Wind turbine generator cloud edge collaborative intelligent diagnosis and dynamic optimization system

Abstract

The invention discloses a wind turbine generator cloud edge cooperative intelligent diagnosis and dynamic optimization system, and relates to the technical field of wind turbine generator optimization control, and the system comprises an edge event collection unit which is used for carrying out the preprocessing of the data of a wind speed, propeller pitch, vibration and temperature multi-source sensor, and generating an event message; the cloud-edge event bus is used for realizing two-way transmission and queue management of edge events based on Kafka; the edge self-adaptive twin subsystem is integrated with a physical model engine which is used for performing real-time simulation based on blade pneumatic and tower drum structure equations; and the hybrid strategy scheduler is used for dynamically adjusting the fusion weight of the physical module and the data-driven model based on the event type and the model error. According to the invention, through event-driven closed-loop adaptive hybrid twinning, including lightweight event bus and physical model and data model online fusion scheduling, real-time adaptive diagnosis and edge control are realized, the problems of model update lag and single model limitation are overcome, the diagnosis response time delay is reduced, and the accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine optimization control, and in particular to a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system. Background Art

[0002] With the large-scale development of wind power, the impact of turbine fault diagnosis and operational optimization on wind farm economics and reliability is becoming increasingly prominent. The traditional "cloud-based centralized analysis + edge data collection" architecture suffers from drawbacks such as delayed model updates, wasted bandwidth, and slow decision-making responses. This makes it difficult to meet the demands of high wind speed fluctuations, rapid local fault resolution, and overall wind farm coordinated optimization.

[0003] Patent CN119267119B discloses a fault diagnosis and processing system for a wind turbine. The above patent improves the reliability of fault diagnosis during operation of a wind turbine, and solves the problem of low reliability of fault diagnosis during operation of a wind turbine in the prior art.

[0004] The above patent obtains the key operating data of the specified wind turbine during operation and obtains the operating status interference score in combination with environmental parameters, then compares the obtained operating status interference score with the operating status threshold range to determine whether to send a fault diagnosis instruction, then receives the fault diagnosis instruction after the fault level is divided and determines whether to perform fault diagnosis based on the obtained response delay score, and finally optimizes the automatic controller according to the statistics on the fault diagnosis report and obtains the optimized evaluation value. However, the fault diagnosis and processing system of the wind turbine disclosed in the above patent has disadvantages such as model update lag, bandwidth waste and slow decision response.

[0005] To this end, this application proposes a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system that can realize real-time adaptive diagnosis and edge control. Summary of the Invention

[0006] The purpose of the present invention is to provide a cloud-edge collaborative intelligent diagnosis and dynamic optimization system for wind turbines to solve the problems of the traditional "cloud-based centralized analysis + edge data collection" architecture proposed in the above background technology, such as model update lag, bandwidth waste and slow decision response, which makes it difficult to meet the technical requirements of high wind speed mutations, rapid handling of local faults and overall collaborative optimization of wind farms.

[0007] To achieve the above objectives, the present invention provides the following technical solutions: a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system, the diagnosis and dynamic optimization system comprising:

[0008] Edge event acquisition unit, used to pre-process wind speed, pitch, vibration, and temperature multi-source sensor data and generate event messages;

[0009] Cloud-edge event bus, based on Kafka, implements bidirectional transmission and queue management of edge events;

[0010] Edge-adaptive twin subsystem, integrating:

[0011] Physical model engine for real-time simulation based on blade aerodynamics and tower structure equations;

[0012] A local online learning module for updating the drive model parameters using the latest sensor data;

[0013] The hybrid strategy scheduler dynamically adjusts the fusion weight of the physical module and the data-driven model based on event type and model error.

[0014] Preferably, the diagnosis and dynamic optimization system further includes:

[0015] A cloud-based panoramic twin subsystem, integrating a high-precision physical simulation platform with a massively parallel data-driven training framework, is used for comprehensive simulation and model updates of the entire fleet.

[0016] Edge closed-loop control unit, used to adjust pitch angle, braking torque and speed in real time based on fault warnings or operation optimization instructions output by the hybrid twin;

[0017] The cloud-based optimization and coordination unit receives edge feedback data, generates a global optimal operation strategy through a federated reinforcement learning algorithm, and sends it to each edge.

[0018] The feedback monitoring module is used to package execution indicators into feedback events and upload them to the cloud-edge event bus after the edge executes cloud policies.

[0019] Preferably, the cloud-edge event bus further includes an event priority management submodule, the event priority management submodule:

[0020] Event messages are classified into three levels: emergency failure, performance degradation, and environmental fluctuation;

[0021] Based on real-time evaluation of network bandwidth and edge computing power, the routing and sharding strategies for different events are dynamically adjusted to ensure that high-priority events are transmitted first.

[0022] The routing decision information is sent to the edge adaptive twin subsystem for adjusting the data reporting frequency on the edge side.

[0023] Preferably, the event priority management submodule has a built-in threshold learning engine, and the threshold learning engine:

[0024] Leveraging historical transmission delay and processing time data, the system automatically optimizes the priority thresholds for each event category based on reinforcement learning.

[0025] When a critical failure event is detected, the "express lane" is triggered and the end-to-end delay is controlled within 50ms;

[0026] The optimized threshold strategy is periodically synchronized to the edge event collection unit.

[0027] Preferably, the hybrid strategy scheduler in the edge adaptive twin system further comprises:

[0028] The granular sub-model management module divides the twin model into aerodynamic sub-model, structural mechanics sub-model, electrical sub-model and control sub-model;

[0029] The sub-model on-demand loading engine loads the corresponding sub-model and unloads the redundant sub-model in real time according to the event type and resource availability;

[0030] The cloud model generation unit automatically generates sub-models of different precision levels based on a meta-learning method and distributes compiled operators to the edge for loading.

[0031] Preferably, the cloud model generation unit further has:

[0032] The multi-task meta-learner is used to train general sub-model meta-parameters under different wind conditions, temperatures and load conditions;

[0033] The automatic compression module generates a lightweight model by pruning and quantizing a high-precision model;

[0034] The secure distribution module distributes sub-model operators and update metadata to each edge node based on an encrypted transmission protocol.

[0035] Preferably, the diagnosis and dynamic optimization system further comprises a self-evolution causal closed-loop reasoning module, which:

[0036] A causal graph database is constructed in the cloud to map the whole process of event-twin prediction-control execution-feedback into causal links;

[0037] The Bayesian update and deep causal inference algorithm are applied to periodically evolve the causal weights and identify new types of fault causal relationships;

[0038] A multi-step optimization strategy is automatically generated based on the graph path and pushed to the edge closed-loop control unit through the cloud-edge event bus.

[0039] Preferably, the self-evolution causal closed-loop reasoning module further comprises a root cause analysis sub-module, which:

[0040] Receives the fault feature vector uploaded by the edge, performs multi-path reasoning in combination with the causal graph, and outputs the most likely fault root cause and probability score;

[0041] The root cause analysis results and optimization strategies are encapsulated as an execution plan and sent to the edge adaptive twin subsystem through the event bus.

[0042] Preferably, the edge closed-loop control unit includes a reconfigurable computing unit, and the reconfigurable computing unit:

[0043] Supports dynamic reconfiguration of FPGA and DPU for deploying quantized or pruned networks under different scheduling strategies;

[0044] A built-in real-time inference monitoring module monitors model inference latency and accuracy, and automatically switches to a higher-precision model when the latency exceeds the limit or the accuracy decreases.

[0045] Preferably, the diagnosis and dynamic optimization system further includes an ultrasonic fatigue crack early detection subsystem, wherein the ultrasonic fatigue crack early detection subsystem:

[0046] Arrange piezoelectric acoustic emission sensor arrays at the main shaft bearings and gearbox;

[0047] Detect ultrasonic frequency band crack signals through multi-channel video analysis module and extract energy, spectrum and envelope characteristics;

[0048] The crack features are encapsulated as high-priority fault events, reported to the cloud-edge event bus in real time, and the closed-loop control strategy is triggered.

[0049] Preferably, the diagnosis and dynamic optimization system further includes a dynamic resource scheduling module based on 5G network slicing, the dynamic resource scheduling module:

[0050] Allocate transmission bandwidth and time slots in different slices according to event priority and data type;

[0051] Monitor slice quality and network latency in real time, and dynamically adjust the transmission strategy of diagnostic data, twin status, and optimization instructions;

[0052] Slice scheduling decisions and network status feedback are synchronized as events to the edge adaptive twin module to achieve end-to-end QoS closed-loop guarantee.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] 1. This invention uses event-driven closed-loop adaptive hybrid twins, including a lightweight event bus and online fusion scheduling of physical and data models, to achieve real-time adaptive diagnosis and edge control, overcoming the problems of model update lag and the limitations of a single model, reducing diagnostic response latency and improving accuracy.

[0055] 2. This invention uses event priority adaptive routing, including multi-level event classification and threshold learning-driven soft routing and 5G slice allocation, to achieve ultra-fast channel protection for critical faults, solve the transmission delay and packet loss problems of high-priority events, reduce end-to-end latency for emergency events, and improve resource utilization;

[0056] 3. This invention uses a model granularity self-adjusting engine, including pluggable sub-models for on-demand loading and cloud-source learning compression and distribution, to achieve adaptive allocation of edge computing power and memory, solving edge computing bottlenecks and model redundancy problems, optimizing computing power utilization, and reducing energy consumption.

[0057] 4. This invention uses self-evolving causal closed-loop reasoning, including cloud-based causal graphs, Bayesian and deep causal inference, and multi-step root cause analysis, to achieve precise root cause location and multi-step optimization path generation, addressing the unclear causal relationship of faults and the limitations of single-step strategies, improving the accuracy of root cause location, and enhancing the stability of overall power output. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a schematic diagram of the diagnosis and dynamic optimization system architecture of the present invention;

[0059] Figure 2 Schematic diagram of the diagnosis and dynamic optimization workflow of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0061] In the description of the present invention, it should be noted that the terms "upper," "lower," "inner," "outer," "front end," "rear end," "both ends," "one end," "the other end," and the like, indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0062] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "provided with," "connected," etc., should be understood in a broad sense. For example, "connected" may refer to a fixed connection, a detachable connection, or an integral connection; it may refer to a mechanical connection or an electrical connection; it may refer to a direct connection or an indirect connection through an intermediate medium; it may refer to internal communication between two components. Those skilled in the art will be able to understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0063] See also Figure 1 and Figure 2 The present invention provides an embodiment of a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system. System hardware deployment: Inside the wind turbine nacelle, along the blade root, the middle section of the tower, and the main shaft bearing, are installed: a wind speed sensor (ultrasonic, sampling frequency 1kHz, accuracy ±0.1m / s), a pitch sensor (photoelectric encoder, resolution 0.01°), a three-axis vibration accelerometer (bandwidth 0-5kHz), and a temperature sensor (thermocouple, accuracy ±0.5°C). The edge gateway uses an ARM Cortex-A72 platform with built-in 2GB of memory, a quad-core CPU, and 1×1Gbps Ethernet port, running a Linux embedded system. After the edge event acquisition unit collects the raw signals from each sensor, it performs the following operations in sequence:

[0064] 1. Low-pass filtering (cut-off frequency 2.5kHz) and denoising;

[0065] 2. Feature extraction: envelope energy, homogeneity, peak-to-peak value;

[0066] 3. Standardize and structure packaging to generate JSON format event messages;

[0067] The cloud-edge event bus deploys a three-node Apache Kafka cluster, with the edge gateway as the producer and the cloud platform as the consumer. Topics are configured for edge-to-cloud and cloud-to-edge. Events such as "fault warning" and "operation optimization" are sent by the edge, and the cloud platform subscribes to these events, triggering corresponding processing.

[0068] Edge-adaptive twin subsystem: The physical model engine implements aerodynamic simulation based on the Blade Element Momentum (BEM) method, inputting wind speed and pitch to output blade force and unit power. The local online learning model deploys a multivariate linear regression model to fit the power-wind speed curve using the most recent 1,000 sets of sampled data. The hybrid strategy scheduler automatically switches the physical model weight to 30% and the data model weight to 70% when the output power error between the local model and the physical model exceeds 5% or a "vibration warning" event is triggered.

[0069] Cloud panoramic twin subsystem is based on high-performance computing cluster, runs CFD software to perform full-field unit wind field simulation, receives edge events and feedback data in batches at regular time (default every 1h), and updates global wind speed distribution and structure response model parameters;

[0070] Closed-loop control and feedback: The edge closed-loop control unit receives the JSON optimization instruction issued by the cloud, and issues it to the PLC through the CAN bus to complete the pitch fine adjustment; after the control is executed, the edge reads the power and vibration and generates an "execution feedback" event and uploads it to the cloud-edge event bus.

[0071] Please refer to Figure 1 and Figure 2 , the application provides an embodiment: a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system, after adding event priority management and soft routing:

[0072] Event priority classification: emergency fault: main shaft vibration > 200 mm / s; performance degradation: power efficiency drop > 3%; environmental fluctuation: wind speed change > 2 m / s;

[0073] Threshold learning engine: the cloud is based on the past 30 days of delay and packet loss rate data, and uses Q-learning algorithm to automatically adjust the priority threshold of each event category; for example, when the average round-trip delay of "emergency fault" is > 80 ms, the priority threshold proportion is increased to 0.8;

[0074] Soft routing and fragmentation: before sending the "emergency fault" event, the edge marks the message as URGENT and uses 5G URLLC slice; performance degradation events use eMBB slice, with a delay requirement < 100 ms; environmental fluctuation events are uploaded in batches through mMTC slice at night;

[0075] Feedback strategy: when the URGENT event reaches the cloud, immediately issue a "safe speed reduction" instruction (reduce the speed to 0.6 times the rated value), and return the local time delay data to the threshold learning engine.

[0076] Please refer to Figure 1 and Figure 2 , the application provides an embodiment: a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system, based on the addition of event priority management and soft routing, realizes hybrid twin pluggable sub-model and cloud meta-learning ability:

[0077] Sub-model division: aerodynamic sub-model (CFD / linear regression), structural mechanics sub-model (finite element / low-order vibration model), vibration modal sub-model (multi-channel spectrum / regression), electrical power sub-model (electromagnetic field simulation / experience formula);

[0078] On-demand loading engine: When a "vibration limit exceeded" event is triggered, the edge only loads the high-precision versions of the vibration and structural sub-models. Under stable working conditions, only the lightweight versions of the pneumatic and electrical sub-models are loaded, and other sub-models are unloaded, reducing computing power utilization by ≥35%;

[0079] Cloud-based meta-learning: The multi-task meta-learner trains sub-model meta-parameters based on different wind conditions, temperatures, and load scenarios. The automatic compression module converts high-precision models into lightweight operators through pruning and quantization tools (such as TensorRT). The secure distribution module encrypts and sends operators and metadata based on the TLS1.3 protocol.

[0080] See also Figure 1 and Figure 2 The present invention provides an embodiment of a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system. Based on the implementation of a hybrid twin pluggable sub-model and cloud-based meta-learning capabilities, it introduces causal graphs and multi-path root cause analysis.

[0081] Causal graph construction: The cloud-based MySQL database stores the causal graph table: nodes (ID, type, attributes), edges (source ID, target ID, weight). The initial weight is set by expert experience, such as 0.7 for "vibration-crack";

[0082] Bayesian Update and Deep Causal Inference: Every night, based on newly uploaded events and feedback, we use Bayesian Network to update edge weights and introduce the GNN-Causal algorithm to explore potential new causal pathways.

[0083] Multi-step optimization strategy generation: If the weight of "crack warning-power reduction" in the cause-effect diagram is greater than 0.8, the system generates a two-step execution plan:

[0084] 1. First fine-tune the braking torque to +10Nm;

[0085] 2. Reduce the pitch by 0.3°;

[0086] The solution is encapsulated as an execution plan and sent to the edge control unit in sequence through the event bus.

[0087] See also Figure 1 and Figure 2 The present invention provides an embodiment of a wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system, wherein the diagnosis and dynamic optimization system further includes a dynamic resource scheduling module based on 5G network slicing. The dynamic resource scheduling module:

[0088] Allocate transmission bandwidth and time slots in different slices according to event priority and data type;

[0089] Monitor slice quality and network latency in real time, and dynamically adjust the transmission strategy of diagnostic data, twin status, and optimization instructions;

[0090] Slice scheduling decisions and network status feedback are synchronized as events to the edge adaptive twin module to achieve end-to-end QoS closed-loop guarantee;

[0091] The introduction of cause-and-effect graphs and multi-path root cause analysis enhances edge hardware reconfigurability and ultrasonic-level early crack warning.

[0092] Reconfigurable computing unit: The edge gateway integrates Xilinx UltraScale+ FPGA and ARM DPU, which can dynamically load quantized CNN or pruned Transformer models. FPGA reconfiguration events are less than 2ms, inference latency is less than 5ms, and it automatically switches back to the backup model when inference latency exceeds 8ms.

[0093] Ultrasonic crack detection subsystem: Four piezoelectric sensors are deployed on the inside of the spindle bearing and two on the gearbox housing, with a sampling rate of 2 MHz. The edge FPGA performs a short-time Fourier transform to extract key crack signal features: high-frequency peaks and energy envelopes. When the feature vector exceeds the "initial threshold + 3σ," a "crack warning" event is reported, triggering edge closed-loop control to reduce the speed to the rated value × 0.6.

[0094] Dynamic 5G slicing scheduling: The edge slice controller monitors UR DCI and SNR in real time. Based on "crack warning" events, it increases the bandwidth of a 5MHz URLLC slice from 10MHz to 15MHz. In the event of performance degradation, it automatically falls back to the eMBB slice, ensuring that critical warning resources are prioritized.

[0095] Working Principle: Each nacelle edge node continuously monitors multi-source sensor data, including wind speed, vibration, and temperature. When an anomaly is detected (such as excessive vibration or performance degradation), a structured event message is immediately generated and published on a lightweight event bus. This drives the local twin model to update and adaptively adjust, quickly providing preliminary diagnosis and safety control recommendations.

[0096] Edge events and diagnostic results are securely aggregated and transmitted to the cloud. The panoramic digital twin platform integrates CFD simulation and group data-driven training to update the global model and causal graph. The cloud then issues optimization strategies and model weights. The edge adaptive twin subsystem automatically selects the left and right physical / data model combinations to achieve bidirectional adaptive evolution of the model.

[0097] The edge closed-loop control unit issues control instructions (such as pitch angle fine-tuning and braking strategy adjustment) according to the priority based on the multi-step optimization plan of the hybrid twin output. After execution, it transmits power, vibration and other indicators back to the cloud as feedback events. The cloud continuously improves the causal graph based on Bayesian and deep causal inference, driving more accurate diagnosis and optimization next time.

[0098] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system, characterized by: The diagnosis and dynamic optimization system includes: Edge event acquisition unit, used to pre-process wind speed, pitch, vibration, and temperature multi-source sensor data and generate event messages; Cloud-edge event bus, based on Kafka, implements bidirectional transmission and queue management of edge events; Edge-adaptive twin subsystem, integrating: Physical model engine for real-time simulation based on blade aerodynamics and tower structure equations; A local online learning module for updating the drive model parameters using the latest sensor data; The hybrid strategy scheduler dynamically adjusts the fusion weight of the physical module and the data-driven model based on event type and model error.

2. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 1 is characterized by: The diagnosis and dynamic optimization system also includes: A cloud-based panoramic twin subsystem, integrating a high-precision physical simulation platform with a massively parallel data-driven training framework, is used for comprehensive simulation and model updates of the entire fleet. Edge closed-loop control unit, used to adjust pitch angle, braking torque and speed in real time based on fault warnings or operation optimization instructions output by the hybrid twin; The cloud-based optimization and coordination unit receives edge feedback data, generates a global optimal operation strategy through a federated reinforcement learning algorithm, and sends it to each edge. The feedback monitoring module is used to package execution indicators into feedback events and upload them to the cloud-edge event bus after the edge executes cloud policies.

3. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 1 is characterized by: The cloud-edge event bus further includes an event priority management submodule, which: Event messages are classified into three levels: emergency failure, performance degradation, and environmental fluctuation; Based on real-time evaluation of network bandwidth and edge computing power, the routing and sharding strategies for different events are dynamically adjusted to ensure that high-priority events are transmitted first. The routing decision information is sent to the edge adaptive twin subsystem for adjusting the data reporting frequency on the edge side.

4. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 3 is characterized by: The event priority management submodule has a built-in threshold learning engine. The threshold learning engine: Leveraging historical transmission delay and processing time data, the system automatically optimizes the priority thresholds for each event category based on reinforcement learning. When a critical fault event is detected, the "ultra-fast channel" is triggered and the end-to-end latency is controlled within 50ms; The optimized threshold strategy is periodically synchronized to the edge event collection unit.

5. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 1 is characterized by: The hybrid strategy scheduler in the edge adaptive twin subsystem further includes: Granular sub-model management module, which divides the twin model into aerodynamic sub-model, structural mechanics sub-model, electrical sub-model and control sub-model; The sub-model on-demand loading engine loads the corresponding sub-model in real time and unloads redundant sub-models based on event type and resource availability; The cloud-based model generation unit automatically generates sub-models of different accuracy levels based on the meta-learning method and sends the compiled operators to the edge for loading.

6. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 5, characterized in that: The cloud model generation unit further comprises: A multi-task meta-learner for training common sub-model meta-parameters under different wind conditions, temperature, and load conditions; Automatic compression module, which converts high-precision models into lightweight models through pruning and quantization; The secure distribution module sends sub-model operators and updated metadata to each edge node based on an encrypted transmission protocol.

7. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 1, characterized in that: The diagnosis and dynamic optimization system further includes a self-evolving causal closed-loop reasoning module: Build a causal graph database in the cloud, mapping the entire process of event-twin prediction-control execution-feedback into a causal chain; Apply Bayesian updating and deep causal inference algorithms to periodically self-evolve causal weights and identify new fault causal relationships; A multi-step optimization strategy is automatically generated based on the graph path and pushed to the edge closed-loop control unit through the cloud-edge event bus.

8. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 7, characterized in that: The self-evolving causal closed-loop reasoning module further includes a root cause analysis submodule, which: Receive fault feature vectors uploaded by the edge, perform multi-path reasoning based on the cause-effect graph, and output the most likely root cause of the fault and its probability score; The root cause analysis results and optimization strategies are encapsulated as an execution plan and sent to the edge adaptive twin subsystem through the event bus.

9. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 2, characterized in that: The edge closed-loop control unit includes a reconfigurable computing unit, which: Supports dynamic reconfiguration of FPGA and DPU for deploying quantized or pruned networks under different scheduling strategies; A built-in real-time inference monitoring module monitors model inference latency and accuracy, and automatically switches to a higher-precision model when the latency exceeds the limit or the accuracy decreases.

10. The wind turbine cloud-edge collaborative intelligent diagnosis and dynamic optimization system according to claim 1, characterized in that: The diagnosis and dynamic optimization system also includes an ultrasonic fatigue crack early detection subsystem, which includes: Arrange piezoelectric acoustic emission sensor arrays at the main shaft bearings and gearbox; Detect ultrasonic frequency band crack signals through multi-channel video analysis module and extract energy, spectrum and envelope characteristics; The crack features are encapsulated as high-priority fault events, reported to the cloud-edge event bus in real time, and the closed-loop control strategy is triggered.

Images