from the past 60 days.

[0025] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. S4: Dynamic prediction and performance evaluation, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data and then predict the state of the next step or the next few steps to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; S5: Early warning and decision support, comparing the prediction results with preset safety thresholds: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0026] Reference Figure 1 Example 3 This embodiment proposes a dynamic prediction method for expansion joints in underground utility tunnels, including the following steps: S1: Multi-source monitoring data acquisition and fusion, firstly, temperature field data: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃; Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 3 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 2 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 50Hz; Time factors: year, month, day, hour, etc., used to capture long-term trends and cyclical patterns; Expansion joint response data (target variable): Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 2 times / hour; Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored, with a range of 10-30mm and a sampling frequency of once per hour; S2: Data preprocessing and feature engineering, cleaning up missing and outlier values ​​caused by sensor malfunctions; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 48 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns; S3: Build a dynamic prediction model using a combination of long short-term memory network and attention mechanism. Model selection: LSTM + Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The Attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions. Model input and output: Input: A time window containing all multi-source feature data from the past 60 days.

[0027] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. S4: Dynamic prediction and performance evaluation, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data and then predict the state of the next step or the next few steps to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; S5: Early warning and decision support, comparing the prediction results with preset safety thresholds: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0028] Reference Figure 1 Example 4 This embodiment proposes a dynamic prediction method for expansion joints in underground utility tunnels, including the following steps: S1: Multi-source monitoring data acquisition and fusion, firstly, temperature field data: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃; Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is once per hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is once per hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 50Hz; Time factors: year, month, day, hour, etc., used to capture long-term trends and cyclical patterns; Expansion joint response data (target variable): Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 2 times / hour; Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored, with a range of 10-30mm and a sampling frequency of 2 times / hour; S2: Data preprocessing and feature engineering, cleaning up missing and outlier values ​​caused by sensor malfunctions; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 48 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns; S3: Build a dynamic prediction model using a combination of long short-term memory network and attention mechanism. Model selection: LSTM + Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The Attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions. Model input and output: Input: A time window containing all multi-source feature data from the past 30 days.

[0029] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. S4: Dynamic prediction and performance evaluation, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data and then predict the state of the next step or the next few steps to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; S5: Early warning and decision support, comparing the prediction results with preset safety thresholds: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0030] Reference Figure 1 Example 5 This embodiment proposes a dynamic prediction method for expansion joints in underground utility tunnels, including the following steps: S1: Multi-source monitoring data acquisition and fusion, firstly, temperature field data: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃; Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 3 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 2 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 180Hz. Time factors: year, month, day, hour, etc., used to capture long-term trends and cyclical patterns; Expansion joint response data (target variable): Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 5 times / hour; Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored, with a range of 10-30mm and a sampling frequency of 5 times / hour; S2: Data preprocessing and feature engineering, cleaning up missing and outlier values ​​caused by sensor malfunctions; Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 24 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns; S3: Build a dynamic prediction model using a combination of long short-term memory network and attention mechanism. Model selection: LSTM + Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The Attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions. Model input and output: Input: A time window containing all multi-source feature data from the past 30 days.

[0031] Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters. S4: Dynamic prediction and performance evaluation, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data and then predict the state of the next step or the next few steps to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination; S5: Early warning and decision support, comparing the prediction results with preset safety thresholds: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

[0032] The following table compares the conventional dynamic prediction method for expansion joints with the dynamic prediction methods for expansion joints obtained in Examples 1 to 5:

[0033] As can be seen from the table above, the dynamic prediction method for deformation joints in underground integrated pipe corridors proposed in this invention has significant improvement, and implementation five is the best embodiment.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for dynamic prediction of expansion joints in underground utility tunnels, characterized in that, Includes the following steps: S1: Multi-source monitoring data acquisition and fusion; S2: Data preprocessing and feature engineering; S3: Construct a dynamic prediction model; S4: Dynamic Prediction and Performance Evaluation; S5: Early warning and decision support.

2. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S1, the temperature field data are first measured: air temperature inside the pipe gallery, concrete temperature of the pipe gallery structure (inner and outer surfaces), and soil temperature outside the gallery, with a range of -40℃ to +80℃. Earth pressure data: Earth pressure gauges are installed at the top, side walls and bottom of the utility tunnel to monitor changes in the surrounding soil pressure. The range is 0-5 MPa and the sampling frequency is 1-5 times / hour. Groundwater level data: Water level monitoring holes are set up around the pipe gallery. Water level changes will affect the effective stress and buoyancy of the soil. The range is 0-10m, and the sampling frequency is 1-5 times / hour. Vibration and ground motion data: For utility tunnels near major traffic arteries or seismically active areas, ground vibration acceleration is monitored, with a range of ±0.5g to ±2g and a sampling frequency of 50Hz to 200Hz. Time factors: year, month, day, hour, etc., are used to capture long-term trends and cyclical patterns.

3. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S1, the expansion joint response data (target variable) is as follows: Crack width variation: Real-time monitoring is performed using a high-precision displacement gauge or crack gauge, with a range of 0-50mm and a sampling frequency of 1-5 times / hour. Three-dimensional misalignment measurement: Using a three-dimensional joint gauge, the settlement, horizontal and vertical misalignment are monitored. The measurement range is 10-30mm, and the sampling frequency is 1-5 times / hour.

4. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S2, missing and abnormal values ​​caused by sensor malfunctions are cleaned up. Data standardization: Normalize data of different dimensions and magnitudes, such as temperature, pressure, and displacement, to the same scale to accelerate model convergence; Feature engineering: Lagged characteristics: Create lagged values ​​of the target variable and historical driving variables as inputs, with the target variable being the narrow gap and the driving variables being the problems of the past 24-48 hours; Sliding statistical characteristics: Calculate the mean, variance, maximum value, minimum value, etc. within the sliding window; Periodic features: Convert timestamps into periodic features to help the model capture daily, weekly, and yearly cyclical patterns.

5. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S3, a model combining a long short-term memory network and an attention mechanism is used. The model selected is LSTM+Attention. LSTM is specifically designed for processing time series data and can effectively capture long-term dependencies in the data. The attention mechanism allows the model to dynamically and selectively focus on the historical moments most relevant to the current prediction when making predictions.

6. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S3, the model input and output are as follows: Input: A time window containing all multi-source feature data from the past 30-60 days.

7. Output: For the next 7 days, the predicted value of the expansion joint width or misalignment. Model training: The dataset is divided into a training set, a validation set, and a test set; The LSTM-Attention model is trained using the training set data, and the error between the predicted value and the true value is minimized through the backpropagation algorithm. Use a validation set to monitor the training process, prevent overfitting, and adjust hyperparameters.

8. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S4, dynamic rolling prediction: After the model is deployed, the system will continuously receive new monitoring data and adopt a rolling prediction method, that is, update the input window with the latest actual data, and then predict the state of the next step or the next few steps, so as to achieve true dynamic and real-time prediction. Performance evaluation metrics: root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination.

9. The method for dynamic prediction of expansion joints in underground integrated utility tunnels according to claim 1, characterized in that, In step S5, the prediction result is compared with a preset safety threshold: Normal range: If the predicted value is within the safe threshold, the system displays normal operation; Warning: If the predicted value approaches the threshold at some point in the future, the system will issue a yellow warning to remind maintenance personnel to pay close attention. Alarm: The predicted value will soon exceed the threshold. The system issues a red alert and recommends intervention.

Images