A real-time monitoring method and system for the verticality of a hybrid tower wind power tower

Abstract

The present application relates to the technical field of monitoring of mixed tower wind power tower, and discloses a real-time monitoring method and system for verticality of mixed tower wind power tower, which comprises a data sensing layer, a communication control layer and a verticality analysis layer; the data sensing layer realizes synchronous measurement and dynamic calibration of multiple points of tower deformation through a displacement sensor array, an inclinometer and the like; the communication control layer completes time synchronization, format standardization and fusion transmission of multi-source data through edge computing, protocol conversion and space-time registration technology; the verticality analysis layer combines design parameters and real-time data to construct a holographic tower space posture model, and realizes deformation prediction, reference correction and abnormal alarm through a dynamic weighting algorithm.The method comprises data acquisition and calibration, processing and transmission, model construction and analysis and result output.The present application realizes real-time and accurate monitoring of tower verticality, improves operation safety and is suitable for state monitoring of mixed tower wind power tower.

Description

Technical Field

[0001] This invention relates to the field of monitoring technology for hybrid wind turbine towers, specifically to a real-time monitoring method and system for the verticality of hybrid wind turbine towers. Background Technology

[0002] With the increasing global demand for clean energy, wind power has been widely promoted as an important renewable energy source. As a key supporting structure for wind turbine generators, the verticality of the hybrid-tower wind turbine directly affects the operational safety, stability, and power generation efficiency of the unit. During actual operation, hybrid-tower wind turbine towers are affected by various factors, such as strong wind loads, temperature changes, mechanical vibration, and foundation settlement. These factors may cause structural deformations such as axial displacement, radial offset, and torsional deformation, leading to verticality deviations. Failure to monitor tower verticality changes in a timely and accurate manner may result in accelerated wear of turbine components, decreased power generation efficiency, and even serious safety accidents, causing significant economic losses and social impact.

[0003] Traditional methods for monitoring tower verticality primarily rely on periodic manual inspections and single-point measuring equipment, such as total stations and theodolites. These methods have significant limitations: manual inspections are inefficient and time-consuming, unable to achieve real-time monitoring, and struggle to detect dynamic tower deformation promptly; single-point measuring equipment can only acquire localized data from limited locations, failing to comprehensively reflect the overall deformation characteristics of the tower, resulting in low monitoring accuracy and reliability. Furthermore, traditional methods typically lack effective fusion and analysis of multi-source monitoring data, making it difficult to establish accurate tower spatial attitude models, scientifically predict and warn of deformation trends, and achieve dynamic calibration and compensation of the monitoring system.

[0004] With the rapid development of intelligent sensing technology, Internet of Things (IoT) technology, big data analytics technology, and digital twin technology, new technical approaches and solutions have been provided for the real-time monitoring of the verticality of hybrid wind turbine towers. Key technical challenges that urgently need to be addressed in the field of hybrid wind turbine tower monitoring include: how to utilize advanced sensor technology to achieve multi-point synchronous measurement of tower structural deformation; how to achieve real-time transmission and fusion processing of multi-source monitoring data through communication technology; how to construct a holographic spatial attitude model of the tower using spatial geometric analysis and dynamic modeling techniques; and how to use intelligent algorithms to predict deformation trends, correct benchmark deviations, and issue alarms for verticality anomalies. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time monitoring method and system for the verticality of hybrid wind turbine towers, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a real-time monitoring system for the verticality of hybrid wind turbine towers, the system comprising:

[0007] Data perception layer, communication control layer, and verticality resolution layer;

[0008] The data sensing layer includes a displacement sensor array, an inclination measuring instrument, a dynamic calibration device, and a reference positioning module, which are used to perform multi-point synchronous measurement of tower structure deformation, reference coordinate calibration, and dynamic calibration compensation operation based on feedback instructions.

[0009] The communication control layer includes edge computing nodes, protocol converters, and cloud processing platforms. It also deploys time synchronization mechanisms and multimodal data fusion strategies to establish redundant transmission links, achieve time alignment and spatial registration of multi-source monitoring data, and use adaptive compression algorithms to extract features and transmit layered data from heterogeneous data according to data accuracy requirements and transmission bandwidth. It also enables dynamic interaction between physical measurement data and 3D modeling data by configuring multi-source data interfaces.

[0010] The verticality analysis layer is used to construct a holographically mapped tower spatial attitude model by dynamically calibrating the tower's three-dimensional deformation model using spatial geometric analysis technology combined with structural design parameters and real-time monitoring datasets. Based on this attitude model, a dynamic weighted algorithm is used to predict deformation trends, correct benchmark deviations, and issue verticality anomaly alarms.

[0011] Preferably, the displacement sensor array includes at least a laser ranging unit, a fiber optic grating sensor, an ultrasonic echo detector, and a millimeter-wave radar, used to acquire the displacement distribution characteristics of the tower cross-section; the dynamic calibration device includes at least a temperature compensation module, a vibration suppressor, a coordinate correction unit, and a data calibrator, used to eliminate environmental errors in the measurement system; the reference positioning module includes at least a Beidou positioning terminal, a total station reference point, and an inertial navigation component, used to execute coordinate calibration commands issued by the verticality resolution layer.

[0012] Preferably, in the communication control layer, the data collected by the displacement sensor array is transmitted from the edge computing node to the protocol converter after being timestamped. The protocol converter then performs data format standardization processing and transmits it together with the spatial attitude information recorded by the tilt measuring instrument to the cloud processing platform for modeling and calculation. The data processing adopts spatiotemporal registration technology, which inputs the collected displacement data into three-dimensional mesh models of different resolutions for joint analysis.

[0013] The method of configuring a multi-source data interface to achieve dynamic interaction between physical measurement data and 3D modeling data includes: configuring a multi-source data interface to achieve dynamic interaction between physical measurement data and 3D modeling data, exchanging real-time tower deformation monitoring data, performing joint analysis on the acquired spatial coordinates, realizing dynamic coordinate system transformation, completing data verification, model update and parameter distribution, and issuing calibration control commands to the data perception layer.

[0014] Preferably, the construction of the holographically mapped tower spatial attitude model includes:

[0015] Establish a parameter correlation channel and a two-way calibration mechanism between physical structure deformation and digital modeling space;

[0016] By using spatial interpolation and coordinate transformation of actual measurement data, and based on the axial displacement, radial offset, and torsional angle obtained by the displacement sensor array, a reference attitude library and deformation feature library for the tower structure are constructed. Based on real-time monitoring data, model parameters are iterated and dynamic characteristics are simulated, transforming the actual structural deformation into a high-precision digital twin model.

[0017] The parameters of the spatial attitude model are optimized by inputting the deformation data monitored in real time into the established attitude model and using the residual correction algorithm to dynamically compensate the output of the model to obtain the optimized tower spatial attitude model.

[0018] The tower spatial attitude model includes a physical measurement space, a digital modeling space, a structural database, and an interaction mechanism between the modules.

[0019] The physical measurement space serves as the data source for the attitude model, containing the original deformation characteristics of the tower structure. The digital modeling space forms a mapping relationship with the physical measurement space, geometrically representing the tower's attitude characteristics through multi-dimensional spatial modeling. The structural database integrates design parameters and real-time monitoring information, providing a basic dataset including a material property library, a load distribution library, and a connection node library. The interaction mechanism enables data communication between modules. The physical measurement space and the structural database achieve parameter acquisition and model updates through a standardized interface. The physical measurement space and the digital modeling space transmit parameters through a data channel. The digital modeling space and the structural database achieve information interaction through middleware.

[0020] Preferably, the step of predicting deformation trend using a dynamic weighted algorithm based on the attitude model includes:

[0021] Based on the tower spatial attitude model, historical operational attitude data, environmental load records, and deformation event characteristic data are obtained to construct a deformation sample set.

[0022] After performing feature enhancement processing on the deformed sample set, it is divided into a training set and a test set;

[0023] A hybrid prediction architecture of ARIMA-LSTM-Prophet is established. The model initialization parameters are set, and the training set is input into the hybrid model for joint training. The ARIMA model is used for trend decomposition, the LSTM network is used for temporal feature extraction, and the Prophet algorithm is used for periodic pattern recognition. An adaptive weighting mechanism is used to balance the prediction errors of different algorithms under specific working conditions until the model fit reaches the set threshold or the predetermined training rounds are completed.

[0024] Input the test set into the trained hybrid model, calculate the overall performance index of the model, and select the optimal deformation prediction model;

[0025] Based on the optimal deformation prediction model, the spatial distribution trend of tower deformation is output, and the evolution path of verticality deviation is determined by combining material mechanical parameters.

[0026] The training process of the hybrid prediction architecture includes:

[0027] The Boosting method is used to generate training sequences for multiple base models, and the predictive performance of each base model is evaluated by sliding window validation.

[0028] The gradient descent algorithm is used to calculate the dynamic fusion coefficients of each base model, and a weighted average mechanism is used to make a comprehensive prediction of the output results of the base models.

[0029] Preferably, the step of correcting the reference deviation using a dynamic weighted algorithm based on the attitude model includes:

[0030] Based on the tower's spatial attitude model, axial deviation, radial offset, and torsional deformation parameters of the structural deformation are extracted, and a set of deviation feature vectors is constructed.

[0031] Independent component analysis was used to denoise the deviation feature vector and obtain the core deviation feature components.

[0032] A biased classification model based on Gaussian mixture model is established, and the optimal number of classifications is determined by the BIC criterion.

[0033] The categorized deviation features are input into a preset correction strategy library to match the optimal correction scheme and generate a targeted set of benchmark correction parameters.

[0034] Preferably, the step of using a dynamic weighted algorithm based on the attitude model to generate a verticality anomaly alarm includes:

[0035] Based on the tower spatial attitude model, structural vibration characteristics, wind speed load spectrum, and foundation settlement data are collected to construct an operational status feature library.

[0036] The data in the runtime status feature library is subjected to overlapping segmentation to generate a spatially distributed sample set.

[0037] A convolutional neural network model is established, the kernel size and pooling parameters are set, and the deep representation of spatial features is obtained through backpropagation.

[0038] Spatial features are input into a classifier for state pattern recognition, and the output results distinguish between normal operating state and verticality abnormal state.

[0039] Preferably, the construction of the holographically mapped tower spatial attitude model further includes:

[0040] A sliding spatial window mechanism is used to divide the continuous monitoring area into blocks, and feature extraction is performed independently on the data blocks within each spatial window;

[0041] Establish a correlation matrix between data block features and structural loads, record the deformation distribution patterns corresponding to different working conditions, incrementally update the attitude model through an online learning algorithm, and trigger a model topology reconstruction mechanism when an unrecorded deformation pattern is detected.

[0042] Preferably, the method for generating the benchmark correction parameter set includes:

[0043] Establish a table of correspondence between deviation types and correction methods, including correction methods for axial deviation corresponding to prestress adjustment and radial offset corresponding to counterweight optimization;

[0044] A genetic algorithm is used to search for the optimal combination of correction parameters, including adjustment magnitude, application location, and correction timing. The correction effect is evaluated in real time through a dual closed-loop control mechanism. When the residual deviation exceeds the threshold, the parameter re-optimization process is triggered.

[0045] Preferably, the present invention also includes a method for real-time monitoring of the verticality of hybrid wind turbine towers, comprising the following steps:

[0046] Multi-point synchronous data acquisition of the tower structure is carried out by a displacement sensor array and an inclination measuring instrument to obtain axial displacement, radial offset and torsional angle measurements. Based on the coordinate calibration results of the reference positioning module, a dynamic calibration device is used to compensate for temperature drift and eliminate vibration interference.

[0047] The collected heterogeneous monitoring data is timestamped through edge computing nodes. A multimodal data fusion strategy is used to synchronize and register displacement and tilt data in time and space. Based on the transmission bandwidth constraints, an adaptive compression algorithm is used to extract displacement distribution features. Finally, the physical measurement data is converted into a standardized data format compatible with the three-dimensional modeling space through a protocol converter.

[0048] The tower structure design parameters are loaded into the cloud processing platform, and the pre-constructed three-dimensional deformation model is dynamically calibrated by combining the real-time monitoring dataset. After optimizing the model parameters by using the residual correction algorithm, a holographically mapped tower spatial attitude model is generated. Based on this model, the axial deformation gradient, radial offset trend and torsional angle change are fused by the dynamic weighting algorithm, and the deformation trend prediction result, the benchmark deviation correction amount and the verticality anomaly alarm signal are output.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] In terms of data acquisition and calibration, the data sensing layer integrates a displacement sensor array, an inclination meter, a dynamic calibration device, and a benchmark positioning module, enabling multi-point synchronous measurement of tower structural deformation. The displacement sensor array includes various sensors such as a laser ranging unit and a fiber optic grating sensor, capable of acquiring the displacement distribution characteristics of the tower cross-section. The dynamic calibration device effectively eliminates environmental errors through components such as a temperature compensation module and a vibration suppressor. The benchmark positioning module uses BeiDou positioning terminals to achieve coordinate calibration, ensuring the high accuracy and reliability of the raw monitoring data and providing a solid foundation for subsequent analysis.

[0051] The communication control layer constructs an efficient data transmission and processing system. Edge computing nodes timestamp data, and protocol converters standardize data formats. Combined with time synchronization mechanisms and multimodal data fusion strategies, it ensures time alignment and spatial registration of multi-source data. Adaptive compression algorithms extract features and perform layered transmission based on data accuracy and transmission bandwidth, ensuring data integrity while improving transmission efficiency. The multi-source data interface enables dynamic interaction between physical measurement data and 3D modeling data, promoting deep integration of monitoring data and models.

[0052] The verticality analysis layer achieves precise analysis of the tower's status through spatial geometric analysis and intelligent algorithms. The constructed holographic mapping tower spatial attitude model, by establishing parameter correlation and a two-way calibration mechanism between the physical structure and the digital modeling space, combined with spatial interpolation, coordinate transformation, and residual correction algorithms, transforms actual deformation into a high-precision digital twin model, comprehensively and in real-time reflecting the tower's spatial attitude. Based on this model's dynamic weighted algorithm, in deformation trend prediction, an ARIMA-LSTM-Prophet hybrid prediction architecture is adopted, integrating the advantages of multiple algorithms to achieve scientific prediction of the spatial distribution trend of tower deformation and the evolution path of verticality deviation. In benchmark deviation correction, independent component analysis and Gaussian mixture models are used to achieve noise reduction, classification, and optimal correction scheme matching of deviation features, improving the pertinence and effectiveness of benchmark correction. In verticality anomaly alarm, a convolutional neural network model is used for in-depth analysis and pattern recognition of operating status characteristics, enabling timely and accurate identification of abnormal states, providing reliable assurance for the safe operation of the tower.

[0053] Furthermore, the system employs a sliding space window mechanism, online learning algorithm, and model topology reconstruction mechanism, enabling the attitude model to perform incremental updates and adaptive adjustments based on real-time monitoring data. This improves the model's adaptability to complex working conditions and novel deformation modes. The application of genetic algorithms and dual closed-loop control mechanisms in the generation of baseline correction parameters achieves the optimization of correction parameters and real-time evaluation of correction effects, further enhancing the system's intelligence and automation level. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the working principle of the real-time monitoring system for the verticality of hybrid wind turbine towers described in this invention.

[0055] Figure 2 A flowchart for data processing and interaction in the communication control layer;

[0056] Figure 3 This is a flowchart of tower deformation trend prediction based on a hybrid algorithm. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Please see Figures 1-3 This invention relates to a real-time monitoring system for the verticality of hybrid wind turbine towers. The core architecture of the system consists of a data sensing layer, a communication control layer, and a verticality resolution layer. The specific implementation steps are as follows:

[0059] The data sensing layer, through a displacement sensor array, tilt meter, dynamic calibration device, and benchmark positioning module, achieves multi-point synchronous measurement of tower structural deformation, benchmark coordinate calibration, and dynamic calibration compensation. Specifically, the displacement sensor array covers key sections of the tower, acquiring multi-dimensional data such as axial displacement and radial offset in real time; the tilt meter synchronously records the tower's spatial attitude parameters; the dynamic calibration device compensates for environmental errors (such as temperature drift and vibration interference) in real time based on feedback commands; and the benchmark positioning module obtains absolute coordinate benchmarks based on BeiDou positioning terminals and total station reference points, providing an initial reference for verticality calculation.

[0060] The communication control layer constructs redundant transmission links through edge computing nodes, protocol converters, and a cloud processing platform to achieve time alignment and spatial registration of multi-source data. Edge computing nodes add timestamps to sensor data, protocol converters standardize data formats, and the cloud processing platform employs a multimodal data fusion strategy, combined with adaptive compression algorithms, to achieve feature extraction and hierarchical transmission of heterogeneous data. It also enables dynamic interaction between physical measurement data and 3D modeling data through multi-source data interfaces.

[0061] The verticality analysis layer, based on spatial geometric analysis technology and combining structural design parameters with real-time monitoring data, dynamically calibrates the tower's three-dimensional deformation model to construct a holographically mapped spatial attitude model of the tower. Based on this model, a dynamic weighted algorithm is used to predict deformation trends, correct benchmark deviations, and issue verticality anomaly alarms, providing real-time decision support for the safe operation of the tower.

[0062] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0063] Example 1:

[0064] The data sensing layer, as the core of the real-time verticality monitoring system for hybrid wind turbine towers, achieves multi-point synchronous measurement of tower structural deformation, benchmark coordinate calibration, and dynamic calibration compensation through the coordinated operation of displacement sensor arrays, tilt measuring instruments, dynamic calibration devices, and benchmark positioning modules. The design of this layer must meet the requirements of high precision, anti-interference, and strong real-time performance to ensure reliable raw data for subsequent data processing and analysis.

[0065] I. Composition and Functional Implementation of Displacement Sensor Array

[0066] The displacement sensor array employs a three-dimensional layout combining multiple sensor types, including at least a laser ranging unit, fiber Bragg grating sensors, ultrasonic echo detectors, and millimeter-wave radar. These sensors are distributed at key locations on different height sections of the tower, based on their measurement principles and applicable scenarios. The laser ranging unit emits a laser beam and receives reflected light from the measured surface. Using the time-of-flight method, it calculates the distance change between the sensor and the tower surface, achieving high-precision measurement of radial offset data. Its measurement accuracy reaches the millimeter level, making it suitable for areas prone to significant deformation, such as the top of the tower and the interfaces of various sections. The fiber Bragg grating sensors are fixed to the surface of the tower's steel structure by pasting or embedding. Utilizing the characteristic that the Bragg wavelength of the fiber Bragg grating changes with strain and temperature, it simultaneously monitors axial strain and ambient temperature parameters. When the tower undergoes axial tensile or compressive deformation, the grating pitch of the fiber Bragg grating changes, causing a shift in the reflected light wavelength. The strain data is obtained by analyzing the wavelength change using a demodulator, and a temperature compensation algorithm is used to eliminate the influence of temperature drift on the measurement results. Ultrasonic echo detectors transmit ultrasonic pulses and receive reflected echoes from the internal structure of the tower. Based on the time delay and amplitude characteristics of the echo signals, they analyze the displacement distribution and internal defects of the tower cross-section, making them suitable for detecting deformation characteristics of complex structures such as flange connection nodes. Millimeter-wave radar, on the other hand, transmits broadband millimeter-wave signals and receives scattered echoes from the tower surface. Using synthetic aperture radar (SAR) technology, it generates two-dimensional or three-dimensional point cloud data of the tower, enabling non-contact, full-area monitoring of the tower's dynamic deformation. This is particularly suitable for continuous monitoring in harsh environments such as strong winds, rain, and snow.

[0067] Various sensors are distributed in arrays across horizontal sections at different heights of the tower. Within the same section, sensors are evenly distributed circumferentially, forming a multi-layered ring monitoring network. For example, 3-5 monitoring sections are set at the bottom, middle, and top of the tower, with 8-12 sensors arranged in each section to ensure the acquisition of radial, axial, and torsional deformation data covering the entire height of the tower. A synchronous triggering mechanism between sensors enables simultaneous data acquisition from multiple points, with synchronization errors controlled at the microsecond level to ensure the spatiotemporal consistency of subsequent data fusion.

[0068] II. Layout and Data Acquisition of the Inclinometer

[0069] The tilt measuring instrument is used to monitor the tower's spatial attitude parameters in real time, including pitch, yaw, and roll angles. This device employs a high-precision MEMS inertial measurement unit (IMU), integrating a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. It calculates the tower's real-time attitude angles using a multi-sensor fusion algorithm. The tilt measuring instruments are installed at the center of the tower's top and at the centroids of each segment to acquire attitude change data for the entire tower and its individual segments. The top tilt measuring instrument primarily monitors the overall verticality deviation of the tower, while the segment tilt measuring instruments analyze the impact of localized tower deformation on the overall attitude. Measurement data is transmitted in real time to the communication control layer via wired or wireless communication links for time synchronization and spatial registration with displacement sensor data.

[0070] III. Error Elimination Mechanism of Dynamic Calibration Device

[0071] The dynamic calibration device is a core component of the data sensing layer for eliminating environmental errors, integrating a temperature compensation module, vibration suppressor, coordinate correction unit, and data calibrator. The temperature compensation module establishes a temperature-deformation error model to correct sensor drift caused by changes in ambient temperature in real time. Specifically, a temperature sensor is deployed near temperature-sensitive devices such as fiber optic grating sensors to monitor ambient temperature changes in real time; a polynomial function relationship between temperature and measurement error is established by fitting historical data. For the axial strain measurement error of the fiber optic grating sensor, it can be expressed as:

[0072] Δε=a·T 2 +b·T+c

[0073] Where T represents temperature, and a, b, and c are fitting coefficients. During data acquisition, error compensation is calculated based on real-time temperature values ​​and superimposed on the original measurement data to achieve dynamic correction of temperature drift. The vibration suppressor employs inertial damping technology, reducing interference from external excitations such as wind loads and mechanical vibrations on the measurement signal by placing damping mass blocks on the sensor mounting base. The mass and stiffness parameters of the damping mass blocks are optimized based on the tower's natural vibration frequency, creating a resonance suppression effect for vibrations at specific frequencies, thereby reducing the impact of vibration noise on displacement and tilt angle measurement data.

[0074] The coordinate correction unit dynamically adjusts the sensor coordinate system based on the coordinate calibration instructions issued by the verticality resolution layer. The calibration instructions are generated based on the coordinate calibration results of the reference positioning module. For example, when the Beidou positioning terminal detects that the tower foundation has settled, causing a shift in the global coordinate system, the coordinate correction unit converts the data in the sensor's local coordinate system into absolute coordinate values ​​in the global coordinate system through translation and rotation transformation matrices, ensuring the spatial reference consistency of data from different sensors.

[0075] The data calibrator verifies the validity of the collected data using a redundancy check algorithm. For displacement sensor arrays, a cross-comparison method is used to compare data from multiple sensors within the same cross-section. When the deviation of a sensor's measurement value from the average value of other sensors within the same cross-section exceeds a preset threshold (e.g., 3 times the standard deviation), the data is determined to be an outlier and discarded, while simultaneously triggering a sensor fault alarm mechanism. For tilt angle measurement data, the reliability of the attitude angle data is ensured by verifying the results of multi-sensor fusion from the inertial navigation component.

[0076] IV. Coordinate Calibration Process of the Base Positioning Module

[0077] The reference positioning module consists of a BeiDou positioning terminal, a total station reference point, and an inertial navigation component, responsible for providing absolute coordinate reference and attitude reference for the entire monitoring system. The BeiDou positioning terminal uses a multi-mode satellite receiver, capable of simultaneously receiving satellite signals from multiple systems such as BeiDou, GPS, and GLONASS. It obtains the three-dimensional absolute coordinates of the tower foundation through real-time dynamic differential (RTK) technology, achieving centimeter-level positioning accuracy. The antenna of the BeiDou positioning terminal is deployed in an open area near the tower foundation to avoid signal quality issues caused by obstruction and multipath effects.

[0078] The total station reference point establishes the initial reference plane for tower verticality monitoring through optical measurement. The specific operation is as follows: Set up 2-3 total station points on stable ground around the tower, ensuring good visibility between each point; measure multiple characteristic points on the reference plane at the bottom of the tower using the total station to establish the equation of the reference plane in a local coordinate system; this reference plane serves as the initial reference for calculating the tower verticality, and all subsequent monitoring data are calculated based on this plane for deviation calculations.

[0079] The inertial navigation component is integrated into the tilt meter. Using real-time measurement data from gyroscopes and accelerometers, it calculates the tower's attitude angle change rate, enabling high-frequency monitoring of the tower's dynamic attitude (sampling frequency can reach over 100Hz). The inertial navigation data, BeiDou positioning data, and total station measurement data are fused using a Kalman filter algorithm to generate globally consistent coordinate references and attitude parameters, providing high-precision reference data for the verticality resolution layer.

[0080] The data perception layer, through the collaborative work of the aforementioned components, forms a complete process from data acquisition and error calibration to benchmark calibration. The displacement sensor array enables multi-point synchronous acquisition of multi-dimensional deformation data, the tilt meter provides attitude parameters, the dynamic calibration device eliminates environmental errors, and the benchmark positioning module establishes a coordinate benchmark. Together, they ensure that the data input to the communication control layer has high precision, spatiotemporal consistency, and physical accuracy, laying the foundation for subsequent real-time analysis and intelligent interpretation of tower verticality. The design of this layer fully considers the structural characteristics and complex operating environment of hybrid wind turbine towers. Through multi-sensor fusion, dynamic calibration, and benchmark calibration technologies, it solves problems such as insufficient accuracy and significant environmental interference associated with traditional single-point monitoring, achieving comprehensive, real-time, and reliable monitoring of tower deformation.

[0081] Example 2:

[0082] The communication control layer serves as the data transmission and processing hub for the real-time verticality monitoring system of hybrid wind turbine towers. Through the collaborative operation of edge computing nodes, protocol converters, and cloud processing platforms, it achieves time synchronization, spatial registration, feature extraction, and model interaction of multi-source heterogeneous data. The design of this layer must meet the requirements of high reliability, low latency, and strong compatibility to ensure the accuracy, integrity, and timeliness of data during transmission and processing.

[0083] Edge computing nodes, acting as near-end data processing units, are deployed at the base of the tower or in nearby cabinets. They directly interface with various sensors in the data sensing layer, undertaking the preprocessing and transmission control of raw data. Their core functions include timestamp marking, noise filtering, and data encapsulation. In the timestamp marking stage, edge computing nodes use built-in high-precision clock modules (such as GPS-based clock synchronization units) to add microsecond-accurate timestamps to each set of data collected by the displacement sensor array, ensuring strict temporal alignment of multi-source data (such as laser ranging data, fiber optic strain data, and millimeter-wave radar point cloud data). This temporal synchronization mechanism is the foundation for subsequent data fusion and modeling, avoiding spatial registration errors caused by time deviations. For noise filtering, edge computing nodes employ a sliding window filtering algorithm to suppress high-frequency vibration noise. For example, for high-frequency tower vibration caused by wind speed excitation, a 50ms sliding window is set, and median or mean filtering is applied to the displacement data within the window to remove abnormal fluctuations and retain the low-frequency trend components reflecting the true deformation of the tower. During the data encapsulation process, edge computing nodes package the processed sensor data according to a preset data frame format. Each data frame contains fields such as sensor number, timestamp, measurement value, and checksum, and is sent to the protocol converter through redundant transmission links (such as parallel industrial Ethernet and 5G wireless communication) to improve the reliability of data transmission. When the main link (such as industrial Ethernet) is interrupted due to a fault, the system automatically switches to the backup link (such as 5G network) to continue transmitting data, ensuring the continuity of the monitoring process.

[0084] As a core component for heterogeneous data format conversion, the protocol converter has a built-in multi-protocol parsing engine, supporting the parsing and conversion of various industrial communication protocols (such as Modbus, OPCUA, MQTT) and sensor-specific data formats. Its processing flow consists of two steps: First, for different types of raw data, such as ASCII distance values ​​output by laser ranging units, wavelength drift from fiber optic grating sensors, and point cloud data (binary format) from millimeter-wave radar, the protocol converter converts them into unified physical quantity units (such as millimeters, microstrain, meters) through the corresponding parsing module and fills them into a standardized data structure. Second, it timestamps the spatial attitude information (pitch angle, yaw angle, roll angle, in degrees) recorded by the tilt meter with the displacement data. A time synchronization algorithm (such as clock calibration based on the NTP protocol) ensures that the timestamp error between the two types of data does not exceed 10 microseconds, forming a spatiotemporally consistent dataset containing spatial position (X, Y, Z axis displacement) and attitude parameters. After format conversion and time alignment, the protocol converter transmits the data to a cloud processing platform via TCP / IP protocol for subsequent modeling and calculation. In addition, the protocol converter also supports data compression. Based on transmission bandwidth constraints and data accuracy requirements, it uses an adaptive compression algorithm (such as a wavelet transform-based compression algorithm) to perform layered compression on large-capacity data such as millimeter-wave radar point cloud data. High-resolution details are preserved for point cloud data of critical areas (such as tower flange connection nodes), while low-resolution compression is performed on data of non-critical areas, thereby reducing transmission bandwidth usage while ensuring data availability.

[0085] The cloud processing platform is the core of data fusion and model interaction in the communication and control layer. Built on a distributed computing architecture, it possesses powerful data storage, processing, and 3D modeling capabilities. The platform first uses spatiotemporal registration technology to jointly analyze multi-source monitoring data with the tower's 3D model. Specifically, the tower structure is divided into 3D mesh models of different resolutions. High-precision meshes (10-50 mm) are used for areas prone to local deformation, such as flange connection nodes and welds, while low-precision meshes (500-1000 mm) are used for regular structures like the tower body. Based on the spatial layout of the displacement sensors, the displacement data collected by each sensor is mapped to the corresponding mesh cells. For example, the radial offset data from the laser ranging unit at the top of the tower is mapped to the Y-axis coordinate of the top mesh cell, and the axial strain data from the fiber optic grating sensor is mapped to the Z-axis strain parameter of the mesh cell in its corresponding section. Through this resolution-based modeling and data mapping method, collaborative analysis of the overall deformation trend and local detailed features of the tower is achieved.

[0086] Regarding the dynamic interaction between physical measurement data and 3D modeling data, the cloud processing platform establishes a bidirectional data transmission channel by configuring multi-source data interfaces (such as RESTAPI and WebSocket interfaces). On one hand, the platform inputs real-time monitored deformation data (such as axial displacement, radial offset, and torsional angle of each section) into the 3D modeling space through the interface, triggering iterative updates of model parameters. For example, when the radial offset of a section exceeds the design threshold, the platform automatically adjusts the stiffness parameters of that section in the 3D model and recalculates the overall stress distribution of the tower to reflect performance changes caused by material fatigue or structural damage. On the other hand, theoretical parameters in the 3D modeling space (such as design load, material elastic modulus, and preset deformation threshold) are fed back to the data perception layer through the interface to optimize sensor layout and dynamic calibration parameters. For example, based on the simulation results of the 3D model, it is determined that the sensor deployment density needs to be increased in high-stress areas, or the temperature compensation coefficient of the dynamic calibration device needs to be adjusted to match the temperature-deformation relationship predicted by the model. During the interaction, the system automatically completes the dynamic transformation of the coordinate system. Through the coordinate transformation matrix (translation, rotation, scaling), the measurement data in the local coordinate system of the sensor is converted into the absolute coordinates in the global coordinate system of the tower. At the same time, the theoretical coordinates of the 3D model are converted into local coordinates that can be recognized by the sensor, ensuring the consistency of data in different spaces.

[0087] During data processing, the cloud-based processing platform also employs a multimodal data fusion strategy to fuse displacement data, tilt data, and environmental load data (such as wind speed and temperature) at multiple levels. In the data layer fusion stage, spatiotemporal registration and format standardization integrate various data types into a unified dataset. In the feature layer fusion stage, dimensionality reduction algorithms such as Principal Component Analysis (PCA) are used to extract core features characterizing tower deformation (such as principal displacement direction and maximum strain area) from the multidimensional data. In the decision-making layer fusion stage, combined with structural design specifications and historical operational data, comprehensive evaluation results (such as verticality deviation level and deformation trend warning) are generated. This multimodal fusion approach fully leverages the complementarity of different data types, improving the accuracy and reliability of monitoring results.

[0088] The communication control layer, through a three-tiered architecture of edge computing nodes, protocol converters, and a cloud processing platform, achieves end-to-end control from data acquisition to model analysis. Edge computing nodes address real-time preprocessing and reliable transmission of near-end data, protocol converters eliminate format barriers between heterogeneous data, and the cloud processing platform leverages powerful computing and modeling capabilities to enable in-depth data mining and dynamic interaction. This layer not only ensures the spatiotemporal consistency and physical accuracy of multi-source data but also provides efficient data support for model construction and dynamic weighting algorithms in the verticality analysis layer, making it a crucial link in achieving intelligent and real-time monitoring of the entire system. Its design fully considers the complexity of mixed-tower wind power tower monitoring scenarios, effectively addressing challenges such as large data volumes, high real-time transmission requirements, and strong environmental interference through redundant transmission links, adaptive compression algorithms, and multimodal data fusion, laying a solid foundation for accurate tower verticality monitoring and safety assessment.

[0089] Example 3:

[0090] The verticality analysis layer constructs a holographically mapped tower spatial attitude model to achieve dynamic calibration and accurate characterization of the tower's three-dimensional deformation. Its core processes include parameter correlation between physical and digital spaces, model construction, parameter optimization, and interaction mechanism design. This layer uses multi-dimensional modeling and data fusion technology to transform actual monitoring data into a digital twin model that can be used for deformation analysis, providing fundamental support for subsequent dynamic weighted algorithms.

[0091] The system first establishes a parameter correlation channel and a two-way calibration mechanism between the physical structural deformation and the digital modeling space. Raw data from the physical measurement space (such as axial strain collected by displacement sensors and yaw angle measured by tilt measuring instruments) are converted into regular mesh data using spatial interpolation algorithms (such as inverse distance weighted interpolation and Kriging interpolation) to adapt to the three-dimensional mesh structure of the digital modeling space. The coordinate transformation process employs a seven-parameter transformation method, using translation parameters (ΔX, ΔY, ΔZ), rotation parameters (ωx, ωy, ωz), and scale parameter (k) to convert the measurement data in the sensor's local coordinate system into absolute coordinates in the tower's global coordinate system, ensuring consistency with the spatial reference of the digital model. The two-way calibration mechanism allows theoretical parameters from the digital modeling space (such as design loads and material Poisson's ratio) to be transmitted back to the physical measurement space. For example, when the digital model simulation shows that the stress in a certain area exceeds the material's yield strength, the system automatically adjusts the sampling frequency or measurement range of the sensors in that area to obtain denser monitoring data.

[0092] Based on fundamental data such as axial displacement, radial offset, and torsional angle acquired by a displacement sensor array, the system constructs a reference attitude library and a deformation feature library for the tower structure. The reference attitude library stores the ideal geometric parameters of the tower under no-load conditions, including design coordinates of each section, verticality tolerance, and material elastic modulus, serving as a benchmark for deformation analysis. The deformation feature library, accumulated through historical monitoring data, records deformation patterns under different operating conditions (such as average wind speed of 5 m / s, strong wind speed of 25 m / s, and seismic conditions), including displacement distribution curves, strain peak locations, and torsional angle variation ranges. For example, under strong wind conditions, the deformation feature library records the nonlinear relationship between the radial offset of the tower top and wind speed, as well as the axial strain distribution at each segment interface. The reference attitude library and deformation feature library are iteratively updated using real-time monitoring data. The system employs an incremental learning algorithm, automatically updating the statistical parameters (such as mean and variance) of the corresponding operating conditions in the library each time new data is acquired, ensuring that the model always reflects the latest state of the tower.

[0093] During model construction, the system uses displacement sensor data as a foundation, combined with structural design parameters (such as tower wall thickness and flange connection stiffness), to generate an initial three-dimensional deformation model of the tower using finite element analysis. This model employs a hybrid modeling approach using shell and beam elements. The tower body uses shell elements to simulate a thin-walled structure, while the flange connection nodes use a combination of beam and spring elements to simulate complex stress characteristics. Subsequently, through spatial interpolation and coordinate transformation, the actual measurement data is mapped to the corresponding nodes of the model, forming a preliminary mapping relationship between the physical measurement space and the digital modeling space. For example, the axial strain value measured by a fiber optic grating sensor at a certain cross-section is applied as a boundary condition to the corresponding element of the model, driving the model to calculate the theoretical displacement value of that cross-section. This value is then compared with the measured value from the laser ranging unit to verify the initial accuracy of the model.

[0094] The parameter optimization process employs a residual correction algorithm to dynamically compensate for the model output. The specific steps are as follows: First, the real-time monitored deformation data (such as the measured radial offset of the tower's center at a certain moment) is compared with the predicted value from the digital model to calculate the residual (measured value - predicted value). Then, the residual is spatially fitted using the least squares method to generate a residual correction matrix, which contains the residual correction coefficients for each model node. Finally, the residual correction matrix is ​​embedded into the model calculation process to compensate for subsequent predictions node-by-node. For example, if the model predicts a radial offset of 20mm for a node, while the measured value is 22mm, the correction coefficient for that node is set to +2mm, and subsequent predictions will automatically increase the calculated result for that node by 2mm. Furthermore, the system uses a sliding spatial window mechanism to divide the continuous monitoring area into blocks. Each data block within a spatial window (such as a cylindrical area with a height of 5 meters and a circumference of 30 degrees) undergoes independent feature extraction. By calculating parameters such as the displacement gradient and mean strain of the area, a correlation matrix between the data block features and the structural load is established. For example, when the displacement gradient within a certain spatial window suddenly increases, the correlation matrix suggests that there may be a local load anomaly in that area, requiring further analysis.

[0095] The tower spatial attitude model comprises four major modules: physical measurement space, digital modeling space, structural database, and interaction mechanism. The physical measurement space serves as the data source, continuously inputting the tower's original deformation characteristics into the model in the form of a data stream through sensors. The digital modeling space visualizes the tower's attitude through multi-dimensional geometric modeling (such as 3D meshes and stress cloud maps) and outputs deformation prediction results. The structural database integrates design parameters and real-time monitoring information. The material property library stores data such as steel grade, yield strength, and elastic modulus; the load distribution library records historical wind speed loads and temperature loads; and the connection node library contains parameters such as flange bolt torque and weld strength. The interaction mechanism achieves data connectivity between modules through standardized interfaces—the physical measurement space and the structural database use API interfaces for parameter acquisition and model updates. For example, the structural database periodically obtains sensor calibration parameters from the physical measurement space and updates the temperature correction coefficient in the material property library. The physical measurement space and the digital modeling space transmit parameters in real time through data channels to ensure that measurement data and model calculations are synchronized. The digital modeling space and the structural database exchange information through middleware. For example, when calculating stress distribution, the digital model retrieves material elastic modulus and load distribution data from the structural database, generates a stress cloud map, and then feeds it back to the structural database for storage.

[0096] When a new deformation pattern (such as an unrecorded combination of torsional angles) is detected, the system triggers a model topology reconstruction mechanism. The specific process is as follows: First, by comparing data block features with the correlation matrix, the spatial window corresponding to the abnormal deformation pattern is identified; then, the model mesh within this spatial window is refined (e.g., the mesh size is reduced from 100mm to 20mm), and the element types are redefined (e.g., shell elements are converted to solid elements); finally, based on the refined mesh, the stress-strain distribution in this region is recalculated, the deformation feature library is updated, and the model parameters are adjusted. This dynamic reconstruction mechanism enables the model to adapt to the complex changes in the tower structure, improving its ability to capture sudden deformations.

[0097] The verticality analysis layer, through the holographic mapping tower spatial attitude model constructed using the aforementioned process, achieves precise alignment between the physical and digital worlds. This model not only reflects the tower's current deformation state in real time but also continuously improves prediction accuracy through parameter iteration and dynamic compensation mechanisms. It provides reliable spatial benchmarks and data support for deformation trend prediction, benchmark deviation correction, and verticality anomaly alarms based on dynamic weighted algorithms. Its design fully integrates spatial geometric analysis, digital twin, and real-time data processing technologies, solving the problem of insufficient adaptability of traditional monitoring models to complex operating conditions and providing an intelligent analysis tool for the safe operation of hybrid wind power towers.

[0098] Example 4:

[0099] The verticality analysis layer, based on the tower's spatial attitude model, uses a dynamic weighted algorithm to predict deformation trends, correct baseline deviations, and issue verticality anomaly alarms. Its core processes encompass data sample construction, hybrid model training, deviation feature processing, and the application of state recognition algorithms. This layer provides decision support for real-time assessment and fault warning of tower verticality through multi-algorithm fusion and intelligent analysis.

[0100] Deformation trend prediction constructs a sample set based on historical data. The system extracts historical operating attitude data (such as axial displacement, radial offset, and torsional angle at various times), environmental load records (such as wind speed, wind direction, and temperature at corresponding times), and deformation event feature data (such as operating parameters when the historical maximum offset occurred) from the tower spatial attitude model, forming a deformation sample set containing multi-dimensional information of time, space, and environment. To improve model training efficiency, feature enhancement processing is performed on the sample set, including data normalization (mapping displacement values ​​to the [-1,1] interval), missing value imputation (using linear interpolation between adjacent times), and outlier removal (based on the 3σ principle). The processed samples are divided into a training set (70%) and a test set (30%), which are used for model training and performance verification, respectively.

[0101] The hybrid prediction model employs an ARIMA-LSTM-Prophet architecture, combining the advantages of these three algorithms to achieve multi-scale trend analysis. During model initialization, key parameters are set, such as the difference order of the ARIMA model (d=1), the number of hidden layer neurons in the LSTM network (128), and the seasonal cycle parameters (annual, monthly, and daily cycles) of the Prophet algorithm. During training, the ARIMA model first performs trend decomposition on the time-series data of the training set, separating long-term trend terms (such as the increasing trend of axial displacement of the tower due to foundation settlement), seasonal cycle terms (such as the thermal expansion and contraction cycle caused by temperature changes), and random fluctuation terms. The LSTM network receives the decomposed random fluctuation terms and extracts nonlinear features from the time-series data through a gating mechanism (such as the abrupt change pattern of displacement under strong wind conditions). The Prophet algorithm then models the long-term trend terms and seasonal cycle terms, identifying periodic patterns in the data (such as the periodic changes in radial displacement caused by daily morning and evening temperature differences). The outputs of the three algorithms are fused through an adaptive weighting mechanism, with the weight coefficients dynamically adjusted according to the current operating conditions. For example, under strong wind conditions, the LSTM network, which excels at capturing nonlinear features, is given higher weights; under normal operating conditions, the weights of the three algorithms tend to be balanced. The training process uses the Boosting method to generate training sequences for multiple base models. A sliding window validation (window size of 100 samples) is used to evaluate the predictive performance of each base model in different time intervals. The gradient descent algorithm is used to optimize the dynamic fusion coefficients of the base models until the model fit (e.g., root mean square error RMSE) reaches a set threshold (e.g., ≤5mm) or a predetermined number of training rounds (e.g., 500 rounds) is completed. After training, the test set is input into the hybrid model, and comprehensive performance indicators (e.g., RMSE, mean absolute error MAE) are calculated. The model with the highest prediction accuracy is selected as the optimal deformation prediction model. Based on the spatial distribution trend of tower deformation output by the model (such as the predicted value sequence of radial offset of each section in the next 48 hours), combined with material mechanical parameters (such as the elastic modulus of steel and allowable stress), the evolution path of deviation is simulated by finite element analysis to determine the development trend and potential risk areas of verticality deviation.

[0102] The baseline deviation correction process begins with feature extraction and denoising. The system extracts real-time parameters of axial deviation, radial offset, and torsional deformation from the tower's spatial attitude model, constructing a feature vector set containing multi-dimensional deviation quantities (e.g., a vector at a certain moment of [ΔZ = 30mm, ΔR = 25mm, θ = 0.8°]). Since the monitoring data may be affected by environmental noise, Independent Component Analysis (ICA) is used to denoise the feature vectors. The core deviation feature components are separated by maximizing the non-Gaussianity criterion, eliminating irrelevant components such as sensor noise and electromagnetic interference. Subsequently, a deviation classification model is established based on a Gaussian Mixture Model (GMM). The optimal number of classifications is determined using the Bayesian Information Criterion (BIC) (e.g., dividing deviations into "slight," "moderate," and "severe" categories). Each category corresponds to a different deviation range and physical meaning—for example, "slight" deviation corresponds to normal deformation during daily operation, while "severe" deviation may indicate structural damage. The classified deviation features are input into a pre-defined correction strategy library, which contains a table showing the correspondence between axial deviation and prestress adjustment, radial offset and counterweight optimization, and other correction methods. Taking radial offset as an example, the system matches the weight and installation position of the counterweight according to the offset magnitude: when the offset is ≤50mm, the first-level counterweight scheme is triggered (adding a 50kg adjustable mass block); when the offset is >50mm and ≤100mm, the second-level scheme is triggered (adding a 100kg mass block and adjusting its position). The correction parameters (adjustment range, application position, and correction timing) are searched for the optimal combination using a genetic algorithm. The algorithm uses minimizing the residual deviation as its objective function, generating multiple generations of candidate solutions through selection, crossover, and mutation operations to ultimately determine the optimal parameter combination. A dual closed-loop control mechanism is employed during the correction process: the inner loop calculates the residual deviation using real-time monitoring data and evaluates the correction effect; the outer loop adjusts the genetic algorithm parameters based on the feedback from the inner loop. If the residual deviation exceeds a threshold (e.g., 10mm), a parameter re-optimization process is triggered until the deviation is controlled within a safe range.

[0103] The verticality anomaly alarm is based on a convolutional neural network (CNN) model to achieve state pattern recognition. The system first collects structural vibration features (such as acceleration time history curves and frequency response functions), wind speed load spectra, and foundation settlement data to construct an operational state feature library containing samples of normal and abnormal states. To improve the model's ability to capture time-series features, the data in the feature library is processed by overlapping segmentation. For example, continuous one-hour vibration data is divided into 500 one-minute samples with 50% overlap, generating a spatially distributed sample set containing time-series information. The CNN model architecture includes an input layer, convolutional layers, pooling layers, and a classification layer. The input layer receives normalized sample data (dimensions of 1×60×3, corresponding to time points and feature dimensions, respectively); the convolutional layer uses 3×3 convolutional kernels to extract local features (such as abrupt changes in vibration amplitude within a certain time window) and introduces nonlinearity through the ReLU activation function; the pooling layer uses max pooling (pooling window 2×2) to reduce feature dimensions and retain key information; the classification layer outputs the probability values ​​of "normal operation" and "verticality anomaly" through the Softmax function. The model training employs a backpropagation algorithm to optimize weight parameters, improving classification accuracy by minimizing the cross-entropy loss function. When the model predicts a "verticality anomaly" for a sample with a probability exceeding a threshold (e.g., 90%), the tower is determined to be in an abnormal state. The system automatically triggers an alarm mechanism, sending warning information to maintenance personnel via audible and visual signals, SMS notifications, etc., while simultaneously recording the time, location, and characteristic parameters of the anomaly (e.g., wind speed of 28 m / s and radial offset of the top at the time of the anomaly) to provide detailed records for fault diagnosis.

[0104] The dynamic weighted algorithm achieves multi-level analysis of tower deformation through multi-model fusion and hierarchical processing: the hybrid prediction model predicts trends based on temporal and spatial characteristics, the benchmark deviation correction process provides closed-loop control for real-time deviations, and the CNN model identifies abnormal patterns from vibration and load data. This algorithm fully utilizes the multi-dimensional data of the tower's spatial attitude model, improving the intelligence level and decision reliability of the monitoring system through a combination of data-driven and physical models. It can promptly identify potential risks and provide targeted correction strategies, providing strong technical support for the safe operation of hybrid wind turbine towers.

[0105] Example 5:

[0106] In constructing the holographically mapped tower spatial attitude model, the verticality resolution layer utilizes a sliding space window mechanism and an online learning algorithm to achieve dynamic optimization and adaptive updating of the model. The specific implementation is as follows:

[0107] The system employs a sliding spatial window mechanism to segment the continuous monitoring area of ​​the tower. The tower is divided into multiple vertical cylindrical regions along its height, and each cylindrical region is further subdivided into several sector-shaped spatial windows in the horizontal cross-section (e.g., one cylindrical region every 2 meters in the vertical direction, and one sector-shaped spatial window every 45 degrees in the horizontal cross-section). Each spatial window corresponds to a local area of ​​the tower, such as a specific orientation of a section of the tower. For the monitoring data within each spatial window (e.g., axial displacement and radial offset data of the displacement sensor array in that area), the system independently extracts features and calculates parameters such as displacement gradient, mean strain, and rate of change of torsional angle for that area. For example, within a certain spatial window, the uniformity of deformation in that area is assessed by calculating the standard deviation of the radial offset data. If the standard deviation suddenly increases, it indicates that there may be local damage or abnormal load in that area.

[0108] Based on feature extraction, the system establishes a correlation matrix between data block features and structural loads. The rows of the correlation matrix represent different spatial windows, and the columns represent load types (e.g., static load, wind load, temperature load) and corresponding characteristic parameters (e.g., wind speed, temperature change, axial force). Through historical data statistics, the deformation distribution patterns of each spatial window are recorded under different operating conditions (e.g., startup, full-power operation, and shutdown maintenance). For example, under full-power operation and a wind speed of 15 m / s, the correlation matrix records a radial offset of 30 mm and a strain of 80 μe for a spatial window in the middle of the tower, corresponding to the wind speed load at the same time. This correlation matrix provides a data foundation for the online learning of the model, enabling the system to identify typical deformation patterns under different load conditions.

[0109] The online learning algorithm optimizes the attitude model through an incremental update mechanism. When new monitoring data is input, the system first assigns the data to the corresponding spatial window, calculates the real-time feature parameters of that spatial window, and then compares them with the historical patterns recorded in the correlation matrix. If the real-time features match a historical pattern (e.g., similarity exceeds 85%), the model parameters are adjusted according to preset update rules. For example, the stiffness parameters of the mesh elements corresponding to that spatial window are updated to reflect the gradual changes in material properties. If the real-time features do not match any historical patterns (i.e., an unrecorded deformation pattern is detected, such as an abnormally large torsional angle in a spatial window under low wind speed), the model topology reconstruction mechanism is triggered. The topology reconstruction process includes: refining the model mesh in the region where the spatial window is located (e.g., reducing the original mesh size from 200mm to 50mm), increasing the number of mesh elements to improve local modeling accuracy; redefining the type of mesh elements (e.g., converting shell elements to solid elements to more accurately simulate complex forces); recalculating the stress-strain distribution based on the refined mesh, and recording the new deformation patterns in the correlation matrix to expand the applicability of the model.

[0110] During model parameter iteration, the system dynamically compensates for the output results using a residual correction algorithm, while simultaneously adjusting the model finely by combining the feature analysis results of the sliding space window. For example, if the residual between the measured real-time radial offset of a certain space window and the model's predicted value consistently exceeds 10 mm, the system determines that there is a deviation in the model parameters for that region. By adjusting the material elastic modulus parameter corresponding to that space window, the residual is gradually reduced. Furthermore, the basic datasets in the structural database, such as the material property library and load distribution library, are also synchronously adjusted based on the feature updates of the space windows. For instance, when abnormal deformation occurs repeatedly in a certain region, the material property library automatically marks that the steel in that region may have fatigue damage, adjusting its elastic modulus and yield strength values ​​to provide more accurate parameter support for subsequent model calculations.

[0111] Interactive mechanisms play a crucial role in spatial window processing and model updates. The physical measurement space and the digital modeling space transmit spatial window characteristic parameters in real time via a data channel, ensuring the digital model reflects the latest state of the physical structure. The digital modeling space interacts with the structural database through middleware; for example, when model topology reconstruction requires new material parameters, the middleware retrieves relevant data from the structural database and feeds the new parameters back to the database after the model update is complete. The physical measurement space and the structural database use standardized interfaces for parameter acquisition and model updates; for instance, the structural database periodically obtains calibration parameters for each spatial window from the physical measurement space to update the environmental load and deformation correlation model in the load distribution library.

[0112] By combining a sliding space window mechanism with online learning algorithms, the system achieves refined local analysis of tower deformation and dynamic optimization of the overall model. This design not only improves the model's adaptability to complex working conditions but also enables timely detection of abnormal changes in the tower structure, providing a more accurate model foundation for subsequent functions such as deformation trend prediction and baseline deviation correction. Simultaneously, the model topology reconstruction mechanism ensures that the system can autonomously evolve in the face of new deformation patterns, avoiding monitoring blind spots caused by model rigidity, thereby enhancing the reliability and intelligence of the entire monitoring system.

[0113] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A real-time monitoring system for the verticality of hybrid wind turbine towers, characterized in that, It includes: Data perception layer, communication control layer, and verticality resolution layer; The data sensing layer includes a displacement sensor array, an inclination measuring instrument, a dynamic calibration device, and a reference positioning module, which are used to perform multi-point synchronous measurement of tower structure deformation, reference coordinate calibration, and dynamic calibration compensation operation based on feedback instructions. The communication control layer includes edge computing nodes, protocol converters, and cloud processing platforms. It also deploys time synchronization mechanisms and multimodal data fusion strategies to establish redundant transmission links, achieve time alignment and spatial registration of multi-source monitoring data, and use adaptive compression algorithms to extract features and transmit layered data from heterogeneous data according to data accuracy requirements and transmission bandwidth. It also enables dynamic interaction between physical measurement data and 3D modeling data by configuring multi-source data interfaces. The verticality analysis layer is used to construct a holographically mapped tower spatial attitude model after dynamically calibrating the tower's three-dimensional deformation model by combining spatial geometric analysis technology with structural design parameters and real-time monitoring datasets. Based on this attitude model, a dynamic weighted algorithm is used to predict deformation trends, correct benchmark deviations, and issue verticality anomaly alarms. The construction of the holographic mapping tower spatial attitude model includes: Establish a parameter correlation channel and a two-way calibration mechanism between physical structure deformation and digital modeling space; By using spatial interpolation and coordinate transformation of actual measurement data, and based on the axial displacement, radial offset, and torsional angle obtained by the displacement sensor array, a reference attitude library and deformation feature library for the tower structure are constructed. Based on real-time monitoring data, model parameters are iterated and dynamic characteristics are simulated, transforming the actual structural deformation into a high-precision digital twin model. The parameters of the spatial attitude model are optimized by inputting the deformation data monitored in real time into the established attitude model and using the residual correction algorithm to dynamically compensate the output of the model to obtain the optimized tower spatial attitude model. The tower spatial attitude model includes a physical measurement space, a digital modeling space, a structural database, and an interaction mechanism between the modules. The physical measurement space serves as the data source for the attitude model, containing the original deformation characteristics of the tower structure. The digital modeling space forms a mapping relationship with the physical measurement space, geometrically representing the tower's attitude characteristics through multi-dimensional spatial modeling. The structural database integrates design parameters and real-time monitoring information, providing a basic dataset including a material property library, a load distribution library, and a connection node library. The interaction mechanism enables data communication between modules. The physical measurement space and the structural database achieve parameter acquisition and model updates through a standardized interface. The physical measurement space and the digital modeling space transmit parameters through a data channel. The digital modeling space and the structural database achieve information interaction through middleware. The deformation trend prediction based on the attitude model using a dynamic weighted algorithm includes: Based on the tower spatial attitude model, historical operational attitude data, environmental load records, and deformation event characteristic data are obtained to construct a deformation sample set. After performing feature enhancement processing on the deformed sample set, it is divided into a training set and a test set; A hybrid prediction architecture of ARIMA-LSTM-Prophet is established. The model initialization parameters are set, and the training set is input into the hybrid model for joint training. The ARIMA model is used for trend decomposition, the LSTM network is used for temporal feature extraction, and the Prophet algorithm is used for periodic pattern recognition. An adaptive weighting mechanism is used to balance the prediction errors of different algorithms under specific working conditions until the model fit reaches the set threshold or the predetermined training rounds are completed. Input the test set into the trained hybrid model, calculate the overall performance index of the model, and select the optimal deformation prediction model; Based on the optimal deformation prediction model, the spatial distribution trend of tower deformation is output, and the evolution path of verticality deviation is determined by combining material mechanical parameters. The training process of the hybrid prediction architecture includes: The Boosting method is used to generate training sequences for multiple base models, and the predictive performance of each base model is evaluated by sliding window validation. The gradient descent algorithm is used to calculate the dynamic fusion coefficients of each base model, and a weighted average mechanism is used to make a comprehensive prediction of the output results of the base models. The holographically mapped tower spatial attitude model also includes: A sliding spatial window mechanism is used to divide the continuous monitoring area into blocks, and feature extraction is performed independently on the data blocks within each spatial window; Establish a correlation matrix between data block features and structural loads, record the deformation distribution patterns corresponding to different working conditions, incrementally update the attitude model through an online learning algorithm, and trigger a model topology reconstruction mechanism when an unrecorded deformation pattern is detected.

2. The real-time monitoring system for the verticality of a hybrid wind turbine tower according to claim 1, characterized in that, The displacement sensor array includes at least a laser ranging unit, a fiber optic grating sensor, an ultrasonic echo detector, and a millimeter-wave radar, used to acquire the displacement distribution characteristics of the tower cross section; The dynamic calibration device includes at least a temperature compensation module, a vibration suppressor, a coordinate correction unit, and a data calibrator, used to eliminate environmental errors in the measurement system; the reference positioning module includes at least a Beidou positioning terminal, a total station reference point, and an inertial navigation component, used to execute coordinate calibration commands issued by the verticality resolution layer.

3. The real-time monitoring system for the verticality of a hybrid wind turbine tower according to claim 1, characterized in that, In the communication control layer, the data collected by the displacement sensor array is transmitted from the edge computing node to the protocol converter after being timestamped. The protocol converter then performs data format standardization processing and transmits it together with the spatial attitude information recorded by the tilt measuring instrument to the cloud processing platform for modeling and calculation. The cloud processing platform uses spatiotemporal registration technology to input the collected displacement data into three-dimensional mesh models of different resolutions for joint analysis. The method of configuring a multi-source data interface to achieve dynamic interaction between physical measurement data and 3D modeling data includes: configuring a multi-source data interface to achieve dynamic interaction between physical measurement data and 3D modeling data, exchanging real-time tower deformation monitoring data, performing joint analysis on the acquired spatial coordinates, realizing dynamic coordinate system transformation, completing data verification, model update and parameter distribution, and issuing calibration control commands to the data perception layer.

4. The real-time monitoring system for the verticality of a hybrid wind turbine tower according to claim 1, characterized in that, The reference deviation correction based on the attitude model using a dynamic weighted algorithm includes: Based on the tower's spatial attitude model, axial deviation, radial offset, and torsional deformation parameters of the structural deformation are extracted, and a set of deviation feature vectors is constructed. Independent component analysis was used to denoise the deviation feature vector and obtain the core deviation feature components. A biased classification model based on Gaussian mixture model is established, and the optimal number of classifications is determined by the BIC criterion. The categorized deviation features are input into a preset correction strategy library to match the optimal correction scheme and generate a targeted set of benchmark correction parameters.

5. The real-time monitoring system for the verticality of a hybrid wind turbine tower according to claim 1, characterized in that, The method for verticality anomaly alarm based on the attitude model using a dynamic weighted algorithm includes: Based on the tower spatial attitude model, structural vibration characteristics, wind speed load spectrum, and foundation settlement data are collected to construct an operational status feature library. The data in the runtime status feature library is subjected to overlapping segmentation to generate a spatially distributed sample set. A convolutional neural network model is established, the kernel size and pooling parameters are set, and the deep representation of spatial features is obtained through backpropagation. Spatial features are input into a classifier for state pattern recognition, and the output results distinguish between normal operating state and verticality abnormal state.

6. The real-time monitoring system for the verticality of a hybrid wind turbine tower according to claim 4, characterized in that, The method for generating the benchmark correction parameter set includes: Establish a table of correspondence between deviation types and correction methods, including correction methods for axial deviation corresponding to prestress adjustment and radial offset corresponding to counterweight optimization; A genetic algorithm is used to search for the optimal combination of correction parameters, including adjustment magnitude, application location, and correction timing. The correction effect is evaluated in real time through a dual closed-loop control mechanism. When the residual deviation exceeds the threshold, the parameter re-optimization process is triggered.

7. A method for real-time monitoring of the verticality of hybrid wind turbine towers, characterized in that, A real-time monitoring system for the verticality of a hybrid wind turbine tower as described in any one of claims 1-6 includes the following steps: Multi-point synchronous data acquisition of the tower structure is carried out by a displacement sensor array and an inclination measuring instrument to obtain axial displacement, radial offset and torsional angle measurements. Based on the coordinate calibration results of the reference positioning module, a dynamic calibration device is used to compensate for temperature drift and eliminate vibration interference. The collected heterogeneous monitoring data is timestamped through edge computing nodes. A multimodal data fusion strategy is used to synchronize and register displacement and tilt data in time and space. Based on the transmission bandwidth constraints, an adaptive compression algorithm is used to extract displacement distribution features. Finally, the physical measurement data is converted into a standardized data format compatible with the three-dimensional modeling space through a protocol converter. The tower structure design parameters are loaded into the cloud processing platform, and the pre-constructed three-dimensional deformation model is dynamically calibrated by combining the real-time monitoring dataset. After optimizing the model parameters by using the residual correction algorithm, a holographically mapped tower spatial attitude model is generated. Based on this attitude model, the axial deformation gradient, radial offset trend and torsional angle change are fused by the dynamic weighting algorithm, and the deformation trend prediction result, the benchmark deviation correction amount and the verticality anomaly alarm signal are output.

Images