Bridge cofferdam construction safety assessment system and method based on data real-time analysis

Abstract

The invention relates to the technical field of bridge construction safety monitoring, discloses a bridge cofferdam construction safety evaluation system and method based on data real-time analysis, and solves the problems of scattered and heterogeneous monitoring data, inaccurate safety state prediction and long risk response time in the prior art. Specifically, multi-dimensional physical parameters of the cofferdam structure and the surrounding environment of the cofferdam structure are collected in real time through a multi-source sensor network; carrying out preprocessing and space-time alignment on the collected original data; inputting the preprocessed fusion data into a dynamic risk assessment model integrated with a random forest and a long-short-term memory network, and predicting a safety state evolution trend of the cofferdam structure in a future time period; and generating graded early warning information and disposal suggestions according to the prediction result, and pushing the information and suggestions to the site construction terminal. According to the technical scheme, fusion and space-time registration of monitoring data are realized, the perspectiveness and accuracy of safety state prediction are improved, the risk response time is shortened, and the intelligent level of engineering management is improved.

Description

Technical Field

[0001] This invention belongs to the field of bridge construction safety monitoring technology, specifically relating to a bridge cofferdam construction safety assessment system and method based on real-time data analysis. Background Technology

[0002] With the continuous advancement of infrastructure construction, bridge engineering, as a key node in transportation networks, has received increasing attention for its construction safety and quality control. Cofferdams, as temporary water-retaining structures in bridge foundation construction, play a crucial role in ensuring the dryness of the foundation pit and the safety of the working environment and construction personnel. Under complex hydrogeological conditions, cofferdam structures endure multiple forces, including water erosion, earth pressure, and construction loads, and their stability directly affects the safe progress of the entire bridge project. However, traditional methods for assessing cofferdam construction safety mainly rely on manual inspections, static monitoring, and experience-based judgment, which are insufficient to meet the demands of modern large-scale bridge engineering for high-precision and timely safety management.

[0003] Besides manual patrols and monitoring, existing technologies also have the following shortcomings:

[0004] First, the monitoring data is collected in a scattered manner and lacks a unified spatiotemporal benchmark, making it difficult to achieve effective alignment and fusion of multi-source heterogeneous data.

[0005] Secondly, safety assessment models are mostly based on static thresholds or simple statistical rules, which cannot dynamically adapt to the stress evolution characteristics of cofferdams at different construction stages.

[0006] Secondly, the risk warning mechanism is slow to respond, often triggering an alarm only after the structure has already shown obvious deformation or leakage, lacking forward-looking predictive capabilities.

[0007] Therefore, there is an urgent need for a real-time security assessment system and method that can achieve high timeliness, high accuracy, and adaptability. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a bridge cofferdam construction safety assessment system and method based on real-time data analysis, which can effectively solve the problems in the background technology. To achieve the above objective, this invention provides the following technical solution:

[0009] On the one hand, a bridge cofferdam construction safety assessment system based on real-time data analysis is provided, which includes the following components:

[0010] A multi-source sensor network module is used to collect multi-dimensional physical parameters of the cofferdam structure and its surrounding environment in real time. The multi-source sensor network module includes strain sensors, tilt sensors, and seepage pressure sensors deployed at key sections of the cofferdam, as well as water level gauges and flow meters deployed in the surrounding waters.

[0011] The data preprocessing and spatiotemporal alignment module is used to clean and filter the collected raw data, i.e., multidimensional physical parameters, and to realize coordinate transformation and time synchronization of multi-source heterogeneous data with a unified spatiotemporal reference.

[0012] The dynamic risk assessment model module is used to fused data after data preprocessing and spatiotemporal alignment. It predicts the evolution trend of the safety status of the cofferdam structure at different construction stages through an integrated learning algorithm. The dynamic risk assessment model module integrates a random forest regressor and a long short-term memory network.

[0013] The adaptive early warning decision module is used to dynamically adjust the early warning threshold based on the risk assessment results of the dynamic risk assessment model module, and generate tiered early warning decision information and handling feedback.

[0014] The construction management closed-loop feedback module is used to push early warning decision information to the on-site construction terminal and receive handling feedback to update model parameters.

[0015] Furthermore, the strain sensor employs fiber Bragg grating technology, and its center wavelength shift... With dependent variable Satisfying Relationships Among them, the strain sensitivity coefficient The value is 1.2 The tilt sensor is based on the MEMS accelerometer principle, with a measurement range covering ±15° and a resolution of 0.001°. The pressure sensor adopts the vibrating wire principle, with its frequency output having a linear relationship with the pore water pressure, a range of 0~1MPa, and an accuracy of 0.1%FS.

[0016] The strain sensor monitoring points are arranged in a 5-meter grid, with 8 measuring points arranged in each section, and the sampling frequency is set to 10Hz; the tilt sensor is installed at the top cap beam and middle support of the cofferdam, and spatial attitude is calculated through triaxial acceleration measurement; the seepage pressure sensor is buried at the contact surface of the soil inside and outside the cofferdam to monitor the distribution of seepage pressure.

[0017] Furthermore, the data preprocessing and spatiotemporal alignment module incorporates a moving average filtering algorithm and a Grubbs criterion outlier removal mechanism. The window width of the moving average filtering is adaptively adjusted according to the acquisition frequency. When the acquisition frequency is 10Hz, the window width is set to 50 data points. The spatiotemporal alignment process uses the Gauss-Kruger projection coordinate system as the spatial reference and the GPS timing signal as the time reference. The discrete monitoring point data is mapped to a unified grid node through bilinear interpolation.

[0018] The moving average filtering algorithm uses a weighted average method, with a weight of 0.6 for near-end data and 0.4 for far-end data; outlier identification uses the 3σ criterion, and data exceeding 3 times the standard deviation are considered outliers; the grid resolution of the spatiotemporal alignment process is set to 1 meter × 1 meter.

[0019] Furthermore, the random forest regressor uses historical strain, tilt, water level, and flow velocity data from the previous 24 hours as input features to predict the maximum displacement of the cofferdam within the next 2 hours. Its decision tree count is set to 100 trees, and its maximum depth is 15 layers. The long short-term memory network contains two hidden layers, each with 64 neurons. Its input sequence length is 60 time steps, and its output is the time series prediction value of the structural safety factor.

[0020] The node splitting criterion for the random forest regressor is the mean squared error, and the hyperparameters are determined through 10-fold cross-validation. The model training of the long short-term memory network adopts the adaptive moment estimation algorithm, with an initial learning rate of 0.001, a batch size of 32, and a training cycle of 1000 rounds.

[0021] Furthermore, the dynamic risk assessment model module integrates a random forest regressor and a long short-term memory network. The random forest regressor uses historical strain, tilt, water level, and flow velocity data from the previous 24 hours as input features to predict the maximum displacement of the cofferdam within the next 2 hours. Its decision trees are set to 100 trees with a maximum depth of 15 layers. The long short-term memory network contains two hidden layers, each with 64 neurons. Its input sequence length is 60 time steps, and its output is a time-series prediction of the structural safety factor.

[0022] Furthermore, the adaptive early warning decision module has a built-in three-level early warning mechanism. When the predicted safety factor is lower than 1.5 but higher than 1.2, a blue warning is triggered, and only a prompt message is sent to the monitoring center. When the predicted safety factor is between 1.2 and 1.0, a yellow warning is triggered, requiring enhanced monitoring and preparation of emergency resources. When the predicted safety factor is lower than 1.0, a red warning is triggered, and personnel evacuation and engineering rescue are immediately carried out. The early warning threshold is dynamically adjusted according to the type of cofferdam and the construction stage. The red warning threshold for steel sheet pile cofferdams is set to 1.05 during the excavation stage, while that for earth-rock cofferdams is set to 1.15 during the backfilling stage.

[0023] Blue alerts are sent to the monitoring center via SMS in JSON format; when a yellow alert is triggered, the monitoring frequency is increased from 10Hz to 20Hz, and an emergency resource allocation list is generated; when a red alert is triggered, an evacuation signal is issued via an audible and visual alarm device, which lasts for 3 minutes and repeats every 30 seconds.

[0024] Furthermore, the construction management closed-loop feedback module sends early warning instructions to the on-site smart terminal via 4G / 5G wireless communication protocol. The smart terminal integrates a QR code scanning function. Construction personnel can scan the QR code of the work order to confirm the receipt and execution status of the instructions. The module also records the response time and effect evaluation data, and updates the feature weight coefficients in the dynamic risk assessment model accordingly.

[0025] In the construction management closed-loop feedback module, the downlink adopts QPSK modulation and the transmission rate is not less than 100kbps; the response time is required to be no more than 5 minutes; the feature weight update cycle is 24 hours, and each update is calculated based on the historical data of the most recent 30 days.

[0026] On the other hand, a method for safety assessment of bridge cofferdam construction based on real-time data analysis is described, with the following specific steps:

[0027] S110 collects strain, tilt, and seepage pressure data of the cofferdam structure, as well as water level and flow velocity data of the surrounding environment in real time through a multi-source sensor network. The data acquisition cycle is 100 milliseconds, and the data reading and buffering of all sensors are completed within each cycle.

[0028] S120 performs moving average filtering and outlier removal on the collected raw data, and completes coordinate transformation and time synchronization of multi-source data based on a unified spatiotemporal benchmark. The filtering algorithm adopts a weighted average method, with a weight of 0.6 for near-end data and a weight of 0.4 for far-end data. Outlier identification adopts the 3σ criterion, and data exceeding 3 times the standard deviation are considered outliers.

[0029] S130, the preprocessed fused data is input into a dynamic risk assessment model that integrates random forest and long short-term memory network to predict the evolution trend of the safety status of the cofferdam structure in the future period. The input data is first normalized to map each feature value to the interval between 0 and 1. The model output is a safety coefficient prediction sequence of one point every 10 minutes in the next 2 hours.

[0030] S140. Based on the prediction results and dynamically adjusted warning thresholds, the risk level is determined, and graded warning information and corresponding handling suggestions are generated. The judgment logic adopts an if-else structure. First, it is determined whether a red warning is triggered, then a yellow warning, and finally a blue warning. Each warning level corresponds to a specific handling suggestion template.

[0031] S150 pushes early warning decision information to on-site construction terminals and updates the parameter configuration of the risk assessment model based on the handling feedback data. The push mechanism adopts a publish-subscribe model, with terminal devices subscribing to the early warning topics of their respective construction areas. The parameter update adopts an incremental learning method, which incorporates new experience data while retaining the original knowledge of the model.

[0032] Furthermore, in step S110, the strain sensor uses fiber optic grating technology, the tilt sensor is based on the MEMS accelerometer principle, the pressure sensor uses the vibrating wire principle, the water level gauge uses a pressure sensor, and the flow meter uses the acoustic Doppler principle.

[0033] Furthermore, in step S120, the spatiotemporal alignment process uses the Gauss-Kruger projection coordinate system as the spatial reference and the GPS timing signal as the time reference. The discrete monitoring point data is mapped to a unified grid node through bilinear interpolation, and the grid resolution is set to 1 meter × 1 meter.

[0034] Furthermore, in step S130, the random forest regressor uses the historical strain, tilt, water level, and flow velocity data of the previous 24 hours as input features to predict the maximum displacement of the cofferdam in the next 2 hours; the long short-term memory network contains two hidden layers, each with 64 neurons, and its input sequence length is 60 time steps, and the output is the time series prediction value of the structural safety factor.

[0035] In summary, this application includes at least one of the following beneficial technical effects:

[0036] 1. By using a data alignment mechanism between a multi-source sensor network and a unified spatiotemporal reference, deep fusion and high-precision spatiotemporal registration of cofferdam monitoring data were achieved, solving the problem of scattered and heterogeneous traditional monitoring data;

[0037] 2. The dynamic risk assessment model combining random forest and long short-term memory network can adapt to the stress evolution characteristics of the cofferdam at different construction stages, significantly improving the foresight and accuracy of safety status prediction;

[0038] 3. A three-level adaptive early warning mechanism based on the predicted safety factor was established, realizing the transformation from passive response to active intervention and effectively shortening the risk response time;

[0039] 4. A closed-loop feedback system for construction management was constructed, which directly links the evaluation results to the on-site handling process, forming a complete "perception-analysis-decision-feedback" control loop, which greatly improves the efficiency of safety measures and the level of intelligence in project management. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall technical architecture of the bridge cofferdam construction safety assessment system based on real-time data analysis proposed in this invention.

[0041] Figure 2 This is a schematic diagram of the core principle framework of the dynamic risk assessment model module in this invention;

[0042] Figure 3This is a schematic diagram of the adaptive early warning decision module in this invention. Detailed Implementation

[0043] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0044] Example 1

[0045] See Figure 1 This is a schematic diagram of the bridge cofferdam construction safety assessment system based on real-time data analysis proposed in this invention. The system includes a multi-source sensor network module, a data preprocessing and spatiotemporal alignment module, a dynamic risk assessment model module, an adaptive early warning decision module, and a construction management closed-loop feedback module.

[0046] The following sections will describe in detail the objects of the bridge cofferdam construction safety assessment system of the present invention:

[0047] In this embodiment, specifically, the multi-source sensor network module is used to collect multi-dimensional physical parameters of the cofferdam structure and its surrounding environment in real time. The multi-source sensor network module includes strain sensors, tilt sensors, and seepage pressure sensors deployed on key sections of the cofferdam, as well as water level gauges and flow velocity meters deployed in the surrounding waters.

[0048] The strain sensors deployed on the upstream and downstream sides of the cofferdam employ fiber Bragg grating technology, and their center wavelength drift is... With dependent variable Satisfying Relationships Among them, the strain sensitivity coefficient The value is 1.2 The strain sensor monitoring points are spaced 5 meters apart, with 8 measuring points arranged at each cross section, and the sampling frequency is set to 10Hz.

[0049] The tilt sensor is installed at the top cap beam and middle support of the cofferdam. Based on the principle of microelectromechanical systems (MEMS) accelerometers, its measurement range covers ±15 degrees, with a resolution of 0.001 degrees. Spatial attitude is calculated through triaxial acceleration measurement. The tilt sensor data undergoes coordinate transformation, converting the acceleration values ​​in the sensor coordinate system to the tilt angle values ​​in the geodetic coordinate system. The transformation matrix contains 3 rows and 3 columns of parameters.

[0050] The pressure sensor is embedded in the soil contact surface inside and outside the cofferdam. It operates on a vibrating wire principle, with its frequency output linearly related to the pore water pressure. The range is 0 to 1 MPa, and the accuracy is 0.1% of full scale. It is used to monitor the distribution of seepage pressure. The pressure sensor's frequency signal is converted into a pressure value through a calibration curve. The calibration curve contains 10 calibration points, with 30 sets of data collected at each point and the average value taken.

[0051] The water level gauges are deployed in the surrounding waters, specifically using pressure sensors, with a measurement range of 0 to 50 meters and an accuracy of ±1 cm.

[0052] The current meter uses the acoustic Doppler principle, with a measurement range of 0 to 5 meters per second and an accuracy of ±0.01 meters per second. Data from the water level gauge and current meter are transmitted wirelessly to the data processing center via 4G. The transmission protocol is TCP / IP, and the data packets contain fields such as timestamps, device numbers, and measured values, collectively forming a monitoring network for the steel sheet pile cofferdam structure and its surrounding hydrological environment.

[0053] The multi-source sensor network module collects multi-dimensional physical parameters of the cofferdam structure and its surrounding environment through the cooperation of various objects.

[0054] The data preprocessing and spatiotemporal alignment module cleans and filters the collected raw data, i.e., multidimensional physical parameters, and achieves coordinate transformation and time synchronization of multi-source heterogeneous data using a unified spatiotemporal benchmark. The module incorporates a moving average filtering algorithm and a Grubbs criterion outlier removal mechanism, as detailed below:

[0055] The data output by the strain sensor is first processed by moving average filtering. The window width of the moving average filter is adaptively adjusted according to the sampling frequency. When the sampling frequency is 10 Hz, the window width is set to 50 data points, that is, the data within a 5-second time window participates in the calculation. The filtered data is then used for outlier identification and removal by the Grubbs criterion, with the confidence level set to 95%.

[0056] The spatiotemporal alignment process using a unified spatiotemporal reference employs the Gauss-Kruger projection coordinate system as the spatial reference. The longitude of the central meridian of the projection zone is set as the longitude of the construction area's center. Discrete monitoring point data are mapped to unified grid nodes using bilinear interpolation, with a grid resolution of 1 meter × 1 meter. Time synchronization is based on GPS timing signals, and the time deviation between the built-in clocks of each sensor and the main control computer is controlled within 1 millisecond.

[0057] The data preprocessing and spatiotemporal alignment module can integrate the collected data and remove abnormal data to prevent erroneous data from interfering with subsequent analysis and prediction.

[0058] See Figure 2 This diagram illustrates the dynamic risk assessment model module, which uses an ensemble learning algorithm to predict the evolution trend of the safety status of the cofferdam structure at different construction stages. Within this module, a prediction model is constructed based on data preprocessed by the data preprocessing and spatiotemporal alignment modules. This module integrates two algorithms: a random forest regressor and a long short-term memory network.

[0059] The random forest regressor uses historical strain, tilt, water level, and flow velocity data from the previous 24 hours as input features. The feature dimension is 120-dimensional, including statistics such as the maximum, minimum, mean, and standard deviation of each monitoring point.

[0060] The number of decision trees was set to 100, with a maximum depth of 15 layers. The node splitting criterion was the mean squared error, and hyperparameters were determined through 10-fold cross-validation. The Long Short-Term Memory (LSTM) network contained two hidden layers, each with 64 neurons. The input sequence length was 60 time steps (6 hours of data), and the output was a time-series prediction of the safety factor of the cofferdam structure for the next 2 hours, with a prediction step size of 12 time steps.

[0061] The structural safety factor is a core quantitative indicator for evaluating the stability of a cofferdam. In this embodiment, the structural safety factor K mainly considers anti-sliding stability, and is defined as the ratio of the anti-sliding force (Fr) to the sliding force (Fs) of the cofferdam structure, i.e., K = Fr / Fs. The sliding force (Fs) is mainly determined based on the water pressure calculated from real-time monitoring data from water level gauges and flow velocity meters, and the earth pressure calculated in conjunction with soil parameters; the anti-sliding force (Fr) is calculated based on the weight of the cofferdam structure itself and the frictional resistance generated by the effective stress at the base, corrected by seepage pressure sensor data. During the model training phase, the historical safety factor values ​​calculated using the above formula serve as training labels for supervised learning of the Long Short-Term Memory network.

[0062] The Long Short-Term Memory (LSTM) network model was trained using the Adaptive Moment Estimation (IME) algorithm, with an initial learning rate of 0.001, a batch size of 32, and a training cycle of 1000 epochs. Model validation employed the hold-out method, with a training set to test set ratio of 8:2, and a required root mean square error (RMSE) of no more than 0.05.

[0063] The dynamic risk assessment model module predicts the evolution trend of the safety status of the cofferdam structure at different construction stages through an ensemble learning algorithm, facilitating timely analysis. Specifically, a random forest regressor and a long short-term memory network run in parallel, forming the dual-core prediction engine of this invention. The "structural safety coefficient" output by the long short-term memory network is the core basis for triggering the subsequent adaptive early warning decision module; while the "maximum displacement of the cofferdam" predicted by the random forest regressor serves as an important physical quantitative indicator for engineering managers, corroborating the safety coefficient and providing multi-dimensional decision support for risk assessment.

[0064] In the adaptive early warning decision module, the adaptive early warning decision module executes a graded response based on the prediction results of the dynamic risk assessment model module.

[0065] A blue alert is triggered when the predicted safety factor is below 1.5 but above 1.2. The blue alert information includes the monitoring point number, the predicted safety factor, and trend analysis. The alert is sent to the monitoring center staff via SMS. The information is in JSON format and includes four fields: time, location, level, and content.

[0066] When the predicted safety factor is between 1.2 and 1.0, a yellow alert is triggered. When a yellow alert is triggered, the system automatically starts the enhanced monitoring mode, and the monitoring frequency is increased from 10 Hz to 20 Hz. At the same time, an emergency resource allocation list is generated, which includes three types of resources: emergency materials, equipment, and personnel. Each type of resource is specified in terms of name, quantity, and storage location.

[0067] When the predicted safety factor is lower than 1.0, a red alert is triggered. When a red alert is triggered, the system immediately executes the personnel evacuation command and issues an evacuation signal through the sound and light alarm device. The signal lasts for 3 minutes and is repeated every 30 seconds. At the same time, the emergency rescue plan is activated, which includes specific details such as the assembly of the rescue team, equipment deployment, and technical solutions.

[0068] The warning threshold is dynamically adjusted according to the type of cofferdam and the construction stage. The red warning threshold for steel sheet pile cofferdams during the excavation stage is set to 1.05, and for earth-rock cofferdams during the backfilling stage it is set to 1.15. The correction coefficient is calculated based on a historical accident database containing 200 historical cases.

[0069] Reference Figure 3 The diagram shown is a schematic of the adaptive early warning decision module in the invention.

[0070] The construction management closed-loop feedback module enables full-process management of early warning information. This module transmits early warning commands to on-site intelligent terminals via 4G / 5G wireless communication protocols, with the downlink using QPSK modulation and a transmission rate of no less than 100kbps. The on-site intelligent terminals integrate QR code scanning functionality; construction personnel confirm command receipt by scanning the QR code on the work order. The QR code contains information such as the work order number, early warning level, handling requirements, and completion deadline.

[0071] The construction management closed-loop feedback module simultaneously records response time and effectiveness evaluation data, and uses this data to update the feature weight coefficients in the dynamic risk assessment model. Specifically, this update process is achieved by constructing a loss function related to the effectiveness of the response. For example, this loss function can be designed to output a large penalty value when the response time exceeds a preset threshold (e.g., 5 minutes), or when the structural state (e.g., displacement, seepage pressure) fails to recover to a safe level within a specified time after the warning is lifted. Subsequently, the system uses gradient descent to calculate the gradient based on this loss value and performs backpropagation and fine-tuning updates to the weight coefficients of the input features in the dynamic risk assessment model. In this way, the model can learn which feature combinations indicate more difficult-to-handle risks, thus giving these features higher attention in future predictions and achieving adaptive optimization of the model.

[0072] The response time is recorded to the second. The time difference between the issuance of the instruction and the confirmation by scanning the code is defined as the response time, and the standard response time requirement is no more than 5 minutes.

[0073] The effectiveness evaluation data includes indicators such as the implementation status of disposal measures, the trend of structural status changes, and the time of warning cancellation. This data is fed back to the dynamic risk assessment model through a feature weight update algorithm. The weight update cycle is 24 hours, and each update is calculated based on the historical data of the most recent 30 days. The feature weight coefficients are optimized using the gradient descent method, with a learning rate set to 0.01. The convergence condition is that the change in the loss function is less than 0.001.

[0074] In summary, this is a detailed description of each object in this system. Specifically, a multi-source sensor network is used to collect multi-dimensional physical parameters of the cofferdam structure and its surrounding environment in real time; the collected raw data is preprocessed and spatiotemporally aligned; the preprocessed fused data is input into a dynamic risk assessment model integrating random forest and long short-term memory networks to predict the evolution trend of the safety status of the cofferdam structure in future periods; based on the prediction results, graded early warning information and disposal suggestions are generated and pushed to the on-site construction terminal, thus achieving real-time and accurate risk assessment.

[0075] In addition, for the bridge cofferdam construction safety assessment method based on real-time data analysis in this application, the method steps are strictly executed in sequence during the specific implementation process.

[0076] Step S110: The strain, tilt, and seepage pressure data of the cofferdam structure, as well as the water level and flow velocity data of the surrounding environment, are collected in real time through a multi-source sensor network. The data acquisition cycle is 100 milliseconds, and the data reading and caching of all sensors are completed in each cycle.

[0077] Step S120: Perform moving average filtering and outlier removal on the collected raw data. The filtering algorithm adopts a weighted average method, with a weight of 0.6 for near-end data and a weight of 0.4 for far-end data. Outlier identification adopts the 3σ criterion, and data exceeding 3 times the standard deviation are considered outliers.

[0078] Step S130: Input the preprocessed fused data into the dynamic risk assessment model that integrates random forest and long short-term memory network. The input data is first normalized to map each feature value to the interval between 0 and 1. The model output is a safety factor prediction sequence for one point every 10 minutes in the next 2 hours.

[0079] Step S140: Determine the risk level based on the prediction results and the dynamically adjusted warning threshold. The determination logic adopts an if-else structure. First, determine whether a red warning is triggered, then a yellow warning, and finally a blue warning. Each warning level corresponds to a specific handling suggestion template.

[0080] Step S150: Push the early warning decision information to the on-site construction terminal. The push mechanism adopts a publish-subscribe mode. The terminal devices subscribe to the early warning topics of their respective construction areas. At the same time, the parameter configuration of the risk assessment model is updated based on the handling feedback data. The parameter update adopts an incremental learning method, which incorporates new experience data while retaining the original knowledge of the model.

[0081] Example 2

[0082] In the construction of bridge cofferdams in the estuary area, a multi-source sensor network module was specially configured to address the significant tidal effects of the environment. Strain sensors were densely deployed in the cofferdam area where tidal changes were drastic, with the spacing between measuring points reduced to 2 meters and the sampling frequency increased to 20 Hz to adapt to the rapid changes in tidal loads.

[0083] The tilt sensor incorporates a dynamic compensation algorithm to eliminate the influence of tidal-induced periodic oscillations on the structural tilt angle measurement. The compensation parameters are obtained by fitting continuous 72 hours of observation data.

[0084] The density of seepage pressure sensors was increased at the bottom of the cofferdam, with 12 measuring points arranged at each cross section to monitor the changes in seepage pressure gradient caused by tidal rise and fall.

[0085] The water level gauge adopts a dual redundancy design with a 50-meter interval between the main and backup water level gauges. When the main water level gauge fails, it automatically switches to the backup device to ensure the continuity of tidal data.

[0086] Three current meters were deployed upstream and downstream of the cofferdam. They captured changes in tidal flow direction using a Doppler velocity array, and the current velocity data was output at a frequency of 1 Hz.

[0087] The data preprocessing and spatiotemporal alignment module optimizes the processing flow for tidal environments. The strain data filtering window adaptively adjusts according to the tidal cycle, with a window width of 30 data points during high tide and 70 data points during low tide. Outlier removal uses a dynamic threshold method, with the threshold adjusted according to the tide level.

[0088] The tilt sensor data is compensated for by coordinate rotation, and the rotation matrix is ​​updated in real time according to the tidal direction, with an update frequency of once per minute.

[0089] The tidal level correction coefficient is introduced into the osmotic pressure sensor data. The correction formula is derived based on Darcy's law and includes parameters such as permeability coefficient and hydraulic gradient.

[0090] The tidal dominant frequency component was extracted from the water level and flow velocity data through spectral analysis. The dominant frequency was identified using Fast Fourier Transform with a frequency resolution of 0.001 Hz.

[0091] The spatiotemporal alignment process takes into account the Earth's curvature correction and introduces elevation anomaly correction based on the Gauss-Kruger projection. The correction amount is calculated using a global elevation model.

[0092] Time synchronization uses dual-mode BeiDou and GPS time service, automatically switching when a single signal fails, with time deviation controlled within 0.5 milliseconds.

[0093] The dynamic risk assessment model module adds tidal influence factors. The input features of the random forest regressor are expanded to 150 dimensions, and new tidal parameters such as tidal phase, tidal range, and tidal velocity are added. The weight of tidal phase in the feature importance ranking is set to 0.15. The hidden layers of the Long Short-Term Memory network are increased to 3 layers, with 128 neurons per layer. The length of the input sequence is extended to 120 time steps to cover the complete tidal cycle. A time decay factor is introduced into the model training, with the weight of recent training samples being 1.5 times that of distant samples, and the decay coefficient decreasing exponentially. The prediction output includes a safety factor for extreme tidal conditions, including special hydrological conditions such as spring tides and storm surges, and the prediction confidence interval is set to 95%.

[0094] The adaptive early warning decision module strengthens the tide-related early warning mechanism. The blue warning threshold is lowered by 0.05 during high tide and raised by 0.03 during low tide, with the adjustment based on the correlation analysis between tide level and structural response. After a yellow warning is triggered, in addition to routine measures, the frequency of tide level monitoring is increased, and the monitoring interval is shortened from 1 hour to 15 minutes. The red warning response plan includes a special tidal emergency plan, which details evacuation routes and the selection of rescue timing under different tide conditions. The dynamic correction of the warning threshold incorporates tidal forecast data, sourced from the National Marine Environmental Forecasting Center, with a forecast lead time of 72 hours and a forecast accuracy requirement of a tide level error not exceeding 0.2 meters.

[0095] The construction management closed-loop feedback module optimizes the feedback mechanism under tidal conditions. Tidal status indicators are added to early warning commands, including current tide level, tidal time, and tidal range. Tidal response guidelines are added to the QR codes on on-site smart terminals, providing specific tidal response measures for different early warning levels. Response time assessments differentiate between tidal conditions, requiring a response time of no more than 3 minutes during high tide and no more than 7 minutes during low tide. Tidal impact analysis is added to the effectiveness evaluation data, analyzing the correlation between tidal level changes and response effectiveness; the correlation coefficient is calculated using the Pearson correlation coefficient. The feature weight update algorithm incorporates tidal cycle matching to ensure the model's experience transfer under similar tidal phases.

[0096] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

[0097] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0098] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A bridge cofferdam construction safety evaluation system based on real-time data analysis, characterized in that, The system comprises the following components: A multi-source sensor network module for real-time collection of multi-dimensional physical parameters of the cofferdam structure and its surrounding environment, the multi-source sensor network module comprising strain sensors, tilt sensors, and osmotic pressure sensors deployed at key sections of the cofferdam, as well as water level gauges and current meters arranged in the surrounding water area; A data preprocessing and space-time alignment module for cleaning and filtering the collected raw data, i.e., multi-dimensional physical parameters, and realizing coordinate transformation and time synchronization of multi-source heterogeneous data based on a unified space-time reference; A dynamic risk assessment model module for fused data processed by the data preprocessing and space-time alignment module, which predicts the safety state evolution trend of the cofferdam structure at different construction stages through an ensemble learning algorithm, the dynamic risk assessment model module integrating a random forest regressor and a long short-term memory network; An adaptive early warning decision module for dynamically adjusting the early warning threshold based on the risk assessment results of the dynamic risk assessment model module, and generating graded early warning decision information and disposal feedback; A construction management closed-loop feedback module for pushing the early warning decision information to the on-site construction terminal and receiving disposal feedback to update the model parameters.

2. The bridge cofferdam construction safety assessment system based on real-time data analysis according to claim 1, characterized in that, The strain sensor adopts a fiber grating technology, and a center wavelength drift amount is proportional to a strain amount satisfies a relationship , wherein a strain sensitivity coefficient is 1.2 The tilt sensor is based on a MEMS accelerometer principle, a measurement range covers ±15°, a resolution reaches 0.001°, the osmotic pressure sensor adopts a vibrating string type principle, a frequency output is linear with a pore water pressure, a range is 0~1MPa, and an accuracy is 0.1%F.S. The strain sensor monitoring points are arranged at a 5-meter grid, with 8 measuring points arranged at each section, and the sampling frequency is set to 10 Hz; the tilt sensors are installed at the top crown beam and the middle support position of the cofferdam, and the spatial attitude solution is realized through three-axis acceleration measurement; the osmotic pressure sensors are buried in the contact surface between the inner and outer sides of the cofferdam to monitor the osmotic pressure distribution.

3. The bridge cofferdam construction safety assessment system based on real-time data analysis according to claim 1, characterized in that, The data preprocessing and space-time alignment module has a moving average filtering algorithm and a Grubbs criterion outlier rejection mechanism built in, the window width of the moving average filtering algorithm is adaptively adjusted according to the collection frequency, when the collection frequency is 10 Hz, the window width is set to 50 data points, the space-time alignment process uses the Gauss-Krueger projection coordinate system as the spatial reference, and the GPS time signal as the time reference, and the discrete monitoring point data is mapped to a unified grid node through a bilinear interpolation method; The moving average filtering algorithm uses a weighted average method, the near-end data weight is 0.6, and the far-end data weight is 0.4; the 3σ criterion is used for outlier identification, data exceeding 3 times the standard deviation is considered abnormal; the grid resolution of the space-time alignment process is set to 1 m x 1 m.

4. The bridge cofferdam construction safety assessment system based on real-time data analysis according to claim 1, characterized in that, The random forest regressor takes the historical strain, tilt, water level, and flow rate data of the previous 24 hours as input features to predict the maximum displacement of the cofferdam within the next 2 hours, the number of decision trees is set to 100, and the maximum depth is 15 layers, the long short-term memory network contains 2 hidden layers, each layer has 64 neurons, the input sequence length is 60 time steps, and the output is a time series prediction value of the structure safety coefficient; The node splitting criterion of the random forest regressor uses mean square error, and the hyperparameters are determined through 10-fold cross-validation; the model training of the long short-term memory network uses the adaptive moment estimation algorithm, the initial learning rate is 0.001, the batch size is set to 32, and the training period is 1000 rounds.

5. The bridge cofferdam construction safety assessment system based on real-time data analysis according to claim 1, characterized in that, The adaptive early warning decision module is internally provided with a three-level early warning mechanism, a blue early warning is triggered when the predicted safety coefficient is lower than 1.5 but higher than 1.2, only sending a prompt information to the monitoring center, a yellow early warning is triggered when the predicted safety coefficient is between 1.2 and 1.0, needing to start enhanced monitoring and prepare emergency resources, and a red early warning is triggered when the predicted safety coefficient is lower than 1.0, immediately executing personnel evacuation and engineering rescue, the early warning threshold is dynamically modified according to the cofferdam type and construction stage, the red early warning threshold of the steel sheet pile cofferdam in the excavation stage is set to 1.05, and the red early warning threshold of the earth-rock cofferdam in the backfill stage is set to 1.15; The blue early warning information is sent to the monitoring center through the short message platform, the information format is JSON structure; the monitoring frequency is increased from 10Hz to 20Hz after the yellow early warning is triggered, and an emergency resource allocation list is generated; the red early warning sends an evacuation signal through the sound and light alarm device, the signal duration is 3 minutes, and it is repeated once every 30 seconds.

6. The bridge cofferdam construction safety assessment system based on real-time data analysis according to claim 1, characterized in that, The construction management closed-loop feedback module sends the early warning instruction to the on-site intelligent terminal through the 4G / 5G wireless communication protocol, the intelligent terminal is integrated with a two-dimensional code scanning function, the construction personnel confirm the instruction receiving and execution state by scanning the disposal work order two-dimensional code, the module records the disposal response time and effect evaluation data at the same time, and updates the feature weight coefficient in the dynamic risk assessment model; In the construction management closed-loop feedback module, the downlink adopts QPSK modulation mode, and the transmission rate is not less than 100kbps; the disposal response time requirement is not more than 5 minutes; the feature weight updating period is 24 hours, and each update is based on the historical data of the recent 30 days.

7. A bridge cofferdam construction safety evaluation method based on real-time data analysis, characterized in that, The method comprises the following steps: S110, real-time acquisition of strain, inclination, and seepage pressure data of the cofferdam structure and water level and flow velocity data of the surrounding environment through a multi-source sensor network, the data acquisition period is 100 milliseconds, and data reading and caching of all sensors are completed in each period; S120, sliding average filtering and outlier rejection are performed on the collected raw data, and coordinate transformation and time synchronization of multi-source data are completed based on a unified space-time reference, the filtering algorithm adopts a weighted average method, the near-end data weight is 0.6, the far-end data weight is 0.4, and the 3σ criterion is used for outlier identification, data exceeding 3 times the standard deviation is regarded as abnormal; S130, the preprocessed fusion data is input into a dynamic risk assessment model integrated with a random forest and a long short-term memory network, the safety state evolution trend of the cofferdam structure in the future period is predicted, the input data is first normalized, and each feature value is mapped to the interval of 0 to 1, and the model output is a safety coefficient prediction sequence of every 10 minutes in the next 2 hours; S140, risk level determination is performed according to the prediction result and the dynamically adjusted early warning threshold, and graded early warning information and corresponding disposal suggestions are generated, the determination logic adopts an if-else structure, first, whether the red early warning is triggered is determined, then the yellow early warning is determined, and finally the blue early warning is determined, each early warning level corresponds to a specific disposal suggestion template; S150, push the early warning decision information to the field construction terminal, and update the parameter configuration of the risk assessment model based on the treatment feedback data. The pushing mechanism adopts the publish-subscribe mode. The terminal device subscribes to the early warning topics of the construction area it is responsible for. The parameter update adopts the incremental learning method. The new experience data is integrated on the basis of preserving the original knowledge of the model.

8. The bridge cofferdam construction safety assessment method based on real-time data analysis according to claim 7, characterized in that, In the step S110, the strain sensor adopts the fiber Bragg grating technology, the tilt sensor is based on the MEMS accelerometer principle, the osmotic pressure sensor adopts the vibrating string principle, the water level meter adopts the pressure type sensor, and the flow meter adopts the acoustic Doppler principle. 9.The bridge cofferdam construction safety assessment method based on real-time data analysis of claim 7, wherein, In the step S120, the space-time alignment process adopts the Gauss-Krueger projection coordinate system as the spatial reference and the GPS time signal as the time reference. The discrete monitoring point data is mapped to the unified grid node through the bilinear interpolation method. The grid resolution is set to 1m x 1m.

10. The bridge cofferdam construction safety assessment method based on real-time data analysis according to claim 7, characterized in that, In the step S130, the random forest regressor takes the historical strain, tilt, water level and flow rate data in the previous 24 hours as input features to predict the maximum displacement of the cofferdam in the next 2 hours. The long short-term memory network contains 2 hidden layers, each layer has 64 neurons, the input sequence length is 60 time steps, and the output is the time series prediction value of the structure safety factor.

Images