Intelligent identification and visual early warning method for construction risk of deepwater immersed tunnel

Abstract

The invention discloses a deepwater immersed tunnel construction risk intelligent identification and visual early warning method, which comprises the following steps: firstly, carrying out multi-modal sensor layout on a construction site, then carrying out data acquisition and processing, establishing a risk identification model, and establishing a grading early warning system. According to the deep water environment early warning system, the multi-mode sensor is introduced, key parameters such as the water pressure, the flow speed, the water temperature and the construction equipment state in the deep water environment can be monitored in real time, and a solid data foundation is provided for comprehensively sensing the construction environment. In combination with a machine learning algorithm, deep fusion of historical data and real-time data can be realized, and a dynamic risk assessment model is constructed, so that potential risks are predicted in real time. Furthermore, through the risk grade grading early warning system, the construction team can grasp the risk grade in time and take corresponding measures, so that the construction safety and efficiency are ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of deep - water immersed tunnel construction, and particularly relates to an intelligent risk identification and visualization warning method for deep - water immersed tunnel construction. Background Technique

[0002] An immersed tunnel is an important form of underwater passage construction and is widely used in major infrastructure projects such as cross - river and cross - sea projects. It forms a continuous underwater passage through the floating, sinking, and docking of precast pipe segments. Compared with bridges and shield tunnels, immersed tunnels have significant advantages such as short construction periods, strong adaptability, and relatively low construction costs, especially being prominent in wide - water - area crossing scenarios.

[0003] Traditional immersed tunnel construction mainly focuses on shallow - water areas, and the construction technology is relatively mature, and the risk control means are relatively perfect. However, with the increasing demand for deep - water environment construction, immersed tunnels are gradually expanding into the 100 - meter deep - water environment. This development not only expands the application scope of immersed tunnels but also brings comprehensive challenges in construction technology and risk management.

[0004] In the 100 - meter deep - water environment, the construction faces the following significant difficulties:

[0005] Complex hydrological conditions: In deep - water areas, the water flow velocity and direction change significantly, and the water temperature gradient and sediment content fluctuate frequently. These factors pose higher requirements for the stability of pipe - segment floating and the sinking accuracy. At the same time, this complexity also increases the unpredictability of construction risks.

[0006] High - water - pressure environment: The pressure at 100 - meter deep water is close to 10 atmospheres, which poses extremely high requirements for the pressure - resistant performance of construction materials, equipment, and personnel. High water pressure may cause problems such as deformation of the pipe - segment structure and failure of equipment seals, and at the same time pose a threat to the health and safety of underwater operators.

[0007] Adaptability of construction equipment and technology: Traditional construction equipment is difficult to adapt to the extreme conditions of the deep - water environment, and conventional technologies show obvious deficiencies in key links such as deep - water sinking, precise docking, and sealing treatment, and special designs and improvements are required for the deep - water environment.

[0008] In response to the special requirements of immersed tunnel construction under 100 - meter deep - water conditions, developing risk identification and warning technologies suitable for the deep - water environment is a key topic in the current field of deep - water immersed tunnel risk control. The innovation of this technology will provide technical support for deep - water construction and at the same time promote the development of immersed tunnel technology to a higher level. Summary of the Invention

[0009] To solve the above problems, the present invention discloses an intelligent risk identification and visual warning method for deep - water immersed tunnel construction, which can significantly improve the safety and efficiency of immersed tunnel construction under the condition of 100 - meter deep water. Through multi - modal sensor fusion technology and an intelligent warning system, the construction team can real - time master the environmental and equipment status, effectively avoid deep - water construction risks, and ensure the smooth progress of the project.

[0010] To achieve the above object, the technical solution of the present invention is as follows:

[0011] An intelligent risk identification and visual warning method for deep - water immersed tunnel construction, comprising the following steps:

[0012] S1. Layout multi - modal sensors in the deep - water area and high - pressure area at the construction site;

[0013] S2. Collect and process the construction process data of the immersed tunnel in the deep - water environment;

[0014] S3. Establish a multi - factor coupling risk identification model based on the construction of deep - water immersed tunnels;

[0015] S4. Establish a hierarchical warning system based on the model output;

[0016] S5. The warning information is synchronously pushed to on - site construction personnel and management personnel through effective channels such as acoustic - optical alarm devices set in the control room and other key positions, and mobile terminals issued to construction personnel;

[0017] S6. Establish a warning MR (Mixed Reality) visualization system under deep - water conditions.

[0018] Further, step S1 is specifically as follows:

[0019] Layout a multi - modal sensor network at the construction site, including:

[0020] Table 1 Sensor Name and Use

[0021]

[0022] The number and type of sensors can be determined as needed according to the site conditions.

[0023] Further, step S2 is specifically as follows:

[0024] The sensors collect the seabed environment data and key - point monitoring data during the construction process of the deep - water immersed tunnel in real - time, and transmit them to the ground control center through a high - precision underwater communication system. The ground control center conducts multi - modal analysis on the transmitted data, and combines the characteristics of data collection and long - distance transmission in the deep - water environment to perform noise removal and signal delay compensation. Subsequently, all data will be uniformly processed with time stamps to eliminate outliers and missing values, ensuring the accuracy and reliability of the data under deep - water conditions.

[0025] Further, step S3 is specifically as follows:

[0026] S3.1 Feature extraction:

[0027] Use data analysis methods to extract key features, including the instantaneous change rate of deep - water pressure, the fluctuation amplitude of water flow velocity, the change trend of the inclination angle of construction equipment, etc.

[0028] S3.2 Construction risk sub - model construction of deep - water immersed tunnel:

[0029] Environmental condition risk sub - model: Based on the parameters of deep - water pressure, flow velocity, water temperature, etc. collected by the aforementioned sensors, use the support vector machine (SVM) algorithm to establish the mapping relationship between hydrological conditions and risk levels;

[0030] Equipment operation risk sub - model: Use the convolutional neural network (CNN) to perform pattern recognition on equipment status data to evaluate whether the equipment vibration and inclination exceed the safe range;

[0031] S3.3 Model integration:

[0032] Through ensemble learning methods (such as random forest or weighted average method), integrate the outputs of each sub - model to generate a comprehensive risk score;

[0033] S3.4 Dynamic correction:

[0034] During the risk assessment process, use real - time monitoring data to adjust the model parameters to improve the prediction accuracy.

[0035] Further, step S4 is specifically as follows:

[0036] 1) Low risk (green): The construction environment is safe and no adjustment is required;

[0037] 2) Medium risk (yellow): Potential risks appear. It is recommended that the construction team take precautions;

[0038] 3) High risk (red): There are major potential safety hazards. Construction must be stopped immediately and the emergency plan must be activated.

[0039] Further, step S6 is specifically as follows:

[0040] Adopt the HoloLens 2 hardware platform to develop a customized MR (mixed reality) application, support gesture interaction to retrieve the risk analysis report, and real - time annotate information such as the monitoring situation of each risk point and the docking error of the pipe section. Superimpose the risk warning information on the three - dimensional model of the construction scene, and project the risk location, level and disposal suggestions in real - time through the Hololens 2 device. High - risk areas are highlighted in red and the emergency plan is automatically pushed.

[0041] The beneficial effects of the present invention are as follows:

[0042] 1. Introduce the multi-modal sensor fusion technology for deep water environment, which is used to collect and analyze various key parameters such as deep water pressure, flow velocity, water temperature, and construction equipment status in real time, providing a solid data basis for comprehensively perceiving potential construction risks in the deep water environment.

[0043] 2. Construct a multi-factor coupled risk assessment model based on machine learning, and through the combination of historical data and real-time monitoring data, realize the intelligent prediction of abnormal conditions during the construction process of deep water immersed tunnels.

[0044] 3. Develop a risk level classification and early warning system based on the output of the coupled model, and dynamically guide the construction plan according to the risk level, improving the safety and efficiency of construction.

[0045] 4. Develop a real-time visualization system for construction risks based on MR (Mixed Reality), improving the visualization level and early warning timeliness level of risk early warning monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of the work of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] The present invention will be further clarified below in conjunction with the drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0048] As shown in the figure, an intelligent identification and visualization early warning method for construction risks of a deep water immersed tunnel according to the present invention includes the following steps:

[0049] S1. Layout of multi-modal sensors: Arrange a multi-modal sensor network at the construction site, including:

[0050] 1) Water depth sensor, used to report the real-time water depth;

[0051] 2) Deep water pressure sensor, used to monitor the water pressure change in the construction area under the condition of 100-meter deep water;

[0052] 3) Deep water flow velocity sensor, used to measure the water flow velocity and direction at the monitoring point;

[0053] 4) Fiber optic sensor, used to monitor the stability of the soil layer around the construction area to prevent landslides or ground settlement;

[0054] 5) Displacement sensor, used to monitor the displacement, settlement, and deformation of the surrounding soil and rock strata;

[0055] 6) Stress and strain sensor, used to monitor the stress and strain conditions of the segments and steel structures of the deep water immersed tunnel;

[0056] 7) Fiber optic crack sensor, used to monitor potential cracks at key points of tunnel structures;

[0057] 8) Air quality sensor, used to monitor the content of hazardous gases in the tunnel;

[0058] 9) Temperature and humidity sensor, used to monitor the temperature and humidity in the tunnel;

[0059] 10) Deep water temperature sensor, used to measure the water temperature at external detection points of deep water;

[0060] 11) Water quality sensor, used to monitor the chemical components in the water outside the tunnel, especially indicators such as salinity, pH value, dissolved oxygen, etc.;

[0061] 12) Triaxial vibration sensor, used to monitor abnormal vibrations during construction;

[0062] 13) Gyroscope sensor, used to monitor the attitude during the installation of immersed tunnels;

[0063] 14) Equipment status sensor, used to monitor the working status of construction equipment and the usage of consumables;

[0064] 15) Marine meteorological sensor, used to monitor the meteorological conditions of the surrounding sea area in real time and give early warnings for adverse meteorological conditions.

[0065] S2. Data collection and processing: The sensors collect the seabed environment data and key point monitoring data during the construction of the deep - water immersed tunnel in real time, and transmit them to the ground control center through a high - precision underwater communication system. The ground control center conducts multi - modal analysis on the transmitted data. Combining the characteristics of data collection and long - distance transmission in the deep - water environment, it performs noise removal and signal delay compensation. Subsequently, all data will be processed with a unified time stamp to eliminate outliers and missing values, ensuring the accuracy and reliability of the data under deep - water conditions.

[0066] S3. Risk identification model: The detailed process of the risk identification model is as follows:

[0067] S3.1 Feature extraction:

[0068] Use data analysis methods to extract key features, including the instantaneous change rate of deep - water pressure, the fluctuation amplitude of water flow velocity, the change trend of the tilt angle of construction equipment, etc.

[0069] S3.2 Construction of two sub - models:

[0070] 1) Hydro - meteorological condition risk sub - model: Based on parameters such as deep - water pressure, flow velocity, water temperature, etc., use the support vector machine (SVM) algorithm to establish the mapping relationship between hydro - meteorological conditions and risk levels;

[0071] Implementation Process Analysis

[0072] 1. Data Preparation:

[0073] Format the hydrological condition data into three-dimensional inputs (pressure, flow velocity, water temperature) and output labels (risk level).

[0074] The data can be collected through on-site construction sensors and then undergo standardization and anomaly processing.

[0075] 2. SVM Model Construction:

[0076] Use the Radial Basis Function (RBF) kernel for non-linear classification.

[0077] The parameters C and gamma can be optimized through grid search to improve the model performance.

[0078] 3. Model Training and Testing:

[0079] The data is divided into a training set and a testing set, and the performance of the model is verified after training the model.

[0080] Output the accuracy rate and classification report to evaluate the classification ability of the model for different risk levels.

[0081] 4. Real-time Prediction:

[0082] After the model training is completed, it can be used for real-time prediction. Input the sensor data into the model and output the risk level in real-time.

[0083] Code Example:

[0084] from sklearn.svm import SVC

[0085] from sklearn.model_selection import train_test_split

[0086] from sklearn.metrics import accuracy_score,classification_report

[0087] import numpy as np

[0088] # Simulated data example (deep water pressure, flow velocity, water temperature and corresponding risk levels)

[0089] data=np.array(

[0090] [100,1.2,14,0],# Pressure 100 kPa, flow velocity 1.2 m / s, water temperature 14 °C, risk level is 0 (low risk)

[0091] [120, 1.5, 15, 0], # Pressure 120 kPa, flow rate 1.5 m / s, water temperature 15 °C, risk level is 0

[0092] [140, 1.8, 16, 0], # Pressure 140 kPa, flow rate 1.8 m / s, water temperature 16 °C, risk level is 0

[0093] [160, 2.1, 17, 1], # Pressure 160 kPa, flow rate 2.1 m / s, water temperature 17 °C, risk level is 1 (medium risk)

[0094] [180, 2.4, 18, 1], # Pressure 180 kPa, flow rate 2.4 m / s, water temperature 18 °C, risk level is 1

[0095] [200, 2.7, 19, 1], # Pressure 200 kPa, flow rate 2.7 m / s, water temperature 19 °C, risk level is 1

[0096] [220, 3.0, 20, 2], # Pressure 220 kPa, flow rate 3.0 m / s, water temperature 20 °C, risk level is 2 (high risk)

[0097] [240, 3.3, 21, 2], # Pressure 240 kPa, flow rate 3.3 m / s, water temperature 21 °C, risk level is 2

[0098] [260, 3.6, 22, 2] # Pressure 260 kPa, flow rate 3.6 m / s, water temperature 22 °C, risk level is 2 )

[0100] X = data[:, :3] # Feature variables: deep water pressure, flow rate, water temperature (for predicting risk level)

[0101] y = data[:, 3] # Target variable: risk level (0: low risk, 1: medium risk, 2: high risk)

[0102] # Divide the data into training set and test set, with the test set accounting for 20%, for validating the model performance

[0103] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 42)

[0104] # Create a support vector machine model

[0105] model = SVC(kernel='rbf', C=1.0, gamma='scale') # Use the radial basis function kernel, C is the penalty coefficient, and gamma is the kernel parameter

[0106] model.fit(X_train, y_train) # Train the model using the training set

[0107] # Make predictions using the test set

[0108] y_pred = model.predict(X_test) # y_pred is the risk level predicted by the model

[0109] # Calculate the accuracy and classification performance of the model

[0110] print("Model accuracy:", accuracy_score(y_test, y_pred)) # Output the accuracy on the test set

[0111] print("Classification report:\n", classification_report(y_test, y_pred)) # Output metrics such as recall and F1-score for each type of risk

[0112] # Use the model to predict the risk level of new data points

[0113] new_data = np.array([[210, 2.8, 19.5]]) # Example new data point: deep water pressure 210 kPa, flow velocity 2.8 m / s, water temperature 19.5 °C

[0114] predicted_risk = model.predict(new_data) # predicted_risk is the predicted risk level of the new data point

[0115] print(f"Predicted risk level of the new data point: {predicted_risk[0]}") # Output the prediction result

[0116] 2) Equipment operation risk sub-model: Use a convolutional neural network (CNN) to perform pattern recognition on equipment status data and evaluate whether equipment vibration and tilt exceed the safe range;

[0117] Implementation process analysis

[0118] 1. Data collection and preprocessing: Format the equipment status data into features and labels.

[0119] 2. Data splitting and format adjustment: Divide into training set and test set, and adjust to the input format supported by CNN

[0120] 3. Model construction: Design a CNN structure to extract device status features.

[0121] 4. Model training and evaluation: Train the model and verify its performance on the test set.

[0122] 5. Real-time prediction: Use new data points to predict the risk level.

[0123] Code example:

[0124] import tensorflow as tf

[0125] from tensorflow.keras import Sequential

[0126] from tensorflow.keras.layers import Conv1D, MaxPooling1D, Flatten, Dense, Dropout

[0127] import numpy as np

[0128] from sklearn.model_selection import train_test_split

[0129] # Simulated data: Device status (vibration, tilt angle, temperature) and risk level

[0130] # Data format: Vibration intensity, tilt angle, temperature -> Risk level (0: low risk, 1: medium risk, 2: high risk)

[0131] data = np.array(

[0132] [0.2, 1.0, 25.0, 0], # Vibration intensity 0.2g, tilt angle 1°, temperature 25°C, risk level 0 (low risk)

[0133] [0.5, 1.5, 26.0, 0], # Vibration intensity 0.5g, tilt angle 1.5°, temperature 26°C, risk level 0

[0134] [1.0, 2.0, 27.0, 1], # Vibration intensity 1.0g, tilt angle 2°, temperature 27°C, risk level 1 (medium risk)

[0135] [1.5, 3.0, 28.0, 1], # Vibration intensity 1.5g, tilt angle 3°, temperature 28°C, risk level 1

[0136] [2.0, 4.0, 30.0, 2], # Vibration intensity 2.0g, tilt angle 4°, temperature 30°C, risk level 2 (high risk)

[0137] [2.5, 5.0, 31.0, 2], # Vibration intensity 2.5g, tilt angle 5°, temperature 31°C, risk level 2

[0138] # Separate features and labels

[0139] X = data[:, :3] # Feature variables: vibration intensity, tilt angle, temperature

[0140] y = data[:, 3] # Target variable: risk level (0: low risk, 1: medium risk, 2: high risk)

[0141] # Split data into training set and test set

[0142] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 42)

[0143] # X_train, X_test: Feature data of the training set and test set respectively

[0144] # y_train, y_test: Label data of the training set and test set respectively

[0145] # Convert the input data into the input format of CNN (number of samples, length of time series, number of features)

[0146] # Assume the length of the time series is 1 (only detect the state at the current moment each time)

[0147] X_train = X_train.reshape(X_train.shape[0], 1, X_train.shape[1]) # Adjust the shape of the training set

[0148] X_test = X_test.reshape(X_test.shape[0], 1, X_test.shape[1]) # Adjust the shape of the test set

[0149] # Create a CNN model

[0150] model = Sequential(

[0151] Conv1D(filters=32, kernel_size=1, activation='relu', input_shape=(1, 3)), # Convolutional layer for feature extraction

[0152] MaxPooling1D(pool_size=1), # Max pooling layer for dimensionality reduction

[0153] Flatten(), # Flatten layer to map features to the fully connected layer

[0154] Dense(64, activation='relu'), # Fully connected layer for comprehensive processing of extracted features

[0155] Dropout(0.5), # Dropout layer to prevent overfitting

[0156] Dense(3, activation='softmax') # Output layer, three risk levels (0: low risk, 1: medium risk, 2: high risk) )

[0158] # Compile the model

[0159] model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) # optimizer: Optimizer (Adam), loss: Loss function (cross entropy), metrics: Evaluation metric (accuracy)

[0160] # Train the model

[0161] model.fit(X_train, y_train, epochs=10, batch_size=2, validation_data=(X_test, y_test))

[0162] # epochs: Number of training epochs, batch_size: Number of samples per training, validation_data: Validation dataset

[0163] # Test model prediction

[0164] new_data = np.array([[1.8, 3.5, 29.0]]) # New data point: Vibration 1.8g, tilt angle 3.5°, temperature 29°C

[0165] new_data = new_data.reshape(1, 1, 3) # Reshape to match CNN input format

[0166] predicted_risk = model.predict(new_data) # Use the model to predict the risk level of the new data point

[0167] print(f"Predicted risk level for the new data point: {np.argmax(predicted_risk)}") # Output the risk level

[0168] S3.3 Sub - model integration

[0169] Integrate the outputs of all sub - models through a weighted voting method, where the weights are determined by the performance of the sub - models (such as cross - validation accuracy).

[0170] Formula:

[0171]

[0172] Where:

[0173] P ensemble (i) is the predicted probability of the integrated model for risk level i;

[0174] w j is the weight of the j - th sub - model;

[0175] is the predicted probability of the j - th sub - model for risk level i.

[0176] Based on the integrated probability P ensemble Select the risk level with the highest probability as the final prediction result.

[0177] S3.4 Dynamic correction:

[0178] During the risk assessment process, use real - time monitoring data to adjust the model parameters to improve prediction accuracy.

[0179] S4. Hierarchical warning system

[0180] 1) Low risk (green): The construction environment is safe, no adjustment is required;

[0181] 2) Medium risk (yellow): Potential risks appear, it is recommended that the construction team take precautions;

[0182] 3) High risk (red): There are major safety hazards, construction must be stopped immediately and the emergency plan must be activated.

[0183] S5. The early warning information is synchronously pushed to on-site construction workers and management personnel through effective channels such as audible and visual alarm devices installed in the control room and other key locations, and mobile terminals issued to construction workers.

[0184] S6. Risk visualization system

[0185] Using the HoloLens 2 hardware platform, a customized MR (mixed reality) application is developed to support gesture interaction for retrieving risk analysis reports, and real-time annotation of information such as the monitoring status of each risk point and the docking error of pipe sections. The risk early warning information is superimposed on the three-dimensional model of the construction scenario, and the risk location, level, and disposal suggestions are projected in real time through the Hololens 2 device. High-risk areas are highlighted in red, and the emergency plan is automatically pushed.

[0186] In view of the special requirements of the construction of a 100-meter-deep immersed tube tunnel, the present invention innovatively proposes a risk identification and early warning method, which can significantly improve the safety and efficiency of the construction of an immersed tube tunnel under 100-meter-deep water conditions. Through the deep-water multi-modal sensor fusion technology and the intelligent early warning system, the construction team can grasp the environmental and equipment status in real time, effectively avoid the deep-water construction risks, ensure the smooth progress of the project, and provide important technical support for the construction of deep-water immersed tube tunnels.

[0187] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can still be made, and these improvements and refinements all fall within the protection scope of the claims of the present invention.

Claims

1. A method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks, characterized in that: The following steps are involved: S1. Install and debug the deepwater multi-modal sensor system at the construction site; S2. Data collection and processing during the construction of immersed tube tunnels in deep water environments; S3. Establish a multi-factor coupling risk identification model based on deep-water immersed tube tunnel construction; S4. Establish a hierarchical early warning system based on model output; S5. Early warning information is pushed to on-site construction personnel and management personnel simultaneously through the sound and light alarm devices installed in the control room and other key locations, and the mobile terminals distributed by construction personnel; S6. Establish an early warning MR visualization system under deep water conditions.

2. The method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks according to claim 1 is characterized by: Step S1 is specifically as follows: A deepwater multimodal sensor network is deployed at the construction site, including: (1) Water depth sensor, used to report real-time water depth; (2) Deepwater pressure sensor, used to monitor water pressure changes in the construction area under 100-meter deep water conditions; (3) Deep water flow velocity sensor, used to measure the water flow velocity and direction at the monitoring point; (4) Fiber optic sensors are used to monitor the stability of soil layers around the construction area to prevent landslides or ground subsidence; (5) Displacement sensors, used to monitor the displacement, settlement, and deformation of surrounding soil and rock formations; (6) Stress and strain sensors, used to monitor the stress and strain of deep-water immersed tunnel segments and steel structures; (7) Fiber optic crack sensors, used to monitor potential cracks at key points in the tunnel structure; (8) Air quality sensor, used to monitor the content of dangerous gases in the tunnel; (9) Temperature and humidity sensor, used to monitor the temperature and humidity in the tunnel; (10) Deep water temperature sensor, used to measure the water temperature of the deep water external detection point; (11) Water quality sensors, used to monitor the chemical composition of water outside the tunnel, especially salinity, pH value, and dissolved oxygen; (12) Three-axis vibration sensor, used to monitor abnormal vibration during construction; (13) Gyroscope sensor, used to monitor the posture of the immersed tunnel during installation; (14) Equipment status sensors, used to monitor the working status of construction equipment and the use of consumables; (15) Marine meteorological sensors are used to monitor the weather conditions in the surrounding sea areas in real time and issue early warnings for adverse weather conditions.

3. The method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks according to claim 1 is characterized by: Step S2 is as follows: after the sensor system collects data in real time, it transmits the data to the ground control center through the deepwater underwater communication system; the ground control center performs preliminary processing on the multimodal data to remove noise and compensate for signal delay; the data is processed with a unified timestamp to remove outliers and missing values ​​to ensure data quality.

4. The method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks according to claim 1 is characterized by: Step S3 is as follows: S3.1 Feature extraction: The key features of deepwater immersed tube tunnel construction risks are extracted using data analysis methods, including the instantaneous rate of change of deepwater pressure, the fluctuation amplitude of water flow velocity, and the trend of the inclination angle of construction equipment. S3.2 Sub-model construction: Hydrological condition risk sub-model: Based on the parameters collected in real time by the sensor system and processed in step S2, the support vector machine algorithm is used to establish the mapping relationship between hydrological conditions and risk levels; Equipment operation risk sub-model: Use convolutional neural networks to perform pattern recognition on equipment status data to assess whether equipment vibration and tilt exceed safety limits; S3.3 Model integration: Through ensemble learning methods, the outputs of each sub-model are combined to generate a comprehensive risk score; S3.4 Dynamic Correction: During the risk assessment process, real-time monitoring data is used to adjust model parameters to improve prediction accuracy.

5. The method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks according to claim 1 is characterized by: Step S4 is specifically as follows: 1) Low risk is marked in green: the construction environment is safe and no adjustments are required; 2) Medium risk is marked in yellow: potential risks have occurred and the construction team is advised to take precautions; 3) High risk is marked in red: there are major safety hazards, construction must be stopped immediately and the emergency plan must be activated.

6. The method for intelligent identification and visual early warning of deep-water immersed tunnel construction risks according to claim 1 is characterized by: The risk visualization system described in step S6 is specifically as follows: Using the HoloLens 2 hardware platform, a customized MR application is developed that supports gesture interaction to retrieve risk analysis reports and real-time annotation of monitoring information of each risk point. The risk warning information is superimposed on the three-dimensional model of the construction scene, and the risk location, level and disposal suggestions are projected in real time through the HoloLens 2 device. High-risk areas are highlighted in red, and emergency plans are automatically pushed.

Images