Wind turbine generator remote monitoring method and system based on PLC

Abstract

The embodiment of the invention provides a wind turbine generator remote monitoring method and system based on a PLC (Programmable Logic Controller), which actively performs real-time feature fusion and analysis on a multi-source sensor data stream from the PLC at an edge side, and inferes the health state of a generator in real time by utilizing a lightweight AI (Artificial Intelligence) model. Specifically, abstract packets, feature packets or snapshot packets with different information granularities are dynamically constructed through different normal, warning or critical health status codes evaluated at an edge end, and the abstract packets, the feature packets or the snapshot packets are matched with differentiated MQTT service quality levels for reporting. According to the method, passive data handling is changed into active information extraction and preferential reporting, network bandwidth occupation is reduced, transmission timeliness and reliability of key alarm information are ensured, and therefore the problem of analysis lag is solved.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a PLC-based remote monitoring method and system for wind turbine generators. Background Technology

[0002] As the core equipment of wind power generation, wind turbines are complex in structure and expensive, and are typically deployed in remote, high-altitude areas with harsh natural environments, posing significant challenges to their operation and maintenance. To ensure the safe and stable operation of wind turbines, improve power generation efficiency, and reduce maintenance costs, real-time and accurate remote monitoring of the health status of key components is crucial. The programmable logic controller (PLC), as the main control unit of the wind turbine, is responsible for executing the unit's control logic and collecting massive amounts of sensor data, including vibration, temperature, and speed data. It is the most direct and essential data source for achieving remote monitoring.

[0003] However, existing remote monitoring methods for wind turbines typically rely on a traditional architecture of passive data acquisition and centralized analysis. In this architecture, the programmable logic controllers or data acquisition units at the wind turbine site merely act as data carriers, indiscriminately transmitting massive streams of raw data—including vibration signals, temperature values, and speed values—collected from sensors to a cloud-based monitoring center via the network. This long-distance transmission of full-volume, high-frequency raw data places enormous pressure on network bandwidth, especially in remote wind farms with poor network conditions, easily creating data transmission bottlenecks. Secondly, because all analysis and calculations are performed in the cloud, there is a significant time delay between data generation and the acquisition of analytical insights.

[0004] This leads to delayed analysis, preventing the system from responding promptly to fault symptoms and potentially missing the optimal maintenance window. Finally, the architecture lacks the ability to perform intelligent processing at the data source, failing to filter and classify data based on the actual health status of devices. This results in critical alarm information being buried under massive amounts of normal data, making it difficult to guarantee the real-time nature and reliability of alerts.

[0005] Therefore, an optimized remote monitoring solution for wind turbines is desired. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides a PLC-based remote monitoring method and system for wind turbines.

[0007] In a first aspect, embodiments of the present invention provide a PLC-based remote monitoring method for wind turbine generators, comprising:

[0008] Sensor data is collected in real time from the PLC of the wind turbine to obtain the PLC raw data stream, which includes vibration signals, temperature values ​​and speed values.

[0009] In the edge computing device, the raw data stream of the PLC is timestamped and interpolated to obtain an aligned data buffer.

[0010] The data stream within a predetermined time window in the aligned data buffer is subjected to feature analysis, and the obtained multiple data features are fused to obtain the fusion representation vector of the health status of unit components;

[0011] The health status fusion representation vector of the unit components is input into the edge computing device to perform health status-based inference operations to obtain health assessment results, which include equipment health status codes and confidence scores.

[0012] Based on the device health status code, the health assessment results are dynamically layered and packaged into layered data packets.

[0013] Using an MQTT client library in an edge computing device, layered data packets are transmitted via the MQTT protocol to generate published MQTT messages;

[0014] When the cloud-based MQTT Broker receives a published MQTT message, it parses the published MQTT message based on the message type, and updates the dashboard, sends alarm notifications, and archives historical data based on the parsed data.

[0015] Furthermore, feature analysis is performed on the data stream within a predetermined time window in the aligned data buffer, and multiple data features are fused to obtain a fusion representation vector of the health status of unit components. This includes: performing time-domain statistical feature analysis on the vibration signal within the predetermined time window to obtain the root mean square (RMS), peak value, margin, and kurtosis of the vibration signal; performing frequency domain transformation analysis on the vibration signal within the predetermined time window to obtain the band energy and characteristic frequency amplitude; performing multivariate operating condition correlation analysis on the temperature and speed values ​​within the predetermined time window to obtain the temperature change rate under the influence of operating conditions; and combining the RMS, peak value, margin, kurtosis, band energy, characteristic frequency amplitude, and temperature change rate under the influence of operating conditions into a feature vectorized combination to obtain the fusion representation vector of the health status of unit components.

[0016] Furthermore, a multivariate working condition correlation analysis is performed on the temperature value and speed value within the predetermined time window to obtain the temperature change rate under the influence of the working condition, including: calculating the temperature change rate within the predetermined time window based on the temperature value within the predetermined time window;

[0017] The temperature change rate and speed value within the predetermined time window are respectively processed by a forward LSTM-based time-series pattern analyzer to obtain the temperature change rate time-series pattern feature vector and the speed time-series pattern feature vector; the temperature change rate time-series pattern feature vector and the speed time-series pattern feature vector are multiplied positionally to obtain the temperature change time-series pattern feature vector under the influence of the operating condition; the temperature change time-series pattern feature vector under the influence of the operating condition is decoded to obtain the temperature change rate under the influence of the operating condition.

[0018] Furthermore, the fused representation vector of the health status of the unit components is input into the edge computing device for inference operations to obtain a health assessment result. The health assessment result includes a device health status code and a confidence score, including: passing the fused representation vector of the health status of the unit components through a lightweight neural network model to obtain a hidden feature vector of the health status of the unit components; and passing the hidden feature vector of the health status of the unit components through a device health status analyzer to obtain a device health status code and a confidence score. The device health status code includes normal, warning, and critical.

[0019] Furthermore, based on the device health status code, the health assessment results are dynamically layered and packaged to obtain a layered data package, including: creating a summary package when the device health status code is normal; creating a feature package when the device health status code is a warning; and creating a snapshot package when the device health status code is critical.

[0020] Furthermore, in the edge computing device, an MQTT client library is used to transmit layered data packets based on the MQTT protocol to generate published MQTT messages. This includes: dynamically determining the quality of service level and publishing topic for the layered data packets based on the device health status code; and publishing the layered data packets as message bodies through the MQTT client to the publishing topic according to the quality of service level to generate the published MQTT messages.

[0021] Furthermore, based on the device health status code, the service quality level and publishing topic are dynamically determined for the layered data packet, including: when the device health status code is normal, the service quality level is determined to be 0, and the publishing topic is determined to be a preset summary status topic; when the device health status code is warning, the service quality level is determined to be 1, and the publishing topic is determined to be a preset characteristic event topic; when the device health status code is critical, the service quality level is determined to be 2, and the publishing topic is determined to be a preset critical event topic.

[0022] Secondly, embodiments of the present invention provide a PLC-based remote monitoring system for wind turbine generators, comprising:

[0023] The sensor data acquisition module is used to acquire sensor data from the PLC of the wind turbine in real time to obtain the PLC raw data stream, which includes vibration signals, temperature values ​​and speed values.

[0024] The PLC raw data processing module is used to perform timestamp alignment and interpolation processing on the PLC raw data stream in the edge computing device to obtain an aligned data buffer.

[0025] The data stream feature analysis module is used to perform feature analysis on the data stream within a predetermined time window in the aligned data buffer, and to fuse the obtained multiple data features to obtain a fusion representation vector of the health status of unit components.

[0026] The health status calculation and reasoning module is used to input the health status fusion representation vector of the unit components into the edge computing device for health status-based reasoning calculation to obtain health assessment results. The health assessment results include the equipment health status code and confidence score.

[0027] The data packaging module is used to dynamically package health assessment results into layered data packets based on the device health status code.

[0028] The MQTT message generation module is used in edge computing devices to use the MQTT client library to transmit layered data packets based on the MQTT protocol to generate published MQTT messages.

[0029] The message parsing module is fast. When the cloud-based MQTT Broker receives a published MQTT message, it parses the published MQTT message based on the message type and updates the dashboard, sends alarm notifications, and archives historical data based on the parsed data.

[0030] Furthermore, the data stream feature analysis module includes:

[0031] The time-domain statistical feature analysis unit is used to perform time-domain statistical feature analysis on the vibration signal within the predetermined time window to obtain the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, and the kurtosis of the vibration signal.

[0032] The frequency domain transformation analysis unit is used to perform frequency domain transformation analysis on the vibration signal within the predetermined time window to obtain the frequency band energy and characteristic frequency amplitude.

[0033] The multivariable operating condition correlation analysis unit is used to perform multivariable operating condition correlation analysis on the temperature value and speed value within the predetermined time window to obtain the temperature change rate under the influence of operating conditions.

[0034] The feature vectorization combination unit is used to combine the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, the kurtosis of the vibration signal, the frequency band energy, the amplitude of the characteristic frequency, and the rate of temperature change under the influence of operating conditions to obtain the fusion characterization vector of the health status of the unit components.

[0035] Furthermore, the multivariate operating condition correlation analysis unit includes:

[0036] The temperature change rate calculation unit is used to calculate the temperature change rate within the predetermined time window based on the temperature value within the predetermined time window.

[0037] The time-series pattern analysis unit is used to pass the temperature change rate and rotational speed value within the predetermined time window through a forward LSTM-based time-series pattern analyzer to obtain the temperature change rate time-series pattern feature vector and the rotational speed time-series pattern feature vector, respectively.

[0038] The timing pattern conversion unit is used to multiply the temperature change rate timing pattern feature vector and the rotation speed timing pattern feature vector by position to obtain the temperature change timing pattern feature vector under the influence of operating conditions.

[0039] The feature vector decoding unit is used to decode the feature vector of the temperature change time sequence pattern under the influence of the operating condition to obtain the temperature change rate under the influence of the operating condition.

[0040] Compared with existing technologies, this invention proposes a PLC-based remote monitoring method and system for wind turbines, which decentralizes data analysis and intelligent decision-making capabilities to edge computing devices closer to the data source. Specifically, this invention no longer transmits all raw data to the cloud; instead, it proactively performs real-time feature fusion and analysis on multi-source sensor data streams from the PLC at the edge, and uses a lightweight AI model to infer the turbine's health status in real time. That is, by evaluating different health status codes (normal, warning, or critical) at the edge, it dynamically constructs summary packets, feature packets, or snapshot packets with different information granularities, and reports them according to differentiated MQTT service quality levels. This method transforms passive data transport into proactive information extraction and optimal reporting, reducing network bandwidth consumption while ensuring the timeliness and reliability of key alarm information transmission, thus effectively solving the problem of analysis lag. Attached Figure Description

[0041] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0042] Figure 1 A flowchart of a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of data flow in a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention.

[0044] Figure 3 This is a flowchart illustrating the process of performing feature analysis on the data stream within a predetermined time window in the aligned data buffer and fusing multiple data features to obtain a fusion representation vector of the health status of unit components in a PLC-based remote monitoring method according to an embodiment of the present invention.

[0045] Figure 4 This is a flowchart illustrating the process of performing multivariate operating condition correlation analysis on temperature and speed values ​​within a predetermined time window to obtain the temperature change rate under the influence of operating conditions in a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention.

[0046] Figure 5 This is a block diagram of a PLC-based remote monitoring system for wind turbines according to an embodiment of the present invention. Detailed Implementation

[0047] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] Unless otherwise specifically stated, the technical or scientific terms used in the embodiments of this invention should be understood in their ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains. The terms "comprising" or "including," as used in the embodiments of this invention, do not limit the shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof mentioned, nor do they exclude the appearance or addition of one or more other different shapes, numbers, steps, actions, operations, components, elements, and / or groups thereof, or the inclusion of these.

[0049] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale, and techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail; however, where appropriate, the illustrated techniques, methods, and apparatus should be considered part of the specification. In all the examples shown and discussed herein, any other specific example may have different values. It should be noted that similar symbols and letters in the following figures denote similar items; therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0050] In the description of the embodiments of the present invention, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In the embodiments of the present invention, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of the present invention, as well as the features of different embodiments or examples.

[0051] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0052] Existing remote monitoring technologies for wind turbines still have inherent shortcomings in practical applications. Traditional monitoring architectures typically follow a passive data acquisition and centralized analysis model. PLCs or data acquisition units at the wind turbine site merely act as data carriers, transmitting massive amounts of unprocessed high-frequency sensor data (such as vibration waveforms) unchanged through the SCADA system to a remote cloud monitoring center, where the central server performs unified storage and analysis. This architecture leads to severe data silos and analytical delays. First, because the wind turbine's main control PLC has extremely limited computing resources to ensure millisecond-level real-time performance and stability, it cannot run complex analysis algorithms at the data generation end, resulting in a lack of intelligent sensing capabilities for field equipment. Second, indiscriminately aggregating all raw data for cloud processing not only puts enormous pressure on the already limited network bandwidth of remote wind farms, but more importantly, there is a significant time delay between data generation and the acquisition of analytical insights. By the time the cloud platform finally discovers an anomaly in the operating characteristics of a key component through calculations of massive historical data, the optimal maintenance window may have already been missed, making it impossible to effectively avoid economic losses caused by unplanned downtime. Therefore, how to overcome the drawbacks of the separation of data acquisition and intelligent analysis in the traditional architecture, and to carry out proactive and intelligent analysis and processing at the data generation end, transforming passive post-event analysis into real-time status early warning, is a technical problem that urgently needs to be solved in the field of wind power remote monitoring.

[0053] To address the technical challenges of analysis lag and network bandwidth bottlenecks caused by the remote transmission of full raw data in traditional wind power remote monitoring, this invention proposes a PLC-based remote monitoring method for wind turbines. This method first acquires raw data streams containing vibration signals, temperature values, and rotational speed values ​​in real time from the wind turbine's PLC at an edge computing device. These streams are then immediately timestamped and interpolated to form a unified time-base aligned data buffer. Subsequently, deep feature engineering is performed on the data within this buffer at the edge. This includes time-domain statistical and frequency-domain transformation analysis of the vibration signals to extract key features, and innovatively, multivariate operating condition correlation analysis of temperature and rotational speed values ​​to obtain the temperature change rate under the influence of operating conditions. Finally, these multidimensional features are vectorized and combined into a fusion vector that comprehensively characterizes the health status of the turbine components. This feature vector is directly input into a lightweight neural network model deployed at the edge for inference operations, providing real-time evaluation results including equipment health status codes such as normal, warning, or critical. Based on these evaluation results, data packets with different information granularities, such as summary packets, feature packets, or snapshot packets, are dynamically generated. Finally, these different types of data packets are intelligently matched with MQTT service quality levels from 0 to 2 and specific publishing topics. Selective and highly reliable data transmission is carried out through the MQTT client. After being received by the cloud, the data can be efficiently updated, alarms notified, and data archived. This enables intelligent analysis from the data source and on-demand reporting, solving the technical problem of analysis lag.

[0054] The present invention proposes a PLC-based remote monitoring method for wind turbine generators. Figure 1 This is a flowchart of a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the data flow in a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention. Figure 1 and Figure 2As shown, the remote monitoring method for wind turbines based on PLC according to an embodiment of the present invention includes the following steps: S100, acquiring sensor data in real time from the PLC of the wind turbine to obtain a raw PLC data stream, the raw PLC data stream including vibration signals, temperature values, and rotational speed values; S200, performing timestamp alignment and interpolation processing on the raw PLC data stream in an edge computing device to obtain an aligned data buffer; S300, performing feature analysis on the data stream within a predetermined time window in the aligned data buffer, and fusing multiple data features to obtain a fusion representation vector of the health status of turbine components; S400, inputting the fusion representation vector of the health status of turbine components into the edge computing device for inference operations based on health status to obtain a health assessment result, the health assessment result including a device health status code and a confidence score; S500, dynamically layering and packaging the health assessment result based on the device health status code to obtain a layered data packet; S600, using an MQTT client library in the edge computing device to transmit the layered data packet based on the MQTT protocol to generate a published MQTT message; S700, when the cloud MQTT... When the Broker receives a published MQTT message, it parses the published MQTT message based on the message type, and then updates the dashboard, sends alarm notifications, and archives historical data based on the parsed data.

[0055] Specifically, in steps S100 and S200, firstly, sensor data is acquired in real time from the PLC of the wind turbine to obtain a raw PLC data stream, which includes vibration signals, temperature values, and rotational speed values. Secondly, the raw PLC data stream is timestamped and interpolated in the edge computing device to obtain an aligned data buffer. It should be understood that due to differences in the data acquisition frequency, transmission protocol, and scanning cycle of the various sensors within the wind turbine PLC, the data points of different physical quantities (such as vibration signals, temperature values, and rotational speed values) in the raw PLC data stream received by the edge computing device inherently exhibit misalignment and asynchrony in terms of time reference. Therefore, in the technical solution of this invention, after acquiring sensor data in real time from the PLC of the wind turbine, the raw PLC data stream is further timestamped and interpolated in the edge computing device to obtain an aligned data buffer, thereby eliminating the time deviation between the data streams and constructing a multidimensional state dataset with a unified time reference. This provides an accurate and reliable data foundation for subsequent multivariate feature fusion analysis that requires high temporal synchronization, ensuring the accuracy of subsequent health status assessments.

[0056] More specifically, in a concrete example of the present invention, the process first establishes a communication connection between an OPC UA client deployed on an edge computing device and the OPC UA server of the wind turbine PLC, and subscribes to preset sensor data nodes. When the edge computing device receives sensor data from the PLC, it immediately appends a high-precision local timestamp synchronized by a network time protocol to each data point. Subsequently, these heterogeneous data carrying high-precision timestamps are sent to a first-in-first-out data queue as a raw data buffer. A data synchronization processing module runs at fixed time intervals, generating a standardized time series reference at these fixed intervals. For each time point on this reference, the processing module searches for the nearest real sensor readings in the raw data buffer and calculates the precise values ​​of each sensor at that standard time point using a linear interpolation algorithm. Finally, a series of data frames consisting of a unified timestamp and corresponding vibration, temperature, and speed values ​​are generated and stored in an aligned data buffer for subsequent modules to access.

[0057] Specifically, in step S300, the data stream within a predetermined time window in the aligned data buffer is subjected to feature analysis, and the resulting multiple data features are fused to obtain a fusion representation vector of the health status of unit components. It should be understood that because the original sensor data stream is high-dimensional, informationally redundant, and contains noise, directly using it for health status assessment would reduce the accuracy and efficiency of the model, and single sensor data is insufficient to comprehensively reflect the true health status of the equipment under complex operating conditions. Therefore, in the technical solution of this invention, the data stream within a predetermined time window in the aligned data buffer is further subjected to feature analysis, and the resulting multiple data features are fused to obtain a fusion representation vector of the health status of unit components. This extracts key information from the multi-dimensional original signal that can accurately represent the early faults and operating status of core components, and eliminates the interference of operating condition changes on status monitoring. In this way, a standardized, high-information-density low-dimensional feature vector can be generated, providing high-quality input for subsequent accurate health status inference.

[0058] Figure 3 This is a flowchart illustrating the process of performing feature analysis on the data stream within a predetermined time window in the aligned data buffer and fusing multiple data features to obtain a fused representation vector of the health status of turbine components, according to an embodiment of the PLC-based remote monitoring method for wind turbines based on an embodiment of the present invention. Figure 3As shown, step S300 includes: S310, performing time-domain statistical feature analysis on the vibration signal within the predetermined time window to obtain the root mean square of the vibration signal, the peak value of the vibration signal, the vibration signal margin, and the vibration signal kurtosis; S320, performing frequency domain transformation analysis on the vibration signal within the predetermined time window to obtain the frequency band energy and characteristic frequency amplitude; S330, performing multivariate operating condition correlation analysis on the temperature value and speed value within the predetermined time window to obtain the temperature change rate under the influence of operating conditions; S340, combining the root mean square of the vibration signal, the peak value of the vibration signal, the vibration signal margin, the vibration signal kurtosis, the frequency band energy, the characteristic frequency amplitude, and the temperature change rate under the influence of operating conditions into a feature vectorized combination to obtain the fusion characterization vector of the health status of the unit components.

[0059] In steps S310 to S320, time-domain statistical feature analysis is performed on the vibration signal within the predetermined time window to obtain the root mean square (RMS), peak value, margin, and kurtosis of the vibration signal; frequency-domain transform analysis is performed on the vibration signal within the predetermined time window to obtain the band energy and characteristic frequency amplitude. It should be understood that the raw vibration signal directly acquired from the PLC is a high-frequency time-series data stream containing a massive number of data points, which includes a large amount of background noise and redundant information. Without processing, it is difficult to directly reveal the true mechanical state inside the equipment, especially its insufficient ability to characterize early, subtle faults such as bearing pitting and gear wear. Therefore, in the technical solution of this invention, time-domain statistical feature analysis is further performed on the vibration signal within the predetermined time window to obtain the RMS, peak value, margin, and kurtosis of the vibration signal, and frequency-domain transform analysis is performed on the vibration signal within the predetermined time window to obtain the band energy and characteristic frequency amplitude. This extracts low-dimensional, high-information-density quantitative indicators that can sensitively reflect changes in the health status of key components from the original high-dimensional data. In this way, the non-stationary raw signal can be transformed into a set of structured features with clear physical meaning. This not only significantly reduces the load on data processing and transmission, but more importantly, it enhances the signal-to-noise ratio of fault features, laying a solid data foundation for accurate health status assessment and fault diagnosis.

[0060] More specifically, in a particular example of the invention, the process is executed on an edge computing device. It first obtains a segment of vibration signal data, for example, containing 2500 sampling points, within a predetermined time window from an aligned data buffer. First, a time-domain statistical feature analysis is performed, calculating the root mean square (RMS) value of the data array to characterize the overall level of vibration energy within the time window. Next, the peak value of the data array, i.e., the maximum absolute value of the signal amplitude, is calculated to reflect the presence of a severe impact event in the signal. Subsequently, the peak value is divided by the RMS value to obtain a signal margin, which quantifies the intensity of the impact component. Finally, the kurtosis of the data array is calculated, which measures the degree to which the probability distribution of the signal deviates from a Gaussian distribution, and is particularly sensitive to early impact failures of components such as rolling bearings. In parallel or sequentially, a frequency-domain transform analysis is performed, first applying a Hanning window function to the same data array to suppress spectral leakage, and then performing a Fast Fourier Transform on the windowed data to obtain the signal's spectrum. Based on prior knowledge of the inherent fault frequencies of the monitored gearbox bearings, characteristic frequency bands related to the inner and outer rings, rolling elements, or cages are located in the spectrum, and the sum of the energy within these frequency bands is calculated to obtain the frequency band energy. Simultaneously, the amplitude corresponding to the characteristic frequency points is precisely extracted to obtain the characteristic frequency amplitude. Finally, the four statistical features obtained from the time-domain analysis are combined with the frequency band energy and characteristic frequency amplitude obtained from the frequency-domain analysis to form a feature set containing multi-dimensional information for subsequent fusion processing.

[0061] In step S330, a multivariate operating condition correlation analysis is performed on the temperature and speed values ​​within the predetermined time window to obtain the temperature change rate under the influence of operating conditions. It should be understood that because the temperature of core components such as the wind turbine gearbox is strongly coupled with the unit's operating conditions such as speed and power, the normal temperature variation range changes dynamically under different operating conditions. If threshold judgments are based solely on isolated temperature values ​​or simple change rates, it is easy to generate a large number of false alarms or miss real faults due to normal fluctuations in operating conditions. Therefore, in the technical solution of this invention, a multivariate operating condition correlation analysis is further performed on the temperature and speed values ​​within the predetermined time window to isolate the interference caused by normal operating condition fluctuations such as speed changes on temperature changes, and to accurately extract temperature change information that truly reflects the abnormal health status of components. This significantly improves the accuracy and reliability of fault warnings and effectively reduces operation and maintenance costs.

[0062] Figure 4 This document describes a flowchart illustrating the multivariate operating condition correlation analysis performed on temperature and rotational speed values ​​within a predetermined time window to obtain the temperature change rate under the influence of operating conditions, based on a PLC-based remote monitoring method for wind turbines according to an embodiment of the present invention. Figure 4As shown, step S330 includes: S331, calculating the temperature change rate within the predetermined time window based on the temperature value within the predetermined time window; S332, passing the temperature change rate and speed value within the predetermined time window through a forward LSTM-based time-series pattern analyzer to obtain a temperature change rate time-series pattern feature vector and a speed time-series pattern feature vector, respectively; S333, multiplying the temperature change rate time-series pattern feature vector and the speed time-series pattern feature vector positionally to obtain a temperature change time-series pattern feature vector under the influence of operating conditions; S334, decoding the temperature change time-series pattern feature vector under the influence of operating conditions to obtain the temperature change rate under the influence of operating conditions.

[0063] In steps S331 and S332, the temperature change rate within the predetermined time window is calculated based on the temperature value within that window. The temperature change rate and rotational speed value within the predetermined time window are then processed by a forward LSTM-based time-series pattern analyzer to obtain a temperature change rate time-series pattern feature vector and a rotational speed time-series pattern feature vector, respectively. It should be understood that since the temperature and rotational speed sensor data of wind turbines are typical time-series signals, their instantaneous values ​​cannot fully reflect the dynamic evolution of the equipment's operating state. For example, a stable heating process and a heating process with shock oscillations, even if their average rate of change is the same, imply drastically different underlying equipment health conditions. Therefore, in the technical solution of this invention, the temperature change rate within the predetermined time window is further calculated based on the temperature value within that window; the temperature change rate and rotational speed value within the predetermined time window are then processed by a forward LSTM-based time-series pattern analyzer to obtain a temperature change rate time-series pattern feature vector and a rotational speed time-series pattern feature vector, thereby deeply mining and quantifying the dynamic behavior patterns and intrinsic correlations of the two key parameters, temperature and rotational speed, in the time dimension. In this way, the original one-dimensional time series data stream can be transformed into a high-dimensional feature vector with fixed dimensions that can characterize its dynamic process, providing high-quality input for subsequent more accurate multivariate working condition correlation analysis.

[0064] More specifically, in a specific example of the present invention, the implementation includes the following steps: First, data panening and preprocessing: For the real-time acquired temperature and rotational speed data streams, sliding sampling is performed within a fixed time window to form a temperature data sequence and a rotational speed data sequence containing multiple consecutive sampling points. Subsequently, a first-order difference operation is performed on the temperature data sequence to obtain a temperature change rate data sequence characterizing its rate of change. Second, temporal feature extraction: The temperature change rate data sequence is input into a pre-trained temperature temporal pattern analyzer based on a feedforward long short-term memory network, while the rotational speed data sequence is input into another independent rotational speed temporal pattern analyzer. Each LSTM network processes its corresponding input sequence sequentially, learning and memorizing long-term dependencies in the sequence through an internal gating mechanism. Third, feature vector generation: After the respective complete data sequences have been processed, the hidden layer output state of the last time step of each LSTM network is taken as the final feature expression. This output is the fixed-dimensional temperature change rate temporal pattern feature vector and rotational speed temporal pattern feature vector, which contains a compact encoding of the dynamic pattern of the data within the entire time window.

[0065] In steps S333 and S334, the time-series feature vector of the temperature change rate and the time-series feature vector of the rotational speed are multiplied positionally to obtain the time-series feature vector of temperature change under the influence of operating conditions; the time-series feature vector of temperature change under the influence of operating conditions is decoded to obtain the temperature change rate under the influence of operating conditions. It should be understood that since the temperature change rate and the rotational speed time-series feature vector generated in the previous step are located in high-dimensional abstract spaces, they independently characterize the temporal dynamics of temperature and operating conditions, but the mutual influence between them has not been quantified and cannot be directly used to isolate the normal influence of operating conditions on temperature rise. Therefore, in the technical solution of this invention, the time-series feature vector of the temperature change rate and the time-series feature vector of the rotational speed are further multiplied positionally to obtain the time-series feature vector of temperature change under the influence of operating conditions; the time-series feature vector of temperature change under the influence of operating conditions is decoded to obtain the temperature change rate under the influence of operating conditions. This allows for deep fusion and nonlinear interaction between the operating condition mode and the temperature change mode in the feature space, generating a fused feature that contains the correlation effect between the two. The decoder then maps this feature back to a scalar value with clear physical meaning. In this way, the net temperature change rate, which has been filtered out from the influence of normal operating condition fluctuations and is caused solely by changes in the equipment's own health status, can be accurately calculated, providing a highly sensitive and reliable basis for subsequent fault diagnosis regarding thermal anomalies.

[0066] More specifically, in a concrete example of this invention, two 16-dimensional feature vectors generated by a time-series pattern analyzer are obtained: a time-series feature vector for the rate of temperature change and a time-series feature vector for the rotational speed. First, these two 16-dimensional vectors are multiplied positionally, that is, the i-th element of the first vector is multiplied by the i-th element of the second vector, resulting in a new 16-dimensional vector, which is the time-series feature vector for the temperature change under the influence of the operating condition. Subsequently, this 16-dimensional product vector is input into a decoder, which is a neural network consisting of a single fully connected layer with 16 neurons in the input layer and 1 neuron in the output layer. Through forward propagation calculation by the decoder, a scalar value, such as 0.08, is finally output, which is the final obtained rate of temperature change under the influence of the operating condition.

[0067] In step S340, the root mean square (RMS) of the vibration signal, peak value of the vibration signal, vibration signal margin, vibration signal kurtosis, band energy, characteristic frequency amplitude, and temperature change rate under operating conditions are combined using feature vectorization to obtain a fusion representation vector of the health status of the unit components. It should be understood that the previously extracted individual physical features, such as vibration kurtosis or temperature change rate, can only reflect the operating status of the equipment from an isolated dimension. Fluctuations in any single feature may be caused by multiple reasons, making it difficult to comprehensively and accurately point to a specific health status or fault mode, and prone to misjudgment or omission when making comprehensive judgments. Therefore, in the technical solution of this invention, the RMS of the vibration signal, peak value of the vibration signal, vibration signal margin, vibration signal kurtosis, band energy, characteristic frequency amplitude, and temperature change rate under operating conditions are further combined using feature vectorization to obtain a fusion representation vector of the health status of the unit components. This effectively integrates heterogeneous information from multiple sources such as vibration, temperature, and speed, constructing a unified data structure that can comprehensively describe the current health status of the components from multiple angles. This allows for the formation of a fusion feature input with higher information density and stronger representational capabilities, providing richer and more robust decision-making basis for subsequent intelligent reasoning models, thereby significantly improving the accuracy and reliability of health status assessment.

[0068] More specifically, in a concrete example of the present invention, after calculating seven key characteristic values ​​of the gearbox component, such as the root mean square of the vibration signal being 0.5g, the peak value being 1.5g, the margin being 3, the kurtosis being 3.8, the target frequency band energy being 0.1J, the characteristic frequency amplitude being 0.08g, and the temperature change rate under the influence of operating conditions being 0.02℃ / min, a 7-dimensional vector is created. Subsequently, these seven values ​​are sequentially filled into predetermined positions in the vector, and normalization processing is performed on each element in the vector, for example, linearly scaling it to the [0,1] interval, finally generating a numerical vector such as [0.55,0.62,0.71,0.78,0.45,0.58,0.66]. This vector is the fusion characterization vector of the health status of the unit components and can be directly used as input for subsequent inference models.

[0069] Specifically, in step S400, the fusion representation vector of the unit component health status is input into an edge computing device for health status-based inference operations to obtain a health assessment result. The health assessment result includes the device health status code and confidence score. It should be understood that since the fusion representation vector of the unit component health status generated in the previous step is a high-dimensional numerical vector containing multi-dimensional physical information, there is a complex nonlinear mapping relationship between its internal features and specific health status. This relationship cannot be accurately and reliably analyzed through simple threshold judgments or linear rules. Therefore, in the technical solution of this invention, the fusion representation vector of the unit component health status is further input into an edge computing device for health status-based inference operations to obtain a health assessment result. The health assessment result includes the device health status code and confidence score. This utilizes an intelligent model deployed on the edge side to automatically learn and identify complex patterns in the fusion vector, achieving accurate mapping and quantitative assessment from underlying data features to high-level health status conclusions. In this way, real-time, automatic health status diagnosis can be completed at the source of data generation, eliminating the latency caused by data transmission to the cloud for analysis, and providing an immediate and reliable decision-making basis for subsequent dynamic hierarchical data reporting strategies.

[0070] More specifically, in this embodiment of the invention, the fusion representation vector of the health status of the unit components is input into the edge computing device for inference operations to obtain a health assessment result. The health assessment result includes a device health status code and a confidence score, comprising: passing the fusion representation vector of the health status of the unit components through a lightweight neural network model to obtain a hidden feature vector of the health status of the unit components; and passing the hidden feature vector of the health status of the unit components through a device health status analyzer to obtain a device health status code and a confidence score. The device health status code includes normal, warning, and critical.

[0071] More specifically, the edge computing device inputs a normalized 7-dimensional health status fusion representation vector of the unit components into a loaded lightweight neural network model. First, this vector passes through the network's input and hidden layers, undergoing a series of weighted sums and nonlinear activation function calculations to generate a latent feature vector of the unit component's health status at the output of the last hidden layer. Subsequently, this latent feature vector is fed into the model's equipment health status analyzer. This analyzer contains three neurons, corresponding to normal, warning, and critical states, respectively. After processing with the Softmax activation function, the analyzer outputs a probability distribution vector, for example, [0.10, 0.85, 0.05]. The system maps the index with the highest probability value to the equipment health status code "Warning," and uses this probability value of 0.85 as the confidence score, ultimately forming a complete health assessment result.

[0072] Specifically, in step S500, based on the equipment health status code, the health assessment results are dynamically layered and packaged into layered data packets. It should be understood that during the long-term operation of wind turbine units, the equipment is in a healthy state for most of the time. If detailed data containing in-depth features is continuously uploaded indiscriminately, it would result in a huge waste of network bandwidth resources and ineffective data storage costs. At the same time, when abnormal symptoms appear, richer and more detailed information than regular data is needed for diagnostic analysis. Therefore, in the technical solution of this invention, the health assessment results are further dynamically layered and packaged into layered data packets based on the equipment health status code. This maximizes the conservation of network resources while ensuring normal equipment operation, and automatically reports the core data required for in-depth diagnosis when potential or urgent equipment failures occur, achieving intelligent and efficient data transmission and significantly improving the effectiveness and relevance of early warnings.

[0073] More specifically, in this embodiment of the invention, based on the device health status code, the health assessment results are dynamically layered and packaged to obtain a layered data package, including: creating a summary package when the device health status code is normal; creating a feature package when the device health status code is a warning; and creating a snapshot package when the device health status code is critical.

[0074] In other words, more specifically, for the health assessment scenario of wind turbine gearboxes, when the health status analyzer in the edge computing device outputs a normal equipment health status code, the data packaging module will call the summary package generation function to create a summary package containing only key performance indicators such as equipment ID, timestamp, status code, and current average power. If the equipment health status code output by the analyzer subsequently changes to a warning, the module will switch to the feature package generation function, adding a complete health status fusion representation vector of the entire unit component that caused the warning to the summary package information, forming a feature package. Once the status escalates to critical, the module will execute the snapshot package generation logic, not only packaging the entire contents of the feature package, but also backtracking from the aligned data buffer and extracting the original vibration and temperature sequence data for 5 seconds before and after the anomaly occurred, compressing it, and encapsulating it into the data package to form a snapshot package containing the most complete on-site information.

[0075] Specifically, in steps S600 and S700, the edge computing device uses an MQTT client library to transmit layered data packets via the MQTT protocol to generate published MQTT messages. When the cloud-based MQTT Broker receives the published MQTT messages, it parses the messages based on message type and updates the dashboard, issues alarm notifications, and archives historical data based on the parsed data. It should be understood that after the edge computing device completes its health status assessment and generates layered data packets locally, it must transmit these time-sensitive results to the remote monitoring center through an efficient and reliable communication mechanism. Especially under the complex and variable network environment and bandwidth-constrained conditions of wind farms, ensuring that critical data is not lost or delayed is a prerequisite for effective remote monitoring. Therefore, in the technical solution of this invention, an MQTT client library is further used in the edge computing device to transmit layered data packets based on the MQTT protocol to generate published MQTT messages. When the cloud MQTT Broker receives the published MQTT messages, it parses the published MQTT messages based on message type, and updates dashboards, issues alarm notifications, and archives historical data based on the parsed data. This leverages the lightweight nature, publish-subscribe model, and quality of service (QoS) level guarantee of the MQTT protocol to build a stable data channel from the edge to the cloud and establish an automated data reception and processing workflow in the cloud. This ensures that the intelligent analysis results generated at the edge are delivered efficiently and reliably and processed in a timely manner, ultimately transforming monitoring insights into real-time dashboards, alarm notifications, and traceable historical data for operations and maintenance personnel, thus forming a complete closed-loop monitoring system from data collection and edge analysis to cloud applications.

[0076] More specifically, in this embodiment of the invention, an MQTT client library is used in an edge computing device to transmit layered data packets via the MQTT protocol to generate published MQTT messages. This includes: dynamically determining the quality of service (QoS) level and publishing topic for the layered data packets based on the device health status code; and publishing the layered data packets as message bodies via the MQTT client to the publishing topic according to the QoS level to generate the published MQTT messages. Specifically, dynamically determining the QoS level and publishing topic for the layered data packets based on the device health status code includes: when the device health status code is normal, determining the QoS level as 0 and the publishing topic as a preset summary status topic; when the device health status code is in a warning state, determining the QoS level as 1 and the publishing topic as a preset characteristic event topic; and when the device health status code is in a critical state, determining the QoS level as 2 and the publishing topic as a preset critical event topic.

[0077] More specifically, in a specific example of this invention, the first step is program module initialization: in the application of the edge computing device, an MQTT client object is instantiated, and the connection initialization settings to the cloud MQTT Broker are completed. The second step is obtaining decision criteria: the application receives the encapsulated layered data packets and their corresponding device health status codes from the previous-level dynamic data packaging module. The third step is executing dynamic parameter mapping: the program internally includes a decision logic module that receives the device health status code as input and makes judgments based on preset rules: when the device health status code is normal, the service quality level variable in the transmission parameters is set to 0, and the published topic string variable is set to a preset summary status topic; when the device health status code is in a warning state, the service quality level variable is set to 1, and the published topic string variable is set to a preset characteristic event topic; when the device health status code is in a critical state, the service quality level variable is set to 2, and the published topic string variable is set to a preset critical event topic. The fourth step is to call the interface to execute the publish. The application calls the publish function of the MQTT client instance, passing in the layered data packet, as well as the service quality level and publish topic string dynamically determined in the previous step as parameters. The client library automatically completes the network communication protocol interaction that conforms to the corresponding service quality level standard, and reliably sends the message to the cloud server.

[0078] In summary, the PLC-based remote monitoring method for wind turbines according to embodiments of the present invention is explained, which decentralizes data analysis and intelligent decision-making capabilities to edge computing devices closer to the data source. Specifically, the present invention no longer transmits all raw data to the cloud, but actively performs real-time feature fusion and analysis on multi-source sensor data streams from the PLC at the edge, and uses a lightweight AI model to infer the health status of the turbine in real time. That is, by evaluating different health status codes such as normal, warning, or critical at the edge, it dynamically constructs summary packets, feature packets, or snapshot packets with different information granularities, and reports them according to differentiated MQTT service quality levels. This method transforms passive data transport into proactive information extraction and optimal reporting, reducing network bandwidth consumption while ensuring the timeliness and reliability of key alarm information transmission, thereby effectively solving the problem of analysis lag.

[0079] This invention also provides a PLC-based remote monitoring system for wind turbine generators.

[0080] Figure 5 This is a block diagram of a PLC-based remote monitoring system for wind turbines according to an embodiment of the present invention. Figure 5 As shown, the PLC-based remote monitoring system 100 for wind turbines according to an embodiment of the present invention includes: a sensor data acquisition module 110, used to acquire sensor data from the PLC of the wind turbine in real time to obtain a raw PLC data stream, the raw PLC data stream including vibration signals, temperature values, and rotational speed values; a PLC raw data processing module 120, used to perform timestamp alignment and interpolation processing on the raw PLC data stream in an edge computing device to obtain an aligned data buffer; a data stream feature analysis module 130, used to perform feature analysis on the data stream within a predetermined time window in the aligned data buffer, and fuse the obtained multiple data features to obtain a fusion representation vector of the health status of the turbine components; and a health status calculation and reasoning module 140, used to calculate the fusion representation vector of the health status of the turbine components. The edge computing device performs health status-based inference operations on the input eigenvectors to obtain health assessment results, including device health status codes and confidence scores. A data packaging module 150 dynamically packages the health assessment results into layered data packets based on the device health status codes. An MQTT message generation module 160 uses an MQTT client library on the edge computing device to transmit the layered data packets via the MQTT protocol to generate published MQTT messages. A message parsing module 170 parses published MQTT messages based on message type when the cloud MQTTBroker receives them, and updates dashboards, issues alarm notifications, and archives historical data based on the parsed data.

[0081] Furthermore, the data stream feature analysis module 130 includes:

[0082] The time-domain statistical feature analysis unit is used to perform time-domain statistical feature analysis on the vibration signal within the predetermined time window to obtain the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, and the kurtosis of the vibration signal.

[0083] The frequency domain transformation analysis unit is used to perform frequency domain transformation analysis on the vibration signal within the predetermined time window to obtain the frequency band energy and characteristic frequency amplitude.

[0084] The multivariable operating condition correlation analysis unit is used to perform multivariable operating condition correlation analysis on the temperature value and speed value within the predetermined time window to obtain the temperature change rate under the influence of operating conditions.

[0085] The feature vectorization combination unit is used to combine the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, the kurtosis of the vibration signal, the frequency band energy, the amplitude of the characteristic frequency, and the rate of temperature change under the influence of operating conditions to obtain the fusion characterization vector of the health status of the unit components.

[0086] Furthermore, the multivariate operating condition correlation analysis unit includes:

[0087] The temperature change rate calculation unit is used to calculate the temperature change rate within the predetermined time window based on the temperature value within the predetermined time window.

[0088] The time-series pattern analysis unit is used to pass the temperature change rate and rotational speed value within the predetermined time window through a forward LSTM-based time-series pattern analyzer to obtain the temperature change rate time-series pattern feature vector and the rotational speed time-series pattern feature vector, respectively.

[0089] The timing pattern conversion unit is used to multiply the temperature change rate timing pattern feature vector and the rotation speed timing pattern feature vector by position to obtain the temperature change timing pattern feature vector under the influence of operating conditions.

[0090] The feature vector decoding unit is used to decode the feature vector of the temperature change time sequence pattern under the influence of the operating condition to obtain the temperature change rate under the influence of the operating condition.

[0091] The specific implementation method of the PLC-based remote monitoring system 100 for wind turbines in this embodiment of the invention can be found in the description of the PLC-based remote monitoring method for wind turbines in this embodiment of the invention, and will not be repeated here.

[0092] The PLC-based remote monitoring system 100 for wind turbines according to embodiments of the present invention can achieve collaborative operation between edge computing devices and cloud platforms, such as industrial PCs, smart gateways, or embedded controllers deployed next to wind turbines, and communicate with remote cloud data centers. In one possible implementation, the PLC-based remote monitoring system 100 for wind turbines according to embodiments of the present invention can be integrated into the monitoring system of wind turbines as a software module and / or hardware module. For example, on the edge computing device side, the system can be a background service program running in its operating system, containing functions such as data acquisition, feature analysis, model inference, and MQTT communication, or it can be a standalone application developed for turbine status monitoring. Of course, the lightweight neural network model used to perform inference operations in the system can also be embedded in dedicated industrial computing hardware, such as the NPU or FPGA of the edge computing device, to accelerate the real-time assessment process of the health status of turbine components.

[0093] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A PLC-based remote monitoring method for wind turbine generators, characterized in that, include: Sensor data is collected in real time from the PLC of the wind turbine to obtain the PLC raw data stream, which includes vibration signals, temperature values ​​and speed values. In the edge computing device, the raw data stream of the PLC is timestamped and interpolated to obtain an aligned data buffer. The data stream within a predetermined time window in the aligned data buffer is subjected to feature analysis, and the obtained multiple data features are fused to obtain the fusion representation vector of the health status of unit components; The health status fusion representation vector of the unit components is input into the edge computing device to perform health status-based inference operations to obtain health assessment results, which include equipment health status codes and confidence scores. Based on the device health status code, the health assessment results are dynamically layered and packaged into layered data packets. Using an MQTT client library in an edge computing device, layered data packets are transmitted via the MQTT protocol to generate published MQTT messages; When the cloud-based MQTT Broker receives a published MQTT message, it parses the published MQTT message based on the message type, and updates the dashboard, sends alarm notifications, and archives historical data based on the parsed data.

2. The PLC-based remote monitoring method for wind turbines according to claim 1, characterized in that, The data stream within a predetermined time window in the aligned data buffer is subjected to feature analysis, and the resulting multiple data features are fused to obtain a fused representation vector of the health status of unit components, including: Time-domain statistical feature analysis is performed on the vibration signal within the predetermined time window to obtain the root mean square, peak value, margin, and kurtosis of the vibration signal. The vibration signal within the predetermined time window is subjected to frequency domain transformation analysis to obtain the frequency band energy and characteristic frequency amplitude; Multivariate working condition correlation analysis is performed on the temperature and speed values ​​within the predetermined time window to obtain the temperature change rate under the influence of working conditions; The root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, the kurtosis of the vibration signal, the frequency band energy, the amplitude of the characteristic frequency, and the rate of temperature change under the influence of operating conditions are combined by feature vectorization to obtain the fusion characterization vector of the health status of the unit components.

3. The PLC-based remote monitoring method for wind turbines according to claim 2, characterized in that, Multivariate correlation analysis of temperature and rotational speed values ​​within the predetermined time window is performed to obtain the rate of temperature change under the influence of operating conditions, including: Calculate the rate of temperature change within the predetermined time window based on the temperature values ​​within the predetermined time window; The temperature change rate and rotational speed value within the predetermined time window are respectively processed by a forward LSTM-based time-series pattern analyzer to obtain the temperature change rate time-series pattern feature vector and the rotational speed time-series pattern feature vector. The temperature change rate time-series pattern feature vector and the speed time-series pattern feature vector are multiplied by position to obtain the temperature change time-series pattern feature vector under the influence of operating conditions. The temperature change time-series pattern feature vector under the influence of the operating conditions is decoded to obtain the temperature change rate under the influence of the operating conditions.

4. The PLC-based remote monitoring method for wind turbine generators according to claim 1, characterized in that, The health status fusion representation vector of the unit components is input into the edge computing device for inference operations to obtain a health assessment result. The health assessment result includes a device health status code and a confidence score, including: The health status fusion representation vector of the unit components is passed through a lightweight neural network model to obtain the hidden feature vector of the health status of the unit components; The hidden feature vectors of the health status of the unit components are passed through the equipment health status analyzer to obtain the equipment health status code and confidence score. The equipment health status code includes normal, warning and critical.

5. The PLC-based remote monitoring method for wind turbines according to claim 1, characterized in that, Based on the device health status code, the health assessment results are dynamically layered and packaged into layered data packets, including: When the device health status code is normal, a digest package is created; When the device health status code is a warning, a feature packet is created; When the device health status code is critical, a snapshot package is created.

6. The PLC-based remote monitoring method for wind turbine generators according to claim 5, characterized in that, Using an MQTT client library in an edge computing device, layered data packets are transmitted via the MQTT protocol to generate published MQTT messages, including: Based on the device health status code, the service quality level and publishing topic are dynamically determined for the layered data packets; According to the service quality level, the layered data packet is used as the message body and published to the publishing topic through the MQTT client to generate the published MQTT message.

7. The PLC-based remote monitoring method for wind turbine generators according to claim 6, characterized in that, Based on the device health status code, dynamically determine the quality of service level and publishing topic for the layered data packets, including: When the device health status code is normal, the service quality level is determined to be 0, and the published topic is determined to be a preset summary status topic; When the device health status code is in a warning state, the service quality level is determined to be 1, and the published topic is determined to be a preset characteristic event topic; When the device health status code is in a critical state, the service quality level is determined to be 2, and the published topic is determined to be a preset critical event topic.

8. A PLC-based remote monitoring system for wind turbine generators, characterized in that, include: The sensor data acquisition module is used to acquire sensor data from the PLC of the wind turbine in real time to obtain the PLC raw data stream, which includes vibration signals, temperature values ​​and speed values. The PLC raw data processing module is used to perform timestamp alignment and interpolation processing on the PLC raw data stream in the edge computing device to obtain an aligned data buffer. The data stream feature analysis module is used to perform feature analysis on the data stream within a predetermined time window in the aligned data buffer, and to fuse the obtained multiple data features to obtain a fusion representation vector of the health status of unit components. The health status calculation and reasoning module is used to input the health status fusion representation vector of the unit components into the edge computing device for health status-based reasoning calculation to obtain health assessment results. The health assessment results include the equipment health status code and confidence score. The data packaging module is used to dynamically package health assessment results into layered data packets based on the device health status code. The MQTT message generation module is used in edge computing devices to use the MQTT client library to transmit layered data packets based on the MQTT protocol to generate published MQTT messages. The message parsing module is fast. When the cloud-based MQTT Broker receives a published MQTT message, it parses the published MQTT message based on the message type and updates the dashboard, sends alarm notifications, and archives historical data based on the parsed data.

9. The PLC-based remote monitoring system for wind turbines according to claim 8, characterized in that, The data stream feature analysis module includes: The time-domain statistical feature analysis unit is used to perform time-domain statistical feature analysis on the vibration signal within the predetermined time window to obtain the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, and the kurtosis of the vibration signal. The frequency domain transformation analysis unit is used to perform frequency domain transformation analysis on the vibration signal within the predetermined time window to obtain the frequency band energy and characteristic frequency amplitude. The multivariable operating condition correlation analysis unit is used to perform multivariable operating condition correlation analysis on the temperature value and speed value within the predetermined time window to obtain the temperature change rate under the influence of operating conditions. The feature vectorization combination unit is used to combine the root mean square of the vibration signal, the peak value of the vibration signal, the margin of the vibration signal, the kurtosis of the vibration signal, the frequency band energy, the characteristic frequency amplitude, and the temperature change rate under the influence of operating conditions to obtain the fusion characterization vector of the health status of the unit components.

10. The PLC-based remote monitoring system for wind turbines according to claim 9, characterized in that, The multivariate operating condition correlation analysis unit includes: The temperature change rate calculation unit is used to calculate the temperature change rate within the predetermined time window based on the temperature value within the predetermined time window. The time-series pattern analysis unit is used to pass the temperature change rate and rotational speed value within the predetermined time window through a forward LSTM-based time-series pattern analyzer to obtain the temperature change rate time-series pattern feature vector and the rotational speed time-series pattern feature vector, respectively. The timing pattern conversion unit is used to multiply the temperature change rate timing pattern feature vector and the speed timing pattern feature vector by position to obtain the temperature change timing pattern feature vector under the influence of operating conditions. The feature vector decoding unit is used to decode the feature vector of the temperature change time sequence pattern under the influence of the operating condition to obtain the temperature change rate under the influence of the operating condition.

Images