Modularized dry-type boost system regulation and control method and system based on multi-physics field perception

Abstract

The invention relates to the technical field of intelligent operation and maintenance of wind power generation equipment, in particular to a modular dry-type boost system regulation and control method and system based on multi-physics field perception. The method is applied to a dry booster station unit which is pre-integrated in a factory and integrally hoisted in a wind power generation tower. The method comprises the following steps: constructing a multi-physical field collaborative sensing network, and collecting temperature, vibration, humidity and electrical characteristic data in real time; performing state evaluation and fault risk prediction on the equipment by using a pre-trained digital twinborn health evaluation model; and according to the evaluation and prediction result, generating and executing a self-adaptive optimization control instruction for cooling, dehumidification, harmonic suppression and vibration control. Through deep integration state perception, artificial intelligence and optimization control, intelligent health management and self-adaptive safe operation of the boosting system in the closed complex environment of the tower drum are realized, and the reliability, the energy efficiency and the operation and maintenance intelligent level are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for wind power generation equipment, specifically to a modular dry booster system control method and system based on multi-physics field sensing. Background Technology

[0002] As a crucial component of clean energy, the efficient and reliable operation of wind power is paramount. The step-up system, a key element of wind turbines, is responsible for raising the generator's output voltage to the level of the busbar. Traditional wind turbine step-up equipment typically employs a decentralized approach, requiring large components such as dry-type transformers and switchgear to be transported separately to the base of the tower, and then repeatedly hoisted and assembled at high altitude using elevators or auxiliary cranes within the tower. This method suffers from numerous drawbacks, including long work cycles, high safety risks, and maintenance difficulties due to limited installation space within the tower. To simplify installation, modular prefabricated step-up substations have emerged, integrating the main equipment in the factory onto a common rigid base, forming a single unit, which is then transported to the site for one-time hoisting. While this significantly improves installation efficiency, it also places all equipment in the unique and harsh operating environment inside the tower. The enclosed space within the tower and poor air circulation result in severe heat dissipation conditions for the equipment, especially for dry-type transformers, where hot spot temperatures directly affect insulation lifespan. Meanwhile, wind-induced tower swaying generates continuous low-frequency vibrations, which are transmitted through the foundation to the entire booster unit. Long-term effects can lead to structural fatigue, bolt loosening, and electrical connection deterioration. Furthermore, condensation easily forms inside the tower in coastal or humid areas, seriously threatening the insulation safety of the primary equipment. Existing solutions primarily focus on structural integration and physical installation improvements. However, a systematic technical solution is lacking for ensuring the long-term reliability of integrated booster systems and achieving intelligent operation and maintenance under such complex and demanding multi-physics (electrical-thermal-mechanical-environmental) coupling effects. Conventional monitoring systems typically only provide threshold alarms for a limited number of electrical parameters and temperatures, lacking in-depth assessment of equipment health status, early fault prediction, and adaptive optimization capabilities for operating parameters. This makes it difficult to meet the requirements of modern wind farms for high equipment availability, low maintenance costs, and full lifecycle management.

[0003] Therefore, the existing technology still needs further development. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a modular dry boost system control method and system based on multi-physics field sensing to solve the problems existing in the prior art.

[0005] To achieve the above-mentioned technical objectives, according to a first aspect of the present invention, the present invention provides a method for controlling a modular dry boost system based on multiphysics sensing, comprising: S100. Construct a multi-physics field collaborative sensing network to acquire multi-dimensional operational status data of the booster station unit in real time; S200. Based on the multi-dimensional operating status data, use a pre-trained health assessment model to assess the equipment health status and predict the failure risk. S300. Based on the results of the equipment health status assessment and fault risk prediction, generate and execute adaptive optimization control commands to adjust the operating status of the booster station unit.

[0006] Specifically, the multi-dimensional operating status data includes at least two of the following: temperature field data, vibration field data, humidity field data, and electrical characteristic data.

[0007] Specifically, the construction of the multi-physics collaborative sensing network includes: deploying distributed temperature sensors at key locations of the transformer windings, cores, and switchgear busbars in the substation unit; deploying vibration sensors on the rigid integrated base and the housings of the main electrical equipment; and deploying humidity sensors in the internal cavities of the substation unit.

[0008] Specifically, the electrical characteristic data is obtained through monitoring devices installed at the incoming and outgoing ends of the booster station unit, and the electrical characteristic data includes current, voltage harmonic components, and partial discharge signals.

[0009] Specifically, the pre-trained health assessment model is a digital twin model, which is trained by integrating historical operating data, multiphysics simulation data, and fault case data, and is used to simulate the real-time health status of the booster station unit.

[0010] Specifically, the equipment health status assessment and fault risk prediction include: inputting the real-time acquired multi-dimensional operating status data into the digital twin model to calculate the current health status score; predicting the thermal aging trend and insulation degradation trend of key components based on the digital twin model; and identifying early fault characteristics using an anomaly detection algorithm based on time-series data and the prediction output of the digital twin model.

[0011] Specifically, the generation and execution of adaptive optimization control instructions includes: dynamically adjusting the operating power of the built-in cooling device of the booster station unit based on the temperature field data and thermal aging trend prediction results; and controlling the start and stop of the active anti-condensation device based on the humidity field data and insulation status assessment results.

[0012] Specifically, the generation and execution of adaptive optimization control instructions further includes: generating compensation instructions for an active filter to suppress harmonic interference based on the harmonic components in the electrical characteristic data; and adjusting the damping parameters of the adaptive damping structure integrated into the rigid base based on the vibration field data and the equipment health status score.

[0013] Specifically, the method further includes: uploading the status assessment results, fault warning information and optimized control logs processed by the edge computing device to the cloud platform; using the cloud platform to aggregate data from multiple booster station units to perform incremental training and parameter optimization of the health assessment model; and generating predictive maintenance strategies for each booster station unit based on the optimized model.

[0014] According to a second aspect of the present invention, a modular dry booster system control system based on multiphysics sensing is provided, the system being deployed in a booster station unit pre-integrated into a rigid base and hoisted integrally within a wind turbine tower, the system comprising: A multi-physics sensing network is configured to collect multi-dimensional operational status data of the booster station unit. An edge intelligent controller is communicatively connected to the multi-physics sensing network and internally stores a pre-trained health assessment model. The edge intelligent controller is configured to: assess the health status of the equipment and predict the failure risk based on the multi-dimensional operating status data and the health assessment model; and generate adaptive optimization control commands based on the assessment and prediction results. An actuator is communicatively connected to the edge intelligent controller and configured to receive and execute the adaptive optimization control command. The actuator includes at least one of a cooling device, an active anti-condensation device, an active filter, and an adaptive damping structure.

[0015] Beneficial effects: The modular dry boost system control method and system based on multi-physics sensing provided by this invention have produced significant and multifaceted beneficial effects compared with the prior art.

[0016] Firstly, at the engineering implementation level, the combination of "modular prefabrication and overall hoisting" with "intelligent control inside the tower" fundamentally reduces the amount and time of high-risk high-altitude operations inside the tower, improving construction safety and efficiency. At the same time, the built-in intelligent system effectively addresses the subsequent operational challenges brought about by the harsh environment inside the tower.

[0017] Secondly, in terms of condition awareness and health management, a multi-physics collaborative sensing network covering the equipment's interior, surface, and operating environment was constructed by deploying distributed temperature, vibration, and humidity sensors and electrical characteristic monitoring devices. This enabled panoramic, high-precision, and synchronous monitoring of the operating status. Utilizing a digital twin model that integrates multi-physics simulation, historical data, and fault case training, quantitative scoring and visualization of equipment health status were achieved. It also enables rolling predictions of key trends such as thermal aging and insulation degradation. Combined with unsupervised anomaly detection algorithms, a three-tiered early warning system of "current score, future trend, and early anomaly" was formed. This transformed the operation and maintenance model from passive response and periodic inspections to condition-based predictive maintenance, significantly improving the timeliness and accuracy of fault warnings. It provided scientific decision support for scheduling optimal maintenance windows and effectively avoided unplanned downtime. Furthermore, in terms of operation optimization and control, the system, based on real-time sensing and model prediction, dynamically adjusts the cooling fan power through a feedforward-feedback composite control strategy, achieving on-demand cooling and significantly reducing auxiliary energy consumption while ensuring the equipment does not overheat. Intelligent anti-condensation control actively prevents insulation from getting damp; rapid harmonic compensation through active filters improves power quality and reduces additional harmonic losses; and semi-active vibration control adaptively suppresses vibration transmitted by the tower, protecting the equipment structure. These adaptive optimization controls work together to create a better operating microenvironment for the booster equipment, directly improving its operational reliability and extending its service life.

[0018] Finally, through an edge-cloud collaborative architecture, the integration of edge intelligence of individual devices and collective intelligence of the wind farm is achieved. The cloud platform can aggregate massive amounts of operational data, continuously optimize and iterate the health assessment model, making the model increasingly intelligent with use, and optimize the scheduling of maintenance resources from a global perspective, thereby maximizing the operational efficiency and economic benefits of the entire wind farm asset.

[0019] In summary, this invention not only solves the problem of long-term reliable operation of modular booster stations after installation inside the tower, but also upgrades them into an intelligent system capable of self-sensing, self-evaluation, and self-optimization through deep integration of sensing, artificial intelligence, and adaptive control, providing an effective technical solution for the intelligent upgrading and cost reduction and efficiency improvement of the wind power industry. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the control method for a modular dry boost system based on multi-physics sensing provided in a specific embodiment of the present invention. Figure 2 This is a schematic diagram of the system composition of the modular dry boost system control system based on multi-physics sensing provided in a specific embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application. Furthermore, directional terms mentioned in the following embodiments, such as "up," "down," "left," and "right," are only for reference to the directions in the accompanying drawings; therefore, the directional terms used are for illustrative purposes and not for limiting the invention.

[0022] The present invention will be further described below with reference to the accompanying drawings and preferred embodiments.

[0023] Please see Figure 1 This invention provides a method for controlling a modular dry boost system based on multi-physics sensing, comprising: The method, applied to a booster station unit pre-integrated into a rigid base and hoisted integrally into a wind turbine tower, includes: S100. Construct a multi-physics field collaborative sensing network to acquire multi-dimensional operational status data of the booster station unit in real time; First, it should be noted that this method applies to a modular dry-type substation unit where all electrical equipment and auxiliary systems are pre-installed and commissioned in a factory. This unit uses a rigid steel structure base as its foundation platform, rigidly connecting and integrating all equipment, including the dry-type transformer, medium-voltage switchgear, control and protection cabinet, and auxiliary power supply cabinet, into a robust whole. During the wind turbine tower hoisting process, this integrated unit is lifted into the pre-cast foundation inside the tower in one go by the main crane, greatly simplifying high-altitude operations. This control method is the core of ensuring the long-term reliable and efficient operation of this integrated unit under the special environment inside the tower (enclosed space, continuous low-frequency vibration, and large temperature and humidity variations). The method constructs a closed-loop intelligent control system of "perception-cognition-decision-execution," whose hardware foundation consists of a multi-physics field sensor network deployed inside the unit, an intelligent controller acting as the "edge brain," and a series of actuators.

[0024] It should be further explained that, at the software level, the execution flow of the method is as follows: First, the sensor network synchronously collects multi-dimensional data at a fixed sampling period (e.g., 1 second for temperature, vibration, and humidity data, and 20 milliseconds for electrical data). After filtering, calibration, and formatting, this data is sent to the edge intelligent controller.

[0025] S200. Based on the multi-dimensional operating status data, use a pre-trained health assessment model to assess the equipment health status and predict the failure risk. It should be further explained that the controller runs a pre-trained digital twin health assessment model for the equipment. This model performs fusion analysis on the input real-time data stream and outputs a quantitative health status score, a status prediction curve for key components (such as transformer winding hotspots), and a preliminary fault risk classification.

[0026] S300. Based on the results of the equipment health status assessment and fault risk prediction, generate and execute adaptive optimization control commands to adjust the operating status of the booster station unit.

[0027] It should be further explained that, based on these high-level cognitive results, the decision logic module within the controller generates specific, executable control commands according to preset optimization strategies (such as safety priority, energy efficiency optimization, etc.). These commands are sent to actuators such as cooling fan inverters, heaters, active filters, and adjustable dampers through digital outputs, analog outputs, or fieldbus messages, thereby dynamically adjusting the operating status of the substation and forming a continuously optimizing closed loop.

[0028] Understandably, this method realizes the transformation from passive response to active prevention, and from single-point monitoring to global optimization. Through deep fusion perception and intelligent decision-making of multi-physics field states, this invention significantly reduces the risk of downtime caused by problems such as overheating, insulation dampness, vibration fatigue, and harmonic pollution, extends equipment life, and reduces auxiliary system energy consumption through precise on-demand control, thereby improving the overall availability and economy of wind turbine units.

[0029] Specifically, the multi-dimensional operating status data includes at least two of the following: temperature field data, vibration field data, humidity field data, and electrical characteristic data.

[0030] It should be further explained that the multi-dimensional operational status data is the foundation for the system's status assessment and decision-making. Preferably, this embodiment simultaneously collects four types of data: temperature, vibration, humidity, and electrical characteristics, to achieve comprehensive monitoring. Temperature field data is used to assess the thermal state and heat dissipation efficiency of the equipment and is the primary input for insulation aging assessment. Vibration field data is used to assess the connection integrity, fatigue state, and impact of external excitations (wind-induced tower sway) of the mechanical structure. Humidity field data is used to assess condensation risk, which is crucial for the insulation of dry equipment in the enclosed space inside the tower. Electrical characteristic data directly reflects the electrical health status of the equipment, including load conditions, power quality, and harmonic pollution levels. These four types of data are synchronized temporally using a high-precision clock from the controller (synchronization accuracy better than 1 millisecond) and spatially by covering key points through the reasonable arrangement of sensors, together forming a multi-dimensional characteristic space that can comprehensively characterize the equipment's operational status. Collecting only one or two types of data, such as only temperature and current, will not be able to effectively distinguish fault modes (e.g., it will not be able to distinguish whether the temperature rise is due to overload or cooling fan failure), nor will it be able to provide early warnings of other risks such as condensation or mechanical loosening. Therefore, collecting at least two different types of data is the minimum requirement for this method to achieve effective evaluation, while collecting all four types of data is a necessary condition for achieving optimal evaluation results. Those skilled in the art can choose to include at least a combination of temperature and vibration, or a combination of temperature and humidity, depending on the specific equipment configuration and cost considerations; however, the best implementation is to collect all types of data.

[0031] Specifically, the construction of the multi-physics collaborative sensing network includes: deploying distributed temperature sensors at key locations of the transformer windings, cores, and switchgear busbars in the substation unit; deploying vibration sensors on the rigid integrated base and the housings of the main electrical equipment; and deploying humidity sensors in the internal cavities of the substation unit.

[0032] It should be further explained that the specific construction of the multi-physics field collaborative sensing network involves detailed design of sensor selection, placement, installation method, and signal transmission. For temperature field sensing, the hot spot temperature of a dry-type transformer is a decisive factor in insulation life. This implementation preferably uses a distributed fiber optic temperature measurement system. The specific steps are as follows: Select a high-temperature resistant, electromagnetic interference-resistant armored temperature-sensing optical cable with a temperature measurement range of -40°C to 300°C, an accuracy of ±1°C, and a spatial resolution of 1 meter. During the transformer winding process, the temperature-sensing optical cable is tightly and evenly wound and fixed in the interlayer insulation of the low-voltage and high-voltage windings to ensure good thermal conductivity contact between the optical cable and the copper conductor. For the iron core, slots are made on the surface of the iron core columns and the yoke, the temperature-sensing optical cable is embedded, and it is fixed by filling with thermally conductive putty. Inside the switchgear, a high-precision, high-response platinum resistance temperature sensor (PT100, Class A accuracy, encapsulated as a stainless steel probe) is installed at each of the A, B, and C phase connection bolts of the main conductive busbar. The probe is attached to the busbar surface via spring clamping or thermally conductive adhesive. Signals from all temperature sensors are connected to a temperature acquisition module located inside the control cabinet via shielded twisted-pair cables. This module converts resistance or optical signals into digital signals, which are then transmitted to the edge intelligent controller via a Modbus RTU bus. For vibration field sensing, to monitor structural vibrations transmitted from the tower foundation to the equipment and the equipment's own electromagnetic vibrations, this implementation uses a triaxial IEPE (integrated piezoelectric) accelerometer with a frequency response range of 0.5Hz to 5kHz and a measurement range of ±50g, sufficient to cover subhertz oscillations of the tower and mid-to-high frequency vibrations of the electrical equipment. One sensor is installed near each of the four mounting holes on the rigid integrated base to monitor foundation input vibration. One sensor is installed at the top center and corresponding bottom position of the dry-type transformer casing to monitor the transformer's vibration response. One sensor is installed at the top and middle of the switchgear cabinet frame. The sensors are fixed with threads or high-strength adhesive, and the signals are connected to a vibration data acquisition card via a coaxial cable. This acquisition card features anti-aliasing filtering and 24-bit high-resolution A / D conversion, and communicates with the controller via an Ethernet interface. For humidity sensing, a temperature-compensated digital output humidity sensor is selected to monitor the absolute humidity and dew point temperature of the air inside the boost unit. The measurement range is 0-100%RH, and the accuracy is ±2%RH (within the 20%-80%RH range). One sensor is placed in each of the following locations inside the boost unit: the upper part (hot air zone), the middle part (equipment core area, near the transformer and switchgear), and the lower part (near the bottom air inlet or cable entry). The sensors are installed in well-ventilated locations to avoid direct heating from the equipment or radiation from cold walls. The sensors communicate directly with the controller via an I2C or RS485 digital interface.

[0033] It is understood that, through the above-described specific deployment, the present invention forms a three-dimensional, multi-scale sensing network from the inside of the device (winding) to the surface (shell, busbar), to the surrounding environment (cavity air), and finally to the basic structure (base), providing accurate, reliable, and spatiotemporally synchronized physical field data for building a high-fidelity digital twin model.

[0034] Specifically, the electrical characteristic data is obtained through monitoring devices installed at the incoming and outgoing ends of the booster station unit, and the electrical characteristic data includes current, voltage harmonic components, and partial discharge signals.

[0035] It should be further explained that acquiring electrical characteristic data requires specialized monitoring equipment. A high-performance Electrical Condition Monitoring Device (IED) is installed at both the incoming (connecting to the wind turbine generator outlet) and outgoing (connecting to the collector line or power grid) ends of the substation unit. The core of this IED is a high-speed synchronous sampling unit, with a sampling rate set to 10kHz per channel to meet the analysis requirements for up to the 50th harmonic (according to IEC 61000-4-30 standard). The IED synchronously acquires the instantaneous values ​​of three-phase voltage (through voltage transformer PT) and three-phase current (through current transformer CT). For harmonic analysis, the controller performs a Fast Fourier Transform (FFT) on the sampled data for each cycle (20ms) to calculate the content of each harmonic (HR_h) and the total harmonic distortion (THD). Specifically, for discrete sampled signals... , its first Amplitude of subharmonic components and phase Calculated using FFT: in, This represents the number of sampling points within one fundamental frequency period (e.g., N=200 for a sampling rate of 10kHz). For harmonic order, These are discrete sampled values ​​of voltage or current. For the first The amplitude of the second harmonic. Its phase. The THD calculation formula is: in, The fundamental amplitude, The highest harmonic order (typically 50) is considered. For partial discharge (PD) monitoring, a combination of ultra-high frequency (UHF) and high-frequency current transformer (HFCT) methods is used. HFCT sensors with a frequency band of 100kHz-30MHz are installed on the grounding wire of the transformer high-voltage bushing end screen and the switchgear busbar grounding wire to detect pulse current signals. Wideband UHF antenna sensors with a frequency band of 300MHz-1.5GHz are installed inside the transformer tank and the switchgear gas chamber to detect electromagnetic wave signals excited by PD. The partial discharge monitoring unit continuously acquires HFCT and UHF signals. When the signal amplitude exceeds a set threshold (the threshold is adaptively set according to the background noise level; for example, the initial threshold is set to 3 times the root mean square value of the background noise), waveform recording is triggered, and the amplitude of a single PD pulse is extracted. Arrival time and the phase angle relative to the power frequency voltage. and form (Phase-amplitude-order) spectrum. These electrical characteristic data (harmonic content, THD, PD pulse characteristics) are aligned with temperature, vibration, and humidity data on the timestamp and then input into the health assessment model.

[0036] Understandably, this invention enables deep perception of electrical conditions, allowing for quantitative assessment of power quality, detection of early insulation degradation (such as winding deformation and partial discharge) and loose connections (increased contact resistance may lead to heating and increased harmonics), providing key electrical dimension information for comprehensive health assessment.

[0037] Specifically, the pre-trained health assessment model is a digital twin model, which is trained by integrating historical operating data, multiphysics simulation data, and fault case data, and is used to simulate the real-time health status of the booster station unit.

[0038] It should be further noted that the pre-trained health assessment model is a deep learning-based digital twin model, and its construction and training are the core steps in implementing this method, as detailed below: 1. Model Architecture: The model employs an encoder-decoder architecture combined with an attention mechanism. The encoder consists of two layers of a bidirectional long short-term memory (Bi-LSTM) network, used to encode the multidimensional time-series features of the input and extract their temporal dependencies. Each LSTM layer has 128 hidden units. The hidden state of the last time step output by the encoder is considered as the context vector of the input sequence. The decoder consists of an attention layer and a fully connected output layer. The attention layer dynamically assigns different weights to the encoder's output at each time step during decoding (prediction), focusing on the historical information most important for the current prediction. The fully connected output layer maps the attention-weighted context vector to the final output. The output includes two parts: a) the current health score (HS, a scalar between 0 and 100); b) key state predictions for multiple future time steps (e.g., the next 24 hours, with a step size of 1 hour), including transformer winding hotspot temperatures. 1. Top oil temperature of the transformer (if it is an oil-immersed type) 2. Maximum effective value of vibration acceleration Insulation status index (A normalized value between 0 and 1).

[0039] 2. Training data preparation: Training data comes from three sources: ① Multiphysics Simulation Data: Using ANSYS or COMSOL Multiphysics software, finite element models of the substation units (focusing on transformers and switchgear) are established. Transient simulations are performed with different boundary and load conditions, such as: ambient temperature (-20°C to 50°C, step size 5°C), load rate (10% to 120% of rated capacity, step size 10%), cooling fan speed (30% to 100%, step size 10%), and foundation vibration excitation (0.1g to 0.5g, different frequencies). Each simulation generates a time series (e.g., 24 hours) of multiphysics data, including winding temperature distribution, structural stress, and internal fluid velocity. Virtual monitoring point data corresponding to the sensor layout are extracted from this data to form "simulation-monitoring data pairs." This part of the data comprises approximately 5000 time series sets.

[0040] ② Historical Operational Data: Collect at least one year of actual operational data from multiple operating wind farms, including SCADA system records and offline monitoring reports, for the same or similar model of booster station units. The data must be cleaned, aligned, and labeled. Normal operating condition data is labeled as a healthy state (HS=100). Data for a period before and after a warning or maintenance event is labeled with corresponding HS values ​​by experts based on the severity of the event (e.g., minor overheating labeled HS=70, severe vibration labeled HS=50). This part of the data should contain no fewer than 10,000 time series sets.

[0041] ③ Fault Case Data: Collect typical fault case reports, such as insulation breakdown, winding overheating, and loose connections. Based on fault mechanism analysis and simulation, generate simulated data of the fault development process and label the HS descent curve from normal to fault. This part of the data is used to enhance the model's ability to identify rare fault modes, and the data volume is approximately 1000 sets. Divide the above three parts of data into training set, validation set, and test set in a 7:2:1 ratio. All data are standardized before being input into the model so that the mean of each feature dimension is 0 and the variance is 1.

[0042] 3. Model Training: The model is trained using the training set. The optimizer chosen is Adam, with an initial learning rate of 0.001 and a learning rate decay strategy (decreasing to 0.9 every 10 epochs). Loss function... Defined as health score regression loss and state prediction regression loss The weighted sum is then added, and an L2 regularization term is added to prevent overfitting: in, To predict the mean squared error (MSE) between the health score and the labeled health score; For the predicted key state quantities ( , , The MSE between the actual value and the true value (or high-fidelity simulation value); and As the weighting coefficient, in this embodiment, we take... , To place greater emphasis on the accuracy of health scores; Let be the regularization coefficient, and take . ; This represents all trainable parameters of the model. The training epochs are set to 200. After each epoch, the model performance is evaluated on the validation set, and the model parameters with the smallest loss on the validation set are retained. Finally, the model performance is evaluated on an independent test set, requiring the mean absolute error (MAE) of the health score prediction to be less than 5, and the MAE of the key state quantity prediction to meet engineering accuracy requirements (e.g., temperature error less than 3°C).

[0043] 4. Model Deployment: The trained model parameters (weights and biases) are stored in the edge intelligent controller's storage unit. During runtime, the controller inputs real-time collected and preprocessed multi-dimensional state data (a time window, e.g., data from the past hour, sampled at 1-minute intervals, i.e., 60 time steps) into the model. After forward propagation calculation, the model outputs the current health score (HS) and a predicted sequence of key states for the next 24 hours. This digital twin model can simulate the state of a physical entity in real time, forming the basis for subsequent intelligent decision-making. Its beneficial effect lies in enabling soft measurement of parameters that cannot be directly measured (such as the temperature of the hottest point in the winding) and rolling prediction of future states, providing crucial input for predictive maintenance.

[0044] Specifically, the equipment health status assessment and fault risk prediction include: inputting the real-time acquired multi-dimensional operating status data into the digital twin model to calculate the current health status score; predicting the thermal aging trend and insulation degradation trend of key components based on the digital twin model; and identifying early fault characteristics using an anomaly detection algorithm based on time-series data and the prediction output of the digital twin model.

[0045] It should be further explained that equipment health status assessment and failure risk prediction is a multi-level and refined process, with the specific steps as follows: Step A: Real-time health score calculation. The edge intelligent controller standardizes multi-dimensional time-series data (temperature, vibration, humidity, harmonics, partial discharge statistics, etc., totaling approximately 20 features) from the past 60 minutes, using a 1-minute cycle, and inputs it into the deployed digital twin model. After the model encoder-attention-decoder network runs, it outputs the comprehensive health score for the current moment. , The scoring logic incorporates expert rules: For "health" (green). "Attention" (yellow) The result is marked as "abnormal" (red). In addition, the model will output sub-scores for each major component (transformer, switchgear) to help pinpoint the problem. The selection of 85 and 70 as thresholds is based on engineering experience: a score below 85 indicates that a parameter deviates from the normal range but is still within the safety margin; a score below 70 indicates that one or more key parameters (such as hotspot temperature, vibration amplitude) are approaching or exceeding 80% of their design safety limits, requiring immediate attention.

[0046] 2. Step B: Trend Forecasting. Trend analysis is performed using the key state prediction sequence for the next 24 hours output by the digital twin model, specifically including: ① Thermal aging trend prediction: The model outputs a predicted sequence of winding hot spot temperatures for the next 24 hours. According to the Arrhenius equation for insulation thermal aging, the relative thermal aging rate... With hot spot temperature The relationship is: in, This is a reference temperature; it is used for common insulating materials. ; It is the thermal aging constant of the insulating material, for Class H insulation (temperature resistance 180°C). Usually taken (That is, the aging rate doubles for every 6°C increase in temperature). Therefore, the cumulative relative loss of aging life over the next 24 hours... for: in, To predict the time step (1 hour). At reference temperature Aging amount after 24 hours of operation (i.e.) By comparing the time spent aging (in hours) with the time spent aging (in hours), the severity of aging in the coming day can be assessed. If... If aging accelerates by more than double, an "accelerated aging" warning will be triggered.

[0047] ② Insulation condition degradation trend prediction: The model outputs a comprehensive insulation condition index. The predicted sequence is calculated. This index is derived by fusing multiple factors, including humidity, partial discharge intensity, and harmonic distortion rate, through internal model weights. The slope of this sequence is then calculated. (For example, using the linear regression slope of the last 6 predicted values) to assess the rate of degradation. If If the insulation condition index drops by more than 0.05 per day, it is determined that the insulation condition is rapidly deteriorating, triggering an "insulation degradation" warning.

[0048] 3. Step C: Early Anomaly Identification. A real-time anomaly detection model based on unsupervised learning, such as Isolation Forest, is run in parallel. This model is trained on historical normal data, learning multi-dimensional data under normal operating conditions (especially vibration spectrum characteristics, partial discharge...). The distribution of statistical characteristics of the spectrum. For each set of data that comes in real time, its anomaly score is calculated. If the anomaly score of a certain feature exceeds a threshold (e.g., an anomaly score greater than 0.65) for multiple consecutive periods (e.g., 5 consecutive periods), it is considered that an early anomalous feature has appeared, even if the digital twin model calculates... The rating may still be within the "attention" range. For example, a sudden and sustained increase in energy in the vibration spectrum around 1000Hz may indicate an early stage of mechanical loosening; partial discharge... The migration of spectral patterns from typical air-gap discharge to surface discharge patterns may indicate the onset of insulation surface contamination. This method can detect subtle early signs of failure that are not yet clearly reflected in the overall health score, providing earlier warning.

[0049] Understandably, this invention achieves a comprehensive, three-dimensional, and forward-looking diagnosis of equipment health status through a three-level assessment system of "current score + future trend + early anomaly", which greatly improves the timeliness and accuracy of fault warning.

[0050] Specifically, the generation and execution of adaptive optimization control instructions includes: dynamically adjusting the operating power of the built-in cooling device of the booster station unit based on the temperature field data and thermal aging trend prediction results; and controlling the start and stop of the active anti-condensation device based on the humidity field data and insulation status assessment results.

[0051] It should be further explained that adaptive optimization control is the execution phase of the decision-making process. For cooling control, the system incorporates multiple forced-air cooling fans driven by frequency converters. The control strategy employs a feedforward-feedback composite control. The controller reads the current winding hot spot temperature output from the digital twin model. And the predicted highest temperature in the next hour The control objective is to maintain the hotspot temperature within a safe threshold. (For example, for Class H insulation, the long-term operating temperature limit is) , Can be set to (With sufficient margin) below, while saving energy as much as possible. Define temperature deviation. ,in Set the desired temperature value (e.g.) Define the predicted temperature deviation. Total fan speed percentage The input is calculated by the fuzzy PID controller. and The fuzzy value. The fuzzy rule base is designed as follows: if If it is "PB" (meaning predicted to exceed the temperature limit), then regardless of Regardless, all output "high" speed; if If it is "zero" (Z) or "negative" (N), then according to The speed is adjusted by the magnitude of the rotation. The precise output is obtained through defuzzification. Values, ranging from 30% (minimum maintenance ventilation) to 100%. Furthermore, if... (Attention threshold), the controller will preemptively... The temperature is increased to over 80% for preventative cooling. This composite control, based on current conditions and short-term forecasts, offers a faster response compared to traditional thermostats, effectively preventing temperature overshoot and saving approximately 15-30% of cooling energy. For anti-condensation control, the system is equipped with PTC heaters and controllable ventilation louvers at the bottom and top of the unit. The controller calculates the dew point temperature within the chamber in real time. The control strategy is as follows: First, compare the internal air dew point temperature. and external ambient air temperature (Provided by the tower weather station). If (If the outside air is relatively dry and the temperature is more than 3°C above the dew point (3°C is a safety margin to account for sensor error and local low temperatures), then open the ventilation louvers to introduce dry outside air for natural dehumidification. If the outside air humidity is higher or does not meet ventilation requirements, then adjust the ventilation based on the internal air temperature. and The difference is controlled. Condensation risk is defined. .when At that time, the PTC heater is started, and the heating power is [not specified]. Through PID control, Maintain at to Within the safe range. Meanwhile, if the insulation condition index... Predicted slope / day, even at present It will also activate low-power heating (e.g., 20% power) for preventative dehumidification.

[0052] Understandably, this control strategy effectively prevents the temperature of any part from dropping below the dew point, eliminating condensation and protecting equipment insulation. This invention achieves on-demand, precise, and proactive control of cooling and anti-condensation, maximizing the optimization of auxiliary energy consumption while ensuring safe equipment operation.

[0053] Specifically, the generation and execution of adaptive optimization control instructions further includes: generating compensation instructions for an active filter to suppress harmonic interference based on the harmonic components in the electrical characteristic data; and adjusting the damping parameters of the adaptive damping structure integrated into the rigid base based on the vibration field data and the equipment health status score.

[0054] It should be further noted that the system is equipped with a parallel active power filter (APF) for harmonic suppression control. The APF's compensation command generation adopts a method based on instantaneous reactive power theory. Detection method. The specific steps are as follows: 1. Collect the three-phase load current on the incoming side of the booster station unit. .

[0055] 2. Obtain the synchronization phase angle of the grid voltage through a phase-locked loop (PLL). .

[0056] 3. Convert the three-phase load current to rotating current using Clark and Park transformations. In coordinate system: in, for Axis current (active current component). for Axis current (reactive current component).

[0057] 4. and The DC component is obtained by passing it through a low-pass filter (LPF, with a cutoff frequency typically set to 20-50Hz to separate the fundamental frequency from the harmonics). and These correspond to the fundamental active and reactive components of the load current.

[0058] 5. Harmonic components are... , .

[0059] 6. and The three-phase harmonic current command to be compensated is obtained by inverse Park transform and inverse Clark transform. .

[0060] 7. The inner-loop current tracking controller of the APF (usually a proportional-resonant controller or predictive current control) generates a PWM signal according to this instruction, driving the IGBT inverter bridge to generate compensation current. This is injected into the grid to offset load harmonic currents. The control objective is to keep the total harmonic distortion (THDi) of the current below 5% to meet power quality standards and reduce harmonic-induced losses. This threshold (5%) is a common requirement for harmonic limits in medium-voltage systems in international standards such as IEEE Std 519.

[0061] It should be further explained that, for vibration suppression control, four magnetorheological (MR) dampers are installed between the rigid integrated base and the tower foundation. The control strategy employs semi-active acceleration feedback control. The controller reads data from acceleration sensors installed at the four corners of the base and calculates the effective value of the combined vibration acceleration at the center of the base. And analyze its main frequency. The control objective of a damper is to reduce the vibration acceleration transmitted to the equipment. Damping force. With the input control current The relationship is approximately linear (before saturation): ,in It is the relative velocity of the damper piston. It is the damping coefficient, which varies with current. It increases as it increases. The control law is designed as follows: .in, For proportional gain, This is the differential gain (frequency-dependent). When the health score... Lower (e.g.) And when the vibration component is prominent, it increases. and This is to increase damping force and quickly suppress vibration. When When the vibration frequency is high and the vibration is low, the gain is reduced to decrease energy consumption. Simultaneously, if the dominant vibration frequency is detected... When the frequency approaches the natural frequency of the equipment or structure (e.g., the natural frequency of transformer windings may be between 100-400Hz), the controller adds an additional resonance suppression term, specifically providing higher damping near that frequency. This control strategy achieves adaptive suppression of broadband vibrations (especially resonance peaks), effectively protecting the equipment from long-term vibration fatigue damage.

[0062] Understandably, this invention enables proactive and dynamic control of electrical harmonics and mechanical vibrations, improving the equipment operating environment from the source and enhancing power quality and mechanical reliability.

[0063] Specifically, the method further includes: uploading the status assessment results, fault warning information and optimized control logs processed by the edge computing device to the cloud platform; using the cloud platform to aggregate data from multiple booster station units to perform incremental training and parameter optimization on the health assessment model; and generating predictive maintenance strategies for each booster station unit based on the optimized model.

[0064] It should be further explained that, to achieve the evolution from single-machine intelligence to swarm intelligence, this method includes a cloud-based collaborative optimization layer. After completing real-time monitoring and control locally, the edge intelligent controller periodically (e.g., every 15 minutes) packages and uploads data summaries to the cloud platform. The data package content includes: timestamp, device ID, and current health score. The cloud platform receives data from hundreds or thousands of substation units in wind farms, including key status predictions, triggered warning codes, and logs of control actions performed over a period of time. After receiving this data, the cloud platform performs the following core tasks: 1. Incremental Model Training and Optimization: A global digital twin model (“teacher model”) is maintained in the cloud, with the same structure as the edge model but optimized parameters. Each week, the cloud randomly selects a certain percentage (e.g., 10%) of new (anonymized) data from all devices and mixes it with historical data to incrementally train (fine-tune) the “teacher model.” During training, a small learning rate (e.g., 0.0001) is used to prevent catastrophic forgetting. After training, the performance of the new model is evaluated on an independent validation set. If the new model's loss on the validation set is lower than that of the old model by a certain percentage (e.g., 5%), the new model is considered to perform better. Subsequently, the cloud securely distributes the parameters of the new model to all online edge devices, which perform online hot updates of the model parameters during idle periods (e.g., nighttime). This mechanism enables the model to continuously learn the degradation patterns of devices under different geographical environments and operating modes, becoming increasingly “intelligent.”

[0065] 2. Generate a predictive maintenance strategy: The cloud platform runs a maintenance scheduling optimization algorithm. This algorithm uses a one-month planning period, and its inputs include: the health score of each booster station unit. The forecast curve, predicted failure risk types, planned maintenance time (e.g., 2 hours for desiccant replacement, 4 hours for tightening checks), maintenance personnel availability, spare parts inventory, and future weather windows (wind speed, rainfall forecasts) are all considered. The objective function is to maximize the expected power generation revenue of the entire wind farm cluster during this cycle, while minimizing maintenance costs and failure risks. Constraints include: maintenance resource constraints, spare parts constraints, and maintenance time window constraints for each device (e.g., ...). (Maintenance must be completed before the predicted value falls below a certain threshold). This optimization problem can be modeled as a mixed integer programming (MIP) problem and solved using commercial solvers (such as Gurobi and CPLEX). The solution generates a suggested maintenance time window, maintenance type, required personnel, and spare parts list for each booster station unit, forming a predictive maintenance work order, which is then pushed to the wind farm management personnel's operation and maintenance system.

[0066] Understandably, this invention achieves continuous self-evolution of model performance through cloud-based data aggregation and intelligent analysis, and performs optimal maintenance scheduling based on a global perspective, thereby maximizing the availability and economic benefits of the entire wind power asset.

[0067] Please see Figure 2 The present invention provides another embodiment, which provides a modular dry boost converter control system based on multiphysics sensing. The modular dry boost converter control system based on multiphysics sensing includes: The system is deployed in a booster station unit pre-integrated into a rigid base and hoisted as a whole into the wind turbine tower. The system includes: A multi-physics sensing network 100 is configured to collect multi-dimensional operational status data of the booster station unit. The edge intelligent controller 200 is communicatively connected to the multi-physics sensing network and internally stores a pre-trained health assessment model. The edge intelligent controller is configured to: perform equipment health status assessment and fault risk prediction based on the multi-dimensional operating status data and the health assessment model; and generate adaptive optimization control commands based on the assessment and prediction results. The actuator 300 is communicatively connected to the edge intelligent controller and configured to receive and execute the adaptive optimization control command. The actuator includes at least one of the following: a cooling device, an active anti-condensation device, an active filter, and an adaptive damping structure.

[0068] Specifically, this system is the physical embodiment of the aforementioned method. The multiphysics sensing network 100 consists of the various sensors, signal conditioners, data concentrators, and communication cables described above. All sensors are connected through a unified industrial IoT gateway, which supports multiple protocol conversions such as Modbus RTU / TCP and IEC61850, and packages the data into unified MQTT or OPCUA messages, which are then sent to the edge intelligent controller via Ethernet. The edge intelligent controller 200 is an industrial-grade embedded computer, whose hardware configuration includes at least: a multi-core ARM or x86 processor (with a main frequency of not less than 1.5GHz), not less than 8GB of RAM, not less than 64GB of eMMC solid-state storage, multiple Ethernet ports, multiple RS485 / RS232 serial ports, digital input / output (DI / DO) modules, and analog input / output (AI / AO) modules. Its operating system is a customized Linux real-time kernel. At the software level, the controller runs multiple containerized or process-based services, including: data acquisition and preprocessing services, digital twin model inference services, health assessment and fault diagnosis services, optimized control decision services, and communication services (communication with sensors, actuators, and the cloud). The pre-trained health assessment model is stored within the controller in an optimized format (e.g., ONNX, Tensor RT). The inference service loads this model and performs real-time calculations on the input data stream. The actuator 300 is a specific physical device: the cooling unit consists of multiple axial fans driven by a frequency converter. The frequency converter's control terminals receive 4-20mA speed command signals from the controller's analog output module. The active anti-condensation device includes a PTC heater group and electrically operated ventilation louvers, whose start and stop are controlled by the controller's digital output module (relay output). The heater power can be adjusted via PWM or a 0-10V analog signal. The active power filter (APF) is a separate cabinet whose control board is connected to the edge intelligent controller via high-speed Ethernet (e.g., EtherCAT) to receive harmonic current compensation commands. The adaptive damping structure consists of four magnetorheological dampers, each with its own current driver. This driver receives a 0-2A current control signal from the controller's analog output module to adjust the damping force. Status feedback signals from all actuators are also returned to the controller via corresponding input modules, forming a closed-loop control system. Through meticulous electrical cabinet wiring, electromagnetic compatibility design, and environmental protection design (IP54 protection rating), the system ensures reliable operation in the harsh environment of tower vibration, temperature and humidity variations, and electromagnetic interference.

[0069] Understandably, this system provides a highly integrated, intelligent, and autonomous hardware and software solution that embeds sensing, analysis, decision-making, and control functions into the booster station unit itself, making it an intelligent entity capable of self-sensing, self-evaluation, and self-optimization. This fundamentally improves the reliability, maintainability, and lifecycle value of the modular dry booster system.

[0070] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the described multiphysics-based modular dry boost system control method. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the method of the present invention.

[0071] This invention can be implemented as a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0072] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor, to perform the steps of this invention when executed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0073] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0074] The specific embodiments of the present invention described above do not constitute a limitation on the scope of protection of the present invention. Any other corresponding changes and modifications made in accordance with the technical concept of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A control method for a modular dry boost converter system based on multiphysics sensing, characterized in that, The method, applied to a booster station unit pre-integrated into a rigid base and hoisted integrally into a wind turbine tower, includes: S100. Construct a multi-physics field collaborative sensing network to acquire multi-dimensional operational status data of the booster station unit in real time; S200. Based on the multi-dimensional operating status data, use a pre-trained health assessment model to assess the equipment health status and predict the failure risk. S300. Based on the results of the equipment health status assessment and fault risk prediction, generate and execute adaptive optimization control commands to adjust the operating status of the booster station unit.

2. The method according to claim 1, characterized in that, The multi-dimensional operating status data includes at least two of the following: temperature field data, vibration field data, humidity field data, and electrical characteristic data.

3. The method according to claim 2, characterized in that, The construction of the multi-physics collaborative sensing network specifically includes: deploying distributed temperature sensors on key parts of the transformer windings, iron cores, and switchgear busbars of the substation unit; deploying vibration sensors on the rigid integrated base and the housings of the main electrical equipment; and deploying humidity sensors in the internal cavities of the substation unit.

4. The method according to claim 3, characterized in that, The electrical characteristic data is obtained through monitoring devices installed at the incoming and outgoing ends of the booster station unit. The electrical characteristic data includes current, voltage harmonic components, and partial discharge signals.

5. The method according to claim 1, characterized in that, The pre-trained health assessment model is a digital twin model, which is trained by integrating historical operating data, multiphysics simulation data, and fault case data, and is used to simulate the real-time health status of the booster station unit.

6. The method according to claim 5, characterized in that, The equipment health status assessment and fault risk prediction specifically include: inputting the real-time acquired multi-dimensional operating status data into the digital twin model to calculate the current health status score; predicting the thermal aging trend and insulation degradation trend of key components based on the digital twin model; and identifying early fault characteristics using an anomaly detection algorithm based on time-series data and the prediction output of the digital twin model.

7. The method according to claim 1 or 6, characterized in that, The generation and execution of adaptive optimization control instructions include: dynamically adjusting the operating power of the built-in cooling device of the booster station unit based on the temperature field data and thermal aging trend prediction results; and controlling the start and stop of the active anti-condensation device based on the humidity field data and insulation status assessment results.

8. The method according to claim 7, characterized in that, The generation and execution of adaptive optimization control instructions further includes: generating compensation instructions for an active filter to suppress harmonic interference based on the harmonic components in the electrical characteristic data; and adjusting the damping parameters of the adaptive damping structure integrated in the rigid base based on the vibration field data and the equipment health status score.

9. The method according to claim 1, characterized in that, The method further includes: uploading the status assessment results, fault warning information and optimized control logs processed by the edge computing device to the cloud platform; using the cloud platform to aggregate data from multiple booster station units to perform incremental training and parameter optimization of the health assessment model; and generating predictive maintenance strategies for each booster station unit based on the optimized model.

10. A modular dry boost converter control system based on multiphysics sensing, used to implement the method of any one of claims 1-9, characterized in that, The system is deployed in a booster station unit pre-integrated into a rigid base and hoisted as a whole into the wind turbine tower. The system includes: A multi-physics sensing network is configured to collect multi-dimensional operational status data of the booster station unit. An edge intelligent controller is communicatively connected to the multi-physics sensing network and internally stores a pre-trained health assessment model. The edge intelligent controller is configured to: assess the health status of the equipment and predict the failure risk based on the multi-dimensional operating status data and the health assessment model; and generate adaptive optimization control commands based on the assessment and prediction results. An actuator is communicatively connected to the edge intelligent controller and configured to receive and execute the adaptive optimization control command. The actuator includes at least one of a cooling device, an active anti-condensation device, an active filter, and an adaptive damping structure.

Images