A wind farm equipment fault intelligent diagnosis method and system

Abstract

The application discloses a kind of wind farm equipment fault intelligent diagnosis method and system, the multidimensional sensor data of wind farm equipment is collected;These data are preliminarily screened using edge analysis algorithm, determine the data that need further diagnosis;The screened data is input into the pre-constructed fault diagnosis model, in the pre-constructed fault diagnosis model, data first pass through multimodal deep learning model fusion, then use transfer learning and domain adaptation and multi-agent reinforcement learning framework to determine the area where fault occurs;At the same time, the causal reasoning engine and the neural symbolic reasoning module enhanced by knowledge graph are responsible for in-depth processing and diagnosing faults.Finally, the determined fault area and diagnosis result are input into the dynamic system health index calculation module to generate the final fault diagnosis result.This method significantly improves the diagnosis efficiency and accuracy of wind farm equipment fault, and provides a strong guarantee for the stable operation of wind farm.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology and relates to an intelligent diagnosis method and system for wind farm equipment faults. Background Technology

[0002] In the field of renewable energy, wind power, as a clean and renewable energy source, is becoming increasingly important. With the rapid development of wind power technology and the continuous expansion of wind farm scale, ensuring the efficient and stable operation of wind farms has become an urgent problem to be solved. The complexity of wind farm equipment, especially wind turbine generators, and the diversity of their operating environments place higher demands on fault diagnosis technology.

[0003] Traditionally, wind farm equipment fault diagnosis has relied primarily on manual inspections and simple threshold-based methods using data from single-type sensors. This approach is not only inefficient but also ill-suited to the increasingly complex fault modes and variable operating environments of modern wind power equipment. The limitation of single-sensor data is that it only provides information on one aspect of the equipment's status and cannot comprehensively reflect its overall health. Furthermore, simple threshold-based methods often fail to accurately detect early signs of equipment failure, leading to delayed fault detection and increased downtime and maintenance costs. Summary of the Invention

[0004] The purpose of this invention is to solve the technical problems of traditional fault diagnosis methods in the prior art, such as limited data, low efficiency, and inability to capture equipment faults in real time and accurately, which makes it difficult to meet the needs of modern wind farms. The invention provides an intelligent fault diagnosis method and system for wind farm equipment.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] The first aspect of this invention provides an intelligent fault diagnosis method for wind farm equipment, comprising the following steps:

[0007] Collect multi-dimensional sensor data from wind farm equipment;

[0008] The multi-dimensional sensor data is preliminarily analyzed using an edge analysis algorithm to determine the multi-dimensional sensor data that requires further diagnosis.

[0009] Input the multi-dimensional sensor data that requires further diagnosis into the pre-built fault diagnosis model to obtain the fault diagnosis results;

[0010] The pre-built fault diagnosis model includes a multimodal deep learning model, a transfer learning and domain adaptation module, a multi-agent reinforcement learning framework, a causal reasoning engine, a knowledge graph-enhanced neural symbolic reasoning module, and a dynamic system health index calculation module. The multi-dimensional sensor data requiring further diagnosis is fused through the multimodal deep learning model and then the region is determined through transfer learning and domain adaptation and the multi-agent reinforcement learning framework. At the same time, the fault is diagnosed through the causal reasoning engine and the knowledge graph-enhanced neural symbolic reasoning module. The determined region and the diagnosed fault are input into the dynamic system health index calculation module to obtain the fault diagnosis result.

[0011] Furthermore, the preliminary analysis of the multi-dimensional sensor data using an edge analysis algorithm to determine the multi-dimensional sensor data requiring further diagnosis specifically involves:

[0012] Preprocess multi-dimensional sensor data;

[0013] An edge analysis algorithm is used to detect outliers in the preprocessed multi-dimensional sensor data; these outliers are the multi-dimensional sensor data that require further diagnosis.

[0014] Furthermore, the multimodal deep learning model includes a spatiotemporal attention mechanism and a multi-scale causal convolutional network.

[0015] Furthermore, the transfer learning and domain adaptation module employs a dynamic weight alignment method and an adversarial domain adaptation method.

[0016] Furthermore, the multi-agent reinforcement learning framework employs a multi-agent soft actor-commentator algorithm.

[0017] Furthermore, the causal reasoning engine employs an improved PC algorithm based on Bayesian networks, combined with a method based on prior probabilities derived from expert knowledge.

[0018] Furthermore, the knowledge graph-enhanced neural symbol reasoning module includes a knowledge graph embedding module and a neural symbol reasoning module; the knowledge graph embedding module adopts the TransE model; and the neural symbol reasoning module adopts a method combining symbolic rules and neural networks.

[0019] A second aspect of the present invention provides an intelligent fault diagnosis system for wind farm equipment, comprising:

[0020] The data acquisition layer is used to collect multi-dimensional sensor data from wind farm equipment.

[0021] An edge computing layer is used to perform preliminary analysis of the multi-dimensional sensor data using an edge analysis algorithm to determine the multi-dimensional sensor data that needs further diagnosis.

[0022] The cloud-based analytics layer is used to input multi-dimensional sensor data that requires further diagnosis into a pre-built fault diagnosis model to obtain fault diagnosis results.

[0023] Furthermore, the intelligent fault diagnosis system for wind farm equipment also includes a network transmission layer, which is used to realize data transmission between the edge computing layer and the cloud analysis layer.

[0024] Furthermore, the intelligent fault diagnosis system for wind farm equipment also includes an application interface layer, which receives data from the cloud analysis layer and is used to provide users with diagnostic results and interactive interfaces.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] This invention discloses an intelligent fault diagnosis method for wind farm equipment. By collecting multi-dimensional sensor data from wind farm equipment, this method can comprehensively capture the operating status information of the equipment, including but not limited to key parameters such as vibration, temperature, current, and voltage, providing a rich and accurate data foundation for subsequent fault diagnosis. An edge analysis algorithm is used to perform preliminary analysis on massive amounts of sensor data, quickly identifying abnormal or potential fault data points, effectively reducing the burden of data transmission and processing, and improving the response speed and accuracy of the diagnostic system. A multimodal deep learning model is used to deeply fuse multi-dimensional data from different sensors, uncovering potential correlations and features between the data, thus improving the accuracy and robustness of fault diagnosis. Simultaneously, by combining transfer learning and domain adaptation modules, historical data and cross-domain knowledge can be effectively utilized to quickly adapt to the characteristics of different wind farms and equipment, further enhancing the generalization ability of the diagnostic model. The introduction of a multi-agent reinforcement learning framework simulates the collaborative decision-making process of multiple agents in complex environments, enabling more precise location of fault locations. Continuous strategy optimization further improves the efficiency and accuracy of fault location. Combined with a causal reasoning engine and a knowledge graph-enhanced neural symbolic reasoning module, not only can faults be accurately classified and identified, but the root causes and propagation paths of faults can also be revealed, providing a scientific basis for fault prevention and maintenance. The introduction of knowledge graphs further enriches the diagnostic system's knowledge base, improving the system's intelligence level and self-learning ability. Through a dynamic system health index calculation module, the health status of wind farm equipment is assessed and predicted in real time, providing operation and maintenance personnel with intuitive health indicators and early warning information. This helps to identify potential faults in advance and take corresponding maintenance measures, reducing the impact of equipment failures on wind farm operations.

[0027] Furthermore, edge analytics algorithms perform initial screening at the data source, ensuring that only sensor data flagged as abnormal is sent to the central server or cloud platform for further diagnosis. This approach significantly reduces the amount of data transmitted, lowers network bandwidth consumption, and accelerates fault response. Because edge analytics algorithms run directly on the edge devices that generate the data (such as wind turbine control cabinets and sensor nodes), they can process sensor data and provide feedback immediately. This real-time feedback mechanism enables the system to respond more quickly to changes in equipment status, improving the timeliness of fault diagnosis. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a block diagram of the intelligent fault diagnosis method for wind farm equipment of the present invention;

[0030] Figure 2 This is a block diagram of the intelligent fault diagnosis system for wind farm equipment of the present invention;

[0031] Figure 3 This is a system architecture diagram of an embodiment of the intelligent fault diagnosis system for wind farm equipment of the present invention.

[0032] Among them: 201-Data Acquisition Layer; 202-Edge Computing Layer; 203-Cloud Analysis Layer. Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and marked in the accompanying drawings can generally be arranged and designed in various different configurations.

[0034] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0035] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0036] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0037] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0038] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0039] The present invention will now be described in further detail with reference to the accompanying drawings:

[0040] See Figure 1 This invention provides an intelligent fault diagnosis method for wind farm equipment, comprising the following steps:

[0041] S1 collects multi-dimensional sensor data from wind farm equipment;

[0042] High-precision sensors are installed in key components of each wind turbine (such as the generator, gearbox, and bearings). These sensors can monitor and collect multi-dimensional data such as temperature, vibration, speed, and current in real time. The collected data is transmitted to the wind farm's data center or edge computing nodes via wired or wireless means.

[0043] S2, The multi-dimensional sensor data is preliminarily analyzed using an edge analysis algorithm to determine the multi-dimensional sensor data that requires further diagnosis;

[0044] Edge analytics algorithms are deployed on edge computing nodes to preprocess the received multi-dimensional sensor data, including data cleaning, noise reduction, and normalization.

[0045] Edge analysis algorithms use statistical or machine learning-based methods (such as threshold detection, isolated forests, etc.) to perform real-time analysis on preprocessed data to detect outliers or abnormal patterns.

[0046] The detected abnormal data (i.e., multi-dimensional sensor data that requires further diagnosis) is marked and prepared to be sent to the central server for in-depth analysis.

[0047] S3, input the multi-dimensional sensor data that needs further diagnosis into the pre-built fault diagnosis model to obtain the fault diagnosis result; the pre-built fault diagnosis model includes a multimodal deep learning model, a transfer learning and domain adaptation module, a multi-agent reinforcement learning framework, a causal reasoning engine, a knowledge graph-enhanced neural symbol reasoning module, and a dynamic system health index calculation module.

[0048] On the central server, a comprehensive fault diagnosis model is pre-built and trained. This model integrates a multimodal deep learning model, a transfer learning and domain adaptation module, a multi-agent reinforcement learning framework, a causal reasoning engine, a knowledge graph-enhanced neural symbolic reasoning module, and a dynamic system health index calculation module.

[0049] The abnormal data transmitted from the edge computing nodes is input into the model. The model first uses a multimodal deep learning model to extract and fuse features from the data, capturing the correlation and complementarity information between different sensors.

[0050] Next, the transfer learning and domain adaptation module adjusts the model's parameters and decision boundaries based on historical fault data and the current equipment status to adapt to changes brought about by different operating conditions and equipment aging.

[0051] Multi-agent reinforcement learning frameworks simulate the collaboration and competition between multiple agents, optimizing fault diagnosis strategies through continuous trial and error and learning, thereby improving the accuracy and efficiency of diagnosis.

[0052] The causal reasoning engine and the knowledge graph-enhanced neural symbolic reasoning module combine domain knowledge and expert experience to conduct in-depth analysis and reasoning of the causes of failures, and construct failure chains and failure trees.

[0053] Finally, the dynamic system health index calculation module calculates the health index of the equipment based on the model's diagnostic results and the equipment's real-time status, providing maintenance personnel with an intuitive basis for equipment health assessment.

[0054] One embodiment of the present invention provides an intelligent fault diagnosis method for wind farm equipment, comprising the following steps:

[0055] Collect multi-dimensional sensor data from wind farm equipment; install high-precision sensors on various key parts of the wind farm equipment to collect multi-dimensional data in real time, and transmit the data to edge computing nodes via wired or wireless means.

[0056] The multi-dimensional sensor data is preliminarily analyzed using an edge analysis algorithm to identify multi-dimensional sensor data that requires further diagnosis. After receiving the data, the edge computing node first performs data preprocessing, including noise reduction, compression, and normalization. Then, it applies an edge analysis algorithm (such as statistical anomaly detection, machine learning models, etc.) to perform preliminary analysis on the data and identify abnormal data. The abnormal data is then marked and transmitted to the central server.

[0057] The multi-dimensional sensor data requiring further diagnosis is input into a pre-built fault diagnosis model to obtain the fault diagnosis results; the pre-built fault diagnosis model includes:

[0058] Multimodal deep learning model: It uses spatiotemporal attention mechanism to capture temporal and spatial features in data, and combines multi-scale causal convolutional network to extract fault features at different scales.

[0059] Transfer learning and domain adaptation module: Employs dynamic weight alignment and adversarial domain adaptation methods to enable the model to adapt to changes in data distribution caused by different operating conditions and equipment aging.

[0060] Multi-agent reinforcement learning framework: Through the multi-agent soft actor-commentator algorithm, the cooperation and competition between multiple agents are simulated to optimize the fault diagnosis strategy.

[0061] Causal reasoning engine: Based on Bayesian networks and using an improved PC algorithm, combined with prior probabilities from expert knowledge, it constructs fault chains and fault trees to deeply analyze the causes of faults.

[0062] Knowledge graph-enhanced neural symbolic reasoning module:

[0063] Knowledge Graph Embedding Module: The TransE model is used to embed entities and relationships in the knowledge graph into a low-dimensional vector space.

[0064] Neural Symbolic Reasoning Module: Combining symbolic rules and neural networks, this module enhances the model's reasoning ability by utilizing structured information from knowledge graphs.

[0065] Dynamic System Health Index Calculation Module: Based on diagnostic results and real-time data, calculates the health index of the equipment, providing maintenance personnel with an intuitive assessment of equipment health.

[0066] See Figure 2 One embodiment of the present invention provides an intelligent fault diagnosis system for wind farm equipment, comprising:

[0067] The data acquisition layer is used to collect multi-dimensional sensor data from wind farm equipment.

[0068] An edge computing layer is used to perform preliminary analysis of the multi-dimensional sensor data using an edge analysis algorithm to determine the multi-dimensional sensor data that needs further diagnosis.

[0069] The cloud-based analytics layer is used to input multi-dimensional sensor data that requires further diagnosis into a pre-built fault diagnosis model to obtain fault diagnosis results.

[0070] See Figure 3 One embodiment of the present invention provides an intelligent fault diagnosis system for wind farm equipment, comprising:

[0071] The data acquisition layer, edge computing layer, network transmission layer, cloud analytics layer, and application interface layer are all included.

[0072] Data Acquisition Layer: Used to collect multi-dimensional sensor data from wind farm equipment. The data acquisition layer is the foundation of the system's perception, responsible for comprehensively and accurately collecting operational data from the wind power equipment. It mainly includes the following components:

[0073] Multi-dimensional sensor networks, including:

[0074] Nano-MEMS sensor array: Deployed in key components such as blades and hubs to collect data on microstructural changes. Employing high-sensitivity triaxial accelerometers and strain sensors, with dimensions of approximately 50μm x 50μm x 10μm and a sensitivity of 0.1μg / √Hz.

[0075] Fiber Bragg grating (FBG) sensor network: Deployed along the wind turbine tower and blades to achieve distributed strain and temperature monitoring. FBG sensors with a wavelength range of 1510nm-1590nm are used, with a temperature sensitivity of approximately 10 pm / ℃ and a strain sensitivity of approximately 1.2 pm / με.

[0076] Acoustic sensor array: Installed around the cabin to capture unusual sound signals. Employs a MEMS microphone array with a frequency response range of 20Hz-20kHz and a signal-to-noise ratio greater than 65dB.

[0077] High-precision vibration sensor: Installed in gearboxes, generators, and other components to monitor vibration conditions. Utilizes a piezoelectric accelerometer with a frequency range of 0.5Hz-10kHz and a sensitivity of 100mV / g.

[0078] Oil monitoring sensor: Used for real-time analysis of gearbox oil condition. Employs a capacitive particle counter, capable of detecting particles in the range of 4μm-100μm.

[0079] An adaptive data acquisition system is used to acquire data from multi-dimensional sensor networks. Dynamic sampling rate adjustment: The sampling frequency is automatically adjusted according to device status and environmental conditions, ranging from 1kHz to 100kHz. Data compression: A real-time data compression algorithm based on compression sensing is used, achieving a compression ratio of up to 10:1. Local caching: Each acquisition node is equipped with a 128GB high-speed SSD, supporting cyclic storage of 72 hours of raw data.

[0080] The data preprocessing unit improves data quality through signal amplification and filtering. Sensor calibration is performed automatically weekly to ensure data accuracy. Simple statistical methods are used for preliminary anomaly detection, marking potentially problematic data.

[0081] The edge computing layer is used for preprocessing and preliminary analysis of the multi-dimensional sensor data. This layer performs preliminary processing and analysis at the data source, reducing the burden on the cloud and improving system real-time performance. It mainly includes:

[0082] The edge computing hardware platform uses an Intel Xeon D-2183IT processor (16 cores, 32 threads); an NVIDIA Tesla T4 GPU (16GB GDDR6) for AI inference acceleration; and a Xilinx Artix-7 XC7A200T FPGA for programmable hardware acceleration. It has 128GB of DDR4-3200 memory and a 2TB NVMe SSD for storage.

[0083] The edge intelligence software stack includes the real-time operating system RT-Linux with the PREEMPT_RT patch. The containerized environment uses lightweight Kubernetes based on K3s. The time-series database is InfluxDB Edge. The stream processing engine is Apache Flink. The edge AI framework uses NVIDIA Triton Inference Server.

[0084] Edge analysis algorithms include the following steps:

[0085] a) Signal processing:

[0086] Adaptive filtering: an adaptive filter based on the minimum mean square error (LMS) algorithm.

[0087] Time-frequency analysis: Short-time Fourier Transform (STFT) and Wavelet Packet Decomposition (WPD).

[0088] b) Feature extraction: Calculate statistical features, spectral features, and time-frequency features.

[0089] c) Lightweight machine learning models:

[0090] Anomaly detection: Local outlier factor (LOF) and isolated forest algorithm are used.

[0091] Fault classification: using quantized decision tree and support vector machine models.

[0092] The network transport layer is used to enable data transmission between the edge computing layer and the cloud analytics layer; it ensures reliable, secure, and efficient data transmission between these two layers. Key features include:

[0093] The primary communication link uses 5G NR technology, employing the 3.5GHz (n78) band, 100MHz bandwidth, and a 64T64R MIMO configuration. The backup link uses low-Earth orbit satellite communication (such as SpaceX Starlink), with an average latency of 20-40ms.

[0094] Network security mechanisms include: data encryption using the TLS 1.3 protocol, supporting post-quantum cryptography (PQC) algorithms; authentication based on two-way authentication using X.509 certificates; intrusion detection deployed via Suricata IDPS for real-time network traffic monitoring; software-defined networking (SDN) technology for flexible routing policies; machine learning algorithms to predict network congestion and dynamically adjust routes; adaptive data compression, dynamically adjusting the Zstandard compression algorithm's compression level based on network bandwidth; and a local caching mechanism using Redis as a high-speed cache to temporarily store data during network instability.

[0095] The cloud analytics layer is used for in-depth analysis and fault diagnosis of transmitted data; it is the core of the system, responsible for complex data analysis and decision support. It mainly includes the following modules:

[0096] (1) Multimodal deep learning models, which mainly include:

[0097] a) Spatiotemporal attention mechanism: used to capture the spatiotemporal dependencies between various components of the wind turbine. The core formula is as follows:

[0098]

[0099] in, It is the final spatiotemporal fusion feature. It is an adaptive parameter. and These are temporal and spatial attention weights, and These are temporal and spatial feature sequences, respectively.

[0100] b) Multi-scale causal convolutional networks: Using causal convolutional layers with different dilation rates to capture failure modes at different time scales.

[0101] (2) Transfer Learning and Domain Adaptation Module:

[0102] Dynamic weight alignment: Minimize the difference in weights between the source and target domain models.

[0103] Adversarial domain adaptation: Training a domain discriminator to reduce the difference in feature distribution between the source and target domains.

[0104] (3) Multi-agent reinforcement learning framework:

[0105] The Multi-Agent Soft Actor-Commentator (MASAC) algorithm is used to model each wind turbine as an agent to optimize the diagnosis and control strategies for the entire wind farm.

[0106] (4) Causal Reasoning Engine:

[0107] Based on Bayesian networks, the interpretability of fault diagnosis is improved. An improved PC algorithm is used for structure learning, incorporating expert knowledge as prior knowledge.

[0108] (5) Knowledge graph-enhanced neural symbolic reasoning module:

[0109] This approach combines knowledge graph embedding from the TransE model with symbolic reasoning from neural networks. The core formula is as follows:

[0110]

[0111] in, This is the output result. It is an activation function. It is a weight matrix. It is entity embedding. This represents the symbolic reasoning results based on the knowledge graph $KG$. It is a bias term.

[0112] (6) Dynamic System Health Index Calculation Module:

[0113] It provides indicators for a comprehensive assessment of the wind turbine's health status. The calculation formula is as follows:

[0114]

[0115] in, It is time Dynamic system health index, It refers to the number of health indicators. It is the first The dynamic weights of each indicator It is a normalization function. It is the first A health indicator over time The value of .

[0116] The Application Programming Interface (API) layer provides diagnostic results and an interactive interface to users; it offers user-friendly and intuitive interfaces for different types of users. It mainly includes:

[0117] (1) Web Application: Front-end framework: React 18 with Next.js. State management: ReduxToolkit. UI component library: Material-UI. Data visualization: D3.js and ECharts.

[0118] (2) Mobile application: Development framework: Flutter 3.0. State management: Riverpod. Local storage: Hive database.

[0119] (3) API Gateway: The core is a service mesh based on Envoy. Authentication uses OAuth 2.0 + OpenIDConnect. Rate limiting is implemented based on an adaptive rate limiting algorithm using a token bucket. Documentation uses the OpenAPI 3.0 specification and is displayed through Swagger UI.

[0120] (4) Mixed Reality Interaction System: The AR device uses Microsoft HoloLens 2. The development framework uses Unity + Mixed Reality Toolkit (MRTK). 3D Modeling: A high-precision wind turbine model is created using Autodesk Inventor. Real-time Data Streaming: Real-time sensor data is transmitted via the WebSocket protocol. Interaction Method:

[0121] Voice commands: Natural language understanding based on Azure Speech Services.

[0122] Gesture control: Uses the built-in gesture recognition provided by MRTK.

[0123] Virtual Touch: A 3D UI interaction based on ray casting.

[0124] The interaction and connection relationships between the various layers of the system in this embodiment are as follows: The communication method between the data acquisition layer and the edge computing layer adopts the EtherCAT real-time industrial Ethernet protocol; sensor data is transmitted to the edge computing node in real time; the edge computing layer can send control commands such as sampling rate adjustment to the data acquisition layer. The edge computing layer and the network transmission layer use the MQTT protocol for data transmission and implement TLS encryption and two-way authentication, while performing adaptive data compression before transmission. The main link between the network transmission layer and the cloud analysis layer uses a 5G network to transmit data; the backup link automatically switches to satellite communication when the main link fails; the communication mode adopts a publish-subscribe model to ensure the real-time performance and reliability of the data. The cloud analysis layer and the application interface layer provide RESTful API and GraphQL interfaces; the data format uses JSON for data exchange; for data that needs to be updated in real time, the WebSocket protocol is used. The overall interaction flow is as follows: the collected data is analyzed by the edge computing layer through the data acquisition layer, transmitted by the network transmission layer to the cloud analysis layer for further detection and analysis, and then transmitted to the application interface layer for display to the user. New models trained in the cloud are periodically distributed to the edge computing layer.

[0125] The system workflow in this embodiment is as follows: First, the data acquisition layer continuously collects wind turbine operation data through a multi-dimensional sensor network. Next, the edge computing layer preprocesses and performs preliminary analysis on the raw data. The processed data is then sent to the cloud via the network transmission layer. The cloud analysis layer performs in-depth analysis and fault diagnosis. The cloud analysis layer is the core of the system. It utilizes various advanced technologies, including a multimodal deep learning model, transfer learning and domain adaptation modules, a multi-agent reinforcement learning framework, a causal reasoning engine, a knowledge graph-enhanced neural symbolic reasoning module, and a dynamic system health index calculation module, to perform in-depth analysis and fault diagnosis on the data from the edge computing layer. The multimodal deep learning model can process multi-source heterogeneous data from different sensors and different types, extracting deep-level feature representations. The transfer learning and domain adaptation module enables knowledge transfer across wind farms and equipment; by utilizing existing fault diagnosis knowledge from wind farms or equipment, it quickly adapts to the diagnostic needs of new wind farms or new equipment, reducing reliance on new data and training time. A multi-agent reinforcement learning framework optimizes the diagnostic strategy for the entire wind farm; it simulates the collaborative work of multiple agents in the wind farm, finding the optimal diagnostic strategy through continuous trial and error and optimization, thereby improving the overall operation and maintenance efficiency and economic benefits of the wind farm. A causal reasoning engine provides interpretable diagnostic results; by hierarchically associating fault causes, it provides interpretable diagnostic results, helping maintenance personnel understand the root causes and propagation paths of faults, providing strong support for subsequent fault repair and prevention. A knowledge graph-enhanced neural symbolic reasoning module integrates domain knowledge and data-driven methods; by constructing a knowledge graph in the wind turbine domain, it enhances the system's reasoning ability and interpretability. A dynamic system health index calculation module dynamically assesses the overall health status of wind turbines based on real-time and historical data, providing timely and accurate references for operation and maintenance decisions. Diagnostic results are presented to users through the application interface layer. Diagnostic results processed by the cloud analysis layer are presented to users in an intuitive and user-friendly manner through the application interface layer. These results may include fault type, fault location, fault severity, maintenance recommendations, and health status assessment reports. Users can access this information through web interfaces, mobile applications, or API interfaces, and perform further operations or decisions as needed. In addition, the system also provides functions such as historical data query and trend analysis to help users fully understand the operating status and performance changes of the wind turbine.

[0126] This invention's cloud-based analysis layer system employs an innovative multimodal deep learning model, effectively fusing heterogeneous data from different sensors. The multimodal deep learning model incorporates a spatiotemporal attention mechanism and a multi-scale causal convolutional network; the spatiotemporal attention mechanism adaptively captures the complex spatiotemporal dependencies between various components of the wind turbine. The specific implementation is as follows:

[0127] Temporal attention: Attention weights in the temporal dimension are computed using a bidirectional long short-term memory network (BiLSTM).

[0128] Spatial attention: Graph attention network (GAT) is used to capture spatial dependencies between components.

[0129] Fusion strategy: Dynamically adjust the importance of temporal and spatial attention through gating mechanisms.

[0130] Multi-scale causal convolutional networks can handle fault modes at different time scales, including:

[0131] Multi-scale receptive field: Achieve multi-scale receptive field using dilated convolutions with different dilation rates.

[0132] Causal convolution: Ensures the temporal causality of the model and avoids information leakage.

[0133] Residual connections: Use residual connections and layer normalization to improve training stability.

[0134] To address the knowledge transfer challenges between different wind farms and wind turbine models, the system employs transfer learning and domain adaptation techniques. Specifically, it utilizes a dynamic weight alignment method to design a learnable mapping matrix, minimizing the difference in model weights between the source and target domains. Regularization techniques, such as L2 regularization and dropout, are used to prevent overfitting. An adversarial domain adaptation method is employed: a domain discriminator is introduced, and adversarial training reduces the difference in feature distributions between the source and target domains. Adversarial training is implemented using a gradient reversal layer.

[0135] To further improve the accuracy of fault source domain identification in wind farms, this invention employs a multi-agent reinforcement learning framework. This framework utilizes the Multi-Agent Soft Actor-Commentator (MASAC) algorithm to model the entire wind farm as a multi-agent system. Its state space includes local observations and state information from neighboring wind turbines. The action space includes continuous actions such as diagnostic threshold adjustment and feature selection. The reward function comprehensively considers diagnostic accuracy, false positive rate, false negative rate, and maintenance cost. The policy network employs a 3-layer fully connected network with hidden layer dimensions of [512, 256]. The value network also employs a 3-layer fully connected network with hidden layer dimensions of [512, 256]. The optimization algorithm uses the Adam optimizer with a learning rate of 1e-4 and a batch size of 256.

[0136] To improve the interpretability of fault diagnosis, a causal inference engine method is used to infer causal relationships from one fault. This invention improves the interpretability of fault diagnosis with a Bayesian network-based causal inference engine, specifically as follows:

[0137] Structure learning: An improved PC algorithm is adopted, and expert knowledge is introduced as prior knowledge.

[0138] Parameter learning: The conditional probability table is estimated using the Expectation-Maximization (EM) algorithm.

[0139] Inference algorithm: Combining variable elimination and importance sampling improves inference efficiency.

[0140] Causal effect estimation: The mean causal effect (ACE) is calculated using do-calculus.

[0141] To further improve the interpretability of fault diagnosis, this invention employs a knowledge graph-enhanced neural symbolic reasoning module, which integrates the advantages of symbolic reasoning and neural networks, specifically as follows:

[0142] Knowledge graph construction: Define domain ontology using OWL 2 Web Ontology Language.

[0143] Graph storage: RDF triples are stored using the Neo4j graph database.

[0144] Knowledge graph embedding: Using the TransE model to map entities and relations to a low-dimensional vector space.

[0145] Neural Symbolic Reasoning: Designing an end-to-end differentiable reasoning network that combines symbolic rules and neural networks.

[0146] The Dynamic System Health Index (DSHI) of this invention provides a comprehensive assessment of the health status of wind turbines:

[0147] Multidimensional health indicators include vibration, temperature, and oil condition, which comprehensively reflect the system's operating status. A min-max normalization method is used to unify indicator data of different dimensions and magnitudes onto the same scale, facilitating subsequent data processing and comparison. An attention mechanism is used to dynamically adjust the importance of each indicator. This method automatically adjusts the weight of each health indicator in the comprehensive evaluation based on the system's current state and operating environment, thereby improving the accuracy and robustness of the assessment. Time series analysis methods such as LSTM (Long Short-Term Memory) are employed to capture the temporal evolution trend of the system's health status. LSTM networks can handle long-term dependencies and have good modeling capabilities for health data with time-series characteristics.

[0148] In its implementation, the diagnostic system of this invention achieved the following performance indicators through innovative technical solutions:

[0149] Fault prediction lead time: An average fault prediction lead time of 7-14 days provides maintenance personnel with ample time to troubleshoot and prepare for repairs.

[0150] Diagnostic accuracy: For known fault modes, the diagnostic accuracy exceeds 95%, ensuring the accuracy and reliability of fault identification;

[0151] False alarm rate: less than 1%;

[0152] Missed Detection Rate: The missed detection rate is less than 0.5%;

[0153] System response time: The response time at the edge is less than 100ms, and the response time for complex analysis in the cloud is less than 1s, ensuring that the system can quickly respond to various operation and maintenance needs;

[0154] Data compression rate: The data compression rate exceeds 90%, effectively reducing the cost of data transmission and storage;

[0155] System availability: The system availability reaches 99.999% (five nines), ensuring that the system can operate continuously and stably;

[0156] Security: Complies with industrial control system security standards such as IEC 62443, ensuring the system's information security and data security.

[0157] The deployment of the intelligent fault diagnosis system for wind farm equipment in this invention adopts an automated deployment script based on Ansible; it uses GitOps workflow and combines ArgoCD to achieve continuous deployment.

[0158] To ensure the stable operation of the intelligent wind power operation and maintenance system and to promptly identify and address potential problems, this invention employs Prometheus + Grafana to build a full-stack monitoring system and sets multi-level alarm thresholds. The system deploys the Prometheus service to collect monitoring data such as system operating status and performance metrics; configures the monitoring targets of Prometheus, including key system components and service interfaces; installs and configures Grafana as a visualization platform for monitoring data; and creates dashboards to display key system metrics and operating status, such as CPU utilization, memory usage, and response time. Alarm rules are defined in Prometheus, and multi-level alarm thresholds are set, such as warnings and critical alarms. Alarm notification channels are configured, including SMS, email, and mobile application push notifications.

[0159] To maintain the advanced nature and stability of the intelligent wind power operation and maintenance system, this invention adopts a blue-green deployment and canary release strategy for system updates.

[0160] Blue-green deployment:

[0161] Prepare a new environment (blue environment) that is completely isolated from the current operating system in the production environment.

[0162] Deploy the new version to the Blue Environment and conduct comprehensive functional and performance testing.

[0163] After the test is passed, the traffic will be switched to the blue environment to achieve zero downtime updates.

[0164] Canary Release:

[0165] Select a small subset of users or devices in the blue environment as canary users (or devices).

[0166] We will gradually switch canary users to the new version, observe its operation, and collect feedback.

[0167] Adjust the new version or optimize the system configuration based on feedback to ensure that the new version is stable and reliable before full rollout.

[0168] To ensure the high availability and data security of the intelligent wind power operation and maintenance system, this invention implements a multi-site active-active architecture and conducts regular disaster recovery drills:

[0169] Implement a geographically distributed active-active architecture to ensure high system availability: deploy system replicas in data centers in different geographical locations to form a geographically distributed active-active architecture; configure a data synchronization mechanism to ensure data consistency and real-time performance between data centers; set up load balancing and failover strategies to ensure automatic switchover to other data centers when a single data center fails.

[0170] Conduct regular disaster recovery drills to verify the effectiveness of backup and recovery processes: Develop detailed disaster recovery plans, including recovery processes, required resources, and expected timelines; conduct regular disaster recovery drills to simulate real disaster scenarios and verify the effectiveness of backup and recovery processes; record problems and areas for improvement during the drills and continuously optimize the disaster recovery plan.

[0171] Backup strategy optimization: Develop appropriate backup strategies based on system characteristics and business needs; regularly back up critical system data and store it in a safe and reliable location; verify the integrity and recoverability of backup data to ensure that the system can be quickly restored when needed.

[0172] This invention presents a wind farm equipment fault intelligent diagnosis system with a flexible architecture, reserving expansion space for future technological development and business needs: It includes interfaces for integration with other renewable energy systems such as photovoltaics and energy storage; a blockchain interface to prepare for future energy trading and carbon credit management; the potential of future quantum computing considered in the algorithm design; and interfaces for continuous enhancement of edge AI capabilities.

[0173] In summary, the intelligent fault diagnosis system for wind farm equipment provided by this invention effectively solves the problems of insufficient diagnostic accuracy, poor interpretability, and difficulty in adapting to different wind farm environments in existing technologies by comprehensively applying advanced technologies such as multimodal data fusion, edge computing, deep learning, transfer learning, reinforcement learning, and causal reasoning. The system's modular design and advanced learning algorithms give it strong adaptability and scalability, enabling it to continuously absorb new data and knowledge and continuously improve its diagnostic capabilities. The application of this system can significantly improve the operational efficiency of wind farms, reduce maintenance costs, and provide strong support for the sustainable development of the wind power industry.

[0174] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent fault diagnosis of wind farm equipment, characterized in that, Includes the following steps: Collect multi-dimensional sensor data from wind farm equipment; The multi-dimensional sensor data is preliminarily analyzed using an edge analysis algorithm to determine the multi-dimensional sensor data that requires further diagnosis. Input the multi-dimensional sensor data that requires further diagnosis into the pre-built fault diagnosis model to obtain the fault diagnosis results; The pre-built fault diagnosis model includes a multimodal deep learning model, a transfer learning and domain adaptation module, a multi-agent reinforcement learning framework, a causal reasoning engine, a knowledge graph-enhanced neural symbolic reasoning module, and a dynamic system health index calculation module. The multi-dimensional sensor data requiring further diagnosis is fused through the multimodal deep learning model, and then the regions are determined by the transfer learning and domain adaptation module and the multi-agent reinforcement learning framework. Simultaneously, the fault diagnosis is processed by the causal reasoning engine and the knowledge graph-enhanced neural symbolic reasoning module. The determined regions and diagnosed faults are then input into the dynamic system health index calculation module to obtain the fault diagnosis result. The multimodal deep learning model includes a spatiotemporal attention mechanism and a multi-scale causal convolutional network; The transfer learning and domain adaptation module employs a dynamic weight alignment method and an adversarial domain adaptation method. The multi-agent reinforcement learning framework employs a multi-agent soft actor-commentator algorithm. The knowledge graph-enhanced neural symbol reasoning module includes a knowledge graph embedding module and a neural symbol reasoning module; the knowledge graph embedding module adopts the TransE model; and the neural symbol reasoning module adopts a method combining symbolic rules and neural networks.

2. The intelligent fault diagnosis method for wind farm equipment according to claim 1, characterized in that, The preliminary analysis of the multi-dimensional sensor data using an edge analysis algorithm to determine the multi-dimensional sensor data requiring further diagnosis specifically involves: Preprocess multi-dimensional sensor data; An edge analysis algorithm is used to detect outliers in the preprocessed multi-dimensional sensor data; these outliers are the multi-dimensional sensor data that require further diagnosis.

3. A wind farm equipment fault intelligent diagnosis system, based on the wind farm equipment fault intelligent diagnosis method according to claim 1, characterized in that, include: The data acquisition layer is used to collect multi-dimensional sensor data from wind farm equipment. An edge computing layer is used to perform preliminary analysis of the multi-dimensional sensor data using an edge analysis algorithm to determine the multi-dimensional sensor data that needs further diagnosis. The cloud-based analytics layer is used to input multi-dimensional sensor data that requires further diagnosis into a pre-built fault diagnosis model to obtain fault diagnosis results.

4. The intelligent fault diagnosis system for wind farm equipment according to claim 3, characterized in that, The intelligent fault diagnosis system for wind farm equipment also includes a network transmission layer, which is used to realize data transmission between the edge computing layer and the cloud analysis layer.

5. The intelligent fault diagnosis system for wind farm equipment according to claim 4, characterized in that, The intelligent fault diagnosis system for wind farm equipment also includes an application interface layer, which receives data from the cloud analysis layer and provides diagnostic results and interactive interfaces to users.

Images