A wind power tower inner voltage boosting system intelligent monitoring and protection method and system

Abstract

The present application relates to the technical field of power system automation and intelligence, in particular to a kind of wind power tower inner boost system intelligent monitoring and protection method and system.The method is by deploying multiple types of sensor acquisition system multidimensional operating parameters;Based on edge computing unit, utilize artificial intelligence model to carry out real-time processing analysis to parameter, generate comprehensive state evaluation result and fault prediction information, and accordingly dynamically adjust protection strategy parameters;Generate intelligent analysis report that fuses prediction information, analysis result and historical event;And with cloud platform collaborative interaction, realize feature data upload and model update reception receive.The present application deeply integrates monitoring, protection, diagnosis and prediction function, realizes the change from passive response to active prediction decision, improves safety through adaptive protection, improves operation efficiency and system sustainability with the help of intelligent fault diagnosis and cloud collaborative evolution.

Description

Technical Field

[0001] This invention relates to the field of power system automation and intelligent technology, specifically to an intelligent monitoring and protection method and system for a step-up system in a wind turbine tower. Background Technology

[0002] As a crucial component of clean energy, the reliable operation of wind power is vital to grid stability. The step-up system within wind turbine towers, containing key equipment such as transformers and switchgear, operates in a confined environment with limited heat dissipation. It also endures electrical and mechanical stresses from turbine start-up and shutdown, and power fluctuations, making it a high-risk area for failures. Currently, monitoring and protection of this system primarily rely on traditional relay protection devices and independent online monitoring systems. Traditional relay protection devices, based on fixed logic and settings, achieve rapid isolation after a fault, but their protection strategies cannot detect changes in the equipment's own health status, posing risks of "overprotection" leading to false trips or "underprotection" leading to failure to trip, and lacking fault early warning capabilities. While independent online monitoring systems can collect state variables such as temperature, vibration, and partial discharge, they typically only provide remote data display and over-limit alarms. Their functions are relatively isolated, and analysis remains at the level of simple threshold judgments, unable to deeply analyze and predict the deterioration trend of equipment status. After a fault occurs, maintenance personnel rely on analyzing discrete fault waveform data and event records; fault location and root cause analysis are time-consuming and laborious, prolonging downtime. Furthermore, in existing solutions, monitoring, protection, and recording modules often come from different vendors, resulting in fragmented systems and a lack of deep data integration and utilization. This creates isolated "information silos," hindering collaborative assessment and closed-loop control of the overall equipment health status. While the development of artificial intelligence has led to attempts to apply individual AI algorithms to fault diagnosis or status prediction, these are mostly added as independent modules to existing systems. They fail to achieve deep collaboration with core protection logic, operational decision-making processes, and cloud resources. Their models also cannot continuously evolve with equipment operation, limiting their level of intelligence and failing to meet the urgent needs of wind farms for minimally staffed and predictive maintenance. Therefore, there is an urgent need for an integrated intelligent monitoring and protection solution that deeply integrates perception, analysis, protection, and decision-making, and possesses self-evolving capabilities.

[0003] Therefore, existing technologies still need further development. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide an intelligent monitoring and protection method and system for the booster system in a wind turbine tower, so as 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 an intelligent monitoring and protection method for a step-up system within a wind turbine tower, comprising: S100: Collects multi-dimensional operating parameters of the booster system deployed in the wind turbine tower through multiple types of sensors; S200: Based on the edge computing unit, the operating parameters are processed and analyzed in real time to generate a comprehensive status assessment result and fault prediction information for the boost system; S300. Based on the comprehensive status assessment results, dynamically adjust the protection strategy parameters of the boost system; S400: Generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results, and historical events; S500, to perform collaborative interaction with the cloud platform, the collaborative interaction including uploading feature data and receiving updated analysis models or optimization strategies issued by the cloud platform.

[0006] Specifically, the real-time processing and analysis of operating parameters includes: The first artificial intelligence model is used to analyze the time-series operating parameters in order to predict the health status trend and early failure risk of key components in the boost system.

[0007] Specifically, the real-time processing and analysis of operating parameters also includes: The second artificial intelligence model is used to identify fault waveform data in the operating parameters in order to determine the fault type and estimate the fault location.

[0008] Specifically, the dynamic adjustment of the protection strategy parameters of the boost system is as follows: Based on the comprehensive status assessment results, the action threshold or action delay of the electrical protection components are adaptively adjusted.

[0009] Specifically, the generation of the intelligent analysis report also includes: The fault prediction information and real-time analysis results are correlated with the pre-stored historical fault records and operation records, and maintenance suggestions are output based on the correlation analysis results.

[0010] Specifically, the collaborative interaction with the cloud platform includes: The feature data processed by the edge computing unit, the comprehensive status assessment results, and the intelligent analysis report are uploaded to the cloud-based digital twin platform.

[0011] Specifically, the collaborative interaction also includes: The system receives an analysis model, trained or optimized based on the feature data and global data, from the cloud-based digital twin platform to update the model running in the edge computing unit.

[0012] Specifically, the method further includes: Based on the comprehensive status assessment results and preset decision rules, preset non-emergency fault handling operations or operation mode switching are automatically executed locally.

[0013] Specifically, the operating parameters include electrical quantities, non-electrical quantity status quantities, and signals acquired by image or acoustic sensors.

[0014] According to a second aspect of the present invention, an intelligent monitoring and protection system for a step-up system within a wind turbine tower is provided, comprising: The data acquisition module is configured to collect multi-dimensional operating parameters of the booster system inside the wind turbine tower through multiple types of sensors; The edge intelligent analysis module has a built-in edge computing unit and at least one artificial intelligence model, and is configured to process the operating parameters to generate comprehensive status assessment results and fault prediction information, and dynamically adjust the protection strategy based on the results. A local protection and control module is connected to the edge intelligent analysis module and configured to execute the adjusted protection strategy. The data management module is configured to generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results, and historical events. The cloud collaboration module is configured to interact with the cloud platform for data and strategies, including uploading feature data and receiving updated models or optimization strategies.

[0015] Beneficial effects: The intelligent monitoring and protection method and system for the booster system inside the wind turbine tower provided by this invention has significant and multifaceted advantages compared to existing technologies: First, by deeply integrating multi-source heterogeneous data acquisition, real-time edge-side AI analysis, dynamic adaptive protection strategies, intelligent report generation, and cloud-based collaborative updates into a unified architecture, the isolated monitoring, protection, and diagnostic functions of traditional systems are completely broken, achieving a fundamental shift from "passive response alarms" to "proactive predictive intervention." Specifically, by utilizing time-series predictive models deployed at the edge to continuously assess the health status of equipment and provide early risk warnings, effective warnings can be provided hours or even earlier before a failure occurs, creating a critical time window for planned maintenance and greatly reducing the risk of unplanned downtime.

[0016] Secondly, the adaptive protection mechanism of this invention innovatively links the protection setting with the real-time health index and operating conditions of the equipment, realizing a leap from "fixed logic protection" to "personalized and precise protection". It can provide more sensitive protection when the equipment is in poor condition, and avoid unnecessary malfunctions when the equipment is in good condition, which significantly improves the intelligence level of the protection system and the safety margin of the equipment.

[0017] Furthermore, the AI-based automatic fault waveform identification and location technology enables rapid and accurate fault type identification and preliminary location estimation. Combined with intelligent analysis reports for correlation mining and root cause analysis of historical data, it greatly shortens fault troubleshooting time and provides highly valuable maintenance suggestions, upgrading the operation and maintenance model from "experience-driven" to "data and knowledge-driven".

[0018] Furthermore, the "edge-cloud" collaborative evolution system constructed in this invention is particularly crucial. The edge side is responsible for real-time agile decision-making, while the cloud-based digital twin performs deep simulation and global optimization, and realizes centralized training and secure distribution and updates of artificial intelligence models. This enables the entire system to have the ability to continuously learn and self-optimize, adapt to different operating environments, equipment aging, and new fault modes, and solves the problem of algorithm solidification in traditional embedded systems.

[0019] Finally, from a system integration perspective, this invention integrates multiple advanced functions into a single device, reducing the number and complexity of on-site equipment, improving system reliability, and providing a solid technical foundation for the intelligent and less-manned operation and maintenance of wind farms. It has significant engineering application value and economic benefits. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the intelligent monitoring and protection method for the booster system inside a wind turbine tower provided in a specific embodiment of the present invention; Figure 2 This is a schematic diagram of the system composition of the intelligent monitoring and protection system for the booster system inside the wind turbine tower 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 an intelligent monitoring and protection method for a step-up system within a wind turbine tower, comprising: S100 collects multi-dimensional operating parameters of the booster system deployed in the wind turbine tower through various types of sensors.

[0024] It should be further noted that the implementation environment of this method is the step-up equipment compartment located inside the wind turbine tower. Its core consists of a 10kV or 35kV dry-type transformer and its associated medium-voltage switchgear, protection, and control units. The various types of sensors specifically include: electrical quantity sensors, namely current transformers (CTs) and voltage transformers (PTs), which synchronously acquire three-phase currents at a sampling rate of 10kHz. Three-phase voltage Zero-sequence current and zero-sequence voltage Non-electrical quantity sensors include fiber optic temperature sensors (temperature range -40~200°C, accuracy ±0.5°C) installed at hot spots in the transformer windings, triaxial vibration acceleration sensors (range ±50g, frequency range 0.5Hz~5kHz) installed on the transformer body, and ultra-high frequency (UHF) partial discharge sensors (detection band 300MHz~1.5GHz). All sensor signals are connected to an edge computing unit located in the control cabinet at the bottom of the tower via shielded cables. This unit uses an industrial-grade embedded platform based on NVIDIA Jetson AGX Orin, with a built-in 256-core GPU and 12-core ARM CPU, running the Ubuntu 20.04 operating system and a Docker container environment.

[0025] S200: Based on the edge computing unit, the operating parameters are processed and analyzed in real time to generate a comprehensive status assessment result and fault prediction information for the boost system.

[0026] It should be further explained that real-time processing and analysis are executed cyclically within the edge computing unit with a 1-second cycle. The software flow is as follows: First, the raw sampled data undergoes Kalman filtering and outlier removal preprocessing. Then, the preprocessed data (the most recent 300 sampling points, i.e., a 30-second time window) is input into two parallel AI model inference containers. The model output is then fused with the judgment results of the rule-based expert system to form the final comprehensive state assessment result (a quantified health index HI and a set of fault probability vectors) and fault prediction information (such as the predicted winding temperature value for the next hour). ).

[0027] S300. Based on the comprehensive status assessment results, dynamically adjust the protection strategy parameters of the boost system.

[0028] It should be further explained that the specific implementation of dynamically adjusting protection strategy parameters is as follows: the edge computing unit, through its built-in IEC61850MMS server, writes the calculated new protection settings (such as the overcurrent protection stage I current setting) to a service. and delay The value is written to the corresponding logic node (LLN0) of a traditional microcomputer protection device (such as Siemens 7SJ series) within the same network to complete the online modification of the setting.

[0029] S400: Generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results, and historical events.

[0030] It should be further noted that the intelligent analysis report is generated in JSON-LD format and includes timestamps, device IDs, health indices, prediction information, a list of associated historical event IDs, and maintenance suggestions in natural language description.

[0031] S500, to perform collaborative interaction with the cloud platform, the collaborative interaction including uploading feature data and receiving updated analysis models or optimization strategies issued by the cloud platform.

[0032] It should be further explained that the collaborative interaction with the cloud platform is achieved through the MQTT protocol based on TLS1.3 encryption. The edge device acts as a publisher, publishing feature data messages to the topic " / edge / device / [device ID] / features" (once every 5 minutes) and subscribing to the topic " / cloud / model / [device ID] / update" to receive model update packages.

[0033] Understandably, this method constructs a complete closed loop from physical perception to cloud collaboration. Its core innovation lies in integrating traditionally discrete protection and monitoring functions into an intelligent hub with real-time analysis, prediction, and adaptive capabilities through edge AI computing units. This method not only achieves deep perception of equipment status but, more importantly, endows the system with "predictability" and "adaptability," enabling dynamic optimization of protection strategies based on the real-time health status of equipment, transforming "passive response" into "active defense," and significantly improving the reliability and economy of wind power system operation. Compared with solutions that simply overlay AI modules, this method, through clearly defined data flows (acquisition, edge analysis, dynamic adjustment, report generation, and cloud interaction) and decision-making logic, forms a complete and implementable technical solution, solving the technical problem in existing technologies where monitoring, protection, and diagnostic functions are isolated and unable to coordinately respond to changes in equipment status.

[0034] Specifically, the real-time processing and analysis of operating parameters includes: using a first artificial intelligence model to analyze time-series operating parameters in order to predict the health status trend and early failure risk of key components in the boost system.

[0035] It should be further noted that the first artificial intelligence model is a Long Short-Term Memory (LSTM) network model. Its input is a time step. (Corresponding to 30 seconds, 3000 original points at a sampling rate of 10kHz, obtained after 100Hz low-pass filtering and 10:1 downsampling), Feature Dimensions Temporal tensor The 12 characteristics include: the effective value of the three-phase current. Three-phase voltage RMS value Total active power Total reactive power Transformer winding temperature Core temperature Effective value of vibration and partial discharge pulse count The model structure consists of two stacked LSTM layers, each with 128 hidden units, followed by a Dropout layer (dropout rate of 0.2) to prevent overfitting, and finally two parallel fully connected output layers. The first output layer is used for regression prediction, outputting the winding temperature sequence for the next 60 time steps (i.e., the next hour). The second output layer is used for health status assessment, outputting a health index. and a failure probability vector , representing the probabilities of overheating, insulation deterioration, and mechanical loosening, respectively.

[0036] Furthermore, the model is trained in the cloud using historical data, which includes at least one year of normal operation data (sampled hourly) and all traceable early failure cases (data snippets from hours prior to the failure). The loss function is a weighted sum of mean squared error (MSE) and cross-entropy. At the edge, the model is converted to a Tensor RT engine to accelerate inference, with each inference taking less than 50 milliseconds. The warning threshold is set as follows: when the predicted future highest winding temperature... (The alarm threshold for Class F insulation is typically 155°C; a 15% warning margin is set here.) An overheat warning is triggered when the health index... When any fault probability is triggered, a sub-health warning is issued; When this occurs, an early fault warning of the corresponding type is triggered.

[0037] Understandably, this LSTM model, through in-depth analysis of multi-dimensional, long-term operating parameters, can capture slow deterioration trends in equipment conditions that are imperceptible to the human eye or simple threshold methods. For example, by analyzing subtle changes in the vibration spectrum and slight increases in current harmonics, it can predict potential overheating risks caused by wear on cooling fan bearings hours in advance, before the temperature has risen significantly. This proactive early warning capability provides a crucial time window for planned maintenance, allowing maintenance personnel to schedule repairs at their convenience, avoiding unplanned downtime, and fundamentally changing the passive "repair after failure" model. It is the core technological support for achieving predictive maintenance.

[0038] Specifically, the real-time processing and analysis of operating parameters also includes: using a second artificial intelligence model to identify fault waveform data in the operating parameters in order to determine the fault type and estimate the fault location.

[0039] It should be further noted that the second artificial intelligence model is a 1D convolutional neural network (1D-CNN) model. When the system detects a sudden change in current ( When the threshold is exceeded or a protection start signal is received, fault recording is automatically triggered, recording the raw sampled data from 4 cycles (80ms@50Hz) before the fault to 20 cycles (400ms) after the fault, with a sampling rate of 20kHz. The recorded data includes 6 channels for three-phase current and three-phase voltage. The preprocessing steps include: first, subtracting the mean (removing the DC component) and dividing by the standard deviation (normalization) of the data in each channel; then, stacking the normalized data of the 6 channels into a 6-row, N-column matrix (N is the number of sampling points) as the input to the CNN. The CNN model structure includes: the first layer has 32 convolutional kernels of size 3 with a stride of 1, using ReLU activation; followed by a max pooling layer (pooling size 2); the second layer has 64 convolutional kernels of size 3 with a stride of 1, ReLU activation, and max pooling (pooling size 2); then the feature map is flattened and connected to a fully connected layer containing 128 neurons, and finally the output layer. The output layer has two heads: a classification head outputs a 7-dimensional vector, which uses the Softmax function to obtain the probabilities of seven fault types, including: single-phase grounding (A phase), two-phase short circuit (BC phase), two-phase ground fault, three-phase short circuit, internal transformer fault, external line fault, and non-faulty disturbance. The regression head outputs a scalar representing the estimated fault distance. (Unit: km) This estimate is based on the traveling wave principle or impedance method and is learned by the model from waveform features. Model training utilizes a large amount of fault waveform data generated by PSCAD / EMTDC simulation and labeled historical fault data from the field. On the edge side, the CNN model is also optimized using Tensor RT, with an analysis time of less than 20 milliseconds for a single waveform data entry. For the location results, the system combines topology information to provide text descriptions such as "transformer side inside the tower" or "collector line, approximately 2.3 km from this tower."

[0040] Understandably, traditional fault recording devices only record waveforms, and fault type identification and location heavily rely on the experience of relay protection personnel, which is time-consuming and prone to misjudgment. This solution applies a specially trained 1D-CNN model to fault recording analysis, achieving automatic, rapid, and high-precision identification of fault types and preliminary estimation of fault location. This not only provides crucial diagnostic information instantly upon fault detection, significantly shortening troubleshooting time, but its identification capabilities are not limited by traditional protection principles (such as impedance circles and directional elements). It can handle more complex and concealed fault types (such as high-resistance grounding), and can even distinguish between genuine faults and disturbances caused by inrush current, resonance, etc., thus significantly improving the intelligence and accuracy of fault diagnosis.

[0041] Specifically, the dynamic adjustment of the protection strategy parameters of the boost system involves adaptively adjusting the action threshold or action delay of the electrical protection components based on the comprehensive status assessment results.

[0042] It should be further explained that the adaptive adjustment is based on a preset dynamic adjustment strategy table, which defines the mapping relationship between the comprehensive state evaluation results and the protection setting adjustment coefficients. For example, for the instantaneous overcurrent protection (overcurrent stage I) of a transformer, its operating current setting... The dynamic adjustment formula is: in, It is a basic setting value, which is determined based on the transformer's rated current and the bus short-circuit current under the maximum operating conditions. For example, it is set to 1.5 times the transformer's rated current. It is the health index adjustment coefficient, and its mapping relationship is: when hour, This indicates that the equipment is healthy and does not require relaxed protection; when hour, This indicates a slight deterioration in the equipment's condition; to provide more sensitive protection, the setpoint will be reduced by 5%. hour, This indicates that the equipment is in poor condition. The setpoint will be reduced by 10% to clear the fault more quickly. It is the temperature adjustment coefficient, and the calculation formula is: in, This is the current winding temperature. It is the winding temperature under rated load. It is an empirical coefficient. Its physical meaning is: when the actual operating temperature of the winding is higher than the rated temperature, the protection setting is allowed to be appropriately increased (because the transformer design allows for overload for a certain period), but the maximum increase should not exceed 1.1 times the basic setting (i.e., ...). For action delay Similar adjustments can be made, for example, when the equipment is in poor condition ( When this is done, the delay can be reduced from the standard 0.1 seconds to 0.05 seconds. The adjustment strategy table, basic settings, and correlation coefficients are set by engineers during system debugging and stored in the configuration file of the edge computing unit. The setting adjustment command is generated by the protection strategy adaptive service of the edge computing unit and sent out in real time through the IEC61850MMS write service.

[0043] Understandably, this adaptive protection strategy breaks the limitations of traditional fixed protection settings, enabling the protection system to "sense" the health status and operating conditions of the protected equipment and make corresponding adjustments. When the equipment is in good condition, the protection settings remain at a normal level to ensure selectivity; when the equipment deteriorates, the protection settings and action delay are automatically reduced to improve the sensitivity and speed of protection, providing "enhanced" protection for vulnerable equipment. This not only effectively prevents the escalation of accidents due to fault current surges when equipment is faulty, but also avoids false trips caused by overly sensitive protection when the equipment is healthy. This "condition-dependent" protection concept represents a leap from "one-size-fits-all" to "personalized and precise protection," and is a significant development in relay protection technology.

[0044] Specifically, the generation of the intelligent analysis report also includes: performing correlation analysis on the fault prediction information, real-time analysis results, and pre-stored historical fault records and operation records, and outputting maintenance suggestions based on the correlation analysis results.

[0045] It should be further explained that the correlation analysis is performed within the data management container of the edge computing unit, which has a built-in time-series database (such as Influx DB) for storing historical data. When a report needs to be generated (e.g., daily scheduled generation, or real-time generation triggered by a high-level alert), the system performs the following steps: 1) Data retrieval: by current time and types of warning events For indexing, retrieve time windows in the historical database. Inside( All relevant event records (up to 7 days) can be collected, including historical warnings, protection actions, operation logs (such as fan start-up and shutdown, power setpoint changes), and the average value of key status quantities of the equipment within the same time period.

[0046] 2) Pattern Mining: Employing Apriori-based association rule algorithms or simple statistical analysis. For example, if the current event is "winding temperature prediction exceeds the limit" (… The system will collect statistics from the past. Within the data, all event sequences related to high winding temperatures were analyzed. A rule was found: {Ambient temperature > 30°C, fan output > 85% of rated output, duration > 2 hours} → {Winding temperature warning}, with a confidence level of 75% and a support level of 8%.

[0047] 3) Root Cause Inference and Recommendation Generation: Based on the mined rules and current operating conditions, and combined with the expert knowledge base (stored in the system as if-then rules), natural language-described analytical conclusions and maintenance recommendations are generated. For example, based on the above rules, the system currently detects an ambient temperature of 32°C, and the fan output has been at 90% for 3 hours. LSTM predicts that the temperature will reach 132°C in 1 hour. The correlation analysis engine will then conclude: "The current warning is highly correlated with high ambient temperature and high load operating mode (confidence level 75%)." The corresponding recommendation in the expert knowledge base is: "Recommended measures: 1. Temporarily reduce the active power setpoint of the unit to 80% of the rated output through the field-level energy management system (SCADA) for at least 2 hours; 2. It is recommended to focus on checking the operating status of the transformer cooling fan and the cleanliness of the radiator surface during the next inspection." Finally, these analytical results, correlation rules, inferred conclusions, and recommended measures are integrated into the intelligent analysis report.

[0048] Understandably, the generation process of intelligent analysis reports places discrete data points (current alerts) within a continuous historical context, revealing potential patterns and correlations through data mining. This elevates simple "status alerts" to "root cause analysis and decision support." This provides operations and maintenance personnel with more than just a signal of "what's wrong" or "what's about to happen"; it delivers a comprehensive decision report containing in-depth information such as "why it might fail," "what it's related to," "has it happened before," and "what should be done now." This significantly reduces reliance on the personal experience of operations and maintenance personnel, enhancing the scientific rigor, relevance, and efficiency of operations and maintenance response decisions. It is a crucial step in achieving intelligent and knowledge-based transformation of operations and maintenance.

[0049] Specifically, the collaborative interaction with the cloud platform includes uploading the feature data processed by the edge computing unit, the comprehensive status assessment results, and the intelligent analysis report to the cloud digital twin platform.

[0050] It should be further noted that the data uploaded to the cloud-based digital twin platform has been carefully designed to balance information content and communication overhead. The feature data consists of the input tensors of the LSTM model. The top 5 principal components after dimensionality reduction by principal component analysis (PCA) These five principal components can retain more than 90% of the variance of the original data, while reducing the amount of data uploaded each time from 12 × 300 = 3600 floating-point numbers to 5. In addition, the uploaded data packet also includes: timestamp, device ID, and health index for the current calculation period. Fault probability vector The system includes fault diagnosis results (if any) and a summary of the intelligent analysis report (in JSON format). This data is published every 5 minutes via MQTT to the topic " / twin / ingest / [Wind Farm ID] / [Equipment ID]". The cloud-based digital twin platform is deployed on servers such as Alibaba Cloud ECS, using a microservice architecture. It maintains a digital twin for each physical boost system; this twin is a high-fidelity simulation model capable of receiving feature data uploaded from the edge. and It can simulate the real-time status of equipment in the cloud by using its built-in physical models (such as transformer thermal circuit model and insulation aging model) and AI models, and can perform "if...then..." type simulations, such as simulating the temperature change and aging accumulation of equipment under different wind speeds and ambient air temperatures around the equipment (such as transformers and switchgear) in the next 24 hours.

[0051] Understandably, uploading processed feature data and high-level analysis results (rather than massive amounts of raw data) to the digital twin platform resolves the contradiction between limited edge computing resources and the need for in-depth analysis of global information in the cloud. The cloud-based digital twin constitutes a virtual mirror of the physical device, reflecting not only the device's status in real time but, more importantly, providing a secure and repeatable "sandbox" that does not affect actual production, enabling advanced analysis, strategy optimization, and fault inversion. For example, operations engineers can test the impact of different protection setting adjustment strategies on the long-term thermal stress of the device on the digital twin, thus providing an optimization basis for adaptive protection rules at the edge. This collaboration between the edge and the cloud achieves a perfect combination of "real-time agile decision-making at the edge" and "deep global optimization in the cloud."

[0052] Specifically, the collaborative interaction also includes: receiving an analysis model trained or optimized based on the feature data and global data from the cloud-based digital twin platform, in order to update the model running in the edge computing unit.

[0053] It's important to further clarify that model updates are a safe and controllable closed-loop process. In the cloud, the digital twin platform integrates a model training service. This service initiates retraining periodically (e.g., every three months) or when model performance drift is detected (e.g., a decrease in accuracy in identifying new types of faults). The training dataset D_global consists of feature data from all devices of the same model across the entire network. It consists of corresponding real-state labels (confirmed by maintenance work orders) and data augmentation samples generated by digital twin simulation. Taking the LSTM prediction model as an example, its retraining steps are as follows: (1) Load the current production version of the model from the model repository. As a pre-trained model.

[0054] (2) Using D_global Fine-tuning focuses on optimizing the parameters of the last one or a few layers to learn new data patterns while retaining the general features it has already learned.

[0055] (3) Evaluate the new model using an independent validation set. The performance is ensured to maintain at least [percentage missing], with all metrics (such as root mean square error of prediction, RMSE, and classification accuracy) no lower than [percentage missing]. Furthermore, the recall rate for all historical failure cases must reach 100%. After verification, [the following will be implemented]. The update package, along with its version number, hash value, and performance report, is then bundled into an update package. This update package is distributed via a secure channel.

[0056] Furthermore, the cloud collaboration module of the edge computing unit subscribes to the topic " / edge / model / update". Upon receiving the update package, it first verifies the signature and hash value, and then temporarily stores it in a backup storage area. The system automatically triggers the update process during preset off-peak hours (such as 2:00 AM): the new model is loaded into an independent container in memory for "shadow mode" operation, that is, real-time data is used to drive the old and new models simultaneously. The output results are compared for a period of time (such as 24 hours). After confirming that there are no errors, production traffic is redirected to the new model through hot-swapping to complete the seamless upgrade.

[0057] Understandably, the online model update mechanism is the core of this system's "self-evolution" capability. It enables the AI ​​model deployed on hundreds or thousands of wind turbines to continuously learn from the network's operational experience, constantly absorbing new fault modes and adapting to the aging characteristics of the equipment and changes in the operating environment. This mechanism effectively overcomes the drawbacks of traditional embedded AI models, which become fixed once deployed and unable to adapt to new situations. The entire wind farm cluster constitutes a "collective intelligence" network, where the operational data of each device nourishes the entire system's AI model, and each device immediately benefits from the more accurate diagnostic and protection capabilities brought about by model optimization, thereby achieving continuous improvement in overall system performance and automatic accumulation of knowledge and experience.

[0058] Specifically, the method further includes: automatically executing preset non-emergency fault handling operations or operating mode switching locally based on the comprehensive status assessment results and preset decision rules.

[0059] It should be further noted that the local automatic decision-making function is implemented by a rule engine (such as the open-source software Drools) within the edge computing unit. The rules are encoded in the form of "WHEN-THEN".

[0060] For example, an automatic handling rule for transformer overheating warnings is as follows: WHEN's overall condition assessment result is: (Health Index) AND (probability of overheating failure) AND (predicted maximum winding temperature for the next hour) ) THEN Execute Action: 1. Send a "power-limited operation" command to the wind turbine main controller via the Modbus TCP protocol to reduce the target active power setpoint to 85% of the current value.

[0061] 2. Start the standby cooling fan on the transformer body (controlled by the relay via the digital output DO point).

[0062] 3. Record the event "AUTO_ACTION_001: Due to overheating risk, automatically limit power to 85% and start the backup fan" in the local operation log, and add this log entry to the intelligent analysis report.

[0063] Another rule for frequent, mild localized discharge is: WHEN Partial discharge pulse count over three consecutive detection cycles All were in the intermediate level range [100, 500] pulses / s, and the fault diagnosis model did not identify a clear fault type.

[0064] THEN Execute Action: 1. Start the automatic dehumidification device installed in the switch cabinet and run it for 1 hour.

[0065] 2. Increase the sampling frequency of partial discharge detection from 1 time / second to 10 times / second for 10 minutes to capture more detailed discharge data.

[0066] 3. Send a "Key Concern" message to the cloud platform, requesting the cloud-based digital twin to perform deep insulation state simulation analysis. All automatically executed actions are designed to be "non-urgent" and "reversible," meaning they will not directly cause equipment downtime and most can be undone via remote commands or system self-recovery mechanisms. Before executing any automatic action, the system performs safety checks, such as ensuring that equipment operating parameters will not exceed safe ranges.

[0067] Understandably, granting limited, rule-based autonomous decision-making power to the edge is a crucial safety backup and capability extension in the event of communication interruptions or delayed responses from the control room. It elevates the system from a simple "monitoring-alarm" mechanism to a level of local autonomy encompassing "perception-analysis-decision-execution." This autonomy enables rapid responses to foreseeable and manageable non-emergency anomalies, suppressing the escalation of faults or buying time for maintenance personnel to arrive, thereby enhancing the overall resilience and automation of the system and reducing the operational burden on the remote monitoring center.

[0068] Specifically, the operating parameters include electrical quantities, non-electrical quantity status quantities, and signals acquired by image or acoustic sensors.

[0069] It should be further explained that the aforementioned multi-dimensional operating parameters constitute a comprehensive perception system, specifically including: 1. Electrical quantities: In addition to basic current and voltage, they also include zero-sequence components, negative-sequence components, harmonic content (such as 2nd-13th harmonics), power, power factor, frequency, etc., which are sampled by high-performance merging units or smart meters at a rate of at least 2kHz.

[0070] 2. Non-electrical state variables: (1) Temperature: including transformer winding hot spot temperature (fiber optic temperature measurement), iron core temperature, low-voltage / high-voltage bushing joint temperature (infrared temperature measurement patch wireless transmission), switch cabinet bus temperature (wireless temperature sensor), and ambient temperature.

[0071] (2) Vibration: The vibration of the transformer body is monitored by a triaxial accelerometer, and its fundamental frequency (100Hz) and harmonic components are analyzed. The vibration signal sampling rate is 5kHz, which is used to analyze the mechanical state of the core and windings.

[0072] (3) Partial discharge: The method of combining ultra-high frequency (UHF) sensor and ultrasonic (AE) sensor is adopted. The UHF sensor detects electrical discharge, and the AE sensor detects the sound wave signal generated by discharge or mechanical loosening. The discharge source can be roughly located by time difference positioning method.

[0073] (4) Gas monitoring (for oil-immersed transformers): An online dissolved gas analyzer (DGA) sensor is used to monitor... , , , , , The content of key gases and their gas production rate.

[0074] 3. Image signal: Deploy wide-angle high-definition cameras to take pictures of the transformer and switchgear appearance regularly (e.g., every hour). The image recognition algorithm built into the edge computing unit (e.g., YOLOv5s) will automatically identify oil level gauge readings, pressure gauge readings, whether there are traces of oil leakage, and the opening and closing status of the cabinet doors.

[0075] 4. Acoustic signals: Deploy a microphone array to collect transformer operating noise, establish a normal operating acoustic baseline through voiceprint recognition technology, and compare it with the real-time acoustic spectrum to detect mechanical anomalies (such as fan blade damage or foreign object impact).

[0076] Understandably, this fusion of multi-source heterogeneous data provides rich and comprehensive information dimensions for subsequent AI analysis. Changes in different physical quantities are often interconnected and mutually corroborating. For example, electrical anomalies (such as increased current harmonics) may occur simultaneously with mechanical vibration anomalies (such as loose iron core), both pointing to a specific fault mode; partial discharge signals and hotspot locations on infrared thermograms can corroborate each other. By fusing and analyzing this multi-source data, the accuracy and reliability of condition assessment and fault diagnosis can be significantly improved, and the false alarm rate can be reduced—something that monitoring systems relying solely on a single type of data (such as electrical quantities) cannot match. It achieves comprehensive and three-dimensional monitoring of the step-up system within wind turbine towers, from "electrical performance" to "mechanical condition" and then to "appearance and environment."

[0077] Please see Figure 2 The present invention provides another embodiment, which provides an intelligent monitoring and protection system for a step-up system within a wind turbine tower. The intelligent monitoring and protection system for a step-up system within a wind turbine tower includes: 1. Data acquisition module 100, configured to acquire multi-dimensional operating parameters of the booster system inside the wind turbine tower through multiple types of sensors.

[0078] It should be further explained that this system is a physical device with tightly integrated hardware and software. The data acquisition module 100 is a distributed sensor network, including: an electrical quantity acquisition unit (such as a 16-bit synchronous sampling ADC board of model ADIAD7606B, connected to the secondary side of the CT / PT), responsible for high-speed synchronous acquisition of electrical signals; a non-electrical quantity acquisition unit, integrating multiple industrial interfaces (such as 4-20mA analog input, RS-485, CAN bus, Zigbee wireless receiver), for connecting to transmitters of sensors such as temperature, vibration, partial discharge, and gas; and an image / acoustic acquisition unit (based on a USB camera and a USB sound card).

[0079] 2. Edge intelligent analysis module 200, which has a built-in edge computing unit and at least one artificial intelligence model, is configured to process the operating parameters to generate comprehensive status assessment results and fault prediction information, and dynamically adjust the protection strategy based on the results.

[0080] It should be further noted that all acquisition units are connected to the core edge intelligent analysis module 200 via an Ethernet switch. The hardware core of this module is the NVIDIA Jetson AGX Orin embedded AI computing platform, and its software architecture is based on Docker containers, running the following key service containers: (1) Data access and preprocessing container: responsible for receiving, synchronizing, filtering and normalizing all sensor data.

[0081] (2) LSTM prediction model inference container: Load and run the optimized TensorRT engine to perform health assessment and prediction.

[0082] (3) CNN fault diagnosis model inference container: triggered when a fault occurs, it performs waveform recognition and localization.

[0083] (4) Protection strategy adaptive calculation container: Based on the model output, new protection settings are calculated and generated in real time.

[0084] 5) Rule Engine Container: Executes local automatic decision rules.

[0085] 3. The local protection and control module 300 is connected to the edge intelligent analysis module and configured to execute the adjusted protection strategy.

[0086] It should be further explained that the local protection and control module 300 can have two forms: one is an external traditional microcomputer protection device, with the edge intelligent analysis module communicating with it through the IEC 61850 MMS / GOOSE service to modify its settings and receive its trip signal; the other is that the protection logic runs directly on the edge computing platform as a software function (compliant with IEC61131-3 or IEC61499 standards), and directly outputs the trip signal to the circuit breaker through its built-in digital output (DO) board.

[0087] 4. Data management module 400, configured to generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results and historical events.

[0088] It should be further explained that the data management module 400 consists of a built-in 1TB NVMe SSD and a time-series database (Influx DB), a relational database (SQLite), and a file system running on it. It is responsible for storing all raw data, processing results, model files, configuration files, and generated reports.

[0089] 5. Cloud Collaboration Module 500, configured to interact with the cloud platform for data and strategies, including uploading feature data and receiving updated models or optimization strategies.

[0090] It should be further explained that the Cloud Collaboration Module 500 is a background service program running on an edge computing platform. It is responsible for establishing and maintaining secure MQTT / TLS connections with the cloud-based digital twin platform via the device's 4G / 5G cellular network interface or fiber optic Ethernet interface, managing data uploads, and receiving and verifying model update packets. All modules are integrated into a 19-inch industrial chassis that meets IP54 protection standards and is suitable for wind turbine tower environments.

[0091] Understandably, this system is not a simple integration of existing protection, monitoring, and communication equipment. Instead, it uses an "edge intelligent analysis module" as the unified computing and control core, deeply integrating and reconstructing traditionally dispersed functions. It organically integrates multi-source data acquisition, AI model inference, adaptive protection calculation, local autonomous decision-making, data management, and cloud collaboration capabilities into a unified hardware platform and software framework. This highly integrated design not only reduces the number, size, and wiring complexity of equipment within the tower, but more importantly, it achieves a closed-loop flow of data, algorithms, and decisions, enabling advanced functions such as predictive maintenance, adaptive protection, and intelligent diagnostics to be implemented efficiently and reliably. This system represents the development direction of a new generation of intelligent monitoring and protection devices for wind power equipment, possessing high integration, intelligence, and evolvability, providing a solid technical foundation for less-manned and efficient operation and maintenance of wind farms.

[0092] In a preferred embodiment, this application also provides an electronic device, the electronic device comprising: The device includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, implement the intelligent monitoring and protection method for the booster system within the wind turbine tower. 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.

[0093] 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.

[0094] 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.

[0095] 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.

[0096] 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 method for intelligent monitoring and protection of a step-up system within a wind turbine tower, characterized in that, include: S100: Collects multi-dimensional operating parameters of the booster system deployed in the wind turbine tower through multiple types of sensors; S200: Based on the edge computing unit, the operating parameters are processed and analyzed in real time to generate a comprehensive status assessment result and fault prediction information for the boost system; S300. Based on the comprehensive status assessment results, dynamically adjust the protection strategy parameters of the boost system; S400: Generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results, and historical events; S500, to interact collaboratively with the cloud platform, the collaborative interaction including uploading feature data and receiving updated analysis models or optimization strategies issued by the cloud platform.

2. The method according to claim 1, characterized in that, The real-time processing and analysis of operating parameters includes: The first artificial intelligence model is used to analyze the time-series operating parameters in order to predict the health status trend and early failure risk of key components in the boost system.

3. The method according to claim 2, characterized in that, The real-time processing and analysis of operating parameters also includes: The second artificial intelligence model is used to identify fault waveform data in the operating parameters in order to determine the fault type and estimate the fault location.

4. The method according to claim 3, characterized in that, The specific steps for dynamically adjusting the protection strategy parameters of the boost system are as follows: Based on the comprehensive status assessment results, the action threshold or action delay of the electrical protection components are adaptively adjusted.

5. The method according to claim 1, characterized in that, The generation of the intelligent analysis report also includes: The fault prediction information and real-time analysis results are correlated with the pre-stored historical fault records and operation records, and maintenance suggestions are output based on the correlation analysis results.

6. The method according to claim 1, characterized in that, The collaborative interaction with the cloud platform includes: The feature data processed by the edge computing unit, the comprehensive status assessment results, and the intelligent analysis report are uploaded to the cloud-based digital twin platform.

7. The method according to claim 6, characterized in that, The collaborative interaction also includes: The system receives an analysis model, trained or optimized based on the feature data and global data, from the cloud-based digital twin platform to update the model running in the edge computing unit.

8. The method according to claim 1, characterized in that, The method further includes: Based on the comprehensive status assessment results and preset decision rules, preset non-emergency fault handling operations or operation mode switching are automatically executed locally.

9. The method according to any one of claims 1-8, characterized in that, The operating parameters include electrical quantities, non-electrical quantity status quantities, and signals acquired by image or acoustic sensors.

10. An intelligent monitoring and protection system for a step-up voltage system within a wind turbine tower, characterized in that, The method according to any one of claims 1-9 comprises: The data acquisition module is configured to collect multi-dimensional operating parameters of the booster system inside the wind turbine tower through multiple types of sensors; The edge intelligent analysis module has a built-in edge computing unit and at least one artificial intelligence model, and is configured to process the operating parameters to generate comprehensive status assessment results and fault prediction information, and dynamically adjust the protection strategy based on the results. A local protection and control module is connected to the edge intelligent analysis module and configured to execute the adjusted protection strategy. The data management module is configured to generate an intelligent analysis report that integrates the fault prediction information, real-time analysis results, and historical events. The cloud collaboration module is configured to interact with the cloud platform for data and strategies, including uploading feature data and receiving updated models or optimization strategies.

Images