Safety Monitoring and Early Warning System and Method Based on Offshore Wind Power Platform Data Fusion

Abstract

This invention belongs to the field of marine engineering monitoring technology, specifically a safety monitoring and early warning system and method based on data fusion of offshore wind power platforms. It collects structural response data and environmental load data of offshore wind power platforms according to time synchronization requirements; employs a hierarchical fusion strategy for data fusion processing, performing data integration and quality control through raw-level fusion to output standardized data; feature-level fusion extracts multi-dimensional feature information from the standardized data; and decision-level fusion fuses the information to output a state assessment result; and a dual-element model combining a data-driven prediction model and a physical mechanism prediction model is used for safety state prediction. This invention solves problems such as insufficient adaptability to the marine environment, difficulty in sensor fault diagnosis, unsatisfactory data fusion effects, and low early warning accuracy in offshore wind power platform monitoring, providing advanced technical support for intelligent operation and maintenance of offshore wind farms.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of offshore engineering monitoring, and particularly relates to a safety monitoring and early warning system and method based on offshore wind power platform data fusion. BACKGROUND

[0002] Offshore wind power, as an important part of clean energy, is developing rapidly. Offshore wind power platforms are in complex marine environments for a long time, facing multiple threats such as wave impact, wind load, salt spray corrosion, and foundation settlement, so structural safety monitoring and early warning are of great significance. Compared with onshore wind power, offshore wind power platforms have significant environmental differences and technical challenges.

[0003] Currently, offshore wind power platform monitoring mainly has the following technical problems:

[0004] 1. Insufficient adaptability to marine environment: The foundation of the offshore wind power platform is subjected to periodic wave impact for a long time, resulting in small periodic displacement and swing of the platform foundation, which makes the sensor measurement reference unstable and affects the accuracy of the monitoring data. Salt spray in seawater causes corrosion to photoelectric sensors, leading to signal attenuation and accelerated equipment aging. The existing system lacks effective salt spray corrosion compensation mechanisms and automatic backup sensor switching strategies. The harsh environmental conditions of high humidity, high salt, and strong ultraviolet light in the ocean pose a serious challenge to the long-term stability of monitoring equipment. When a sensor fails, the system often cannot respond automatically, and monitoring interruptions are likely to occur.

[0005] 2. Difficulty in multi-source data fusion: Offshore wind power platform monitoring involves GNSS positioning data, InSAR deformation data, distributed fiber strain data, vibration acceleration data, and other multi-source heterogeneous data. There are significant differences in time resolution, spatial scale, and data format between each data source. Traditional monitoring systems lack effective hierarchical data fusion algorithms, and data utilization efficiency needs to be improved. Offshore wind power platforms have unique tower-foundation-blade coupled vibration characteristics, and their dynamics are fundamentally different from onshore wind power.

[0006] 3. Difficulty in sensor fault identification and compensation: The failure rate of sensors in harsh marine environments is significantly higher than that on land, including signal attenuation caused by salt spray corrosion, drift caused by temperature and humidity changes, and loosening caused by mechanical vibration. Existing monitoring systems lack effective sensor health status evaluation mechanisms and cannot timely detect sensor performance degradation or failure. They also lack intelligent fault compensation strategies.

[0007] 4. Lack of three-dimensional visualization technology: Existing wind power platform monitoring systems mostly use two-dimensional charts and lists to display monitoring data, which cannot realize real-time correlation display of monitoring parameters and three-dimensional structure models of wind power platforms, making it difficult for operation and maintenance personnel to intuitively judge the spatial position and real-time state of key components such as wind turbine towers, blades, and foundations.

[0008] 5. Early warning model of poor marine environment adaptability: The existing wind power platform early warning model is mainly based on the land wind power operation experience, mainly considers the wind load effect, and is insufficient in considering the wave load, salt spray corrosion and sea current scour of the marine environment. The fatigue damage mechanism under the marine environment is more complex, involving wind wave combined action, corrosion fatigue coupling, marine biological attachment and other factors. The applicability of the traditional early warning threshold and judgment criterion under complex sea conditions is limited, false alarm or missed alarm is easy to occur, which affects the reliability and practicality of the early warning system.

[0009] Therefore, it is urgent to develop a marine wind power platform multi-source heterogeneous data fusion safety monitoring and early warning system specially adapted to the characteristics of the marine environment to solve the above technical problems. SUMMARY

[0010] The purpose of the present application is to provide a safety monitoring and early warning system and method based on data fusion of offshore wind power platforms, to improve the reliability, accuracy and practicality of offshore wind power platform monitoring.

[0011] To achieve the above purpose, according to the first aspect of the present application, a safety monitoring and early warning method based on data fusion of offshore wind power platforms is provided, comprising the following steps:

[0012] S1, collecting the structural response data and environmental load data of the offshore wind power platform according to the time synchronization requirement;

[0013] S2, data fusion processing is carried out by adopting a hierarchical fusion strategy, the hierarchical fusion strategy includes raw level fusion, feature level fusion and decision level fusion; wherein the raw level fusion is used for data integration and quality control, and the standardized data is output, the feature level fusion extracts multi-dimensional feature information from the standardized data, and the decision level fusion fuses the multi-dimensional feature information and outputs the state evaluation result;

[0014] S3, safety state prediction is carried out based on a dual model combining a data-driven prediction model and a physical mechanism prediction model, the data-driven prediction model takes the data obtained by the hierarchical fusion strategy as input and takes the safety state evaluation result as output, and the physical mechanism prediction model takes the environmental load data as input and takes the safety state evaluation result as output.

[0015] Further, in step S2, the data integration includes establishing a unified time reference and spatial coordinate system, and aligning the data in time and space; the quality control includes establishing a data integrity index η=(N_valid / N_total)×100%, triggering a data compensation mechanism when η<95%; wherein N_valid represents the number of valid data, and N_total represents the total number of data;

[0016] The feature-level fusion uses multi-model Kalman filtering to establish displacement state estimation model, strain state estimation model and vibration state estimation model respectively. The multi-dimensional feature information includes displacement state vector, strain state vector and vibration state vector.

[0017] The decision-level fusion uses DS evidence theory to fuse evidence and outputs the overall credibility of the platform status.

[0018] Furthermore, the feature-level fusion also includes: fusing the results of the displacement state estimation model, strain state estimation model, and vibration state estimation model using a dual adaptive weighting strategy; the dual adaptive weighting strategy includes model weights and basic weights;

[0019] No. A state estimation model at time 1 The weights w_i(k) = f(P_i(k)), where P_i(k) is the weight of the k-th ... The model at time... The error covariance matrix, i=1,2,3;

[0020] The base weight w_i_base(k) is automatically adjusted based on real-time sea conditions, sensor health, and historical performance data.

[0021] Final weights for each model

[0022] .

[0023] Furthermore, in step S3, the inputs to the data-driven prediction model include: multi-dimensional feature information, state evaluation results, environmental parameters, historical state sequences of the multi-dimensional feature information, and sensor health status; the data-driven prediction model uses a trained LSTM-CNN neural network model.

[0024] Furthermore, the physical mechanism prediction model is a marine fatigue damage accumulation model D_ocean=D_wind+D_wave+D_coupling, where D_wind is wind load fatigue damage D_wind=∑(n_i / N_i), n_i is the actual number of cycles under the i-th stress level, N_i is the fatigue life under the i-th stress level, ∑(n_i / N_i) is the linear cumulative damage degree, D_wave is wave load fatigue damage, and D_coupling is wind-wave coupled fatigue damage. An early warning is triggered when D_ocean≥1. .

[0025] Furthermore, the final prediction result of the safety status prediction is W(t) = α × P(data) + β × P(physics) + γ × P(trend), where α, β, and γ are weighting coefficients, α + β + γ = 1, and typical values ​​are α = 0.4, β = 0.4, and γ = 0.2. P_data(t) is the prediction probability of the data-driven prediction model (0-1), P_physics is the prediction probability of the physical mechanism prediction model (0-1), P_trend is the prediction probability of the trend analysis (0-1), and W(t) is the comprehensive early warning index (0-1).

[0026] Furthermore, in step S1, the structural response data includes: displacement state, strain state, and vibration state; the environmental load data includes: temperature, humidity, wave height, and wind speed; and the monitoring units used for data acquisition include: GNSS monitoring unit, InSAR monitoring unit, distributed fiber optic sensor unit, vibration sensor unit, and environmental monitoring unit.

[0027] Furthermore, the safety monitoring and early warning method also includes: constructing a three-dimensional visualization scene based on the BIM model, and mapping the monitoring data fused in step S2 onto the three-dimensional model components in real time.

[0028] According to a second aspect of the present invention, a safety monitoring and early warning system based on offshore wind power platform data fusion is provided, comprising:

[0029] The data acquisition module is used to collect structural response data and environmental load data of offshore wind power platforms;

[0030] The data fusion processing module is used to integrate and control data quality through raw-level fusion, output standardized data, extract multi-dimensional feature information from the standardized data through feature-level fusion, fuse the multi-dimensional feature information through decision-level fusion, and output state evaluation results.

[0031] The early warning module is used to predict the safety status based on a dual model that combines a data-driven prediction model and a physical mechanism prediction model. The data-driven prediction model takes the data obtained by the hierarchical fusion strategy as input and the safety status assessment result as output. The physical mechanism prediction model takes the environmental load data as input and the safety status assessment result as output.

[0032] Furthermore, the data acquisition module includes a GNSS monitoring unit, an InSAR monitoring unit, a distributed fiber optic sensor unit, a vibration sensor unit, and an environmental monitoring unit;

[0033] The safety monitoring and early warning system also includes a 3D visualization module, which is used to construct a 3D model of the wind power platform based on the BIM model, and realize the real-time fusion display of monitoring data and the 3D model.

[0034] Furthermore, the security monitoring and early warning system also includes:

[0035] The marine environment adaptive calibration module is used to establish a marine environment calibration model T_cal(t) = T_base + α × Wave_displacement(t) + β × Foundation_settlement(t), where T_cal(t) is the calibrated measurement reference, T_base is the initial reference, Wave_displacement(t) is the instantaneous displacement caused by waves, Foundation_settlement(t) is the long-term foundation settlement, α is the wave displacement influence coefficient, and β is the foundation settlement influence coefficient, which is adaptively adjusted according to real-time wave parameters, sea state level, and platform structural characteristics. The marine environment adaptive calibration module is also used to establish a signal attenuation compensation model based on the corrosive effect of seawater salt spray on fiber optic sensors.

[0036] Furthermore, the sensor fault self-diagnosis module is used to establish a sensor health status assessment mechanism, including a data consistency verification algorithm and a signal quality assessment algorithm, and automatically activates a data compensation mechanism when a sensor fault is detected.

[0037] The system anomaly handling module is used to handle abnormal situations such as sensor failure, abnormal system load, and extreme sea conditions. It includes automatic backup switching and data compensation when sensor failure occurs, adaptive performance adjustment when the system load is too high, activation of backup strategies when the early warning model is abnormal, and automatic recovery mechanism when data acquisition is interrupted.

[0038] The data fusion processing module includes a data standardization module (which converts data from different sensors into a unified format, establishes a WGS84 coordinate system spatial reference, and establishes a unified time reference using GPS / UTC time), a data quality control module (which ensures data quality through mechanisms such as data integrity verification, time sequence consistency verification, and validity determination), and a multi-source data fusion module (which uses algorithms such as Kalman filtering to achieve multi-sensor data fusion and improve monitoring accuracy).

[0039] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0040] 1. The safety monitoring and early warning method based on offshore wind power platform data fusion provided by this invention effectively integrates multi-source heterogeneous data through a hierarchical fusion strategy and an offshore wind power structure-specific fusion algorithm. It fully considers the structural characteristics and marine load characteristics of offshore wind power platforms, significantly improving data utilization, reducing data redundancy, and lowering fusion processing latency compared to traditional methods. By combining the advantages of both data-driven and physical mechanisms through a dual-prediction model architecture, and considering fatigue damage and corrosion models specific to the marine environment, it can provide more accurate early warning results, effectively reducing false alarms and missed detections, and improving early warning reliability.

[0041] 2. Significantly Improved Marine Environmental Adaptability: Through a marine environment adaptive calibration module and a salt spray impact compensation algorithm, the system effectively addresses the instability of measurement benchmarks caused by wave impacts on offshore wind power platform foundations, significantly improving monitoring accuracy in complex marine environments. A comprehensive sensor anomaly handling mechanism has been established, automatically switching to a backup network when GNSS sensors fail, activating data compensation when fiber optic sensors experience salt spray corrosion, and activating sensor protection mode under extreme sea conditions, greatly enhancing system availability.

[0042] 3. Significantly enhanced system reliability: The sensor fault self-diagnosis module and fault adaptive compensation strategy ensure the continuous operation of the monitoring system under partial sensor failure. The multi-redundant transmission architecture and multi-sensor fusion mechanism can improve the system's operational reliability in harsh marine environments and reduce the impact of single-point failures.

[0043] 4. Improved visualization: Based on lightweight BIM 3D visualization technology, the correspondence between monitoring parameters and platform structure can be displayed intuitively, supporting multi-scale display and improving the fault location efficiency of operation and maintenance personnel.

[0044] 5. Supports intelligent operation and maintenance: The intelligent early warning system can predict equipment failure trends in advance, support preventive maintenance, help reduce operation and maintenance costs, and extend equipment life.

[0045] 6. Facilitates widespread application: Standardized data interfaces and modular architecture design support rapid deployment and system integration of different types of offshore wind power platforms. Attached Figure Description

[0046] Figure 1 This is the overall architecture diagram of the offshore wind power platform data fusion security monitoring and early warning system of the present invention.

[0047] Figure 2 This is a schematic diagram of the layout of monitoring points on the offshore wind power platform of the present invention.

[0048] Figure 3 This is a schematic diagram illustrating the working principle of the marine environment adaptive calibration module of this invention.

[0049] Figure 4 This is a flowchart of the algorithm for the sensor fault self-diagnosis module of the present invention.

[0050] Figure 5 This is a technical framework diagram of the offshore wind power structure-specific fusion algorithm of the present invention.

[0051] Figure 6 This is a technical architecture diagram of the three-dimensional visualization layer of this invention.

[0052] Figure 7 This is a schematic diagram of the real-time monitoring interface of the wind power platform based on BIM according to the present invention.

[0053] Figure 8 This is a flowchart of the monitoring and early warning method of the present invention.

[0054] Figure 9 This is a flowchart of the offshore wind power platform data fusion security monitoring and early warning method of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0056] Please see Figure 9 This invention provides a safety monitoring and early warning method based on offshore wind power platform data fusion, comprising the following steps:

[0057] S1. Collect structural response data and environmental load data of offshore wind power platforms in accordance with time synchronization requirements;

[0058] S2. A hierarchical fusion strategy is adopted for data fusion processing. The hierarchical fusion strategy includes raw-level fusion, feature-level fusion, and decision-level fusion. The raw-level fusion is used for data integration and quality control, and outputs standardized data. The feature-level fusion extracts multi-dimensional feature information from the standardized data. The decision-level fusion fuses the multi-dimensional feature information and outputs the state evaluation result.

[0059] S3. A dual-model approach combining a data-driven prediction model and a physical mechanism prediction model is used to predict the safety status. The data-driven prediction model takes the data obtained by the hierarchical fusion strategy as input and the safety status assessment result as output. The physical mechanism prediction model takes the environmental load data as input and the safety status assessment result as output.

[0060] This invention enables reliable monitoring and intelligent early warning of offshore wind power platforms in complex marine environments. It includes the effective fusion of multi-source heterogeneous monitoring data, dynamic calibration of sensor measurement benchmarks, self-diagnosis and compensation of sensor faults, automatic processing and recovery of system anomalies, three-dimensional visualization, and intelligent early warning of the structural specificity of offshore wind power, ensuring that the monitoring system can maintain continuous and reliable operation under various abnormal conditions.

[0061] A safety monitoring and early warning system based on offshore wind power platform data fusion includes:

[0062] The data acquisition module is used to collect structural response data and environmental load data of offshore wind power platforms;

[0063] The data fusion processing module is used to integrate and control data quality through raw-level fusion, output standardized data, extract multi-dimensional feature information from the standardized data through feature-level fusion, fuse the multi-dimensional feature information through decision-level fusion, and output state evaluation results.

[0064] The early warning module is used to predict the safety status based on a dual model that combines a data-driven prediction model and a physical mechanism prediction model. The data-driven prediction model takes the data obtained by the hierarchical fusion strategy as input and the safety status assessment result as output. The physical mechanism prediction model takes the environmental load data as input and the safety status assessment result as output.

[0065] The present invention will be further illustrated by specific embodiments below.

[0066] Example 1: Deployment of Multi-Source Sensors and Adaptive Calibration for Marine Environment

[0067] The data acquisition layer of this invention is used to collect multi-source monitoring data from offshore wind power platforms, including a GNSS monitoring unit (using RTK-GNSS technology, deploying 5 receivers at the four corners and center of the wind power platform foundation, achieving a positioning accuracy of ±1-5cm, a data sampling frequency of 1Hz, and supporting BDS / GPS / GLONASS multi-constellation positioning); an InSAR monitoring unit (using C-band synthetic aperture radar interferometry technology, with a monitoring cycle of 12 days, deformation monitoring accuracy of ±1mm / year, and a coverage range of 100km×100km); and a distributed fiber optic sensor unit (in wind... The tower is equipped with a set of FBG fiber optic grating sensors every 10m along its height, with strain measurement accuracy of ±1με, temperature monitoring accuracy of ±0.1℃, and sampling frequency of 100Hz; a vibration sensor unit (triaxial MEMS accelerometers are installed in the wind turbine nacelle, tower top, and blade root, with a measurement range of ±16g, frequency response range of 0.1-1000Hz, and sampling frequency of 2048Hz); and an environmental monitoring unit (including an ultrasonic anemometer with accuracy of ±0.1m / s and ±1°, a high-precision ocean wave radar wave with accuracy of ±0.1m, and a temperature and humidity sensor with accuracy of ±0.1℃ and ±1%RH).

[0068] Specifically, such as Figure 1 and Figure 2 As shown, in the deployment of the monitoring system for a certain offshore wind power platform, the GNSS monitoring network deploys five RTK-GNSS receivers at the four corners (A1-A4) and the center position (C) of the wind power platform foundation. The coordinates are as follows: A1 (E121°35'12.345", N31°22'45.678", H15.234m), A2 (E121°35'18.456", N31°22'45.678", H15.198m), A3 (E121°35'18.456", N31°22'39.567", H15.167m), A4 (E121°35'12.345", N31°22'39.567", H15.201m), and C (E121°35'15.401", H15.234m). (N31°22'42.623", H15.200m). Dual-frequency L1 / L2 signal reception is used, with a baseline length of <5km, achieving centimeter-level positioning accuracy.

[0069] Eight FBG sensor arrays are deployed along the 80m tower height, located at 10m, 20m, 30m, 40m, 50m, 60m, 70m, and 80m above sea level. Each array contains four strain sensors (0°, 90°, 180°, and 270° orientations) and one temperature sensor. Vibration monitoring sensors include: three triaxial MEMS accelerometers (measurement range ±16g) at the top of the nacelle, one triaxial accelerometer and one tilt sensor at the top of the tower, and two strain sensors at the blade roots on each blade.

[0070] The data transmission layer adopts a triple-redundant transmission architecture of "fiber optic leased line + 5G wireless + Beidou short message". The main transmission is submarine optical cable (transmission bandwidth 1Gbps, latency <20ms), the backup transmission is 5G private network (transmission bandwidth 100Mbps, latency <50ms), and the emergency transmission is Beidou short message (supports 120-byte data packets, global coverage).

[0071] The marine environment adaptive calibration module is specifically designed to address the instability of measurement references caused by wave impacts on offshore wind power platform foundations. It includes: a real-time foundation motion monitoring unit, which uses GNSS receivers and high-precision tilt sensors deployed on the platform foundation to monitor the six degrees of freedom motion of the foundation in real time, including three translational components (x, y, z displacements) and three rotational components (pitch, roll, and yaw angles); and a dynamic reference calibration algorithm, establishing a marine environment calibration model T_cal(t) = T_base + α × Wave_displacement(t) + β × Foundation_settlement(t), where... T_cal(t) is the calibrated measurement reference, T_base is the initial reference, Wave_displacement(t) is the instantaneous displacement caused by waves, Foundation_settlement(t) is the long-term foundation settlement, α is the wave displacement influence coefficient, and β is the foundation settlement influence coefficient (which can be determined based on marine and foundation engineering experience). The algorithm is adaptively adjusted according to real-time wave parameters, sea state level, and platform structural characteristics. The salt spray effect compensation algorithm establishes a signal attenuation compensation model to address the corrosive effect of seawater salt spray on fiber optic sensors. Through regular standard signal calibration and environmental parameter correction, the measurement accuracy of the fiber optic sensors is maintained.

[0072] The implementation process of the marine environment adaptive calibration algorithm is as follows: Establish a GPS time reference T_GPS = T_UTC + 18s (leap second compensation), where T_UTC is Coordinated Universal Time; calculate the time drift compensation Δt_drift = k × (T_current - T_init), where the drift coefficient k = 1 × 10^25. -6T_current is the current system time, T_init is the system initialization time. The network latency compensation Δt_network = (RTT / 2) + jitter_buffer is calculated, where RTT (Round Trip Time) is the network round-trip time, and jitter_buffer is the jitter buffer to compensate for network latency variations. The final synchronization time is T_sync = T_GPS + Δt_drift + Δt_network. The calibration model is: T_cal(t) = T_base + α × Wave_displacement(t) + β × Foundation_settlement(t). The calibration coefficients are determined based on sea state: when wave height H < 1m, α = 0.8, β = 1.0; when 1m ≤ H < 3m, α = 1.0, β = 1.2; when H ≥ 3m, α = 1.2, β = 1.5, as shown below. Figure 3 As shown.

[0073] To address the impact of seawater salt spray on fiber optic sensors, a salt spray corrosion compensation model was established. Through regular calibration with a standard light source, the signal attenuation of the fiber optic sensor was monitored, an attenuation curve was established, and real-time compensation was performed based on environmental parameters such as temperature, humidity, and salinity. The compensation algorithm effectively extends the lifespan of the fiber optic sensor and maintains its measurement accuracy.

[0074] For the system anomaly handling layer, considering the complexity of the marine environment, a multi-dimensional anomaly detection, processing, and recovery mechanism is established. This includes a sensor anomaly handling module (automatically switching to a backup network when a GNSS sensor fails, maintaining positioning accuracy in the short term through inertial navigation; activating a backup FBG array and resistance strain gauges for data compensation when a fiber optic sensor fails due to salt spray corrosion; activating sensor protection mode under extreme sea conditions (wave height > 4m), shutting down non-critical sensors, retaining only core monitoring functions, and ensuring the system's survivability in harsh environments), a visualization system adaptive module (automatically reducing the LOD level when the system load exceeds 85%, activating streaming processing when memory is insufficient, and supporting seamless 2D / 3D switching), an early warning system fault tolerance module (activating a backup strategy when the LSTM model degrades, performing intelligent correction when physical constraints are violated, and possessing self-verification and continuous optimization capabilities), and a monitoring process adaptive module (automatically switching transmission channels when data acquisition is interrupted, and dynamically adjusting strategies when processing capacity is insufficient).

[0075] Example 2: Sensor Fault Self-Diagnosis Algorithm

[0076] like Figure 4 As shown, the sensor fault self-diagnosis module adopts a multi-level diagnostic strategy, specifically including:

[0077] (1) Data Consistency Verification: For multiple sensors of the same type, the system determines whether there are any anomalies by calculating the data consistency between them. The consistency index between sensors of the same type is calculated as Consistency_index = Σ|Data_i - Data_predicted_i| / n, where Data_i is the measured value of the i-th sensor, Data_predicted_i is the expected value predicted based on data from other sensors, and n is the total number of sensors. When the consistency index exceeds a set threshold, abnormal sensors are automatically marked. For example, for strain sensors deployed at different heights on a tower, according to structural mechanics principles, there should be a certain correlation between them. When the measured value of a sensor deviates too much from the expected value, the system automatically marks the sensor as suspicious.

[0078] (2) Signal quality assessment: A comprehensive evaluation system was established, encompassing multiple indicators such as signal-to-noise ratio (SNR), signal stability, and data continuity. SNR reflects the clarity of the sensor signal, stability reflects signal consistency, and continuity reflects the integrity of data transmission. These three indicators comprehensively assess the sensor's operational status. Specifically, a comprehensive signal quality index was established: Signal_quality = (SNR × Stability × Continuity) / 3, where SNR is the signal-to-noise ratio, Stability is the signal stability, and Continuity is the data continuity. This multi-dimensional assessment of the sensor's operational status is used to evaluate its performance.

[0079] (3) Environmental Factor Impact Assessment: The impact of environmental factors such as temperature, humidity, and vibration on sensor performance is considered. By establishing a correlation model between environmental parameters and sensor performance, it is possible to distinguish whether the performance change is caused by sensor malfunction or environmental factors.

[0080] A sensor health index is established: Health_index = w1 × Data_quality + w2 × Signal_strength + w3 × Environmental_factor, where Data_quality is the data quality score (0-1), Signal_strength is the signal strength score (0-1), and Environmental_factor is the environmental impact factor (0-1). The weight coefficients w1, w2, and w3 are adjusted according to the sensor type and importance, and satisfy w1 + w2 + w3 = 1.

[0081] (4) Fault adaptive compensation strategy: When a sensor fault or performance degradation is detected (e.g. When interference occurs, the system first attempts to eliminate it through filtering algorithms; if this fails, the weight of the sensor in data fusion is reduced; for sensors in critical locations, interpolation algorithms based on neighboring sensors are used for data compensation; in extreme cases, backup sensors are activated or the monitoring strategy is adjusted.

[0082] Example 3: Specific Implementation of Offshore Wind Power Structure-Specific Fusion Algorithm

[0083] like Figure 5 As shown, the offshore wind power structure-specific fusion algorithm is specifically designed for the coupled vibration characteristics of offshore wind power platform tower-foundation-blade and the characteristics of marine loads. It adopts an improved hierarchical fusion strategy to process multi-source heterogeneous data, including three levels: raw-level fusion, feature-level fusion, and decision-level fusion. The specific implementation includes the following key steps:

[0084] (1) Raw-level fusion: Based on the adaptive calibration of the marine environment, data integration and quality control are carried out. In the data preprocessing stage, multi-source heterogeneous data are uniformly processed. First, a unified time reference is established, using GPS / UTC time as the system reference. The time synchronization algorithm T_sync=T_GPS+Δt_drift+Δt_network is used to achieve accurate alignment of sensor data with different sampling frequencies, and the time synchronization accuracy reaches ±1ms. A unified spatial reference is established, using the WGS84 coordinate system, and centimeter-level spatial registration is achieved through the coordinate transformation matrix. Data quality control is carried out by establishing a data integrity index η=(N_valid / N_total)×100%. When η<95%, a data compensation mechanism is triggered.

[0085] (2) Feature-level fusion: In the feature extraction stage, a multi-model adaptive Kalman filter algorithm is used for sensor fault diagnosis and multi-dimensional feature extraction. The hierarchical modeling strategy effectively handles the state estimation problem of wind power platforms. Multi-source heterogeneous data are classified according to physical attributes, and displacement, strain and vibration state estimation models are constructed respectively.

[0086] (21) Displacement state estimation model (state vector: ,in For three-dimensional displacement deviation, For the corresponding velocity components. State transition equation: , ,in, A 1 represents a 6×6 state transition matrix, employing a displacement-velocity coupled model; H 1 represents the observation matrix, which can be dynamically configured as a 3×6 or 2×6 matrix based on GNSS availability.

[0087] (22) Strain state estimation model (establishing a 6-dimensional strain state vector: ,in For normal strain components, For shear strain components. State transition equation:

[0088] , ,in A 2 is a diagonal matrix, considering the viscoelastic effect of the material; H 2. The arrangement angle and number of fiber optic sensors are determined, and the relationship is established based on strain conversion.

[0089] (23) Vibration state estimation model (establishing a 6-dimensional vibration state vector:

[0090] ,in For acceleration components, Let be the vibration velocity component. State transition equation: , ,in Considering vibration damping characteristics, (This is a 3×6 observation matrix, directly mapping acceleration measurements).

[0091] Design a Kalman filter for each subsystem, and perform a prediction step ( , ,in For the first The state estimates of the model, For the error covariance matrix, and the update steps , , Perform state estimation, where For Kalman gain, Q i , The covariance matrix of process noise and observation noise, (These are sensor observations).

[0092] (24) Multi-model fusion (using an adaptive weight strategy to fuse the results of each sub-model: establish an adaptive weight strategy w_i(k)=f(P_i(k)), where w_i(k) is the weight of the first sub-model. Each model (including displacement state estimation model, strain state estimation model, and vibration state estimation model) at time... The weight, It is the first The model at time... The error covariance matrix is ​​calculated. The fusion weights are dynamically adjusted based on the estimation accuracy P_i(k) of each sub-model; sensors with high accuracy and reliability receive greater weights, while faulty or degraded sensors automatically have reduced weights. The contribution of each sub-model is determined by its weight. The larger the weight, the greater the influence of the state estimate of the sub-model on the final result.

[0093] (25) Adaptive weight adjustment mechanism: Under harsh marine environmental conditions, the system automatically adjusts the weight allocation strategy based on real-time sea state, sensor health, and historical performance data. When the wave height H ≥ 3m, the weights of GNSS and InSAR monitoring are increased to capture large-amplitude structural responses; when salt spray corrosion is detected, the weights of the affected fiber optic sensors are reduced accordingly and backup sensors are activated; through this multi-level, adaptive fusion strategy, accurate and reliable structural state assessment results can still be obtained in complex marine environments. For example, a specific adjustment strategy can be: if the wave height >= 3m: modify the base weights of the GNSS sub-model in the first layer w_GNSS_base = w_GNSS_base × 1.2, increasing by 20%; w_InSAR_base = w_InSAR_base × 1.1, increasing by 10%; if salt spray corrosion is detected: directly reduce the weights of the fiber optic sensor sub-model w_fiber_base = w_fiber_base × 0.5, reducing by 50%; activate backup sensors.

[0094] Comprehensive Status Assessment = , This is the state estimate or output of the i-th model (including the displacement state model, strain state model, and vibration state model) at time k. This formula obtains the final fusion result by weighted averaging of the outputs of all sub-models. w_final(k) represents the final weight of each sub-model, determined by... This is obtained together with w_i_base(k).

[0095] In (24), the adaptive weights in the multi-model fusion belong to the algorithm level, w_i(k)=f(P_i(k)), which is based on the Kalman filter error covariance matrix. It only acts within the Kalman filter algorithm of feature-level fusion and is automatically calculated at each sampling time. In (25), the marine environment weight adjustment belongs to the system level. It is the top-level control of the entire fusion system. It is adjusted according to marine environmental conditions, sensor health status, and engineering experience. The adjustment frequency is triggered by environmental changes. Through the collaborative working mechanism of the two weights, the final fusion weight = algorithm base weight × system adjustment factor w_final_i(k) = w_i(k) × system_factor_i, where w_i(k) is the Kalman filter base weight in (24), and system_factor_i is the marine environment system adjustment factor. The specific workflow is as follows:

[0096] Algorithm layer: Kalman filtering automatically calculates the basic weights based on P_i(k);

[0097] System layer: Adjust weights based on sea state, sensor health, etc.

[0098] Final fusion: State estimation is performed using the corrected weights.

[0099] Practical application examples

[0100] Scenario: Deteriorating sea conditions + salt spray corrosion of fiber optic sensors

[0101] Step 1: The algorithm layer automatically calculates (normal operation) w_GNSS(k)=f(P_GNSS(k))=0.4, based on error covariance; w_fiber(k)=f(P_fiber(k))=0.3, based on error covariance; w_vibration(k)=f(P_vibration(k))=0.3, based on error covariance.

[0102] Step 2: System layer environment adjustment. If the wave height is >= 3m, increase the GNSS weight, system_factor_GNSS=1.2; otherwise, system_factor_GNSS=1.0. If salt spray corrosion is detected, adjust system_factor_fiber = 1.0 to system_factor_fiber = 0.5, and reduce the fiber weight.

[0103] Step 3: Final weight calculation: w_final_GNSS = 0.4 × 1.2 = 0.48; w_final_fiber = 0.3 × 0.5 = 0.15; w_final_vibration = 0.3 × 1.0 = 0.30; total weight = 0.48 + 0.15 + 0.30 = 0.93, w_norm_GNSS = 0.48 / 0.93 = 0.52, w_norm_fiber = 0.15 / 0.93 = 0.16, w_norm_vibration = 0.30 / 0.93 = 0.32.

[0104] By using algorithmic layer weights, mathematical optimality is ensured, enabling rapid response to changes in data quality. By using system layer weights, engineering knowledge is incorporated to address physical phenomena that the algorithm cannot perceive. For example, when a fiber optic sensor begins to be corroded by salt spray, the error covariance P_i(k) may not have changed significantly yet. At this time, the algorithmic layer weight w_i(k) is still a normal value, but the system layer has already detected the corrosion and can proactively reduce the system_factor to mitigate its impact, thus preventing it in advance.

[0105] (3) Decision-level fusion: Multi-source evidence fusion is performed using DS evidence theory, and basic confidence values ​​are assigned. Where K is the conflict coefficient, used to handle conflicts between different sources of evidence, and m(A) is the basic confidence assignment after fusion. This represents all possible combinations of sources of evidence. Indicates that source 1 of evidence relates to the proposition. The basic confidence assignment is as follows. Specifically, the confidence assignment functions m1, m2, and m3 represent the probability distributions of the platform's state under historical data, physical models, and real-time monitoring, respectively, and are used to quantify the credibility of the platform being in a normal, abnormal, or unknown state. For example, m1(normal) = 0.8, m1(abnormal) = 0.1, m1(unknown) = 0.1; m2(normal) = 0.7, m2(abnormal) = 0.2, m2(unknown) = 0.1; m3(normal) = 0.9, m3(abnormal) = 0.05, m3(unknown) = 0.05. To handle conflicts between different evidence sources, a conflict coefficient K = 0.105 is calculated. This coefficient is used to adjust the weights of each evidence source. Finally, the overall credibility of the platform's state is obtained through fusion calculation, where the credibility of the platform being in a normal state is 99.2%, and the credibility of the abnormal state is 0.8%. When the credibility m(abnormal) of the abnormal state in the fusion result is greater than 0.3, the system will trigger an early warning mechanism. This algorithm can effectively integrate evidence from multiple sources, providing support for real-time monitoring, fault diagnosis, and early warning of wind power platforms.

[0106] In summary, this invention achieves data preprocessing and alignment, time synchronization, spatial registration, and quality control through raw-level fusion, thereby outputting standardized multi-source data; through feature-level fusion and Kalman filtering, the three sub-models process displacement, strain, and vibration data respectively, thereby outputting three state vectors X_1(k), X_2(k), and X_3(k); through decision-level fusion using DS evidence theory, the results of the three sub-models are fused to obtain a comprehensive state assessment result.

[0107] Example 4: Implementation of 3D Visualization Based on BIM

[0108] like Figure 2 and Figure 6 As shown, the 3D visualization system is based on lightweight BIM technology, enabling real-time fusion display of monitoring data and 3D models.

[0109] Lightweight BIM Model Processing: Utilizing Level of Detail (LOD) optimization technology, the wind power platform BIM model is simplified in stages according to different application scenarios, establishing a four-level accuracy standard from LOD100 to LOD400. A mesh simplification algorithm compresses the model file size to 20-30% of its original size. Specifically, multiple accuracy standards are established: LOD100 (conceptual design level, displaying the overall platform outline), LOD200 (schematic design level, displaying major components), LOD300 (construction drawing level, displaying detailed structure), and LOD400 (manufacturing and installation level, displaying all details). Mesh simplification algorithms and texture compression technology are employed to dynamically adjust the model accuracy based on viewing distance and display requirements.

[0110] Real-time data mapping: Establishing a precise mapping relationship between monitoring data and BIM model components. Each monitoring point corresponds to a specific component in the model, and the monitoring data is displayed in real time on the 3D model through color coding, numerical annotation, and animation effects. The color coding uses a gradient of green-yellow-orange-red to intuitively reflect the safety status.

[0111] Multi-scale visualization: Supports multi-scale display from the overall layout of the wind farm to the structure of a single platform, and then to the details of individual components. Users can smoothly switch between different scales using mouse operation to obtain comprehensive monitoring information. The macro scale displays the overall operating status of the wind farm, the meso scale displays the structural response of a single platform, and the micro scale displays detailed information of key components.

[0112] The 3D visualization system features an adaptive mechanism: a comprehensive performance monitoring and degradation strategy. When the system load exceeds 85%, the BIM model's LOD level is automatically reduced to prioritize real-time display of monitoring data. When memory is insufficient, streaming data processing and time window management are enabled to prevent system crashes. The system supports seamless switching between 2D and 3D interfaces and can still provide basic monitoring data display functions even in extreme conditions.

[0113] Example 5: Implementation of Intelligent Early Warning Algorithm

[0114] The intelligent early warning layer adopts a dual model that combines data-driven approaches with physical mechanisms.

[0115] (1) Data-driven prediction model: LSTM-CNN hybrid neural network model is adopted.

[0116]

[0117] Where f(t) is the forget gate and i(t) is the input gate. Let Ct be the activation value of the output gate, and C̃(t) be the candidate cell state. Output in hidden state It is the sigmoid function. It is the hyperbolic tangent activation function. , , , It is the weight matrix of each gate. , , , These are the bias vectors for each gate. It is the hidden state from the previous moment. and current input splicing.

[0118] A deep learning prediction model is built based on historical monitoring data. The input layer includes multi-dimensional monitoring data (displacement, strain, vibration, environmental parameters, etc., totaling 64 feature parameters), the hidden layer uses a 3-layer LSTM network, each layer containing 128 neurons, and the output layer provides a safety status assessment value (a continuous value between 0 and 1). The model is trained on a large amount of historical data and is able to identify complex nonlinear patterns.

[0119] (2) Physical mechanism prediction model: Based on the structural mechanics and materials science theories in the marine environment. The fatigue damage model adopts Miner's linear cumulative damage theory and combines the load spectrum characteristics of the marine environment; the corrosion model considers the combined effects of electrochemical corrosion, erosion corrosion and stress corrosion; the structural response model is based on finite element analysis and considers material nonlinearity and geometric nonlinearity.

[0120] The cumulative fatigue damage model for the ocean is D_ocean = D_wind + D_wave + D_coupling, where D_wind represents wind load fatigue damage (D_wind = ∑(n_i / N_i), n_i is the actual number of cycles at the i-th stress level, N_i is the fatigue life at the i-th stress level, ∑(n_i / N_i) is the linear cumulative damage degree, D_wave represents wave load fatigue damage, and D_coupling represents wind-wave coupled fatigue damage. An early warning is triggered when D_ocean ≥ 1; P_(physics)(t) = 0.1 + 0.9(P_(physics)) 2 .

[0121] (3) Integrated early warning decision-making: Establish a four-level early warning system. When the monitoring parameters of the blue warning are within the normal range of 0-70% threshold, it indicates that the system is in good condition. When the monitoring parameters of the yellow warning are close to the warning value of 70-85% threshold, it indicates that attention is needed. When the monitoring parameters of the orange warning exceed the warning value of 85-95% threshold, it indicates that inspection is needed. When the monitoring parameters of the red warning reach the danger value of >95% threshold, it indicates that immediate action is needed.

[0122] Early warning decision function Where α, β, and γ are weighting coefficients (α+β+γ=1, typical values ​​are α=0.4, β=0.4, and γ=0.2), P_data(t) is the prediction probability of the data-driven model (0-1), P_physics(t) is the prediction probability of the physical mechanism model (0-1), P_trend(t) is the prediction probability of the trend analysis (0-1), and W(t) is the comprehensive early warning index (0-1). A yellow warning is issued when W(t)>0.7, an orange warning is issued when W(t)>0.85, and a red warning is issued when W(t)>0.95.

[0123] like Figure 7As shown, the early warning decision function comprehensively considers the results of the data-driven model, the physical mechanism model, and trend analysis, and derives the final early warning level through evidence fusion theory.

[0124] (4) Fault-tolerant mechanism of intelligent early warning system: A multi-level early warning reliability assurance system has been established. When the performance of the LSTM model degrades, the system automatically activates the backup model and ensemble learning; when the data-driven model output violates physical constraints, physical consistency correction is activated; when a large number of sensors fail, risk assessment is conducted based on weather forecasts and historical statistics. The early warning system has self-verification capabilities and continuously optimizes the early warning strategy through cross-validation and historical accuracy evaluation.

[0125] The data flow of this invention includes three stages, as detailed below:

[0126] Phase 1: Multi-source data fusion → State vector

[0127] Raw data → Kalman filter → Fusion state vector: GNSS data → Displacement state vector X_1(k) [6D], Fiber optic data → Strain state vector X_2(k) [6D], Vibration data → Vibration state vector X_3(k) [6D], Weighted fusion:

[0128] X_comprehensive(k) = .

[0129] Phase 2: State Vector → LSTM Features

[0130] X_Comprehensive(k) → Feature Extraction → LSTM Input Features, LSTM Input = [Current State Vector X_Comprehensive(k), 18-dimensional, Historical State Sequence [X_Comprehensive(k-1)...X_Comprehensive(kn)], 18 The system comprises 64 features: n-dimensional (n=2), environmental parameters [temperature, humidity, wave height, wind speed] (4-dimensional), sensor health [h_1, h_2, h_3] (3-dimensional), and DS fusion confidence [normal, abnormal, unknown] (3-dimensional). The DS fusion result, as one of the input features of the LSTM, enhances the model's ability to perceive uncertainty.

[0131] Phase 3: LSTM Output → Final Warning

[0132] LSTM output P_data(t) → Binary model fusion → Early warning level W(t) = α P_data(t)+β P_physics(t)+γ P_trend(t), where: P_data(t) is the output of the LSTM model (based on the fused state vector), P_physics(t) is the output of the physical model (based on mechanical calculations), and P_trend(t) is the output of the trend analysis (based on historical statistics).

[0133] Specific examples:

[0134] Assume at time t: the fiber optic sensor experiences salt spray corrosion, and its health status drops from 1.0 to 0.6 - the weights are adjusted from w_2=0.4 to w_2=0.24 - the weights of the strain components in the fused state vector X_comprehensive(k) decrease - the influence of strain-related features in the feature vector received by the LSTM weakens - at the same time, the sensor health status features indicate that the fiber optic sensor is abnormal - the LSTM adjusts its predictions based on this information, which may increase the probability of early warning.

[0135] Example 6: Monitoring and Early Warning Method Flow

[0136] like Figure 8 As shown, the safety monitoring and early warning method for offshore wind power platforms includes the following steps:

[0137] Step S1: Continuously collect monitoring data through multiple types of sensors deployed at key locations on the wind power platform. The data collection frequency is GNSS (1Hz), fiber optic sensor (100Hz), and vibration sensor (2048Hz). The data storage format is timestamp + sensor ID + value + status identifier. All data has a precise timestamp to ensure time synchronization.

[0138] Step S2: Data Preprocessing Stage. Quality control is performed on the raw monitoring data, including outlier detection and removal, data format standardization, and spatiotemporal benchmark alignment. A unified time benchmark and spatial coordinate system are established to ensure consistency between data from different sources.

[0139] Step S3: An improved hierarchical fusion strategy is adopted to fuse the preprocessed data. Data integration and quality control are performed at the raw level, multi-dimensional feature information is extracted at the feature level, and deep fusion and state assessment are performed at the decision level.

[0140] Step S4: 3D Visualization Stage. 3D visualization is achieved based on a lightweight BIM model, with monitoring data updated every 1 minute. The user interface supports multiple view modes, including overall view, partial view, and time-series view, providing maintenance personnel with intuitive monitoring information.

[0141] Step S5: Based on a dual-model approach, predict and issue early warnings for the safety status. Utilize data-driven and physical mechanism models to predict future states, comprehensively consider historical data trends, physical mechanism constraints, and real-time monitoring information to assess risk levels, provide accurate and reliable early warning results, and issue corresponding level early warning information. The early warning information includes detailed information such as the warning level, risk causes, and recommended measures, and is sent to relevant personnel through multiple channels.

[0142] Adaptive execution of monitoring and early warning methods: The entire monitoring and early warning process has comprehensive interruption recovery and adaptive adjustment capabilities. When data acquisition is interrupted, the system automatically switches the transmission channel and enables local caching; when processing capacity is insufficient, it dynamically adjusts processing accuracy and parallelization strategies; when extreme anomalies occur, it activates the minimum functional mode to ensure uninterrupted core monitoring.

[0143] This invention provides a multi-source heterogeneous data fusion safety monitoring and early warning system and method for offshore wind power platforms. The system includes a data acquisition layer, a data transmission layer, a data fusion processing layer, a 3D visualization layer, and an intelligent early warning layer. The data fusion processing layer includes a marine environment adaptive calibration module, a sensor fault self-diagnosis module, and an offshore wind power structure-specific fusion algorithm module, specifically designed to address the monitoring challenges of offshore wind power platforms in complex marine environments. Effective integration of multi-source data is achieved through dynamic benchmark calibration algorithms, multi-model adaptive Kalman filtering, and DS evidence fusion theory. The intelligent early warning layer employs a dual-element model combining data-driven and physical mechanisms to establish a four-level early warning mechanism. The system possesses comprehensive anomaly handling capabilities, including automatic sensor fault switching and data compensation, adaptive performance management of the visualization system, a multi-layer fault-tolerant mechanism for the early warning system, and monitoring process interruption recovery capabilities, ensuring continuous and reliable monitoring even in extreme sea conditions and equipment failures. This invention effectively solves problems such as insufficient marine environment adaptability, difficulty in sensor fault diagnosis, unsatisfactory data fusion effects, and low early warning accuracy in offshore wind power platform monitoring, providing advanced technical support for intelligent operation and maintenance of offshore wind farms.

[0144] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A safety monitoring and early warning method based on offshore wind power platform data fusion, characterized in that, The method comprises the following steps: S1, collecting structural response data and environmental load data of the offshore wind power platform according to time synchronization requirements; S2, performing data fusion processing by using a hierarchical fusion strategy, the hierarchical fusion strategy comprising raw-level fusion, feature-level fusion and decision-level fusion; wherein the raw-level fusion is used for data integration and quality control, and outputs standardized data, the feature-level fusion extracts multi-dimensional feature information from the standardized data, and the decision-level fusion fuses the multi-dimensional feature information and outputs a state evaluation result; The data integration comprises establishing a unified time reference and a spatial coordinate system, and aligning data in time and space; the quality control comprises establishing a data integrity index η=(N_valid / N_total)×100%, triggering a data compensation mechanism when η<95%; wherein N_valid represents the number of valid data, and N_total represents the total number of data; The feature-level fusion uses a multi-model Kalman filter to respectively establish a displacement state estimation model, a strain state estimation model and a vibration state estimation model, and the multi-dimensional feature information comprises a displacement state vector, a strain state vector and a vibration state vector; The decision-level fusion uses D-S evidence theory for evidence fusion, and outputs a comprehensive credibility of the platform state; S3, performing safety state prediction based on a dual model combining a data-driven prediction model and a physical mechanism prediction model, the data-driven prediction model taking data obtained by the hierarchical fusion strategy as input and taking a safety state evaluation result as output, and the physical mechanism prediction model taking environmental load data as input and taking a safety state evaluation result as output; the input of the data-driven prediction model comprises multi-dimensional feature information, a state evaluation result, environmental parameters, a historical state sequence of the multi-dimensional feature information and sensor health degree; the data-driven prediction model uses a trained LSTM-CNN neural network model.

2. The safety monitoring and early warning method based on offshore wind power platform data fusion according to claim 1, characterized in that, The feature-level fusion further comprises: using a dual adaptive weight strategy to fuse results of the displacement state estimation model, the strain state estimation model and the vibration state estimation model; the dual adaptive weight strategy comprises a model weight and a basic weight; No. A state estimation model at time 1 The weights w_i(k) = f(P_i(k)), where P_i(k) is the weight of the k-th ... A state estimation model at time 1 The error covariance matrix, i=1,2,3; The basic weight w_i_base(k) is automatically adjusted according to real-time sea conditions, sensor health degree and historical performance data; The final weights for each model: .

3. The safety monitoring and early warning method based on offshore wind power platform data fusion according to claim 1, characterized in that, In step S3, the physical mechanism prediction model is a sea fatigue damage accumulation model D ocean = D wind + D wave + D coupling, wherein D wind is a wind load fatigue damage D wind =∑(n_i / N_i), n_i is an actual cycle number at an i th stress level, N_i is a fatigue life at the i th stress level; D wave is a wave load fatigue damage, and D coupling is a wind wave coupling fatigue damage. When D ocean ≥1, an early warning is triggered. .

4. The safety monitoring and early warning method based on offshore wind power platform data fusion according to claim 3, characterized in that, The final prediction result of the safety state prediction is W(t)=α×P_(data)+β×P_(physics)+γ×P_(trend), wherein α, β and γ are weight coefficients, α+β+γ=1, typical values are α=0.4, β=0.4 and γ=0.2, P_data is a prediction probability (0-1) of the data-driven prediction model, P_physics is a prediction probability (0-1) of the physical mechanism prediction model, P_trend is a prediction probability (0-1) of trend analysis, and W(t) is a comprehensive early warning index (0-1).

5. The safety monitoring and early warning method based on offshore wind power platform data fusion according to claim 1, characterized in that, In step S1, the structural response data includes displacement state, strain state and vibration state, and the environmental load data includes temperature, humidity, wave height and wind speed; the monitoring unit for data collection includes GNSS monitoring unit, InSAR monitoring unit, distributed optical fiber sensor unit, vibration sensor unit and environmental monitoring unit.

6. The safety monitoring and early warning method based on offshore wind power platform data fusion according to any one of claims 1-5, characterized in that, The safety monitoring and early warning method further comprises: constructing a three-dimensional visualization scene based on the BIM model, and mapping the monitoring data fused in step S2 to the three-dimensional model components in real time.

7. A safety monitoring and early warning system based on offshore wind power platform data fusion, characterized in that, Comprise: a data collection module for collecting structural response data and environmental load data of the offshore wind power platform; a data fusion processing module for data integration and quality control through original level fusion, outputting standardized data, extracting multi-dimensional feature information from the standardized data through feature level fusion, and fusing the multi-dimensional feature information through decision level fusion to output state evaluation results; the data integration comprises establishing a unified time reference and spatial coordinate system, and aligning data in time and space; the quality control comprises establishing a data integrity index η=(N_valid / N_total)×100%, triggering a data compensation mechanism when η<95%; wherein N_valid represents the number of valid data, and N_total represents the total number of data; the feature level fusion adopts multi-model Kalman filtering to respectively establish displacement state estimation model, strain state estimation model and vibration state estimation model, and the multi-dimensional feature information includes displacement state vector, strain state vector and vibration state vector; the decision level fusion adopts D-S evidence theory for evidence fusion to output the comprehensive credibility of the platform state; an early warning module for safety state prediction based on a dual model combining a data-driven prediction model and a physical mechanism prediction model, wherein the data-driven prediction model takes the data obtained by the data fusion processing module as input and takes the safety state evaluation result as output, and the physical mechanism prediction model takes the environmental load data as input and takes the safety state evaluation result as output; the input of the data-driven prediction model includes multi-dimensional feature information, state evaluation result, environmental parameters, historical state sequence of the multi-dimensional feature information and sensor health degree; the data-driven prediction model adopts a trained LSTM-CNN neural network model.

8. The safety monitoring and early warning system based on offshore wind power platform data fusion according to claim 7, characterized in that, The data collection module comprises GNSS monitoring unit, InSAR monitoring unit, distributed optical fiber sensor unit, vibration sensor unit and environmental monitoring unit; The safety monitoring and early warning system further comprises a three-dimensional visualization module for constructing a three-dimensional model of the wind power platform based on the BIM model, and realizing real-time fusion display of the monitoring data and the three-dimensional model.

9. The safety monitoring and early warning system based on offshore wind power platform data fusion according to claim 8, characterized in that, The safety monitoring and early warning system further comprises: The marine environment self-adaptive calibration module is used to establish a marine environment calibration model T_cal(t)=T_base+α× Wave_displacement(t)+β×Foundation_settlement(t), wherein T_cal(t) is a calibrated measurement reference, T_base is an initial reference, Wave_displacement(t) is a wave-induced instantaneous displacement, Foundation_settlement(t) is a foundation long-term settlement, and α is a wave displacement influence coefficient; β is a foundation settlement influence coefficient, which is adaptively adjusted according to real-time wave parameters, sea state levels and platform structure characteristics; the marine environment self-adaptive calibration module is also used to establish a signal attenuation compensation model according to the corrosion influence of seawater salt mist on the optical fiber sensor; And / or, the sensor fault self-diagnosis module is used to establish a sensor health state evaluation mechanism, including a data consistency verification algorithm and a signal quality evaluation algorithm, and a data compensation mechanism is automatically started when a sensor fault is detected.

Images