A Method and System for Regional Monitoring and Early Warning of Offshore Wind Farms Based on Multi-Source Information

Abstract

This invention proposes a method and system for regional monitoring and early warning of offshore wind farms based on multi-source information, belonging to the field of offshore wind farm safety monitoring technology. The method includes: collecting vibration response data of pile foundation structures, seabed topography data, submarine cable tension data, and marine environmental parameters through a multi-type sensor network; acquiring bio-optical response signals through a biosensor monitoring unit; fusing the multi-source information using a multi-stage process including spatiotemporal registration, feature extraction, correlation mapping, and conflict resolution to generate a reliable comprehensive sensing result; performing trend extrapolation based on a neural network predictive analysis model to generate predicted data on pile foundation scour depth and submarine cable displacement; and generating an early warning signal and initiating a response through an early warning level mapping mechanism. This invention integrates physical and biological response information, significantly improving the accuracy, reliability, and timeliness of monitoring and early warning.

Description

Technical Field

[0001] This invention relates to the field of offshore wind farm safety monitoring technology, and in particular to a method and system for regional monitoring and early warning of offshore wind farms based on multi-source information. Background Technology

[0002] Offshore wind farms, as crucial clean energy infrastructure, operate in harsh marine environments. Their pile foundations and submarine cable systems are susceptible to structural damage and displacement risks due to multiple factors, including ocean currents, geological shifts, and biological activity. Traditional monitoring methods often rely on single-type sensors, failing to comprehensively capture the multi-physics coupling effects in complex environments. This results in low early warning accuracy, high false alarm rates, and difficulty in timely identification of potential risks. Furthermore, existing methods lack effective utilization of the correlation between biological behavior and geomechanical responses, hindering deep fusion and reliability assessment of multi-source information, thus limiting the improvement of offshore wind farm safety monitoring. Summary of the Invention

[0003] To address the aforementioned problems in existing technologies, the first aspect of this invention proposes a method for regional monitoring and early warning of offshore wind farms based on multi-source information, comprising:

[0004] S1: Based on a multi-type sensor network deployed in the offshore wind farm area, collect multi-source monitoring information including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters;

[0005] S2: Based on biosensing monitoring units deployed in the seabed area and along the submarine cable route, acquire bio-optical response signals;

[0006] S3: Based on multi-source monitoring information and bio-optical response signals, multi-source data fusion processing is performed through a multi-stage data processing flow that includes spatiotemporal registration, feature extraction, correlation mapping and conflict resolution to generate a comprehensive perception result including credibility.

[0007] S4: Based on the comprehensive sensing results, the trend extrapolation is performed through a neural network prediction and analysis model to generate a set of predicted data on pile foundation scour depth and submarine cable displacement.

[0008] S5: Based on the predicted data set, generate a warning level signal through a warning level mapping mechanism and activate the corresponding warning response mechanism.

[0009] In some implementations, S3 includes:

[0010] S31: Based on multi-source monitoring information and bio-optical response signals, spatiotemporal registration is performed through time synchronization and spatial coordinate alignment to generate a registered multi-source dataset.

[0011] S32: Based on the registered multi-source dataset, physical features from the multi-source monitoring information and biological features from the biological optical response signal are extracted by feature extraction algorithms to generate a multi-dimensional feature set. Among them, physical features include acoustic terrain features and mechanical response features, and biological features include spectral intensity features and spatial distribution features.

[0012] S33: Based on a multi-dimensional feature set, establish the correlation mapping relationship between physical features and biological features through feature association analysis, and generate a feature mapping relationship table;

[0013] S34: Based on the feature mapping relationship table, the consistency judgment and credibility calibration of abnormal areas are performed through conflict detection and data re-verification mechanisms to generate comprehensive perception results.

[0014] In some implementations, S33 includes:

[0015] S331: Based on a multi-dimensional feature set, an anomaly region identification algorithm is used to extract the range of anomaly regions indicated by multi-source monitoring information and generate a physical anomaly region map.

[0016] S332: Based on a multi-dimensional feature set, the stress region range displayed by the bio-optical response signal is extracted through stress response analysis to generate a bio-stress region map;

[0017] S333: By analyzing spatial overlap, determine the consistency in distribution between the physical anomaly area map and the biological stress area map, and generate consistency judgment results;

[0018] S334: Based on the consistency judgment results, the credibility of consistent regions is enhanced through a credibility weighting mechanism, and conflict detection and data re-verification are performed on inconsistent regions.

[0019] In some implementations, S32 includes:

[0020] S321: Based on the multi-source monitoring information in the registered multi-source dataset, acoustic terrain features and mechanical response features are extracted through frequency domain analysis and mode decomposition to generate a subset of physical features;

[0021] S322: Based on the bio-optical response signals in the registered multi-source dataset, spectral intensity features and spatial distribution features are extracted through spectral decoupling and spatial clustering to generate a subset of bio-features;

[0022] S323: Merge and standardize the subsets of physical features and biological features to generate a multi-dimensional feature set.

[0023] In some implementations, S34 includes:

[0024] S341: Based on the feature mapping relationship table, a consistency check algorithm is used to identify target regions where physical and biological features differ, and conflict region identifiers are generated.

[0025] S342: Based on the conflict area identification, initiate the re-acquisition command, and re-acquisition data of the target area through a multi-type sensor network and biosensing monitoring unit to generate a re-acquisition dataset;

[0026] S343: Based on the re-collected dataset, the authenticity of the data is determined through redundancy verification and expert rule base, and verified regional status data is generated.

[0027] S344: Integrate the verified regional state data into the comprehensive perception results to complete data re-verification and credibility calibration.

[0028] In some implementations, S4 includes:

[0029] S41: Based on the comprehensive sensing results, the pile foundation scour depth and submarine cable displacement are standardized through the data preprocessing module to generate standardized time series data.

[0030] S42: Based on standardized time series data, a time series prediction analysis is performed using a long short-term memory neural network model to generate a development trend sequence of pile foundation scour depth and submarine cable displacement.

[0031] S43: Based on the development trend sequence, the changes in pile foundation scour depth and submarine cable displacement in future periods are inferred through the trend extrapolation algorithm to generate a predictive data set.

[0032] In some implementations, S5 includes:

[0033] S51: Based on the predicted dataset, the pile foundation scour depth and submarine cable displacement are compared with their respective preset thresholds to generate comparison results;

[0034] S52: Based on the comparison results, the risk status is divided into multiple warning levels, and warning level identifiers are generated;

[0035] S53: Based on the warning level identifier, generate the corresponding warning level signal and trigger the warning response mechanism.

[0036] In some implementations, the multi-type sensor network includes acoustic sensors, mechanical sensors, and hydrological sensors, and S1 includes:

[0037] S11: Acquisition of seabed topographic data around the pile foundation based on acoustic sensors;

[0038] S12: Collect vibration response data of pile foundation structure and tension data of submarine cable based on mechanical sensors;

[0039] S13: Generate marine environmental parameters based on flow velocity, water temperature and turbidity data collected by hydrological sensors.

[0040] In some implementations, the training process of a neural network predictive analytics model includes:

[0041] Based on historical multi-source monitoring information, historical bio-optical response signals and the corresponding actual results of pile foundation scour depth and submarine cable displacement, a training sample set with ground truth labels is constructed through sample annotation.

[0042] Based on the training sample set, a long short-term memory network model is trained using the backpropagation algorithm to generate a preliminary prediction model.

[0043] Based on online acquisition of multi-source monitoring information and bio-optical response signals, incremental learning and adaptive parameter adjustment are performed on the preliminary prediction model to generate a neural network prediction and analysis model.

[0044] Secondly, the present invention provides a regional monitoring and early warning system for offshore wind farms based on multi-source information. The system employs the method provided in any of the above embodiments, and the system includes:

[0045] The multi-source information acquisition module is used to collect multi-source monitoring information, including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters, based on a multi-type sensor network deployed in the offshore wind farm area.

[0046] The biosignal acquisition module is used to acquire bio-optical response signals based on biosensing monitoring units deployed in the seabed area and along the submarine cable path;

[0047] The multi-source data fusion processing module is used to perform multi-source data fusion processing based on multi-source monitoring information and bio-optical response signals through a multi-stage data processing flow including spatiotemporal registration, feature extraction, correlation mapping and conflict resolution, to generate a comprehensive perception result including credibility.

[0048] The trend prediction and analysis module is used to perform trend extrapolation processing based on comprehensive perception results and through a neural network prediction and analysis model to generate a set of predicted data on pile foundation scour depth and submarine cable displacement.

[0049] The early warning generation and response module is used to generate early warning level signals based on the predicted data set through an early warning level mapping mechanism, and to activate the corresponding early warning response mechanism.

[0050] Compared with existing technologies, the advantages of this invention are as follows: By employing a multi-source information-based method for regional monitoring and early warning of offshore wind farms, it effectively solves the problems of narrow coverage and low reliability of traditional monitoring methods. This method first collects vibration response data of the pile foundation structure, seabed topography data, submarine cable tension data, and marine environmental parameters through a multi-type sensor network, forming a foundation for multi-source monitoring information. Simultaneously, it utilizes a biosensor monitoring unit to acquire bio-optical response signals, expanding the monitoring dimensions. Subsequently, a multi-stage data processing workflow is used to fuse the multi-source monitoring information and bio-optical response signals, including spatiotemporal registration, feature extraction, correlation mapping, and conflict resolution, generating a comprehensive perception result with reliability assessment, significantly improving the reliability and consistency of the data. Based on this, a neural network predictive analysis model is used to extrapolate the trends of pile foundation scour depth and submarine cable displacement, generating a predictive dataset and achieving a forward-looking transformation from data to early warning. Finally, an early warning level signal is generated through an early warning level mapping mechanism, and a response mechanism is activated, achieving refined differentiation of risk levels and timely response. The entire process is closely integrated, with multi-source information and biological signals complementing each other, and neural network models enhancing predictive capabilities, thus improving the overall accuracy, reliability, and timeliness of monitoring offshore wind farm areas. Attached Figure Description

[0051] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0052] Figure 1 The diagram shown is a flowchart of a method for regional monitoring and early warning of offshore wind farms based on multi-source information provided in an embodiment of the present invention.

[0053] Figure 2 The diagram shown is a structural schematic of an offshore wind farm area monitoring and early warning system based on multi-source information provided in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0055] The specific embodiments of the present invention will be described below.

[0056] Example 1

[0057] like Figure 1 As shown, this invention proposes a regional monitoring and early warning method for offshore wind farms based on multi-source information, including:

[0058] S1: Based on a multi-type sensor network deployed in the offshore wind farm area, collect multi-source monitoring information including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters;

[0059] S2: Based on biosensing monitoring units deployed in the seabed area and along the submarine cable route, acquire bio-optical response signals;

[0060] S3: Based on multi-source monitoring information and bio-optical response signals, multi-source data fusion processing is performed through a multi-stage data processing flow that includes spatiotemporal registration, feature extraction, correlation mapping and conflict resolution to generate a comprehensive perception result including credibility.

[0061] S4: Based on the comprehensive sensing results, the trend extrapolation is performed through a neural network prediction and analysis model to generate a set of predicted data on pile foundation scour depth and submarine cable displacement.

[0062] S5: Based on the predicted data set, generate a warning level signal through a warning level mapping mechanism and activate the corresponding warning response mechanism.

[0063] Specifically, this method first relies on a multi-type sensor network deployed within the offshore wind farm area. This network is responsible for collecting multi-source monitoring information, including vibration response data of the pile foundation structure, seabed topography data, submarine cable tension data, and marine environmental parameters. These data reflect the structural and environmental conditions from different physical dimensions. Simultaneously, bio-optical response signals are acquired through biosensor monitoring units deployed in the seabed area and along the submarine cable path, utilizing the sensitivity of marine organisms to environmental changes to supplement the limitations of physical sensing.

[0064] Multi-source monitoring information and bio-optical response signals then enter a multi-stage data processing flow. This flow encompasses key steps such as spatiotemporal registration, feature extraction, correlation mapping, and conflict resolution. By coordinating and fusing multi-source heterogeneous data, a comprehensive perception result with credibility evaluation is generated, significantly improving the accuracy of state identification. Based on this result, a neural network predictive analysis model is further used to extrapolate trends, outputting a set of predicted data on pile foundation scour depth and submarine cable displacement, enabling effective inference from the current state to future evolution. Finally, using an early warning level mapping mechanism, the predicted values ​​are compared with preset thresholds to classify risk levels and trigger corresponding early warning response mechanisms, achieving tiered management and control.

[0065] The core of the "acquiring biological optical response signals" mechanism in S2 lies in utilizing the sensitivity of marine benthic or attached organisms to the physicochemical changes in their environment. Through the optical characteristics exhibited by changes in their behavior, physiology, or population distribution, they indirectly reflect the structural health status and environmental disturbances of the offshore wind farm area.

[0066] Specifically, this mechanism relies on biosensing monitoring units pre-deployed along the seabed and cable routes. These units typically consist of: first, monitoring equipment with optical sensing capabilities, such as underwater spectrometers or high-sensitivity optical cameras; second, carriers that induce or capture biological responses, such as specific disturbance-sensitive biological communities (e.g., certain filter-feeding mollusks, polychaete annelids, or microbial membranes) or their biomimetic substitutes; and third, support structures and signal transmission units. These units are deployed in arrays in key areas, especially around pile foundations and vulnerable points of the cable.

[0067] Its working principle is as follows: When the marine environment changes due to pile foundation erosion, submarine cable displacement, or geological shifts, such as sediment disturbance, water flow changes, micro-deformation of the seabed, or increased turbidity, it directly induces stress responses in organisms inhabiting that environment. These responses can manifest as changes in organism movement behavior (such as shell closure in mollusks and escape in annelids), changes in population distribution (such as biofilm shedding and changes in aggregation), or changes in physiological characteristics (such as the release of fluorescent substances and changes in pigmentation). The biosensing monitoring unit continuously or triggerably acquires the optical signals corresponding to these biological responses. For example, it identifies abnormal biological behavior by observing changes in spectral absorption / reflection characteristics in specific wavelength bands, or identifies changes in biological distribution patterns through spatial image analysis, thereby obtaining biological optical response signals.

[0068] The effectiveness of this mechanism stems from the early and sensitive response of organisms to environmental changes, often enabling them to detect subtle or early anomalies more effectively than traditional physical sensors. For example, slight erosion of the seabed may cause benthic organisms to flee their original habitats, altering their distribution density in optical images; minute displacements or vibrations of submarine cables may induce stress behaviors in attached organism communities, thereby changing their apparent optical properties. Through these changes in optical signals, the system can acquire biological behavioral information that complements and verifies physical monitoring data, providing crucial input for subsequent multi-source data fusion.

[0069] This method effectively overcomes the inherent problems of limited monitoring range, delayed response, and high false alarm rate of single sensors through the synergistic effect of multi-source information acquisition, biosignal assistance, deep data fusion, intelligent prediction, and hierarchical early warning. A multi-type sensor network provides rich physical parameters, and bio-optical response signals introduce ecological behavior indicators, expanding the sensing dimensions. Multi-stage data processing ensures consistency of information at the spatiotemporal and feature levels, while a conflict resolution mechanism further enhances output credibility. The neural network prediction model, trained using historical and real-time data, adapts to complex nonlinear changes and achieves high prediction accuracy. Early warning level mapping realizes the transformation from data to decision, enhancing the system's practicality. Overall, this method demonstrates outstanding effectiveness in improving monitoring comprehensiveness, early warning accuracy, and response timeliness.

[0070] In some implementations, S3 includes:

[0071] S31: Based on multi-source monitoring information and bio-optical response signals, spatiotemporal registration is performed through time synchronization and spatial coordinate alignment to generate a registered multi-source dataset.

[0072] S32: Based on the registered multi-source dataset, physical features from the multi-source monitoring information and biological features from the biological optical response signal are extracted by feature extraction algorithms to generate a multi-dimensional feature set. Among them, physical features include acoustic terrain features and mechanical response features, and biological features include spectral intensity features and spatial distribution features.

[0073] S33: Based on a multi-dimensional feature set, establish the correlation mapping relationship between physical features and biological features through feature association analysis, and generate a feature mapping relationship table;

[0074] S34: Based on the feature mapping relationship table, the consistency judgment and credibility calibration of abnormal areas are performed through conflict detection and data re-verification mechanisms to generate comprehensive perception results.

[0075] Specifically, this section includes four sub-steps: spatiotemporal registration, feature extraction, correlation mapping, and conflict resolution. First, spatiotemporal registration is performed, which synchronizes the time and aligns the spatial coordinates of multi-source monitoring information and bio-optical response signals. This eliminates deviations caused by differences in sensor distribution and acquisition times, forming a multi-source dataset under a unified spatiotemporal benchmark, providing a consistent data foundation for subsequent analysis. Feature extraction, on the registered multi-source dataset, extracts physical features (such as acoustic topographic features and mechanical response features) from the multi-source monitoring information and biological features (such as spectral intensity features and spatial distribution features) from the bio-optical response signals, constructing a multi-dimensional feature set to comprehensively characterize the physical and biological state of the monitored object. Next, correlation mapping is performed, establishing the correspondence between physical and biological features through feature association analysis, forming a feature mapping relationship table. This reveals the inherent connections between different information sources, enhancing the basis for state judgment. Finally, conflict resolution is performed, using the feature mapping relationship to perform consistency judgment and credibility calibration on abnormal areas, resolving discrepancies between different source data and outputting the final comprehensive perception result.

[0076] This process effectively integrates multi-source information through progressively refined data processing, compensating for the shortcomings of a single data source and improving the reliability of monitoring results and the effectiveness of decision-making. Spatiotemporal registration ensures data comparability, feature extraction achieves information dimensionality reduction and enhancement, correlation mapping uncovers cross-domain associations, and conflict resolution improves output robustness. The overall process is logically rigorous and progressively advanced, significantly improving the system's perception capabilities and reliability.

[0077] In some implementations, S33 includes:

[0078] S331: Based on a multi-dimensional feature set, an anomaly region identification algorithm is used to extract the range of anomaly regions indicated by multi-source monitoring information and generate a physical anomaly region map.

[0079] S332: Based on a multi-dimensional feature set, the stress region range displayed by the bio-optical response signal is extracted through stress response analysis to generate a bio-stress region map;

[0080] S333: By analyzing spatial overlap, determine the consistency in distribution between the physical anomaly area map and the biological stress area map, and generate consistency judgment results;

[0081] S334: Based on the consistency judgment results, the credibility of consistent regions is enhanced through a credibility weighting mechanism, and conflict detection and data re-verification are performed on inconsistent regions.

[0082] Specifically, spatial consistency judgment among features is achieved through anomaly region identification and stress response analysis. This process first uses anomaly region identification algorithms to identify areas of physical anomaly from multi-source monitoring information based on a multi-dimensional feature set, generating a physical anomaly region map to indicate the location of potential structural damage or environmental variations. Simultaneously, stress response analysis extracts areas of abnormal biological behavior from bio-optical response signals, generating a biological stress region map reflecting the organism's response to environmental disturbances. Subsequently, spatial overlap analysis compares the physical anomaly region map and the biological stress region map to determine their consistency in distribution, forming a consistency judgment result. High overlap indicates mutual corroboration between physical anomalies and biological responses, increasing credibility; inconsistency suggests potential false alarms or unidentified interference factors requiring further processing. Finally, based on the consistency judgment result, a confidence-weighted mechanism is used to enhance the confidence of consistent regions, while conflict detection and data re-verification processes are initiated for inconsistent regions to ensure the high reliability of the final comprehensive perception result.

[0083] This method effectively improves the accuracy of fault identification, reduces false alarms and missed alarms, and enhances the system's adaptability to complex marine environments by combining the spatial correlation between physical and biological anomalies.

[0084] In some implementations, S32 includes:

[0085] S321: Based on the multi-source monitoring information in the registered multi-source dataset, acoustic terrain features and mechanical response features are extracted through frequency domain analysis and mode decomposition to generate a subset of physical features;

[0086] S322: Based on the bio-optical response signals in the registered multi-source dataset, spectral intensity features and spatial distribution features are extracted through spectral decoupling and spatial clustering to generate a subset of bio-features;

[0087] S323: Merge and standardize the subsets of physical features and biological features to generate a multi-dimensional feature set.

[0088] Specifically, this step extracts representative features from multi-source monitoring information and bio-optical response signals, and then fuses and standardizes them. For multi-source monitoring information, frequency domain analysis and mode decomposition are used to process acoustic and mechanical data, extracting acoustic topographic features and mechanical response features. These features reflect the vibration characteristics of the pile foundation structure, seabed topography changes, and the stress state of the submarine cable, forming a subset of physical features. For bio-optical response signals, spectral decoupling technology is used to separate light intensity information in different bands, and spatial clustering methods are combined to identify the distribution patterns of biological activities, extracting spectral intensity features and spatial distribution features, forming a subset of biological features. Subsequently, the physical and biological feature subsets are merged, and standardization is used to eliminate the influence of dimensional differences and numerical ranges, generating a multi-dimensional feature set, providing a unified and comparable data foundation for subsequent correlation analysis. This feature extraction and fusion process fully utilizes the complementarity of multi-source information; physical features reveal mechanical and topographic changes, while biological features reflect ecological responses. The combination of the two significantly enhances the ability to characterize complex working conditions and improves the comprehensiveness and accuracy of monitoring.

[0089] In some implementations, S34 includes:

[0090] S341: Based on the feature mapping relationship table, a consistency check algorithm is used to identify target regions where physical and biological features differ, and conflict region identifiers are generated.

[0091] S342: Based on the conflict area identification, initiate the re-acquisition command, and re-acquisition data of the target area through a multi-type sensor network and biosensing monitoring unit to generate a re-acquisition dataset;

[0092] S343: Based on the re-collected dataset, the authenticity of the data is determined through redundancy verification and expert rule base, and verified regional status data is generated.

[0093] S344: Integrate the verified regional state data into the comprehensive perception results to complete data re-verification and credibility calibration.

[0094] Specifically, the process first uses a consistency check algorithm based on a feature mapping table to identify target areas where physical and biological characteristics differ significantly, generating conflict area markers to indicate areas requiring focused verification. Then, a re-acquisition command is triggered based on the conflict area markers, controlling a multi-type sensor network and biosensing monitoring unit to perform targeted data re-acquisition of the target area, obtaining more detailed and targeted data to form a re-acquisition dataset. Next, the authenticity of the re-acquisition data is determined by combining redundancy check and an expert rule base. Redundancy check eliminates random errors through cross-validation of multi-sensor data, while the expert rule base incorporates domain knowledge to identify abnormal patterns, jointly generating verified area status data. Finally, the verified area status data is integrated into the comprehensive sensing results, completing the re-verification and credibility calibration of the conflict area data, improving the overall reliability of the output results.

[0095] This mechanism effectively resolves contradictions among multiple sources of information through proactive re-collection and multi-method verification, reduces the risk of misjudgment, and enhances the robustness of the system and the credibility of decision-making in complex environments.

[0096] In some implementations, S4 includes:

[0097] S41: Based on the comprehensive sensing results, the pile foundation scour depth and submarine cable displacement are standardized through the data preprocessing module to generate standardized time series data.

[0098] S42: Based on standardized time series data, a time series prediction analysis is performed using a long short-term memory neural network model to generate a development trend sequence of pile foundation scour depth and submarine cable displacement.

[0099] S43: Based on the development trend sequence, the changes in pile foundation scour depth and submarine cable displacement in future periods are inferred through the trend extrapolation algorithm to generate a predictive data set.

[0100] Specifically, based on the comprehensive sensing results, a neural network predictive analysis model is used for trend extrapolation. This process first standardizes the pile scour depth and submarine cable displacement using a data preprocessing module, eliminating dimensional differences and noise interference to generate standardized time-series data suitable for neural network input. The Long Short-Term Memory (LSTM) neural network model then performs time-series predictive analysis on the standardized time-series data. This model effectively captures long-term dependencies and nonlinear variation patterns in the time series, generating a trend sequence for the pile scour depth and submarine cable displacement. Based on this, a trend extrapolation algorithm extrapolates changes over future periods, generating a predictive dataset by fitting existing trends and considering the continuity of environmental factors.

[0101] This prediction method fully leverages the powerful learning capabilities of neural networks and the scientific nature of time-series extrapolation, significantly improving the accuracy and reliability of prediction results. Data preprocessing ensures data quality, providing clean input for the model. Long Short-Term Memory (LSTM) neural networks selectively remember and forget information through their gating mechanism, effectively handling long-term dependencies in time-series data. The trend extrapolation algorithm scientifically extends current trends into the future, providing a basis for forward-looking judgments. The entire prediction process realizes the transformation from multi-source fusion data to future state estimates, providing direct support for early warning decisions. This method can identify potential risk trends early, facilitating preventative measures and avoiding further structural damage. The prediction results, based on multi-source fusion data, have high reliability and practicality. The neural network model can be continuously trained and optimized using historical data to adapt to different sea areas and environmental conditions. Trend extrapolation considers the inherent patterns of time series, avoiding the shortcomings of simple linear prediction. Overall, this prediction method enhances the foresight and intelligence level of offshore wind farm monitoring and early warning.

[0102] In some implementations, S5 includes:

[0103] S51: Based on the predicted dataset, the pile foundation scour depth and submarine cable displacement are compared with their respective preset thresholds to generate comparison results;

[0104] S52: Based on the comparison results, the risk status is divided into multiple warning levels, and warning level identifiers are generated;

[0105] S53: Based on the warning level identifier, generate the corresponding warning level signal and trigger the warning response mechanism.

[0106] Specifically, the mechanism first compares the predicted data of pile foundation scour depth and submarine cable displacement with preset thresholds. These thresholds are determined based on structural safety specifications and historical operational data. The comparison results directly reflect the degree to which the parameters deviate from the safe range. Based on the comparison results, the system classifies the risk status into multiple warning levels, such as normal, caution, warning, and danger, each corresponding to a different degree of risk and urgency. After a warning level identifier is generated, the system automatically triggers the corresponding warning response mechanism. Response measures may include issuing alarm signals, generating maintenance reports, adjusting operating parameters, or activating emergency procedures.

[0107] This early warning mechanism achieves refined risk management and differentiated responses through multi-level classification, avoiding a one-size-fits-all approach. Threshold comparison provides objective risk assessment standards, ensuring the scientific basis of decision-making. Multi-level early warning adapts to risk situations of varying severity, optimizing resource allocation. An automatic response mechanism ensures timely implementation of measures after an early warning, reducing human delays. The entire process forms a closed-loop management system from prediction to response, significantly improving the safety assurance capabilities of wind farms. Early warning levels can be dynamically adjusted according to actual conditions, maintaining system adaptability. The response mechanism can be integrated with existing management systems for collaborative operation. This design enhances the practicality and operability of early warning information, providing effective support for the safe maintenance of wind farms.

[0108] In some implementations, the multi-type sensor network includes acoustic sensors, mechanical sensors, and hydrological sensors, and S1 includes:

[0109] S11: Acquisition of seabed topographic data around the pile foundation based on acoustic sensors;

[0110] S12: Collect vibration response data of pile foundation structure and tension data of submarine cable based on mechanical sensors;

[0111] S13: Generate marine environmental parameters based on flow velocity, water temperature and turbidity data collected by hydrological sensors.

[0112] Specifically, the network comprises three main types of sensors: acoustic sensors, mechanical sensors, and hydrological sensors. Acoustic sensors collect seabed topographic data around the pile foundation based on the principle of sound wave detection, measuring seabed elevation changes and geomorphic features by emitting sound waves and receiving echoes. Mechanical sensors collect vibration response data of the pile foundation structure and tension data of the submarine cable. Vibration response data reflects the dynamic characteristics of the pile foundation under the action of waves and ocean currents, while tension data directly indicates the stress state of the submarine cable. Hydrological sensors collect marine environmental parameters such as current velocity, water temperature, and turbidity, which directly affect the pile foundation scour process and the service condition of the submarine cable.

[0113] Multiple types of sensors work collaboratively to acquire state information about the wind farm area from different dimensions, forming a comprehensive monitoring data foundation. Acoustic sensors provide information on seabed topographic changes, mechanical sensors capture structural mechanical responses, and hydrological sensors record environmental conditions; these three complement each other to form a complete monitoring system. This multi-source data acquisition method overcomes the limitations of single sensor types, providing more comprehensive state perception. Data can be cross-verified, improving the reliability of monitoring results. The sensor network layout can cover key areas, achieving comprehensive spatial monitoring. Acquired data is transmitted in real-time or near real-time, ensuring timely information. This sensor system provides high-quality input for subsequent data fusion and early warning analysis, serving as the fundamental support for the entire monitoring and early warning system.

[0114] In some implementations, the training process of a neural network predictive analytics model includes:

[0115] Based on historical multi-source monitoring information, historical bio-optical response signals and the corresponding actual results of pile foundation scour depth and submarine cable displacement, a training sample set with ground truth labels is constructed through sample annotation.

[0116] Based on the training sample set, a long short-term memory network model is trained using the backpropagation algorithm to generate a preliminary prediction model.

[0117] Based on online acquisition of multi-source monitoring information and bio-optical response signals, incremental learning and adaptive parameter adjustment are performed on the preliminary prediction model to generate a neural network prediction and analysis model.

[0118] Specifically, the training process first constructs a training sample set with ground truth labels based on historical multi-source monitoring information, historical bio-optical response signals, and corresponding actual results of pile scour depth and submarine cable displacement. The sample labeling process requires the participation of domain experts to ensure the accuracy of the labeled data. Based on the training sample set, a long short-term memory network model is trained using an error backpropagation algorithm. This algorithm calculates the error between the predicted output and the true value, and then backpropagates the error to adjust the network parameters, gradually approximating the true input-output relationship. After generating the preliminary prediction model, incremental learning and adaptive parameter adjustment are performed on the preliminary prediction model based on online-collected multi-source monitoring information and bio-optical response signals, enabling the model to adapt to environmental changes and system evolution.

[0119] The entire training process emphasizes the combination of historical and real-time data, ensuring the model possesses both basic predictive capabilities and adaptability. Historical data provides rich training samples, ensuring the model learns sufficient pattern features. Expert annotation guarantees the quality of training data, providing the model with reliable learning targets. The error backpropagation algorithm efficiently optimizes network parameters, improving model accuracy. The incremental learning mechanism enables the model to continuously improve and adapt to new situations. Adaptive parameter adjustment maintains the model's timeliness and accuracy. The scientific and standardized training process provides a reliable model foundation for predictive analysis. The model can maintain optimal performance through periodic retraining. The entire training scheme ensures the reliability and practicality of the neural network predictive analysis model.

[0120] Example 2

[0121] like Figure 2 As shown, this invention provides a monitoring and early warning system for offshore wind farm areas based on multi-source information. The system employs the method provided in any of the above embodiments and includes:

[0122] The multi-source information acquisition module is used to collect multi-source monitoring information, including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters, based on a multi-type sensor network deployed in the offshore wind farm area.

[0123] The biosignal acquisition module is used to acquire bio-optical response signals based on biosensing monitoring units deployed in the seabed area and along the submarine cable path;

[0124] The multi-source data fusion processing module is used to perform multi-source data fusion processing based on multi-source monitoring information and bio-optical response signals through a multi-stage data processing flow including spatiotemporal registration, feature extraction, correlation mapping and conflict resolution, to generate a comprehensive perception result including credibility.

[0125] The trend prediction and analysis module is used to perform trend extrapolation processing based on comprehensive perception results and through a neural network prediction and analysis model to generate a set of predicted data on pile foundation scour depth and submarine cable displacement.

[0126] The early warning generation and response module is used to generate early warning level signals based on the predicted data set through an early warning level mapping mechanism, and to activate the corresponding early warning response mechanism.

[0127] This system corresponds to the method provided in Example 1, and will not be described in detail here.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for regional monitoring and early warning of offshore wind farms based on multi-source information, characterized in that, include: S1: Based on a multi-type sensor network deployed in the offshore wind farm area, collect multi-source monitoring information including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters; S2: Based on biosensing monitoring units deployed in the seabed area and along the submarine cable route, acquire bio-optical response signals; S3: Based on the multi-source monitoring information and the bio-optical response signal, multi-source data fusion processing is performed through a multi-stage data processing flow including spatiotemporal registration, feature extraction, correlation mapping and conflict resolution to generate a comprehensive perception result including credibility. S4: Based on the comprehensive perception results, a set of predicted data on pile foundation scour depth and submarine cable displacement is generated by performing trend extrapolation processing through a neural network prediction and analysis model. S5: Based on the predicted data set, generate a warning level signal through a warning level mapping mechanism, and activate the corresponding warning response mechanism; S3 includes: S31: Based on the multi-source monitoring information and the bio-optical response signal, spatiotemporal registration is performed through time synchronization and spatial coordinate alignment processing to generate a registered multi-source dataset; S32: Based on the registered multi-source dataset, physical features in the multi-source monitoring information and biological features in the biological optical response signal are extracted by feature extraction algorithm to generate a multi-dimensional feature set. The physical features include acoustic terrain features and mechanical response features, and the biological features include spectral intensity features and spatial distribution features. S33: Based on the multi-dimensional feature set, establish the correlation mapping relationship between physical features and biological features through feature association analysis, and generate a feature mapping relationship table; S34: Based on the feature mapping relationship table, the consistency judgment and credibility calibration of abnormal areas are performed through conflict detection and data re-verification mechanisms to generate the comprehensive perception result.

2. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, S33 includes: S331: Based on the multi-dimensional feature set, the abnormal area range indicated by multi-source monitoring information is extracted by the abnormal area identification algorithm to generate a physical abnormal area map; S332: Based on the multi-dimensional feature set, extract the stress region range displayed by the bio-optical response signal through stress response analysis, and generate a bio-stress region map; S333: By analyzing spatial overlap, determine the consistency in distribution between the physical anomaly area map and the biological stress area map, and generate a consistency judgment result; S334: Based on the consistency judgment result, the credibility of consistent regions is enhanced through a credibility weighting mechanism, and conflict detection and data re-verification are performed on inconsistent regions.

3. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, S32 includes: S321: Based on the multi-source monitoring information in the registered multi-source dataset, acoustic terrain features and mechanical response features are extracted through frequency domain analysis and mode decomposition to generate a subset of physical features; S322: Based on the bio-optical response signals in the registered multi-source dataset, spectral intensity features and spatial distribution features are extracted through spectral decoupling and spatial clustering to generate a subset of bio-features; S323: Merge and standardize the physical feature subset and the biological feature subset to generate the multi-dimensional feature set.

4. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, S34 includes: S341: Based on the feature mapping relationship table, a consistency check algorithm is used to identify target regions where physical features and biological features differ, and conflict region identifiers are generated. S342: Based on the conflict area identifier, initiate a re-acquisition command, and re-acquisition data of the target area through the multi-type sensor network and the biosensing monitoring unit to generate a re-acquisition dataset; S343: Based on the re-collected dataset, the authenticity of the data is determined by redundancy verification and expert rule base, and verified regional status data is generated; S344: The verified regional state data is fused into the comprehensive perception result to complete data re-verification and credibility calibration.

5. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, S4 include: S41: Based on the comprehensive sensing results, the pile foundation scour depth and submarine cable displacement are standardized by the data preprocessing module to generate standardized time series data. S42: Based on the standardized time series data, a time series prediction analysis is performed using a long short-term memory neural network model to generate a development trend sequence of pile foundation scour depth and submarine cable displacement. S43: Based on the development trend sequence, the changes in pile foundation scour depth and submarine cable displacement in future periods are extrapolated using a trend extrapolation algorithm to generate the predicted data set.

6. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, S5 include: S51: Based on the predicted data set, the pile foundation scour depth and submarine cable displacement are compared with their respective preset thresholds to generate comparison results; S52: Based on the comparison results, the risk status is divided into multiple warning levels, and a warning level identifier is generated; S53: Based on the warning level identifier, generate a corresponding warning level signal and trigger the warning response mechanism.

7. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, The multi-type sensor network includes acoustic sensors, mechanical sensors, and hydrological sensors. S1 includes: S11: Acquisition of seabed topography data around the pile foundation based on acoustic sensors; S12: Collect vibration response data of pile foundation structure and tension data of submarine cable based on mechanical sensors; S13: Generate marine environmental parameters based on flow velocity, water temperature and turbidity data collected by hydrological sensors.

8. The method for regional monitoring and early warning of offshore wind farms based on multi-source information according to claim 1, characterized in that, The training process of the neural network prediction and analysis model includes: Based on historical multi-source monitoring information, historical bio-optical response signals and the corresponding actual results of pile foundation scour depth and submarine cable displacement, a training sample set with ground truth labels is constructed through sample annotation. Based on the training sample set, a long short-term memory network model is trained using the backpropagation algorithm to generate a preliminary prediction model. Based on the multi-source monitoring information and the bio-optical response signal collected online, the preliminary prediction model is incrementally learned and the parameters are adaptively adjusted to generate a neural network prediction analysis model.

9. A regional monitoring and early warning system for offshore wind farms based on multi-source information, characterized in that, The system employs the method described in any one of claims 1 to 8, the system comprising: The multi-source information acquisition module is used to collect multi-source monitoring information, including pile foundation structure vibration response data, seabed topography data, submarine cable tension data, and marine environmental parameters, based on a multi-type sensor network deployed in the offshore wind farm area. The biosignal acquisition module is used to acquire bio-optical response signals based on biosensing monitoring units deployed in the seabed area and along the submarine cable path; The multi-source data fusion processing module is used to perform multi-source data fusion processing based on the multi-source monitoring information and the bio-optical response signal through a multi-stage data processing flow including spatiotemporal registration, feature extraction, correlation mapping and conflict resolution, to generate a comprehensive perception result including credibility. The trend prediction and analysis module is used to perform trend extrapolation processing based on the comprehensive perception results and through a neural network prediction and analysis model to generate a set of predicted data on pile foundation scour depth and submarine cable displacement. The early warning generation and response module is used to generate an early warning level signal based on the predicted data set through an early warning level mapping mechanism, and to activate the corresponding early warning response mechanism.

Images