Tower drum dynamic inclination early warning method and device based on inclination angle sensor cluster network

Abstract

The invention discloses a tower drum dynamic inclination early warning method and device based on an inclination angle sensor cluster network. The tower drum dynamic inclination early warning method comprises the following steps: acquiring real-time sampling data of an inclination angle sensor cluster; and preprocessing the real-time sampling data to generate long sequence data. And inputting the long sequence data into an inclination trend model, and outputting a tower drum inclination prediction value. An optimization request is sent to the cloud, and the cloud generates optimization model parameters and dynamic thresholds according to the optimization request. And receiving an optimization model parameter and a dynamic alarm threshold value issued by the cloud, and updating the inclination trend model. And outputting a new tower drum inclination prediction value by using the updated inclination trend model, and when the tower drum inclination prediction value is greater than a dynamic alarm threshold value, generating an alarm signal. Through a closed-loop design of cluster perception-trend prediction-cloud evolution-hierarchical response, the inclination trend of the tower drum is accurately, reliably and effectively predicted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tower safety early warning, and in particular to a tower dynamic inclination early warning method and device based on a cluster network of inclination sensors. BACKGROUND

[0002] Offshore wind farms are often subjected to the coupling effects of wind and waves, ocean currents, seabed erosion and geological subsidence, etc., and the tower may produce invisible slow inclination or local buckling deformation, which may easily lead to serious accidents such as wind turbine toppling. In order to ensure the safe and stable operation of the wind farm and avoid economic losses and safety accidents, the most direct measure currently taken is to install inclination sensors on the tower to measure the changes in the inclination angle of the tower and discover potential safety hazards in a timely manner.

[0003] The angle change of the tower reflects the inclination and deformation state of the tower, and the inclination and deformation angle of the tower structure can be measured by using the inclination sensor. The existing technology can monitor the change in the inclination angle of the tower, but lacks accurate and reliable effective prediction of the inclination trend of the tower. SUMMARY

[0004] To solve the technical problem in the prior art that although the change in the inclination angle of the tower can be monitored, the inclination trend of the tower cannot be accurately and reliably effectively predicted, the purpose of the present application is to provide a tower dynamic inclination early warning method and device based on a cluster network of inclination sensors, which accurately and reliably effectively predicts the inclination trend of the tower through the closed-loop design of "cluster perception-trend prediction-cloud evolution-hierarchical response", and upgrades the inclination monitoring of the offshore wind tower from "passive response" to "active prevention", thereby reducing the economic losses and safety accident risks caused by the tower overturning.

[0005] The purpose of the present application is achieved by the following technical solutions:

[0006] In a first aspect, the present application provides a tower dynamic inclination early warning method based on a cluster network of inclination sensors, which has a cluster of inclination sensors based on LoRa communication and an edge computing terminal, and the tower dynamic inclination early warning method comprises the following steps:

[0007] Obtaining real-time sampling data of the cluster of inclination sensors;

[0008] Preprocessing the real-time sampling data to generate long sequence data;

[0009] Inputting the long sequence data into an inclination trend model to output a tower inclination prediction value;

[0010] sending an optimization request to the cloud, the optimization request including real-time model parameters of the tilt trend model, the cloud being configured to generate optimization model parameters and dynamic thresholds according to the optimization request;

[0011] receiving the optimization model parameters and dynamic alarm thresholds issued by the cloud, and updating the tilt trend model;

[0012] outputting a new tower tilt prediction value using the updated tilt trend model, and generating an alarm signal when the tower tilt prediction value is greater than the dynamic alarm threshold.

[0013] In combination with the first aspect, the present application also provides a first specific embodiment of the first aspect. Specifically, the real-time sampling data is preprocessed to generate long sequence data, specifically including:

[0014] Based on an outlier detection algorithm, the outliers in the real-time sampling data are marked;

[0015] Based on a Kalman filter algorithm, the environmental noise in the real-time sampling data is eliminated;

[0016] Corresponding GPS data is obtained and spatio-temporal alignment is performed with the real-time sampling data to construct a spatio-temporal feature matrix;

[0017] The spatio-temporal feature matrix is converted into long sequence data.

[0018] In combination with the first aspect, the present application also provides a second specific embodiment of the first aspect. Specifically, the outlier detection algorithm includes the following steps:

[0019] Based on the statistical model of the mu+3sigma standard deviation method, the mean mu and the standard deviation sigma of the real-time sampling data are dynamically calculated to construct an abnormality judgment threshold:

[0020] The real-time sampling data is subjected to sliding window mean filtering to eliminate transient noise;

[0021] The abnormal threshold is dynamically updated, specifically, the mean mu and the standard deviation sigma of the data in the window are recalculated according to a preset time period, and mu+3sigma is set as the abnormality judgment threshold;

[0022] When the real-time sampling data exceeds the abnormality judgment threshold, it is marked as an outlier and a data review process is triggered.

[0023] In combination with the first aspect, the present application also provides a third specific embodiment of the first aspect. Specifically, the tilt trend model adopts a double-layer LSTM network architecture; the tilt trend model extracts time sequence features and environmental coupling features through feature engineering;

[0024] The timing feature includes a statistical quantity of historical data within a preset time, and the statistical quantity includes a mean value, a variance, an extreme value, and a change rate; and the environment coupling feature is working condition data including fused wind speed and wind turbine power.

[0025] In combination with the first aspect, the present application further provides a fourth specific implementation of the first aspect, in particular, the cloud is configured to perform the following steps:

[0026] The global prediction model of the preset deployment is updated using the real-time model parameters of the inclination trend model;

[0027] Based on historical data and real-time offshore environment data, the preset reinforcement learning algorithm is used to output optimized model parameters and dynamic alarm thresholds;

[0028] The optimized model parameters and dynamic alarm thresholds are issued to the edge computing terminal.

[0029] In combination with the first aspect, the present application further provides a fifth specific implementation of the first aspect, in particular,

[0030] The inclination sensor cluster network includes a plurality of inclination sensors and acceleration sensors deployed at wind turbine test reference points, and the inclination sensors and acceleration sensors each have a LoRa module, and form an inclination sensor cluster through self-organization of the LoRa module;

[0031] The layout scheme of the inclination sensor cluster network is specifically that, taking the wind turbine as a coordinate origin, three test reference points are formed by equally distributing the first section and the second section of the wind turbine at an angle of 120°, and the inclination sensor and the acceleration sensor are deployed at each test reference point; and the height of the second section is greater than that of the first section.

[0032] In combination with the first aspect, the present application further provides a sixth specific implementation of the first aspect, in particular, the tower dynamic inclination early warning method further includes the following steps:

[0033] Based on the sensor data of a plurality of test reference points of the wind turbine, a first circle center coordinate and a second circle center coordinate are fitted, and the first circle center coordinate is located at the first section of the wind turbine, and the second circle center coordinate is located at the second section of the wind turbine;

[0034] The first circle center coordinate and the second circle center coordinate are anti-interference optimized based on a random sample consensus algorithm;

[0035] The tower inclination state of the tower is calculated using the first circle center coordinate and the second circle center coordinate.

[0036] A tower structure vibration behavior model based on a nonlinear state estimation model is constructed, SCADA data of the wind turbine generator is input, and a mapping relationship between the SCADA data and the acceleration signal is learned, and an abnormal state of the tower is judged by comparing a residual error between a measured acceleration signal and a model prediction value.

[0037] An optimal fitting point is output by the tower structure vibration behavior model, the optimal fitting point is used to iteratively optimize a layout scheme of the cluster network of inclination sensors, and the first step is executed in a loop until a preset optimization condition is met.

[0038] In a second aspect, the application further provides a tower dynamic inclination early warning device based on a cluster network of inclination sensors, the cluster network of inclination sensors having a cluster of inclination sensors based on LoRa communication and an edge computing terminal, and the tower dynamic inclination early warning device comprising:

[0039] An acquisition module for acquiring real-time sampling data of the cluster of inclination sensors;

[0040] A preprocessing module for preprocessing the real-time sampling data to generate long sequence data;

[0041] A processing module for inputting the long sequence data into an inclination trend model to output a tower inclination prediction value;

[0042] A sending module for sending an optimization request to a cloud, the optimization request including real-time model parameters of the inclination trend model, the cloud being configured to generate optimization model parameters and a dynamic threshold value according to the optimization request;

[0043] A receiving module for receiving the optimization model parameters and the dynamic alarm threshold value issued by the cloud and updating the inclination trend model;

[0044] An alarm module for outputting a new tower inclination prediction value using the updated inclination trend model, and generating an alarm signal when the tower inclination prediction value is greater than the dynamic alarm threshold value.

[0045] In a third aspect, the application provides an electronic device comprising a memory storing executable program code, and a processor coupled to the memory, wherein the processor invokes the executable program code stored in the memory to execute the tower dynamic inclination early warning method based on the cluster network of inclination sensors as described in the first aspect and the first aspect and the first aspect 1-6 specific embodiments.

[0046] In a fourth aspect, the application further provides a computer readable storage medium storing a computer program, wherein the computer program causes a computer to execute the tower dynamic inclination early warning method based on the cluster network of inclination sensors as described in the first aspect and the first aspect and the first aspect 1-6 specific embodiments.

[0047] Compared with the prior art, the application has at least the following beneficial effects:

[0048] The application provides a tower tube dynamic inclination early warning method based on an inclination sensor cluster network, the inclination sensor cluster network having an inclination sensor cluster and an edge computing terminal based on LoRa communication. The tower tube dynamic inclination early warning method comprises the following steps: acquiring real-time sampling data of the inclination sensor cluster; preprocessing the real-time sampling data to generate long sequence data. The long sequence data is input into an inclination trend model to output a tower tube inclination prediction value. An optimization request is sent to the cloud, the optimization request comprising real-time model parameters of the inclination trend model, the cloud being configured to generate optimized model parameters and a dynamic threshold value according to the optimization request. The optimized model parameters and the dynamic alarm threshold value issued by the cloud are received, and the inclination trend model is updated. The updated inclination trend model is used to output a new tower tube inclination prediction value, and an alarm signal is generated when the tower tube inclination prediction value is greater than the dynamic alarm threshold value.

[0049] The application cooperatively collects tower tube data through the inclination sensor cluster based on LoRa communication, inputs the preprocessed data into the inclination trend model through the edge technology terminal to online predict the future inclination trend, uploads the model parameters to the cloud through the edge technology terminal, and returns the optimized parameters and the dynamic threshold value based on the historical data and the weather data of the whole field, so that the model is continuously evolved. When the prediction value exceeds the dynamic threshold value, an automatic early warning is generated, the inclination trend of the tower tube is accurately and reliably predicted through the closed-loop design of "cluster sensing-trend prediction-cloud evolution-classified response", and the economic loss and the safety accident risk caused by the overturning of the tower tube are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A schematic diagram of an inclination sensor cluster network to which the technical solution of the embodiment of the application can be applied is shown.

[0051] Figure 2 A flowchart of the tower tube dynamic inclination early warning method based on the inclination sensor cluster network of one embodiment of the application is shown.

[0052] Figure 3 A flowchart of the tower tube dynamic inclination early warning method based on the inclination sensor cluster network of one embodiment of the application is shown.

[0053] Figure 4 A structural schematic diagram of an electronic device of the application is shown. DETAILED DESCRIPTION

[0054] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0055] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0056] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0057] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0058] Figure 1 A schematic diagram of a tilt sensor cluster network to which the technical solutions of the embodiments of this application can be applied is shown.

[0059] like Figure 1 As shown, the tilt sensor cluster network 100 may include an edge computing terminal 103, a cloud server 104, and a first tilt sensor cluster 101, a second tilt sensor cluster 102, etc. The edge computing terminal 103 serves as the medium for providing a communication link between the cloud server 104 and the first tilt sensor cluster 101 and the second tilt sensor cluster, etc. The edge computing terminal 103 may include a LoRa wireless communication link, etc. Both the first tilt sensor cluster 101 and the second tilt sensor cluster 102 have low-power long-range wireless LoRa communication capabilities, such as sensors equipped with LoRa communication modules. The cloud server 104 is a server with wireless communication capabilities.

[0060] It should be understood that Figure 1The numbers of the first inclination sensor cluster 101, the second inclination sensor cluster 102, the edge computing terminal 103 and the cloud server 104 in FIG. 1 are only illustrative. According to the implementation needs, there can be any number of the first inclination sensor cluster 101, the second inclination sensor cluster 102, the edge computing terminal 103 and the cloud server 104. For example, the edge computing terminal 103 can be multiple, one edge computing terminal 103 is responsible for a wind turbine in a region, and is based on LoRa ad hoc network with the inclination sensor cluster in the region; for example, the cloud server 104 can be a server cluster composed of multiple servers, etc.

[0061] The first inclination sensor cluster 101 and the second inclination sensor cluster 102 interact with the cloud server 104 through the edge computing terminal 103. For example, the sampling data of the first inclination sensor cluster is uploaded to the cloud server 104 through the edge computing terminal 103.

[0062] In some implementations, the edge computing terminal 103 can be an intelligent gateway.

[0063] It should be noted that the tower dynamic inclination early warning method based on the inclination sensor cluster network provided by the embodiments of the present application is generally executed by the edge computing terminal 103, and accordingly, the tower dynamic inclination early warning device based on the inclination sensor cluster network is generally arranged in the edge computing terminal 103. However, in other embodiments of the present application, the cloud server 104 can also have similar functions, so as to execute the scheme of the tower dynamic inclination early warning method based on the inclination sensor cluster network provided by the embodiments of the present application.

[0064] The implementation details of the technical solutions of the embodiments of the present application are described in detail as follows:

[0065] Figure 2 A flowchart of a tower dynamic inclination early warning method based on an inclination sensor cluster network according to an embodiment of the present application is shown, which can be executed by an edge computing terminal, which can be the edge computing terminal 103 shown in FIG. 1. Referring to FIG. 2, the tower dynamic inclination early warning method includes at least steps S210 to S260, which are described in detail as follows: Figure 1 Figure 2

[0066] Step S210: Obtain real-time sampling data of the inclination sensor cluster.

[0067] In some possible implementations, the inclination sensor cluster is a sensor with LoRa communication function, for example, it can be an inclination sensor, an acceleration sensor, etc. provided with a LoRa module.

[0068] ​​Exemplarily, the stress concentration parts on the tower cylinder cluster of the offshore wind farm, such as the tower base and the tower body, are installed with inclination sensors, acceleration sensors and LoRa modules. The inclination sensors collect real-time inclination angle data of the tower cylinder. The sensor group forms a network topology through the LoRa module ad hoc network, supporting relay transmission between nodes. According to the quality of the offshore channel, the transmission power is dynamically adjusted to balance the transmission rate and reliability. Real-time data and sensor data can be sent to the nearest edge computing terminal through the LoRa network.

[0069] In the present application, a plurality of inclination sensors can be deployed at key positions of the tower cylinder to form a sensor cluster. These sensors can measure the inclination angle of the tower cylinder in real time. Real-time sampling data is periodically obtained from the inclination sensor cluster. The sampling frequency can be adjusted according to the dynamic characteristics of the tower cylinder and the monitoring requirements.

[0070] Step S220: preprocessing the real-time sampling data to generate long sequence data.

[0071] In the present application, through preprocessing, the real-time sampling data is generated into high-quality long sequence data, improving the prediction accuracy of the subsequent model. In one possible implementation, the implementation process of this step S220 includes:

[0072] Step S221: labeling abnormal points in the real-time sampling data based on an outlier detection algorithm.

[0073] In some possible implementations, the real-time sampling data is analyzed using an outlier detection algorithm to label abnormal points in the data. Abnormal points are usually those data points that significantly deviate from the normal range. Exemplarily, the outlier detection algorithm can use Z-Score, IQR (Interquartile Range) or machine learning-based anomaly detection algorithms (such as IsolationForest).

[0074] In another possible implementation, the outlier detection algorithm includes the following steps: based on the statistical model, the μ±3σ standard deviation method is used to dynamically calculate the mean μ and standard deviation σ of the real-time sampling data, and an abnormality judgment threshold is constructed. The real-time sampling data is subjected to sliding window mean filtering to eliminate transient noise. The abnormal threshold is dynamically updated, specifically, the mean μ and standard deviation σ of the data in the window are recalculated according to the preset time period, and μ±3σ is set as the abnormality judgment threshold. When the real-time sampling data exceeds the abnormality judgment threshold, it is labeled as an abnormal point and triggers a data review process.

[0075] In some possible implementations, the original data collected by the inclination sensor is subjected to sliding window mean filtering to eliminate transient noise interference.

[0076] In some possible implementations, the mean μ and the standard deviation σ of the data within the window are recalculated every 15 minutes, and μ ± 3σ is set as the abnormality determination threshold.

[0077] In some possible implementations, step S221 further includes offline calibration optimization, and periodic data calibration is performed in combination with the box plot method: a box plot distribution feature is generated daily, and the interquartile range (IQR) and the upper and lower limits (Q1-1.5IQR, Q3+1.5IQR) are calculated. When the standard deviation method and the box plot method have conflicting detection results, the determination conclusion of the box plot method is preferred.

[0078] By using the μ ± 3σ standard deviation method based on the statistical model, in combination with the sliding window mean filtering and the dynamic updating of the abnormality threshold, the abnormal points in the real-time sampling data can be effectively detected and marked. This method not only can adapt to the dynamic changes of the data, but also can eliminate transient noise through the sliding window filtering, thereby improving the accuracy and reliability of the abnormality detection.

[0079] Step S222: Eliminate environmental noise in the real-time sampling data based on the Kalman filtering algorithm.

[0080] In the present application, the environmental noise is effectively removed by using the Kalman filtering algorithm, the signal-to-noise ratio of the data is improved, and clearer data is provided for subsequent processing.

[0081] Step S223: Obtain corresponding GPS data, and perform spatio-temporal alignment with the real-time sampling data to construct a spatio-temporal feature matrix.

[0082] In one possible implementation, GPS data corresponding to the tilt sensor can be obtained through a GPS sensor, including position information (longitude, latitude) and a timestamp. The GPS data is spatio-temporally aligned with the real-time sampling data to ensure consistency in time and space. The alignment can be achieved through timestamp matching and spatial interpolation. The aligned GPS data and the real-time sampling data are combined into a spatio-temporal feature matrix, each row of which represents the data at a time point, and each column represents a feature (such as tilt angle, longitude, latitude, etc.).

[0083] Step S224: Convert the spatio-temporal feature matrix into long sequence data.

[0084] In one possible implementation, the data in the spatio-temporal feature matrix is spliced into a long sequence data in chronological order. The long sequence data can be a one-dimensional array or a time series, in which each element represents a feature value at a time point. For example, the long sequence data is formatted into a format suitable for subsequent model processing, such as converting the data into a time series object or a tensor.

[0085] Through the implementation process of step S220, including outlier detection, Kalman filtering, spatio-temporal alignment and long sequence data generation, real-time sampling data can be effectively preprocessed to generate high-quality long sequence data. These data provide a reliable basis for subsequent tilt trend prediction, improving the accuracy and reliability of the prediction model.

[0086] Step S230: inputting the long sequence data into the tilt trend model to output a tower tilt prediction value.

[0087] In some possible implementations, the tilt trend model adopts a double-layer LSTM network architecture; and the tilt trend model extracts time sequence features and environment coupling features through feature engineering, so as to improve the prediction accuracy and robustness of the model.

[0088] In some possible implementations, the time sequence features include statistical quantities of historical data within a preset time, and the statistical quantities include mean, variance, extreme value and change rate. For example, the statistical quantities of historical data in the past 1 hour are extracted:

[0089] Mean: calculating the mean of historical data within a preset time window.

[0090] Variance: calculating the variance of historical data within a preset time window, reflecting the fluctuation of data.

[0091] Extreme value: calculating the maximum and minimum values of historical data within a preset time window.

[0092] Change rate: calculating the change rate of historical data within a preset time window, reflecting the change trend of data.

[0093] In some possible implementations, the environment coupling features are working condition data including fused wind speed and wind turbine power. The working condition data fuses wind speed and wind turbine power data to generate features reflecting the working condition of the wind turbine.

[0094] In some possible implementations, the dynamic early warning mechanism of the tilt trend model includes:

[0095] (1) Local threshold setting: determining a dynamic alarm threshold according to the 99.7% percentile of historical data.

[0096] (2) Edge computing response: when the prediction value exceeds the threshold, the edge device triggers an audible and visual alarm within 50 ms, and uploads an alarm log through LoRaWAN.

[0097] In some possible implementations, the network structure of the tilt trend model includes:

[0098] (1) Hidden layer and neuron configuration: initially using 64 hidden units, and increasing to 300 units for complex time sequence tasks.

[0099] (2) Number of layers selection: LSTM model adopts 3-layer network architecture, combined with residual connection to prevent gradient vanishing.

[0100] (3) Input sequence length optimization: According to the characteristics of data cycle, set the sliding window length: input data is set to 60 minutes of measured data.

[0101] Learning rate and optimizer configuration of the tilt trend model: the learning rate is dynamically adjusted method: the initial value is set to 1e-3. The optimizer selection: the adaptive learning rate characteristic Adam method is adopted. In the Dropout configuration: 0.2-0.5 Dropout is randomly inserted between LSTM layers, distinguishing dropout (input gate) and recurrent_dropout (recurrent gate). Weight constraint, add L2 regularization (λ=0.005) in the fully connected layer, and use gradient clipping (threshold 1.0) to prevent gradient explosion. Early stopping strategy: monitor the validation set loss, if there is no improvement for 8-10 consecutive rounds, terminate the training. Prioritize adjusting network structure and learning rate (impact>60%), then gradually optimize regularization parameters, and finally introduce intelligent algorithms for global optimization. The wind power prediction task is recommended to combine PSO to optimize the initial weight of LSTM, which can reduce MAE by about 15-30%.

[0102] Step S240: send an optimization request to the cloud, the optimization request including real-time model parameters of the tilt trend model, the cloud being configured to generate optimized model parameters and dynamic threshold values according to the optimization request.

[0103] In the present application, the cloud can generate optimized model parameters and dynamic alarm thresholds according to historical data and real-time data, ensuring the accuracy and timeliness of the alarm.

[0104] In some possible implementations, the cloud is configured to perform the following steps:

[0105] Step S241: update the pre-deployed global prediction model using the real-time model parameters of the tilt trend model.

[0106] In some possible implementations, the cloud receives real-time model parameters from the edge node cluster. These parameters include the structure parameters, weights, biases, etc. of the tilt trend model. Update the global prediction model in combination with the real-time model parameters of the edge node cluster. Model fusion techniques such as federated learning (Federated Learning) can be used to weight average the parameters of multiple edge models to generate global model parameters. Use the global model parameters to retrain the global prediction model to ensure that the model can adapt to the characteristics of different edge nodes and improve prediction accuracy.

[0107] Step S242: Based on historical data and real-time marine environment data, the preset reinforcement learning algorithm is used to output the optimized model parameters and dynamic alarm threshold.

[0108] In some possible embodiments, the cloud server extracts historical data from the database, including past inclination angles, alarm records, environmental parameters, etc. Real-time environmental parameters such as wind speed, wind direction, and wave height are obtained from the marine environment monitoring system. Then the historical data and real-time environmental parameters are integrated into a unified data set to provide data support for subsequent optimization.

[0109] The edge model and alarm threshold are optimized by reinforcement learning algorithm. Define the reinforcement learning environment, including state space (such as current inclination angle, environmental parameters), action space (such as model parameter adjustment, alarm threshold adjustment) and reward function (such as prediction accuracy, alarm accuracy). Choose appropriate reinforcement learning algorithms, such as Q-learning, DQN (Deep Q-Network) or PPO (Proximal Policy Optimization). These algorithms can dynamically adjust model parameters and alarm thresholds according to the reward signals fed back by the environment. Use reinforcement learning algorithm to train edge model and alarm threshold.

[0110] In some possible embodiments, during the training process, the algorithm evaluates the performance of different parameter settings according to the reward function, and selects the optimal parameter combination. According to the training results, dynamically adjust the parameters of the edge model and the alarm threshold, to ensure that the model can adapt to the real-time changes of the marine environment.

[0111] Step S243: The optimized model parameters and dynamic alarm threshold are issued to the edge computing terminal.

[0112] Step S250: Receive the optimized model parameters and dynamic alarm threshold issued by the cloud, and update the inclination trend model.

[0113] In this application, the edge computing terminal receives the optimized model parameters and dynamic alarm threshold issued by the cloud, and updates the local inclination trend model using the optimized model parameters. By updating the model parameters and alarm threshold, the prediction accuracy of the local model and the accuracy of the alarm are ensured.

[0114] Step S260: Use the updated inclination trend model to output a new tower tube inclination prediction value. When the tower tube inclination prediction value is greater than the dynamic alarm threshold, an alarm signal is generated.

[0115] In this application, the updated tilt trend model is used to output new tower tilt prediction values. The new tower tilt prediction values are compared with the dynamic alarm threshold, and if the prediction value is greater than the dynamic alarm threshold, an alarm signal is generated. The edge computing terminal is transmitted to the monitoring platform through the network, and the tilt trend curve, alarm level and maintenance suggestion are displayed on the monitoring platform.

[0116] The tower dynamic tilt early warning method realizes real-time monitoring, prediction and early warning of the tower tilt state through a series of steps. The specific steps include obtaining real-time sampling data, preprocessing to generate long sequence data, inputting the tilt trend model for prediction, sending optimization request to the cloud, receiving optimization parameters and dynamic threshold, updating the model and generating an alarm signal. These steps work together to ensure the efficient operation and reliability of the tower dynamic tilt early warning system.

[0117] Reference Figure 3 , Figure 3 A flowchart of a tower dynamic tilt early warning method based on an inclination sensor cluster network according to an embodiment of the present application is shown, which includes a plurality of inclination sensors and acceleration sensors deployed at wind turbine test reference points, and the inclination sensors and acceleration sensors each have a LoRa module to form an inclination sensor cluster through LoRa module ad hoc networking. The layout scheme of the inclination sensor cluster network is as follows: taking the wind turbine as the coordinate origin, three test reference points are evenly distributed at a 120° angle at the first cross section and the second cross section of the wind turbine, and an inclination sensor and an acceleration sensor are deployed at each test reference point; the height of the second cross section is greater than that of the first cross section. The tower dynamic tilt early warning method based on the inclination sensor cluster network further includes steps S310 to S330, which are described in detail as follows.

[0118] Regarding the layout scheme of the inclination sensor cluster network: three test reference points are evenly distributed at a 120° angle position with the wind turbine as the coordinate origin, and the reference points are 1.5-2.0D (D is the height of the wind turbine tower) away from the wind turbine, and a closed wire point network is formed. The inclination sensor and the acceleration sensor are erected at the test reference points, and the polar coordinate method is used to observe the top and bottom of the wind turbine tower. The bottom observation point is the connection between the foundation ring and the first section of the tower (i.e. the first cross section), and the top observation point is the junction between the nacelle and the tower (i.e. the second cross section).

[0119] As shown in Figure 3 , the tower dynamic tilt early warning method based on the inclination sensor cluster network further includes:

[0120] Step S310: based on the sensor data of the plurality of test reference points of the wind turbine generator, a first circle center coordinate and a second circle center coordinate are fitted, the first circle center coordinate is located at a first cross section of the wind turbine generator, and the second circle center coordinate is located at a second cross section of the wind turbine generator.

[0121] In some possible implementations, spatial scattered point data (i.e. sensor data) is collected from each test reference point. The spatial scattered point data includes position information (such as X-axis and Y-axis coordinates) of each sensor and measured inclination or acceleration values.

[0122] In some possible implementations, in the spatial state identification process of the wind turbine tower structure, first, spatial scattered point data on several key cross sections of the tower (such as the first cross section and the second cross section) is collected. Based on the data set, a least squares geometric reconstruction algorithm is applied to identify the circular trajectory of the tower cross section, and then the corresponding trajectory radius and the spatial position of the geometric center on the two-dimensional coordinate axis (X-axis and Y-axis) are estimated. Thus, the circle center coordinates of each measured cross section can be obtained as the basic data for subsequent structure offset judgment.

[0123] For example, the collected spatial scattered point data is preprocessed, including removing noise, filling missing values, etc. A least squares geometric reconstruction algorithm is applied to identify the circular trajectory of the tower cross section. The algorithm finds the best fitting circle by minimizing the deviation between the scattered point data and the fitting circle. Through the least squares geometric reconstruction algorithm, the radius of the circular trajectory of each key cross section and the spatial position of the geometric center on the two-dimensional coordinate axis (X-axis and Y-axis) are estimated.

[0124] Specifically, according to the results of the least squares geometric reconstruction algorithm, the circle center coordinates of each key cross section are calculated. The calculated circle center coordinates are recorded as the basic data for subsequent structure offset judgment.

[0125] Step S320: based on the random sample consensus algorithm, the first circle center coordinate and the second circle center coordinate are optimized against interference.

[0126] In some possible implementations, to improve the robustness of the circle center coordinate estimation and suppress the interference of abnormal data points introduced by measurement errors, installation deviations or environmental factors, a model optimization algorithm based on the random consistency verification mechanism (RANSAC, Random Sample Consensus) is introduced. In the point cloud data set containing outliers, by setting the error tolerance, the minimum support sample number and the upper limit of iterations, the effective point set most consistent with the geometric features is selected in multiple rounds of modeling verification, thereby improving the accuracy and stability of the geometric center estimation. The circle center coordinates optimized by the RANSAC algorithm are more accurate and can resist noise and abnormal value interference in the data.

[0127] Step S330: Calculate the tower tilt state of the tower drum using the first and second circle center coordinates.

[0128] In this application, by identifying the geometric center positions of the top and bottom representative sections of the wind turbine tower drum, calculating the axial projection offset in the spatial coordinate system, and combining the vertical distance between the two sections, the overall tilt angle or directional offset of the tower drum structure can be further analyzed for quantitative monitoring and evaluation of the operating state.

[0129] For example, set the first circle center coordinate , and the second circle center coordinate . In the spatial coordinate system, project the geometric centers of the top and bottom sections onto the axial direction of the tower drum. Assuming the axial direction of the tower drum is the z-axis direction, the projection coordinates of the top section geometric center on the z-axis are (0, 0, z1), and the projection coordinates of the bottom section geometric center on the z-axis are (0, 0, z2). Calculate the offset Δx and Δy between the two projection points, where Δx = x 1- x2, Δy = y 1- y2. These two offsets reflect the axial offset of the tower drum in the horizontal direction. Combined with the vertical distance between the top section and the bottom section, the overall tilt angle θ of the tower drum can be calculated using trigonometric functions. The directional offset can be determined by calculating the angle between the offset direction and a reference direction (such as the north direction).

[0130] Step S340: Construct a tower drum structure vibration behavior model based on a nonlinear state estimation model, input the SCADA data of the wind turbine generator and learn the mapping relationship between them and the acceleration signal, and judge the abnormal operating state of the tower drum by comparing the residual error between the measured acceleration signal and the model prediction value.

[0131] For example, when dynamically modeling the vibration response of the wind turbine tower drum, a multi-parameter learning framework based on nonlinear state estimation technology (such as the NSET model) is constructed. This model takes the SCADA data collected during normal operation of the unit as input, and learns the implicit mapping relationship between the tower drum vibration acceleration signal and the external excitation parameters (including wind speed, wind direction, effective power, reactive power, voltage, current, main shaft and impeller speed, grid frequency, etc.).

[0132] In specific implementation, after the tower drum structure vibration behavior model is built, the residual error trend between the measured vibration signal and the model prediction value is analyzed to determine whether the structure has an abnormal operation. When the mean residual error exceeds the set threshold interval, the system determines that there is a potential vibration abnormality, and further adjusts the operation strategy or performs structure maintenance evaluation accordingly.

[0133] Step S350: outputting the optimal fitting point position by using the tower drum structure vibration behavior model, the optimal fitting point position being used to iteratively optimize the layout scheme of the inclination sensor cluster network, and jumping to the first step loop for execution until a preset optimization condition is met.

[0134] Figure 4 A structural diagram of an electronic device 10 that can be used to implement embodiments of the present application is shown. The electronic device is intended to represent various forms of digital computers, such as smart gateways, desktops, workstations, servers, cloud servers, mainframes, and other appropriate computers. The components shown here, their connections, and their functions, as described above, are meant to be examples only, and are not intended to limit the implementations of the present application described and / or claimed herein.

[0135] As shown in Figure 4 The electronic device 10 includes at least one processor 11, and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11, where the memory stores computer programs that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer programs stored in the read-only memory (ROM) 12 or loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0136] A plurality of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network, such as the Internet, and / or various telecommunications networks.

[0137] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs various methods and processes described above, such as a tower drum dynamic inclination early warning method based on an inclination sensor cluster network.

[0138] In some embodiments, a tower dynamic tilt early warning method based on a cluster network of tilt sensors can be implemented as a computer program tangibly embodied in a computer readable storage medium, e.g., storage unit 18. In some embodiments, parts or all of the computer program can be loaded and / or installed onto electronic device 10 via, e.g., ROM 12 and / or communication unit 19. When the computer program is loaded onto RAM 13 and executed by processor 11, one or more steps of a tower dynamic tilt early warning method based on a cluster network of tilt sensors as described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform a tower dynamic tilt early warning method based on a cluster network of tilt sensors by way of other any suitable means, e.g., by way of firmware.

[0139] The various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, specially designed application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0140] Computer programs used to implement the methods of the application can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the computer program running on the processor implements the functions / operations specified in the flowcharts and / or the block diagrams. The computer program can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package and partially on an intelligent gateway, or entirely on a cloud server.

[0141] In the context of the present application, a computer-readable storage medium can be a tangible medium that can contain or store the computer program for use by or in connection with an instruction execution system, apparatus, or device. Computer-readable storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of a machine-readable storage medium will include one or more lines of a program of instructions in a transitory signal, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a cloud server), or that includes a middleware component (e.g., an edge computing terminal), or that includes a front end component (e.g., a sensor), or any combination of the three. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0143] The embodiments of the present application further provide a computer program product, which comprises a computer program, wherein the computer program, when executed by a processor, implements the tower dynamic inclination early warning method based on a cluster network of inclination sensors as provided by the embodiments of the present application.

[0144] The computer program code implementing the processes of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++, or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on an intelligent gateway or a cloud server, or entirely on the intelligent gateway or the cloud server.

[0145] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for dynamic tilt early warning of a tower based on a tilt sensor cluster network, characterized in that, The tilt sensor cluster network has a LoRa-based tilt sensor cluster and an edge computing terminal. The tower dynamic tilt early warning method includes the following steps: Acquire real-time sampling data from the tilt sensor cluster; The real-time sampled data is preprocessed to generate long sequence data; The long-series data is input into the tilt trend model, and the tower tilt prediction value is output. An optimization request is sent to the cloud, the optimization request including real-time model parameters of the tilt trend model, and the cloud is configured to generate optimization model parameters and dynamic thresholds based on the optimization request; Receive optimized model parameters and dynamic alarm thresholds from the cloud and update the tilt trend model; The updated tilt trend model is used to output a new tower tilt prediction value. When the tower tilt prediction value is greater than the dynamic alarm threshold, an alarm signal is generated.

2. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 1, characterized in that, The real-time sampled data is preprocessed to generate long-sequence data, specifically including: Based on an outlier detection algorithm, outliers in the real-time sampled data are marked. Environmental noise in the real-time sampled data is eliminated based on the Kalman filter algorithm; Acquire the corresponding GPS data and align it spatiotemporally with the real-time sampled data to construct a spatiotemporal feature matrix; The spatiotemporal feature matrix is ​​converted into long sequence data.

3. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 2, characterized in that, The outlier detection algorithm includes the following steps: Based on the μ±3σ standard deviation method of the statistical model, the mean μ and standard deviation σ of the real-time sampled data are dynamically calculated to construct an anomaly detection threshold. The real-time sampled data is subjected to sliding window mean filtering to eliminate instantaneous noise; The abnormal threshold is dynamically updated. Specifically, the mean μ and standard deviation σ of the data in the window are recalculated according to a preset time period, and μ±3σ is set as the abnormal judgment threshold. When the real-time sampled data exceeds the anomaly detection threshold, it is marked as an anomaly and the data review process is triggered.

4. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 1, characterized in that: The tilt trend model adopts a two-layer LSTM network architecture; the tilt trend model extracts temporal features and environmental coupling features through feature engineering; The time-series features include statistics on historical data within a preset time period, including mean, variance, extreme values, and rate of change; the environmental coupling features include operating data that integrates wind speed and wind turbine power.

5. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 1, characterized in that, The cloud is configured to perform the following steps: Update the pre-deployed global prediction model using the real-time model parameters of the tilt trend model; Based on historical data and real-time marine environmental data, a preset reinforcement learning algorithm is used to output optimized model parameters and dynamic alarm thresholds. The optimized model parameters and dynamic alarm thresholds are sent to the edge computing terminal.

6. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 1, characterized in that: The tilt sensor cluster network includes multiple tilt sensors and acceleration sensors deployed at the wind turbine test benchmark point. Both the tilt sensors and acceleration sensors have LoRa modules, and they form a tilt sensor cluster by self-organizing a network through the LoRa modules. The deployment scheme of the tilt sensor cluster network is as follows: taking the wind turbine as the coordinate origin, three test reference points are evenly distributed at a 120° angle on the first and second cross sections of the wind turbine. Each test reference point is equipped with a tilt sensor and an acceleration sensor; the height of the second cross section is set to be greater than the height of the first cross section.

7. The method for dynamic tilt early warning of a tower based on a tilt sensor cluster network as described in claim 6, characterized in that, The tower dynamic tilt early warning method also includes the following steps: Based on sensor data from multiple test reference points of the wind turbine, the coordinates of the first and second circles are fitted to obtain the coordinates of the first circle center. The first circle center coordinates are located at the first cross section of the wind turbine, and the second circle center coordinates are located at the second cross section of the wind turbine. The coordinates of the first and second center circles are optimized to resist interference based on the random sampling consensus algorithm. The tower tilt state is calculated using the coordinates of the first and second center circles. A vibration behavior model of the tower structure based on a nonlinear state estimation model is constructed. The SCADA data of the wind turbine is input and its mapping relationship with the acceleration signal is learned. The abnormal operating state of the tower is judged by comparing the residual between the measured acceleration signal and the model prediction value. The optimal fitting point is output using the tower structure vibration behavior model. The optimal fitting point is then iteratively optimized using the layout scheme of the tilt sensor cluster network. The process is repeated from the first step until the preset optimization conditions are met.

8. A tower dynamic tilt early warning device based on a tilt sensor cluster network, characterized in that, The tilt sensor cluster network has a LoRa-based tilt sensor cluster and an edge computing terminal. The tower dynamic tilt early warning device includes: The acquisition module is used to acquire real-time sampling data from the tilt sensor cluster; The preprocessing module is used to preprocess the real-time sampled data to generate long sequence data; The processing module is used to input the long-sequence data into the tilt trend model and output the tower tilt prediction value; The sending module is used to send optimization requests to the cloud. The optimization requests include real-time model parameters of the tilt trend model. The cloud is configured to generate optimization model parameters and dynamic thresholds based on the optimization requests. The receiving module is used to receive the optimized model parameters and dynamic alarm thresholds sent from the cloud, and to update the tilt trend model. The alarm module is used to output a new tower tilt prediction value using the updated tilt trend model. When the tower tilt prediction value is greater than the dynamic alarm threshold, an alarm signal is generated.

9. An electronic device, characterized in that, include: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the tower dynamic tilt early warning method based on tilt sensor cluster network as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein the computer program causes a computer to execute a method for early warning of dynamic tilt of a tower based on a tilt sensor cluster network as described in any one of claims 1 to 7.

Images