Cyclone Foundation Cumulative Deformation Prediction System under Typhoon Load

Abstract

This invention discloses a system for predicting the cumulative deformation of a barrel foundation under typhoon load, comprising a data acquisition module, a core prediction module, and a result output module. The core prediction module employs a prediction framework based on dynamic data assimilation. Through an integrated ensemble Kalman filter algorithm, it iteratively fuses the physical model of the barrel foundation with real-time monitoring data, dynamically updating the model state and key soil parameters, thereby achieving adaptive and high-precision prediction of future cumulative deformation. Simultaneously, the system is supplemented by a multi-source data fusion mechanism that considers data quality to ensure the robustness of the prediction process. This invention effectively solves the technical problem that traditional methods struggle to accurately predict the nonlinear cumulative deformation of barrel foundations during typhoons in real time.

Description

Technical Field

[0001] This invention belongs to the field of marine engineering technology, specifically relating to a system for predicting the cumulative deformation of a barrel foundation under typhoon load. Background Technology

[0002] Offshore wind power plants, offshore platforms, and other marine engineering structures face severe challenges in typhoon-prone areas. Barrel foundations, a common foundation type for these structures, experience significant cumulative deformation under the complex cyclic loads of strong typhoons, directly impacting the overall structural safety and service life. Therefore, developing a method and system capable of accurately predicting the cumulative deformation of barrel foundations during typhoons is of great engineering importance for achieving early warning of structural safety and guiding operation and maintenance decisions.

[0003] Currently, existing technologies in this field suffer from the following shortcomings: First, most methods rely on offline finite element simulations after typhoons or estimations based on simplified empirical formulas, failing to provide real-time prediction and early warning during typhoons. Second, existing real-time monitoring systems are often limited to perceiving the current structural state, lacking the ability to predict future deformation trends, and generally do not fully consider the softening effect of soil strength and stiffness under cyclic loading, resulting in significant long-term prediction deviations. Third, some prediction methods that attempt to combine physical models with data-driven approaches typically employ simple linear superposition or static calibration strategies, which struggle to effectively handle the strong nonlinear and time-varying characteristics between typhoon loads and foundation responses. Fixed model parameters cannot be dynamically adjusted based on real-time observation data, leading to model mismatch when load conditions change abruptly, resulting in insufficient prediction accuracy and adaptability.

[0004] Therefore, a prediction system for the cumulative deformation of a barrel foundation under typhoon load is proposed to address the above-mentioned problems. This system solves the key issues in the existing technology, namely, how to achieve real-time and high-precision prediction of the cumulative deformation of a barrel foundation under typhoon load, and overcomes the shortcomings of insufficient model prediction ability and poor adaptability caused by the nonlinear behavior of soil and the time-varying nature of load. Summary of the Invention

[0005] This invention provides a system for predicting the cumulative deformation of a barrel foundation under typhoon loads to solve the aforementioned technical problems. Specifically, the technical solution is as follows: A system for predicting the cumulative deformation of a barrel foundation under typhoon load includes: The data acquisition module is used to acquire real-time data on strain, displacement, wind speed, and wave height of the barrel foundation under typhoon load. The core prediction module predicts cumulative deformation by executing a dynamic update process for the model state. This process includes: establishing a basic physical model describing the interaction between the bucket and the soil, wherein the state vector of the basic physical model includes the horizontal displacement and vertical settlement of the bucket and memory variables characterizing the cyclic softening effect of the soil; using the basic physical model, predicting the system state at the current moment based on the state estimate value at the previous moment; comparing the measured data at the current moment with the corresponding predicted value, and adjusting the system state estimate value in reverse according to the difference and the preset update rules, thereby outputting the cumulative deformation prediction result at the current moment. The results output module is used to graphically display the cumulative deformation prediction results, including the cumulative deformation development curve within a predetermined time period, the deformation prediction range, and warning information triggered based on a preset threshold.

[0006] Furthermore, the specific process for dynamically updating the model state is as follows: Maintain a set of system state vectors, with a size of 50–200; In each prediction period, the temporal evolution of all state vectors in the set is first advanced based on the basic physical model to obtain the state prediction set; Subsequently, the measurement data at the current moment is compared with the observation forecast values ​​obtained by mapping from the state forecast set. By calculating the statistical correlation between the state forecast set and the measurement data, a state analysis set is generated. The mean of the state analysis set is used as the optimized system state estimate.

[0007] Furthermore, the basic physical model describes the dynamic behavior of the barrel foundation through a set of differential equations, wherein the characteristic of soil resistance decreasing with the number of cyclic loading is characterized by the evolution equation of the memory variable, and the evolution rate of the memory variable is controlled by two adjustable parameters.

[0008] Furthermore, the two adjustable parameters used to regulate the evolution rate of the memory variables are extended to the system state vector and are simultaneously corrected in real time with the displacement and settlement variables during the model state dynamic update process.

[0009] Furthermore, before using the measurement data for state updates, a data optimization step is performed: dynamic weight values ​​are assigned to different types of measurement data, the calculation of which takes into account both the instantaneous signal-to-noise ratio estimate of the data source and its deviation from the long-term observation average level of the system.

[0010] Furthermore, the cumulative deformation prediction system for barrel foundations under typhoon load is also configured to: monitor the stability of data inflow and the utilization rate of computing resources in real time; when abnormal data flow or system load exceeds a preset threshold is detected, automatically simplify the calculation method of dynamic weights and adopt a state update strategy with higher computational efficiency.

[0011] Furthermore, the data acquisition module includes: a fiber optic strain sensor array deployed at key locations on the barrel structure, a GNSS displacement monitoring unit installed on the top of the barrel, an ultrasonic anemometer set on the top of the platform, and a wave radar located near the foundation.

[0012] Furthermore, the result output module displays the prediction range with a 90% confidence level as the default setting, and the warning information includes a primary warning triggered when the primary threshold is exceeded and an emergency warning triggered when a higher-level threshold is exceeded.

[0013] Furthermore, the cumulative deformation prediction system for barrel foundations under typhoon load also integrates an offline learning unit. The learning unit is automatically activated during non-real-time prediction periods, calls up the stored historical typhoon event dataset, and performs inversion and calibration of key empirical coefficients in the basic physical model by minimizing the prediction error function.

[0014] Furthermore, the core prediction module is deployed entirely on an edge computing device, which is connected to the data acquisition module via an industrial Ethernet and is equipped with processor and memory resources that meet the requirements of real-time prediction.

[0015] The advantage of this invention lies in the provision of a cumulative deformation prediction system for barrel foundations under typhoon loads. This system constructs a real-time prediction framework based on dynamic data assimilation, combining a physical model describing the barrel-soil interaction with real-time observation data from sensors. By iteratively executing model prediction and observation updates, and utilizing an ensemble Kalman filter algorithm to continuously correct the model state and key parameters, the physical model can adaptively track the dynamic response of the structure under real typhoon loads. This approach effectively overcomes the mismatch problem of traditional fixed-parameter models under complex nonlinear conditions, significantly improving the adaptability and reliability of the prediction model throughout the typhoon process, and providing a more accurate and timely decision-making basis for structural safety assessment.

[0016] The advantages of this invention also lie in the provided system for predicting the cumulative deformation of barrel foundations under typhoon loads. By introducing a multi-source data fusion mechanism that considers both data quality and system load, the system dynamically assigns weights to measured data based on their instantaneous signal-to-noise ratio and deviation from the system's normal state before inputting the measured data into the prediction model. During operation, this mechanism automatically selects high-reliability data and adaptively simplifies the fusion strategy to ensure the stable operation of the core prediction task when the system detects data flow anomalies or computational resource constraints. This not only improves data utilization efficiency and the robustness of the assimilation process but also ensures the continued effectiveness and practicality of the prediction system under severe typhoon conditions and limited computational resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the cumulative deformation prediction system for barrel foundations under typhoon loads proposed in this application. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0020] In the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection" and "linkage" should be interpreted broadly, and can refer to mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0021] The accompanying drawings of the embodiments disclosed in this invention only involve structures relevant to the embodiments disclosed in this invention. Other structures can be referred to with common designs. Unless otherwise specified, the same embodiment and different embodiments of this invention can be combined with each other.

[0022] like Figure 1 The image shows a cumulative deformation prediction system for a barrel foundation under typhoon load according to this application, which includes: a data acquisition module, a core prediction module, and a result output module.

[0023] The data acquisition module is used to acquire real-time data on strain, displacement, wind speed, and wave height of the bucket foundation under typhoon loads, preferably with a sampling frequency of no less than 1Hz. The core prediction module achieves cumulative deformation prediction by executing a dynamic model state update process. This process includes: establishing a basic physical model describing the interaction between the bucket and the soil; the state vector of the basic physical model includes the horizontal displacement and vertical settlement of the bucket, as well as memory variables characterizing the cyclic softening effect of the soil; using the basic physical model, predicting the system state at the current moment based on the state estimate from the previous moment; comparing the measured data at the current moment with the corresponding predicted value, and adjusting the system state estimate in reverse according to the differences and preset update rules, thereby outputting the cumulative deformation prediction result for the current moment. The result output module is used to graphically display the cumulative deformation prediction result, including the cumulative deformation development curve over a preset future time period, the predicted deformation range, and warning information triggered based on a preset threshold. This preset time period can be set as needed, such as 72 hours.

[0024] In the embodiments of this application, the dynamic update process of the model state is specifically as follows: Maintain a set of system state vectors, with a size of 50–200.

[0025] In each prediction period, the temporal evolution of all state vectors in the set is first advanced based on the fundamental physical model to obtain the state prediction set.

[0026] Subsequently, the measurement data at the current moment is compared with the observation forecast values ​​obtained by mapping from the state forecast set. By calculating the statistical correlation between the state forecast set and the measurement data, a state analysis set is generated.

[0027] The mean of the state analysis set is used as the optimized system state estimate.

[0028] In the embodiments of this application, the basic physical model describes the dynamic behavior of the barrel foundation through a set of differential equations, wherein the characteristic of soil resistance decreasing with the number of cyclic loading is characterized by the evolution equation of the memory variable, and the evolution rate of the memory variable is controlled by two adjustable parameters.

[0029] In the embodiments of this application, two adjustable parameters used to regulate the evolution rate of memory variables are extended to the system state vector and are synchronously corrected in real time with displacement and settlement variables during the dynamic update process of the model state.

[0030] In an embodiment of this application, before using measurement data for state updates, a data optimization step is performed: assigning dynamic weight values ​​to different types of measurement data, the calculation of which takes into account both the instantaneous signal-to-noise ratio estimate of the data source and its deviation from the long-term observation average level of the system.

[0031] In the embodiments of this application, the cumulative deformation prediction system for barrel foundations under typhoon load is also configured to: monitor the stability of data inflow and the utilization rate of computing resources in real time; when abnormal data flow or system load exceeds a preset threshold is detected, automatically simplify the calculation method of dynamic weights and adopt a state update strategy with higher computational efficiency.

[0032] In the embodiments of this application, the data acquisition module includes: a fiber optic strain sensor array deployed at key locations of the barrel structure, a GNSS displacement monitoring unit installed on the top of the barrel, an ultrasonic anemometer set on the top of the platform, and a wave radar located near the foundation.

[0033] In the embodiments of this application, the result output module displays the prediction range with a 90% confidence level as the default setting, and the warning information includes a primary warning triggered when the primary threshold is exceeded and an emergency warning triggered when a higher-level threshold is exceeded.

[0034] In the embodiments of this application, the cumulative deformation prediction system for barrel foundations under typhoon load also integrates an offline learning unit. The learning unit is automatically activated during non-real-time prediction periods, calls up the stored historical typhoon event dataset, and performs inversion and calibration of key empirical coefficients in the basic physical model by minimizing the prediction error function.

[0035] In the embodiments of this application, the core prediction module is deployed on an edge computing device. The edge computing device is connected to the data acquisition module via an industrial Ethernet and is equipped with processor and memory resources that meet the requirements of real-time prediction.

[0036] This system achieves accurate prediction of the cumulative deformation of barrel foundations through the collaborative work of three core modules. The structure and principle of each module of the barrel foundation cumulative deformation prediction system under typhoon load of this application are described in detail below.

[0037] The data acquisition module adopts a distributed sensor network, including 16 fiber optic strain sensors, 2 sets of GNSS displacement monitoring units, 1 three-dimensional ultrasonic anemometer, and 1 set of wave radar.

[0038] These sensors operate at different sampling frequencies: 10Hz for strain data, 1Hz for displacement data, and 2Hz for environmental data. All data is transmitted via industrial Ethernet in a uniform binary data format, with each data packet containing a 4-byte timestamp, a 2-byte sensor ID, a 1-byte quality identifier, and a 4-byte data value.

[0039] The core prediction module constructs an eight-dimensional state vector. The state vector is strictly limited to eight dimensions, with γ treated as a fixed parameter and not involved in state estimation, ensuring system dimensionality consistency and reducing computational complexity. Here, x represents horizontal displacement (m), dx / dt represents horizontal velocity (m / s), θ represents the barrel rotation angle (rad), dθ / dt represents angular velocity (rad / s), s represents vertical settlement (m), M represents soil memory variables (dimensionless), and α and β represent empirical soil parameters (dimensionless).

[0040] This module executes a complete forecast-update cycle every 10 seconds: first, it performs state forecasting based on a nonlinear dynamic model, and then uses the fourth-order Runge-Kutta method to solve the differential equations, with a time step of Δt=10s; Then, using real-time observation data, the state vector is optimally estimated through an ensemble Kalman filter algorithm, and the cumulative deformation prediction value for the next 72 hours is finally output with a time resolution of 10 minutes.

[0041] Two time scales need to be clearly distinguished: the basic calculation time step Δt=10s is used for fourth-order Runge-Kutta numerical integration calculations. This shorter step ensures the numerical stability and accuracy of solving differential equations; the prediction output time resolution of 10min is used for the final result output. This longer interval balances the real-time requirements with the burden of data storage and transmission, while providing a reasonable time granularity for engineering decisions.

[0042] The results output module provides prediction services through a RESTful API interface with a response time of less than 2 seconds. The output data is in JSON format and includes the predicted time series, 90% confidence interval, and warning level. It also provides a web graphical display interface to display the deformation development trend curve and warning information in real time.

[0043] Furthermore, the implementation of the ensemble Kalman filter algorithm includes a complete initialization, prediction, and update process. In the initialization phase, a set of 120 state vectors is constructed. Each vector is generated by adding a Gaussian random perturbation to the initial state estimate. The standard deviations of the perturbations for each variable are carefully designed: displacement 10 mm, velocity 5 mm / s, rotation angle 0.01 rad, angular velocity 0.005 rad / s, settlement 5 mm, memory variable 0.1, parameters α = 0.05, and β = 0.005.

[0044] The prediction phase processes each state vector independently, using the fourth-order Runge-Kutta method to solve the system of differential equations. Four calculation phases are completed within each 10-second time step. First, calculate the slope. ; Then calculate at half step size ; Calculate again at half a step length ; Finally, calculate at the full step length. ; The final forecast value is obtained by weighted averaging. .

[0045] During the update phase, when new observation data arrives, the statistical characteristics of the state forecast ensemble are first calculated, including the mean x. f Covariance P f Then calculate the Kalman gain. , where P xy P is the cross-covariance of state and observation. yy Add the observation error covariance R to the observation forecast covariance.

[0046] Anomaly detection mechanism calculates Mahalanobis distance in real time. When D M Exceed A robust update strategy is initiated when the distribution reaches the 95th percentile, where y obs This is the current observation data vector, containing the latest measurements from all sensors; y f D is the mean vector of observed and predicted values, obtained by averaging the observed and predicted sets; M The calculated Mahalanobis distance statistic is used to measure the degree of statistical difference between observed data and forecast values; The 95th percentile of the distribution is determined based on the dimension of the observed data, when D M Exceeding this threshold indicates a significant discrepancy between the current observation data and the model forecast, which may be due to sensor anomalies or model mismatch.

[0047] Furthermore, the nonlinear dynamic model precisely describes the mechanical behavior of the barrel foundation through a coupled system of differential equations. Displacement equations In this context, m represents the equivalent mass (kg) of the foundation and superstructure, calculated using structural design parameters; c represents the damping coefficient (N·s / m), characterizing the energy dissipation capacity of the foundation when moving within the surrounding soil; k represents the initial stiffness coefficient (N / m), reflecting the resistance of undamaged soil to foundation deformation; α is the stiffness attenuation coefficient, controlling the influence of memory variables on stiffness; and M is the soil memory variable, characterizing the degree of damage to the soil due to accumulated cyclic loading.

[0048] Wave load F wave Calculated using the Morrison equation: , where ρ w Seawater density (kg / m³) 3 ), C d U is the drag force coefficient, D is the diameter of the barrel (m), and u is the velocity of water particles (m / s).

[0049] Wind load F windCalculated using the wind pressure formula: , where ρ a air density (kg / m³) 3 ), C wind Where A is the wind force coefficient and A is the windward area (m²). 2 V is the wind speed (m / s).

[0050] Evolutionary equations of memory variables In this model, β controls the decay rate of the memory variable, reflecting the sensitivity of soil damage accumulation; γ determines the nonlinear characteristics of the decay process, influencing the damage development pattern.

[0051] The model is solved using the fourth-order Runge-Kutta method, which completes four computational stages within each time step to ensure the accuracy and stability of the numerical solution.

[0052] Furthermore, dynamic calibration of empirical parameters achieves complete parameter optimization through a state augmentation method. First, an augmented state vector is constructed. The parameter vector to be calibrated Combined with the physical state vector.

[0053] The parametric evolution model assumes that the parameters change slowly over time: , , Where w(t) is the system noise, and the noise intensity coefficient is set to η. α =0.001、η β =0.0001、η γ =0.01.

[0054] During the correction process, the state and parameters are updated simultaneously based on the difference between the observations and the forecasts, Δy: The Kalman gain matrix K determines the strength of the corrections made by the observed data to the state and parameter estimates.

[0055] The constraint handling mechanism uses a projection method to ensure the physical meaning of the parameters: when the updated parameters exceed the feasible region, , , The system also includes a parameter change rate limit to prevent excessive parameter changes within a single time step and ensure the stability of the calibration process.

[0056] Furthermore, the multi-source data fusion quality control mechanism implements a complete data quality assessment and processing workflow. Quality indicators are calculated using a comprehensive evaluation formula. The coefficients 0.6 and 0.4 represent the emphasis placed on data quality and model consistency, respectively. These ratios are determined through statistical analysis of historical typhoon data to ensure an optimal balance between signal-to-noise ratio and forecast consistency. SNR iGiven the current signal-to-noise ratio (SNR), calculate the ratio of signal power to noise power using a sliding window; SNR max This represents the highest signal-to-noise ratio ever recorded for this type of sensor; σ represents the deviation between the observed and predicted values. i The deviation Δ between this type of observation data and the model forecast. i The historical standard deviation, calculated based on data from the past 30 days, is used to measure the long-term stability of this type of observational data.

[0057] The observation error covariance is dynamically adjusted using the formula. Implementation, in which The baseline observation error variance is determined through sensor calibration experiments; k=2.0 is the sensitivity coefficient. The data selection strategy sets the quality threshold Q. min =0.3, when Q = 0.3 for three consecutive periods of a certain type of data i< Q min When this occurs, the data suspension mechanism is triggered, and the system automatically performs sensor status diagnosis, including checking the number of data jumps, verifying the communication link status, and testing the sensor response characteristics.

[0058] The abnormal data processing workflow monitors sudden data jumps in real time. If the time stamp is marked as suspicious data, cross-validation with other relevant sensor data is initiated. If it is confirmed to be abnormal data, its weight is reduced to 10% of the normal value within the current period, and the abnormal event is recorded for subsequent system maintenance.

[0059] Furthermore, the system load adaptive mechanism ensures system stability through multi-indicator monitoring and dynamic adjustment. The load monitoring indicator system includes: CPU utilization calculated as the average utilization over 1 minute by sampling from the / proc / stat file; memory usage monitored by calculating the proportion of memory used in / proc / meminfo; data reception latency recorded as the difference between the data timestamp and the reception timestamp; and queue backlog length monitored as the number of pending data packets.

[0060] The load status determination rules define three levels: normal state (CPU < 70%, memory < 80%, latency < 1s, queue < 10), high load state (CPU ≥ 70% for 30s or memory ≥ 80% for 20s), and overload state (CPU ≥ 90% for 15s or memory ≥ 90% for 10s).

[0061] The simplified mode switching strategy is activated under high load conditions: the set size is reduced from 120 to 60, and a hierarchical clustering method is used to select representative members; the time step is increased from 10s to 30s; the covariance matrix is ​​changed from a complete structure to a block diagonal structure; and the observation data is reduced from 1Hz to 0.5Hz.

[0062] Emergency mode is activated under overload conditions: the collection size is further reduced to 30; a simplified physical model is adopted to ignore higher-order nonlinear terms; and non-core tasks such as data archiving are suspended. The resource recovery mechanism adopts a gradual recovery strategy. When CPU < 60% and memory < 70% for 60 seconds, it gradually restores to normal configuration in three stages, with each stage running for 10 minutes for stability assessment.

[0063] Furthermore, the sensor network employs an optimized layout scheme to ensure the comprehensiveness and reliability of the monitoring data. Three key monitoring sections are arranged along the height of the barrel structure, located at 1 / 4, 1 / 2, and 3 / 4 of the height from the top of the barrel, respectively. These locations correspond to the critical areas of maximum earth pressure and bending moment. Four fiber optic strain sensors are evenly distributed circumferentially in each section, totaling 12 measuring points, with a measurement range of ±1500με, an accuracy of ±1με, and a sampling frequency of 10Hz.

[0064] Two GNSS monitoring units are installed at the center of the tank top, forming a redundant design with a horizontal accuracy of ±8mm, an elevation accuracy of ±15mm, and an output frequency of 1Hz. The environmental monitoring system includes a three-dimensional ultrasonic anemometer mounted on the top of the platform, with a measurement range of 0-60m / s and an accuracy of ±0.1m / s; and a wave radar deployed near the foundation, with a wave height measurement range of 0-20m, an accuracy of ±0.1m, and a wave period measurement range of 3-25s. All sensors meet the requirements for marine environments, with an IP68 protection rating, an operating temperature range of -20°C to +60°C, and corrosion resistance and anti-interference characteristics. Data transmission adopts the industrial Ethernet protocol, and the network topology uses a hybrid star and ring network structure to ensure that a single point of failure does not affect the overall system operation.

[0065] Furthermore, the early warning output mechanism ensures the accuracy and timeliness of early warnings through a multi-level threshold system and intelligent evaluation. The early warning threshold system establishes a three-level early warning system based on historical data and engineering analysis: the primary early warning (attention level) threshold is set at 50mm horizontal displacement, 30mm settlement, and 0.02rad rotation angle; the intermediate early warning (action level) threshold is 80mm horizontal displacement, 50mm settlement, and 0.035rad rotation angle; and the advanced early warning (emergency level) threshold is 120mm horizontal displacement, 80mm settlement, and 0.05rad rotation angle.

[0066] The early warning assessment process is executed every 10 minutes: First, current and 6-hour forecast data are read, the maximum predicted value of each indicator is calculated, the predicted value is compared with the thresholds at each level, and the early warning level is determined. The early warning information generation includes four core contents: current status summary (real-time values ​​of each monitoring indicator), predicted trend (future change trend and time of extreme value occurrence), early warning level (current early warning level and triggering reason), and recommended measures (targeted engineering disposal suggestions).

[0067] A multi-channel dissemination mechanism ensures timely information delivery: the web interface updates the alert status in real time and provides a visual display; mobile devices push alert information via a dedicated app; SMS notifications send brief alerts to key personnel; and email reports generate detailed alert analysis reports. The alert confirmation and escalation mechanism has a 30-minute confirmation deadline; if no confirmation is received within this timeframe, the notification will automatically escalate. The alert status is assessed every 2 hours, and the alert must be cleared after three consecutive cycles of assessment and confirmation.

[0068] Furthermore, the offline learning unit optimizes model parameter configuration through historical data analysis of the system. During the data preparation phase, complete data from at least three historical typhoon events are collected, including typhoon monitoring records and corresponding basic deformation data, ensuring the data encompasses monitoring information throughout the entire typhoon process. Parameter inversion employs the Tikhonov regularized least squares method, with the objective function being: , where θ is the parameter vector to be inverted, θ prior For prior parameter estimation, λ=0.01 is the regularization coefficient, used to balance data fit and parameter rationality.

[0069] The optimization process employs a quasi-Newton method, updating the approximate Hessian matrix via BFGS to ensure convergence speed and stability. Model validation utilizes k-fold cross-validation, randomly dividing historical data into five subsets. Four subsets are used alternately as the training set, and one subset as the validation set, repeated five times to ensure each subset is used as the validation set. The parameter set is only adopted when the prediction errors on all validation sets are less than the allowable value. The system also incorporates parameter sensitivity analysis, calculating the gradients of each parameter with respect to the objective function to identify key parameters and optimize the inversion strategy, thereby improving the accuracy of parameter identification.

[0070] Furthermore, the system deployment architecture adopts a distributed design to balance computing load and real-time requirements. Industrial-grade edge computing devices are deployed at the edge layer, configured with Intel i7 processors, 16GB of RAM, and 512GB of SSD storage, supporting a wide operating temperature range of -20°C to +60°C, and an IP67 protection rating. These devices are responsible for real-time data acquisition, preprocessing, and core prediction tasks, ensuring a prediction cycle of no more than 8 seconds.

[0071] The communication layer employs a redundant network design. The primary link uses single-mode fiber optic transmission with a bandwidth of 1000Mbps and a latency of <5ms; the backup link uses a 4G / 5G wireless network with a bandwidth of 100Mbps and a latency of <50ms. The primary and backup links adopt an automatic switching mechanism based on heartbeat detection. When three consecutive heartbeat packets (1s apart) are lost or the network latency exceeds 500ms, the system automatically switches to the backup link within 1 second.

[0072] The cloud platform is deployed in a remote data center, employing a microservice architecture to provide data storage, in-depth analytics, and system management functions. Data synchronization utilizes an incremental transmission strategy, transmitting only changed data to reduce network bandwidth consumption. System security management includes data transmission encryption, access control, and operation log auditing to ensure the reliability and security of system operation.

[0073] The complete workflow of this invention is as follows: 1. System initialization and parameter settings After system startup, initialization configuration is performed, setting core algorithm parameters including an 8-dimensional state vector, specifically containing horizontal displacement x, horizontal velocity dx / dt, rotation angle θ, angular velocity dθ / dt, settlement s, memory variable M, and soil parameters α and β. The ensemble Kalman filter is set to a 120-member ensemble with a time step Δt = 10s. A base time step of Δt = 10s is used for numerical integration calculations to ensure algorithm stability. A prediction time resolution of 10 minutes is used for result output, balancing real-time performance and data volume. The total prediction time is 72 hours.

[0074] Initial values ​​for the physical model parameters: equivalent mass m = 1.5 × 10⁻⁶ 6 kg, damping coefficient c = 2.0 × 10⁶ N·s / m, initial stiffness k = 5.0 × 10⁶ N·s / m 8 N / m, empirical soil parameters α=0.5, β=0.05, γ=1.2.

[0075] Simultaneously, three warning thresholds are set: 50mm for primary warning, 80mm for intermediate warning, and 120mm for advanced warning. The system reads historical typhoon databases and pre-calibrates model parameters using the Tikhonov regularization method through offline learning units. The objective function is: The regularization coefficient γ = 0.01 ensures the accuracy of the initial state estimation.

[0076] 2. Real-time data acquisition and preprocessing The data acquisition module begins operation, with 16 fiber optic strain sensors acquiring strain data at 10Hz, two GNSS displacement monitoring units acquiring three-dimensional displacement data at 1Hz, and environmental monitoring equipment acquiring wind speed and wave height data at 2Hz. All data is transmitted via industrial Ethernet, encapsulated in a unified binary format. Each data packet contains a 4-byte timestamp (Unix time format), a 2-byte sensor ID, a 1-byte quality identifier, and a 4-byte IEEE 754 floating-point data value. The preprocessing unit performs real-time verification of the raw data, using a sliding window method to detect outliers. The window size is set to 30 samples. Data exceeding 3 sigma is automatically marked and removed. Missing data is completed using linear interpolation. All data types are uniformly converted to the North-East-Earth coordinate system, and time synchronization is achieved using the PTP protocol, ensuring that the time deviation of all data is less than 10ms.

[0077] 3. Multi-source data fusion and quality assessment The data processing module performs a pre-fusion quality assessment on the collected multi-source data. Quality indicators are calculated for each type of observation data. SNR i The current signal-to-noise ratio (SNR) is obtained by calculating the ratio of signal power to noise power, Δ. i σ represents the deviation between the observed and predicted values. i This is the historical standard deviation of this type of data, calculated based on data from the past 30 days.

[0078] The observation error covariance matrix R is dynamically adjusted based on quality indicators, specifically through the formula... Implementation, where the reference error variance The sensitivity coefficient k was determined to be 0.2 through sensor calibration experiments. When Q... i When Q < 0.3, the weight of the data source is automatically reduced. This applies to three consecutive periods of Q. i When the threshold is reached, the use of the data source is suspended and the sensor diagnostic process is triggered.

[0079] 4. Ensemble Kalman Filter State Prediction The core prediction module begins executing the ensemble Kalman filter prediction step. Starting with the initial set of states, it performs single-step predictions for each of the 120 state vectors.

[0080] The nonlinear dynamic equations are solved using the fourth-order Runge-Kutta numerical integration method: .

[0081] The calculation is completed in four stages within each 10-second time step: first, the slope K1=f(t,X) is calculated, and then at half a step, the calculation is performed. Calculate again at half a step size. Finally, calculate at the full step length. Finally, the state prediction value is obtained. .

[0082] Simultaneously update the evolution equation of the memory variables. The time progression of all state vectors is completed, where β controls the decay rate of the memory variable, with a value range of [0.001, 0.1], reflecting the sensitivity of soil damage accumulation. The larger the value, the faster the soil softens under cyclic loading; γ determines the nonlinear characteristics of the decay process, with a value range of [0.5, 2.0], affecting the damage development pattern. When γ > 1, it exhibits accelerated decay characteristics. It represents the absolute value of the deformation rate, reflecting the intensity of the load.

[0083] 5. Observational data assimilation and state update When new observation data arrives, the system enters the update phase. First, the statistical characteristics of the state forecast ensemble are calculated, including the mean x. f Covariance P f Then calculate the Kalman gain matrix. , where P xy P is the cross-covariance of state and observation. yy Add the observation error covariance R to the observation forecast covariance.

[0084] Update each state vector: , where v i For observational perturbations conforming to an N(0,R) distribution, an anomaly detection mechanism calculates the Mahalanobis distance in real time. When D M Exceed Robust updates are initiated at the 95th percentile of the distribution, and the Kalman gain is recalculated after the observation error covariance matrix is ​​expanded by a factor of 5.

[0085] 6. Parameter Dynamic Calibration and Constraint Processing During the state update process, the system simultaneously performs dynamic calibration of the soil's empirical parameters. Parameters α, β, and γ are incorporated as part of the augmented state vector during the data assimilation process. Based on the difference between observed data and predicted values, the parameter estimates are automatically adjusted using the Kalman gain matrix.

[0086] The parameter constraints are handled using a projection method to ensure that the updated parameters fall within the physically permissible range: α∈[0.1,1.0], β∈[0.01,0.1], γ∈[0.5,2.0]. When the parameter estimate exceeds the feasible region, the system automatically projects it to the nearest boundary. To ensure the rationality of parameter estimation, the parameter constraint range is determined based on physical meaning and experimental data: α∈[0.1,1.0] ensures that the stiffness attenuation is within a reasonable range; β∈[0.01,0.1] is based on the observational statistics of the attenuation rate of soil memory variables; γ∈[0.5,2.0] covers the theoretical range of soil nonlinear characteristics.

[0087] At the same time, a parameter change rate limit is set, so that the parameter change amplitude within a single time step does not exceed 20%, to prevent parameter abrupt changes.

[0088] 7. Cumulative Deformation Prediction and Uncertainty Quantification Based on the updated state vector, the system generates a cumulative deformation prediction for the next 72 hours. By advancing the temporal evolution of the state vector, the expected value of the deformation and the confidence interval at each prediction time point are calculated. The Monte Carlo method is used to quantify the prediction uncertainty. 1000 samples are randomly selected from the analysis set for forward simulation to generate the prediction interval [Y] at a 90% confidence level. min ,Y max ].

[0089] Simultaneously, the deformation development trend curve is calculated, and the acceleration phase and extreme value occurrence times are identified, providing complete time series prediction results for early warning decision-making. The prediction results are output in JSON format, including timestamps, predicted values, upper and lower bounds of confidence intervals, and probability density function parameters.

[0090] 8. Load monitoring and adaptive adjustment The system monitors computing resource usage in real time, including CPU utilization (calculated as a 1-minute average via / proc / stat), memory usage (monitored via / proc / meminfo), data reception latency (difference between data timestamp and reception timestamp), and queue backlog length (number of pending data packets).

[0091] When CPU usage exceeds 70% for 30 consecutive seconds or memory usage exceeds 80%, a simplified operating mode is automatically activated: the set size is reduced from 120 to 60, k-means clustering is used to select representative members, the time step is adjusted from 10 seconds to 30 seconds, the covariance matrix adopts a block diagonal approximation, and the observation data is reduced from 1Hz to 0.5Hz.

[0092] When the load increases further (CPU ≥ 90% for 15 seconds), emergency mode is enabled to further reduce the computing scale to 30 set members, ensuring that the system maintains core prediction functions under resource constraints.

[0093] 9. Early Warning Assessment and Information Dissemination The results output module performs an early warning assessment every 10 minutes. It compares the predicted deformation with thresholds at each level. A primary warning is triggered when the predicted value exceeds 50mm, a secondary warning when it exceeds 80mm, and a high-level warning when it exceeds 120mm. The warning information includes a summary of the current status, the predicted trend, the warning level, and recommended measures. It is output in JSON format via a RESTful API and simultaneously published through multiple channels, including a web interface, mobile app, SMS, and email.

[0094] The system is equipped with an early warning confirmation mechanism. If no confirmation is received within 30 minutes, the notification will be automatically upgraded. The early warning status will be reassessed every 2 hours. The early warning can be lifted after 3 consecutive cycles of assessment and confirmation to ensure that the early warning information is delivered in a timely manner and to avoid false alarms.

[0095] 10. System maintenance and continuous learning During system downtime, the offline learning unit automatically starts, using historical typhoon event data to optimize model parameters. The Tikhonov regularized least squares method is employed to solve the parameter inversion problem, with the objective function... .

[0096] The generalization ability of the parameters was evaluated using 5-fold cross-validation. Historical data was randomly divided into 5 subsets, and 4 subsets were used for training and 1 subset for validation in turn, repeated 5 times to ensure that each subset was used as the validation set. The prior parameter library was updated, and a system operation report was generated, including prediction accuracy statistics, sensor performance evaluation, and system stability analysis, to provide data support for system optimization.

[0097] 11. Troubleshooting and System Recovery The system has a comprehensive fault detection and recovery mechanism. When a sensor fault is detected (data quality index Q for 5 consecutive cycles), the system will take action. i If the value is less than 0.2, the system will automatically switch to a backup sensor or enable a data reconstruction algorithm (Kriging interpolation based on spatiotemporal correlation).

[0098] When the network is interrupted, local caching is enabled to continue operation, with a cache capacity of 24 hours' worth of data. Once the network is restored, an incremental transmission strategy is used to synchronize data. In the event of a system crash, the current state is automatically saved to non-volatile memory, and operation resumes from the most recent valid state upon restart. Regular system self-checks are performed, including sensor calibration, algorithm verification, and performance testing, to ensure long-term stable operation of the system.

[0099] 12. Performance optimization and iterative updates Based on data and experience accumulated during operation, the system undergoes continuous performance optimization. The sensitivity settings of model parameters are adjusted by analyzing the prediction error distribution. Based on actual early warning results, ROC curve analysis is used to optimize early warning thresholds and triggering conditions.

[0100] Based on user feedback, we improved the result display and user experience. We regularly update the physics model and algorithm modules, incorporating the latest research findings and technological advancements, with a major version update every six months to maintain the system's advanced nature and accuracy. Simultaneously, we established an A / B testing mechanism, running new algorithms in parallel in shadow mode to verify their effectiveness and ensure the reliability of system updates.

[0101] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A system for predicting the cumulative deformation of a barrel foundation under typhoon load, characterized in that, include: The data acquisition module is used to acquire real-time data on strain, displacement, wind speed, and wave height of the barrel foundation under typhoon load. The core prediction module predicts cumulative deformation by executing a dynamic update process for the model state. This process includes: establishing a basic physical model describing the interaction between the bucket and the soil, wherein the state vector of the basic physical model includes the horizontal displacement and vertical settlement of the bucket and memory variables characterizing the cyclic softening effect of the soil; using the basic physical model, predicting the system state at the current moment based on the state estimate value at the previous moment; comparing the measured data at the current moment with the corresponding predicted value, and adjusting the system state estimate value in reverse according to the difference and the preset update rules, thereby outputting the cumulative deformation prediction result at the current moment. The results output module is used to graphically display the cumulative deformation prediction results, including the cumulative deformation development curve within a preset time period, the deformation prediction range, and warning information triggered based on a preset threshold.

2. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 1, characterized in that, The specific process for dynamically updating the model state is as follows: Maintain a set of system state vectors, with a size of 50–200; In each prediction period, the temporal evolution of all state vectors in the set is first advanced based on the basic physical model to obtain the state prediction set; Subsequently, the measurement data at the current moment is compared with the observation forecast values ​​obtained by mapping from the state forecast set. By calculating the statistical correlation between the state forecast set and the measurement data, a state analysis set is generated. The mean of the state analysis set is used as the optimized system state estimate.

3. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 2, characterized in that, The basic physical model describes the dynamic behavior of the barrel foundation through a set of differential equations. The characteristic that the soil resistance decreases with the number of cyclic loading is characterized by the evolution equation of the memory variable, and the evolution rate of the memory variable is controlled by two adjustable parameters.

4. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 3, characterized in that, The two adjustable parameters used to regulate the evolution rate of the memory variables are extended to the system state vector and are simultaneously corrected in real time with the displacement and settlement variables during the dynamic update process of the model state.

5. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 2, characterized in that, Before using the measurement data for state updates, a data optimization step is performed: dynamic weight values ​​are assigned to different types of measurement data. The calculation of these dynamic weights takes into account both the instantaneous signal-to-noise ratio estimate of the data source and its deviation from the long-term observation average of the system.

6. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 5, characterized in that, The typhoon load cumulative deformation prediction system for barrel foundations is also configured to: monitor the stability of data inflow and the utilization rate of computing resources in real time; when abnormal data flow or system load exceeds a preset threshold is detected, automatically simplify the calculation method of dynamic weights and adopt a state update strategy with higher computational efficiency.

7. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 1, characterized in that, The data acquisition module includes: a fiber optic strain sensor array deployed at key locations on the barrel structure, a GNSS displacement monitoring unit installed on the top of the barrel, an ultrasonic anemometer located on the top of the platform, and a wave radar located near the foundation.

8. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 1, characterized in that, The result output module displays the prediction range with a 90% confidence level as the default setting. The warning information includes a primary warning triggered when the primary threshold is exceeded and an emergency warning triggered when a higher-level threshold is exceeded.

9. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 1, characterized in that, The cumulative deformation prediction system for barrel foundations under typhoon loads also integrates an offline learning unit. This learning unit is automatically activated during non-real-time prediction periods, calls up the stored historical typhoon event dataset, and performs inversion and calibration of key empirical coefficients in the basic physical model by minimizing the prediction error function.

10. The cumulative deformation prediction system for barrel foundations under typhoon load according to claim 1, characterized in that, The core prediction module is deployed on an edge computing device, which is connected to the data acquisition module via an industrial Ethernet and is equipped with processor and memory resources to meet the requirements of real-time prediction.

Images