Wind power PLC data preprocessing method and system based on edge computing

Abstract

The present application relates to the field of wind power generation and industrial data processing, and provides a wind power PLC data preprocessing method and system based on edge computing, which first analyzes the local original PLC data stream on the edge side of the wind turbine generator set, extracts key parameters to generate a local working condition abstract; then, a hybrid similarity score is used to dynamically select a candidate unit from the wind farm that is highly consistent with the current operating state of the local unit to construct a final peer group, and establish a precise local comparison benchmark; in view of the interference of the wake effect on similarity evaluation, an aerodynamic model is introduced to quantify the wind speed loss generated by the upstream unit, and the similarity is calculated based on the intrinsic performance after the wake compensation, so as to eliminate the false deviation caused by the difference in physical environment; finally, the relative abnormal score of the local unit is calculated based on the statistical characteristics of the final peer group and is packaged and output. This realizes the change from extensive macro statistics to fine intrinsic performance comparison, and improves the sensitivity and reliability of early fault detection of the wind turbine.

Description

Technical Field

[0001] This invention relates to the field of wind power generation and industrial data processing technology, and in particular to a wind power PLC data preprocessing method and system based on edge computing. Background Technology

[0002] Modern large-scale wind turbines are equipped with complex sensor networks and programmable logic controllers, generating massive amounts of operational status data in real time. This data forms the basis for enabling turbine condition monitoring, fault diagnosis, and predictive maintenance. With the expansion of wind farms and the deepening of digital transformation, transmitting all high-frequency raw data to the cloud or central server for processing in real time faces enormous network bandwidth pressure and storage cost challenges.

[0003] Existing edge computing-based data processing methods still suffer from limitations in practical applications due to their coarse-grained processing. Current local processing solutions often tend to adopt a macro-statistical strategy at the entire wind farm level, directly comparing the operating parameters of a single wind turbine with the average or statistical baseline of the operating parameters of all wind turbines in the entire wind farm. This coarse comparison benchmark ignores the individual differences of each wind turbine in terms of manufacturing tolerances, aging, and specific operating environments. When a specific component of a single wind turbine shows early, weak signs of abnormality, these minute signals are easily smoothed or masked by the averaging of the entire field's data, making it difficult for the system to sensitively capture early fault characteristics. A deeper problem is that even when attempting to find wind turbines with similar operating conditions at the edge for comparison, existing technologies usually only rely on apparent operating parameters such as power and speed to determine similarity, ignoring the complex aerodynamic coupling relationships within the wind farm, especially the impact of wake effects. Wind turbines located in the wake region of upstream units experience physical interference with the wind speed and turbulence intensity they receive, resulting in lower performance compared to similar units in free flow. Due to the lack of quantification and compensation for the influence of this physical field, existing methods have difficulty distinguishing between performance differences caused by wake effects and actual equipment failures, resulting in biases in the constructed comparison benchmarks and thus affecting the accuracy and robustness of anomaly detection. Summary of the Invention

[0004] The present invention aims to solve at least one of the problems existing in the prior art, and provides a wind power PLC data preprocessing method and system based on edge computing.

[0005] One aspect of the present invention provides a wind power PLC data preprocessing method based on edge computing, comprising: Acquire the original PLC data stream of the wind turbine generator set; Extract the operating condition dimension parameters and monitoring target parameters from the original PLC data stream to obtain a summary of the operating condition of the machine; Dynamic peer grouping based on hybrid similarity scoring is performed on the local operating condition summary and candidate peer list to obtain the final peer group; Calculate the relative anomaly score between the final peer group and the machine condition summary; The relative anomaly score, the final peer group, and the local condition summary are encapsulated into a preprocessed peer comparison data packet.

[0006] Another aspect of the present invention provides a wind power PLC data preprocessing system based on edge computing, comprising: The raw PLC data stream acquisition module is used to acquire the original PLC data stream of the wind turbine generator set. The local operating condition summary acquisition module is used to extract operating condition dimension parameters and monitoring target parameters from the local PLC raw data stream to obtain a local operating condition summary. The dynamic peer group construction module is used to construct dynamic peer groups based on hybrid similarity scoring from the local operating condition summary and the candidate peer list to obtain the final peer group; The relative anomaly score calculation module is used to calculate the relative anomaly score between the final peer group and the local operating condition summary; The packet encapsulation module is used to encapsulate the relative anomaly score, the final peer group, and the local condition summary into preprocessed peer comparison packets.

[0007] Compared with existing technologies, this invention first parses the original PLC data stream of the wind turbine generator at its edge side, extracting key parameters to generate a summary of the generator's operating conditions. Next, to address the problem of weak fault characteristics being masked by field-wide aggregation, a hybrid similarity scoring method is used to dynamically select candidate turbines within the wind farm that are highly consistent with the generator's current operating state, constructing a final peer group and establishing a precise local comparison benchmark. To address the interference of wake effects on similarity assessment, an aerodynamic model is introduced to quantify the wind speed loss generated by upstream turbines. Similarity is calculated based on the intrinsic performance after wake compensation, thereby eliminating spurious biases caused by differences in the physical environment. Finally, the relative anomaly score of the generator is calculated based on the statistical characteristics of the high-purity peer group, i.e., the final peer group, and then packaged and output. This achieves a shift from coarse macroscopic statistics to refined intrinsic performance comparison, improving the sensitivity and reliability of early fault detection for wind turbines. Attached Figure Description

[0008] One or more embodiments are illustrated by way of example with the corresponding pictures in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0009] Figure 1 This is a flowchart of a wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the data flow in the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the process of extracting operating condition dimension parameters and monitoring target parameters from the original PLC data stream to obtain a local operating condition summary, according to an embodiment of the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 4 The flowchart illustrates the dynamic peer group construction based on hybrid similarity scoring of the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention to obtain the final peer group. Figure 5 This is a flowchart illustrating the calculation of the relative anomaly score between the final peer group and the local operating condition summary in the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 6 This is a block diagram of a wind power PLC data preprocessing system based on edge computing according to an embodiment of the present invention. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and with various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0011] As indicated in the specification and claims of this invention, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0012] While this invention makes various references to certain modules in systems according to embodiments of the invention, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0013] This invention uses flowcharts to illustrate the operations performed by the system according to embodiments of the invention. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps can be processed in reverse order or simultaneously, as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0014] Existing edge processing technologies for wind power programmable logic controllers (PLCs) often employ coarse-grained, field-wide aggregation statistics. This macroscopic comparison benchmark can easily mask early, subtle fault characteristics of specific components in a single wind turbine and ignore the interference of complex physical fields such as wake effects on turbine performance, resulting in insufficient sensitivity and accuracy in anomaly detection. Therefore, this invention proposes a wind power PLC data preprocessing method based on edge computing. This method aims to achieve on-demand allocation of computing resources through intelligent assessment of system status. Specifically, this invention first parses and aggregates massive amounts of high-frequency raw PLC data streams at the edge of the wind turbine generator set, extracting key operating parameters and monitoring targets to generate a structured summary of the generator's operating conditions. Then, instead of simply relying on apparent parameters, it dynamically filters candidate peers based on a hybrid similarity score. In this process, an aerodynamic model is specifically introduced to quantify the wake effect generated by the upstream unit. By calculating the intrinsic performance similarity after wake compensation, interference from differences in the physical environment is eliminated, thereby constructing a final peer group with highly consistent intrinsic health states. Finally, a local benchmark is established based on the statistical characteristics of this final peer group, and the standardized relative anomaly score of the monitored target is calculated. This relative anomaly score and related context are encapsulated into a preprocessed data packet, thus realizing the transformation from data to high-value fault symptoms at the edge, improving the detection rate of minor faults.

[0015] Figure 1 This is a flowchart of a wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the data flow in a wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 1 and Figure 2According to an embodiment of the present invention, a wind power PLC data preprocessing method based on edge computing includes the following steps: S100, acquiring the local raw PLC data stream of the wind turbine generator set; S200, extracting operating condition dimension parameters and monitoring target parameters from the local raw PLC data stream to obtain a local operating condition summary; S300, constructing a dynamic peer group based on a hybrid similarity score from the local operating condition summary and the candidate peer list to obtain a final peer group; S400, calculating the relative anomaly score between the final peer group and the local operating condition summary; S500, encapsulating the relative anomaly score, the final peer group, and the local operating condition summary into a data packet to obtain a preprocessed peer comparison data packet.

[0016] Specifically, in step S100, the original PLC data stream of the wind turbine generator set is acquired. It is understood that modern wind turbine generator sets continuously generate massive amounts of multi-source heterogeneous monitoring data during operation, especially high-frequency signals involving drivetrain vibration and instantaneous power fluctuations, which have extremely high time resolution and data throughput. If all the original data is directly transmitted in real-time to a remote server for centralized processing, it not only faces a huge network bandwidth bottleneck but also inevitably introduces transmission delays, leading to the failure to capture transient fault characteristics. Therefore, in the technical solution of this invention, a data channel is directly established at the edge of the wind turbine nacelle or tower base to acquire the uncompressed and unthinned original PLC data stream of the wind turbine generator set. This ensures that subsequent processing stages can be analyzed based on full, high-fidelity physical layer data. This maximizes the preservation of original waveform details and timing correlation information reflecting early, subtle faults in the equipment, providing a lossless data foundation for high-precision condition monitoring, while avoiding the latency and packet loss risks associated with long-distance data transmission.

[0017] More specifically, in a specific example of the present invention, during step S100, the main control system of the wind turbine generator is directly connected using an industrial fieldbus protocol or Ethernet communication interface to read the numerical changes in various register addresses and sensor feedback signals in real time. This data acquisition process encompasses parallel acquisition and time synchronization of signals at different sampling frequencies, including but not limited to rapidly changing dynamic operating parameters such as generator speed and grid frequency sampled at the millisecond level, and slowly changing thermodynamic state parameters such as gearbox oil temperature and ambient temperature sampled at the second or minute level. Each captured data packet strictly includes a generation timestamp accurate to the millisecond, a unique sensor identifier, and the corresponding physical measurement value, thus forming a continuous time-series data stream. For example, for the vibration acceleration signal of the main bearing, it is continuously read at a preset high-frequency sampling rate, while simultaneously reading the current yaw angle and pitch angle settings, ensuring that high-frequency data reflecting the mechanical vibration characteristics of the equipment and low-frequency data reflecting the operating background are completely captured under the same time reference and temporarily stored in a high-speed memory buffer on the edge side for subsequent real-time parsing.

[0018] Specifically, in step S200, operating condition dimension parameters and monitoring target parameters are extracted from the original PLC data stream to obtain a local operating condition summary. It is understood that the original PLC data stream contains massive amounts of heterogeneous data with different sampling frequencies, and is mixed with instantaneous high-frequency noise. Directly using the full amount of original data for real-time interaction and similarity matching across units would consume enormous network bandwidth and edge computing resources, resulting in processing delays that would be difficult to meet the timeliness requirements of online monitoring. Therefore, in the technical solution of this invention, operating condition dimension parameters, such as wind speed, active power, and pitch angle, which determine the unit's operating environment and load level, and monitoring target parameters, such as main bearing temperature or gearbox oil temperature, which characterize the health status of key components, are further precisely screened and extracted from the continuous data stream. These parameters are then time-aligned and aggregated to form a structured local operating condition summary, thereby generating a lightweight and high-information-density unit status description object for subsequent broadcasting and exchange within the wind farm's local area network. This enables significant dimensionality reduction and feature extraction of data at the edge, greatly reducing network transmission load while providing a unified, standardized, and physically meaningful efficient comparison benchmark for quickly and accurately identifying peers with highly consistent operating backgrounds among numerous units.

[0019] Figure 3 This is a flowchart illustrating the process of extracting operating condition dimension parameters and monitoring target parameters from the original PLC data stream to obtain a local operating condition summary, according to an embodiment of the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 3As shown, step S200 includes: S210, parsing the original PLC data stream based on the parameter configuration table to obtain a mapping table for the instant value buffer and the time series buffer; S220, performing sliding time window parameter aggregation calculation on the original PLC data stream based on the query aggregation function and window size in the parameter configuration table to obtain an aggregated value mapping table; S230, encapsulating the instant value buffer, the aggregated value mapping table, the local identity identifier, and the current timestamp into a working condition summary to obtain a local working condition summary.

[0020] In step S210, the original PLC data stream is parsed based on the parameter configuration table to obtain the instant value buffer and time series buffer mapping table. It is understood that the original PLC data stream is a continuous and unclassified mixed sequence, containing both discrete parameters that only require attention to the current state and continuous parameters that need to be combined with historical intervals to calculate statistical characteristics. Without pre-processing logical splitting and targeted caching, subsequent operation condition summary generation will face problems of chaotic data retrieval and low computational efficiency. Therefore, in the technical solution of this invention, the accessed original PLC data stream is further parsed frame by frame according to the preset parameter configuration table, routing different types of monitoring data to the instant value buffer or time series buffer mapping table respectively, thereby constructing a dedicated data storage structure for different processing needs. This provides a clear and efficient data foundation for subsequent fast snapshot extraction and sliding window statistical calculations, ensuring the orderliness and real-time performance of edge computing resources when processing high-throughput data.

[0021] More specifically, in a specific example of the present invention, the parsing process in step S210 first establishes or loads a parameter configuration table that defines in detail the identifiers of each sensor and their corresponding processing attributes. This parameter configuration table explicitly specifies that parameters such as the unit operating mode status word and emergency stop signal belong to the instant update category, while parameters such as wind turbine speed, generator active power, and ambient temperature belong to the time-series accumulation category. As each data point in the original PLC data stream arrives, attribute matching is performed in the parameter configuration table based on its unique identifier. For parameters marked as instant updates, their latest measured values ​​are directly written into the instant value buffer to maintain the real-time nature of the data by overwriting the original values, ensuring that a snapshot of the unit's current instantaneous state can be obtained at any time when the buffer is read. For parameters marked as time-series accumulation, a tuple containing the timestamp and the measured value is appended to the corresponding queue structure in the time series buffer mapping table, thereby forming a historical data sequence with time depth, preserving a complete time-series context for subsequent statistical operations such as mean, variance, or trend based on a sliding time window.

[0022] In step S220, based on the query aggregation function and window size in the parameter configuration table, a sliding time window parameter aggregation calculation is performed on the original PLC data stream to obtain an aggregated value mapping table. It is understood that the original monitoring data of wind turbine generators, such as wind speed and power, have extremely high instantaneous fluctuations and random noise. Sampled values ​​at a single moment often fail to accurately reflect the stable operating conditions of the unit over a period of time, and direct use for state comparison can easily lead to misjudgments. Therefore, in the technical solution of this invention, a sliding time window parameter aggregation calculation is further performed on the original PLC data stream temporarily stored in the time series buffer mapping table based on the predefined query aggregation function and window size in the parameter configuration table to construct an aggregated value mapping table. This smooths the high-frequency fluctuating data and extracts statistically representative feature indicators. This effectively eliminates the interference of transient noise and generates robust parameters such as minute-level average wind speed or power standard deviation that accurately characterize the macroscopic operating state of the unit, providing high signal-to-noise ratio data support for subsequently finding comparable wind turbines with similar operating conditions in the wind farm.

[0023] More specifically, in a specific example of the present invention, step S220 is driven by periodic calculation logic. First, it iterates through each entry to be processed in the parameter configuration table, reads its associated aggregation algorithm type (e.g., arithmetic mean, standard deviation calculation, or extreme value extraction), and the corresponding time window length (e.g., an observation period of sixty seconds). Then, based on the current timestamp, it backtracks from the time series buffer mapping table and extracts all historical data points falling within the time window range, forming a time series data slice to be processed. Next, this time series data slice is input into a specified aggregation function for calculation, for example, calculating the arithmetic mean of the wind speed data series within the window period to characterize the environmental flow field intensity, or calculating the standard deviation of the power data series to characterize the output stability. After the calculation is completed, a key-value mapping relationship is established between the generated statistical result value and its corresponding parameter name, and updated in the aggregation value mapping table, thereby completing the transformation from discrete high-frequency raw data streams to high-level state characteristic values.

[0024] In step S230, the instant value buffer, aggregated value mapping table, local identifier, and current timestamp are encapsulated into a working condition summary to obtain a local working condition summary. It is understood that monitoring parameters scattered across different buffers lack a unified semantic association and a clear spatiotemporal context. If these parameters are transmitted independently in a loose form to the wind farm communication network, the receiving edge devices will face high parsing complexity and timing matching errors when aligning multi-machine data. Therefore, in the technical solution of this invention, the instant value buffer, aggregated value mapping table, local identifier, and current timestamp are further encapsulated into a working condition summary to obtain a local working condition summary. This integrates multi-source data with different physical meanings but synchronized time into a standardized information object with atomicity and self-description. This ensures that subsequent data packets transmitted in the network not only carry complete operating status information but also strictly bind the source location and exact time of data generation, thus providing a reliable data carrier for accurate working condition matching and difference analysis of other units in the wind farm.

[0025] More specifically, in a specific example of the present invention, the encapsulation process in step S230 first instantiates a structured data container conforming to a preset communication protocol standard in memory, such as a JSON object or a compact binary structure; then, the local identity identifier embedded in the device configuration and the current timestamp generated by the system clock are written into the header metadata field of the structured data container to establish the spatiotemporal coordinates of the data; subsequently, according to the classification definition in the parameter configuration table, the instantaneous value buffer and the aggregated value mapping table are traversed in parallel, mapping parameters representing the current operating environment, such as average wind speed and average active power, to the operating condition dimension field of the structured data container, while mapping parameters representing the health status of key components, such as real-time bearing temperature, to the monitoring target field of the structured data container. Through this structured assembly operation, the originally discrete values ​​are given a hierarchical logical relationship, ultimately generating a local operating condition summary that can be immediately used for network broadcasting, realizing the form transformation from low-level data fragments to high-level application objects.

[0026] Specifically, in step S300, a dynamic peer group is constructed based on a hybrid similarity score for the local operating condition summary and the candidate peer list to obtain the final peer group. It is understandable that in a wind farm environment, existing methods for constructing dynamic peer groups for relative anomaly detection suffer from a physical blindness problem, meaning they fail to fully consider the strong aerodynamic coupling relationships between wind turbine generators. Specifically, existing schemes directly compare the apparent operating condition parameters of each unit when assessing wind turbine similarity, neglecting the most critical physical phenomenon in the field—the wake effect. For example, a healthy wind turbine located in the wake region of an upstream unit will naturally receive lower wind speeds and increased turbulence, resulting in predictably lower power output and other performance indicators compared to similar units in free flow. If judged solely based on raw data, existing methods would incorrectly interpret this difference, determined by physical laws, as a dissimilar operating condition, thus excluding truly comparable peers and leading to a systematic bias in the similarity measurement standard. Meanwhile, traditional technologies use coarse, discontinuous, discrete weights to measure the similarity of static attributes such as physical distance between units, losing fine information about spatial relationships. Furthermore, by combining dynamic operating conditions and static attributes in a rigid linear manner, they fail to capture the complex nonlinear coupling relationship between the two. This makes it difficult to accurately identify true similar units in complex actual flow field environments, seriously affecting the sensitivity and reliability of subsequent anomaly detection.

[0027] Therefore, in the technical solution of this invention, a peer group similarity calculation method based on wake compensation and spatial continuity is further applied. This method dynamically constructs peer groups based on a hybrid similarity score for the machine's operating condition summary and candidate peer list to obtain the final peer group. This transforms the comparison from apparent operating conditions to comparing intrinsic performance after wake compensation. Instead of directly comparing the wind turbine's surface operating parameters, it first uses an aerodynamic model to remove the influence of the wake effect, and then performs a fair comparison under a unified, interference-free virtual benchmark. This ensures that even if the wind turbine is in a complex turbulent or wake-obstructed area, a true healthy peer can still be found based on the restored physical essence, eliminating comparison errors caused by environmental interference.

[0028] Figure 4 This is a flowchart illustrating the dynamic peer group construction based on hybrid similarity scoring of the wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention, to obtain the final peer group. Figure 4As shown, step S300 includes: S310, based on the wind turbine layout database and wake model parameters, quantifying the wake field influence of candidate peers on the local operating condition summary and the candidate peer list to obtain a wind speed loss mapping table; S320, based on the wind speed loss mapping table and the wind turbine layout database, calculating the wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list to obtain an intrinsic similarity mapping table; S330, performing weighted sorting and final determination of peer groups on the intrinsic similarity mapping table to obtain the final peer group.

[0029] In step S310, based on the wind turbine layout database and wake model parameters, the influence of the wake field of the candidate peers is quantified on the local operating condition summary and the candidate peer list to obtain a wind speed loss mapping table. It is understandable that the wake field influence of the candidate peers is quantified first because in the complex wind farm flow field, this key physical interference factor of the wake must be quantified first to provide a basis for compensation in subsequent fair comparisons. If this physical mechanism is ignored and the wind turbine is treated as an isolated data point, the perspective of placing the wind turbine in an mutually influencing physical flow field will be missing, leading to downstream units blocked by the wake being misjudged as having abnormal performance due to natural wind speed attenuation. Therefore, in the technical solution of this invention, the influence of the wake field of the candidate peers is further quantified on the local operating condition summary and the candidate peer list based on the wind turbine layout database and wake model parameters to obtain a wind speed loss mapping table, thereby embedding the aerodynamic model into the peer group construction algorithm for the first time. This allows a key physical compensation factor to be generated for each candidate wind turbine. This physical compensation factor accurately quantifies the degree of pollution of the wind resources it receives due to the wake effect, thereby ensuring that subsequent similarity assessments are based on the true physical state after removing environmental interference.

[0030] More specifically, in a concrete example of the present invention, for any wind turbine in a wind farm as a candidate peer, based on real-time wind direction across the entire farm, the geographical coordinates of each turbine, and thrust coefficient, a preset aerodynamic wake model is used to calculate the comprehensive wind speed loss caused by all upstream turbines. First, based on the current real-time wind direction data, the set of interfering turbines upstream of each candidate turbine is dynamically retrieved from the turbine layout database. For example, using a combination of the classic Katic and Jensen models, the candidate turbines can be calculated. Total wind speed loss rate That is, based on the wind turbine layout database and wake model parameters, the influence of the wake field of the candidate peers is quantified in the local operating condition summary and the candidate peer list to obtain the wind speed loss mapping table, including: quantifying the influence of the wake field of the candidate peers in the local operating condition summary and the candidate peer list using the following formula: ; in, Representative candidate wind turbine Total wind speed loss rate Representatives are located in the candidate wind turbines A collection of wind turbines upstream that have a wake effect on them; It is an upstream wind turbine The real-time thrust coefficient changes dynamically with the operating conditions of the upstream unit; It is an upstream wind turbine The rotor radius determines the initial width of the wake; It is the wake expansion coefficient, which describes the diffusion rate of the wake as it propagates downstream. It is an upstream wind turbine With candidate wind turbines The projected distance in the wind direction directly determines the degree of wake recovery when the wake reaches the downstream area. By traversing the candidate peer list and applying the above formula to each wind turbine to calculate the total wind speed loss rate, a mapping table containing the real-time wind speed loss rates of all candidate turbines—the wind speed loss mapping table—is generated. This provides accurate quantitative correction values ​​for intrinsic performance restoration in subsequent steps. For example, in a typical wind farm operation scenario, assuming the prevailing wind direction is northerly, turbine A05, located upstream, forms a direct wake obstruction relationship with turbine B08, located 500 meters directly downstream. At this time, real-time readings show that turbine A05 is operating at rated power with a high real-time thrust coefficient. Subsequently, the real-time thrust coefficient of Unit A05 and the projected distance between the two units were calculated. (i.e., 500 meters) and the preset wake expansion coefficient Substituting into the above formula, the wind speed loss component caused by Unit A05 to Unit B08 is quantified through calculation. If Unit B08 is also affected by part of the wake effect of Unit C12 to the side and in front, these multiple wake effects are superimposed through the summation term in the formula, finally yielding the total wind speed loss rate of Unit B08 at that moment. For example, a calculation result of 0.15 indicates a wind speed loss of 15%. This result is then updated in the wind speed loss mapping table, meaning that when evaluating whether Unit B08 is a qualified peer, it is already known that its input wind energy resources have been reduced by 15%, thus avoiding misjudging it as a power drop caused by a fault and ensuring the physical objectivity of subsequent intrinsic performance similarity calculations.

[0031] In step S320, based on the wind speed loss mapping table and the wind turbine layout database, the intrinsic performance similarity after wake compensation between the unit's operating condition summary and the candidate peer list is calculated to obtain an intrinsic similarity mapping table. It is understood that in actual wind farm operating environments, simply relying on direct comparison of operating data such as apparent power or rotational speed cannot eliminate the physical suppression of unit output performance by the wake effect. Furthermore, traditional discrete spatial weights are difficult to accurately characterize the complex spatial coupling relationships between units, leading to healthy units in the wake region being easily misjudged as having abnormal performance and excluded from the peer group. Therefore, in the technical solution of this invention, based on the wind speed loss mapping table generated in the previous steps and the static wind turbine layout database, a hybrid scoring model with embedded fluid dynamics principles is used to perform a deep calculation of the similarity between the unit's operating condition summary and the candidate peer list. This is used to calculate the intrinsic performance similarity after wake compensation, and based on the physical compensation factor obtained in the first step, a new similarity metric that reflects the unit's intrinsic performance and eliminates external flow field interference is constructed. In this way, environmentally disturbed observation data can be restored to indicators that reflect the essential health status of the equipment, ensuring that the system can penetrate the complex flow field fog and identify the reference unit that truly has physical homogeneity when constructing peer groups.

[0032] More specifically, in a particular example of the invention, step S320 aims to establish a fair virtual comparison benchmark through mathematical transformations. Specifically, for each candidate wind turbine... Calculate its relationship with the local machine Comprehensive intrinsic similarity score : Read local machine With candidate wind turbines The real-time measured power, combined with the physical distance between the two turbines and the total wind speed deficit rate calculated in the previous steps, is substituted into a preset hybrid similarity model for calculation. Specifically, it is obtained by multiplying the spatial proximity weight by the wake-compensated performance similarity. That is, based on the wind speed deficit mapping table and the turbine layout database, the wake-compensated intrinsic performance similarity between the turbine's operating condition summary and the candidate peer list is calculated to obtain the intrinsic similarity mapping table, including: The wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list is calculated using the following formula: ; in, For this machine With candidate wind turbines The intrinsic similarity score. It is the spatial proximity weight. This is the local machine. With candidate wind turbines The physical distance between them; It is the spatial attenuation coefficient, which replaces the original discrete weights with a continuous function. For example, when the distance between the local unit and the candidate wind turbine is extremely close, Approaching 1, it exhibits an exponentially smooth decay as the distance between the local unit and the candidate wind turbine increases, thus preserving fine gradient information in the spatial relationship. It is the performance similarity after wake compensation, where, and These are the local machines. Measured power and candidate wind turbines The measured power These are the candidate wind turbines calculated in the previous step. The total wind speed loss rate. In addition, This item, based on the physical principle that wind energy is proportional to the cube of wind speed, selects candidate wind turbines. Measured power affected by wake This allows us to deduce the equivalent free-flow power without wake interference. For example, if a candidate wind turbine is located downstream... Its actual measured power is only 1500kW, but it can withstand a 10% wind speed loss. This calculation allows us to restore it to a theoretical power of approximately 2057 kW. If the machine is located upstream of a free-flowing system at this point... The measured power of the unit is 2000kW. Therefore, the two units are highly similar under the restored virtual benchmark, thus being judged as having high similarity. This mechanism directly embeds fluid dynamics principles into the data processing algorithm, transforming a contaminated observation that cannot be directly compared into an intrinsic performance index that can be fairly compared with units in free flow. This results in a new intrinsic performance similarity score with clear physical meaning. This score integrates continuous spatial relationships and physically compensated performance, accurately measuring the similarity in the intrinsic health and performance of the two wind turbines, effectively avoiding misjudgments caused by differences in physical location. The intrinsic performance similarity scores between this unit and each candidate wind turbine constitute the intrinsic similarity mapping table.

[0033] In step S330, the intrinsic similarity mapping table is weighted, sorted, and the peer group is finally determined to obtain the final peer group. It is understood that simply obtaining the intrinsic performance similarity scores between the machine and each candidate wind turbine is insufficient to directly construct a reliable comparison benchmark. Directly introducing all candidate wind turbines or high-scoring candidate wind turbines with potential faults would contaminate the statistical characteristics of the benchmark value with noise, thereby reducing sensitivity to small deviations. Therefore, in the technical solution of this invention, the intrinsic similarity mapping table is further weighted, sorted, and the peer group is finally determined to obtain the final peer group, thus completing the closed loop from similarity measurement to final decision-making and constructing a high-purity peer group. This ensures that the members within the peer group not only have similar current operating conditions, but more importantly, their intrinsic performance state is highly consistent with the machine. This means that a small temperature rise of only 0.5 degrees Celsius caused by an early equipment fault will no longer be obscured by field-level averaging or erroneous operating condition comparisons, but will instead manifest as a deviation relative to the high-precision peer group benchmark, thus being sensitively detected.

[0034] More specifically, in a specific example of the present invention, step S330 sorts the candidate wind turbines in descending order according to the intrinsic performance similarity scores between the local turbine and each candidate wind turbine recorded in the intrinsic similarity mapping table, and selects the top N wind turbines to form the final peer group. For example, in a wind farm containing dozens of wind turbines, the five wind turbines with the highest intrinsic performance similarity scores are selected as members of the peer group to form a preliminary reference set, i.e., the peer group. Subsequently, as a measure to enhance robustness, the internal consistency verification of the target values ​​to be monitored for the selected peer group members can be performed to eliminate potential abnormal individuals within the group. This process involves statistical outlier detection on key parameters such as bearing temperature of the selected N wind turbines. If a wind turbine has a high intrinsic performance similarity score, but its actual monitored value has a significant statistical difference from other peer group members, it is determined that the wind turbine itself may have an anomaly, and it is removed from the peer group, thereby making the peer group after removing the wind turbine the final peer group. This rigorous screening mechanism enables the precise identification of a set of counterparts with highly consistent internal health status and performance for any wind turbine, regardless of whether it is in free flow or complex wake region, thus providing a high-precision benchmark for subsequent relative anomaly detection.

[0035] Specifically, in step S400, the relative anomaly score between the final peer group and the local operating condition summary is calculated. It is understood that individual physical monitoring values ​​are often subject to significant fluctuations due to drastic environmental conditions, and simple difference comparisons cannot quantify the statistical significance of deviations. Especially when the current group's dispersion cannot be determined, subtle fault characteristics can easily be confused with normal background noise. Therefore, in the technical solution of this invention, the relative anomaly score between the final peer group and the local operating condition summary is further calculated. This is used to construct a dynamic statistical baseline based on the real-time distribution characteristics of high-purity peer group members, and to transform the absolute value of the local monitored target into a standardized metric reflecting its deviation from the group. This eliminates common-mode interference caused by operating condition fluctuations, amplifies minor physical quantity changes caused by early faults into significant statistical anomaly signals, and thus achieves accurate assessment of equipment health status with high sensitivity and low false alarm rate at the edge.

[0036] Figure 5 This is a flowchart illustrating the calculation of the relative anomaly score between the final peer group and the local operating condition summary in a wind power PLC data preprocessing method based on edge computing according to an embodiment of the present invention. Figure 5 As shown, step S400 includes: S410, constructing a peer group statistical baseline model for the final peer group and the local operating condition summary based on the target parameter name to obtain the peer group statistical characteristics; S420, calculating the standardized relative anomaly score for each local target value to be monitored in the local operating condition summary based on the peer group statistical characteristics to obtain the relative anomaly score.

[0037] In step S410, based on the target parameter name, a statistical baseline model of the peer group is constructed for the final peer group and the machine's operating condition summary to obtain the statistical characteristics of the peer group. It is understood that although each member of the final peer group, i.e., each candidate wind turbine, is highly similar in macroscopic operating conditions and intrinsic performance, at the microscopic level, due to random factors such as manufacturing tolerances and wear, its monitoring data still exhibits a reasonable discrete distribution. If only simple numerical comparisons are performed while ignoring the statistical regularity of the group distribution, the boundaries of normal fluctuations cannot be defined, making it difficult to accurately quantify the severity of deviations. Therefore, in the technical solution of this invention, a statistical baseline model of the peer group is further constructed based on the target parameter name for the final peer group and the machine's operating condition summary to obtain the statistical characteristics of the peer group. This abstracts the discrete peer group observation samples into a probability distribution model with a central tendency and discrete metric description. This provides a dynamic reference benchmark containing uncertainty information for subsequent anomaly judgment, ensuring that the evaluation results reflect both the common behavior of the group and accommodate reasonable individual differences.

[0038] More specifically, in a concrete example of the present invention, the statistical baseline construction process in step S410 begins with the targeted extraction and cleaning of data samples. First, based on the specified target parameter name, such as main bearing temperature, the operating condition summary object of each member (i.e., each candidate wind turbine) in the final peer group is traversed, and the corresponding real-time physical measurement values ​​are indexed and extracted, thereby constructing a numerical vector reflecting the current state of the healthy group. While ensuring the validity of the data within the numerical vector, statistical moment calculations are performed on this numerical vector. On the one hand, the arithmetic mean of all sample values ​​is calculated as the mean, establishing the theoretical expected center of the parameter under the current operating condition; on the other hand, the standard deviation of the samples is calculated to quantify the fluctuation range and consistency level of the target parameter within the healthy group. Finally, the calculated mean and standard deviation are combined and encapsulated into a peer group statistical characteristic object. This object constitutes a floating scale used to measure whether the machine's state is abnormal at the current moment, clarifying the statistical boundary from normal to abnormal.

[0039] In step S420, based on the statistical characteristics of peer groups, the relative anomaly score is calculated for each monitored target value in the local operating condition summary to obtain a relative anomaly score. It is understood that the monitoring data of wind turbine generators is subject to significant dynamic changes due to environmental conditions, and the dispersion of healthy peer generator groups under different operating conditions is not constant. Simple numerical differences cannot objectively reflect the statistical significance of deviations, especially in extreme cases where the standard deviation approaches zero due to extremely high peer group consistency. Direct division can lead to numerical instability or even calculation overflow. Therefore, in the technical solution of this invention, based on the statistical characteristics of peer groups, the relative anomaly score is further calculated for each monitored target value in the local operating condition summary to obtain a relative anomaly score. This introduces a standardization algorithm with numerical clamping protection, transforming absolute physical deviations into dimensionless statistical indicators reflecting the degree of deviation from the group distribution. This effectively avoids the computational risk of a zero denominator, while shielding common-mode interference caused by fluctuations in operating conditions. It ensures that the generated abnormal scores have uniform dimensions and comparability under different speeds, power levels, and ambient temperatures, thereby achieving robust quantification of subtle fault symptoms.

[0040] More specifically, in a specific example of the present invention, the calculation process in step S420 is performed by an edge computing unit, aiming to evaluate the degree of deviation of the local state from the healthy group through mathematical statistics. First, the mean and standard deviation of the statistical characteristics of the peer group are read. Then, to prevent calculation divergence caused by excessively small standard deviations due to high data consistency among peer group members, a preset minimum noise threshold, i.e., a minimum standard deviation threshold, is introduced. The standard deviation is corrected, and the relative anomaly score is calculated using the corrected standard deviation. In other words, step S420 includes: calculating the standardized relative anomaly score for each monitored target value in the machine condition summary using the following formula: First, calculate the effective standard deviation using the following formula. : ; Next, the difference between the local observation and the population mean is standardized and scaled using this effective standard deviation, and the relative outlier score is calculated according to the following formula. : ; in, For relative outlier scores, and These are the mean and standard deviation, respectively, of the statistical characteristics of the equivalent groups. The target value to be monitored on this machine. The minimum standard deviation threshold, The standard deviation after threshold correction is the effective standard deviation. express Function. For example, in a scenario of monitoring the temperature of the rear bearing of a wind farm generator, assuming that the final matched groups selected under the current operating conditions exhibit extremely high consistency, the calculated mean... 60 degrees Celsius, standard deviation The temperature is only 0.1 degrees Celsius, while the preset minimum standard deviation threshold is... The value is 0.05. At this point, if the measured value of this machine is the same as the target value to be monitored by this machine... It is 60.5 degrees Celsius, although it is higher than A temperature 0.5 degrees Celsius above the threshold is easily overlooked in traditional absolute threshold monitoring, but the relative anomaly score calculated using this embodiment... The score will reach 5.0. This significant score of up to 5 standard deviations can keenly alert maintenance personnel that the bearing temperature of this machine has shown an unusual outlier trend relative to the current ultra-steady-state group, thus successfully capturing early overheating risks.

[0041] Specifically, in step S500, the relative anomaly score, the final peer group, and the local machine condition summary are encapsulated into data packets to obtain a preprocessed peer comparison data packet. It is understood that while a single relative anomaly score intuitively indicates the existence of a state deviation, without the contextual basis for generating the relative anomaly score—that is, without specific peer group composition information as a comparison benchmark and the detailed operating background of the local machine at the time—the remote monitoring center or cloud-based expert system will find it difficult to trace and verify the authenticity of the alarm, and will also be unable to support subsequent model optimization and fault reproduction. Therefore, in the technical solution of this invention, the relative anomaly score, the final peer group, and the local machine condition summary are further encapsulated into data packets to obtain a preprocessed peer comparison data packet, thereby constructing a self-contained, full-element, and traceable standardized diagnostic information unit. This allows for the complete preservation of the evidence chain used to interpret the judgment logic with extremely low network bandwidth usage, enabling maintenance personnel not only to know that an anomaly has occurred on the local machine, but also to instantly check under what operating conditions and relative to which groups the anomaly was exhibited to what extent, thereby significantly improving the efficiency and reliability of remote fault diagnosis.

[0042] More specifically, in a specific example of the present invention, the data packet encapsulation process in step S500 is executed by the protocol stack processing module of the edge device. First, it allocates a structured data container in memory that conforms to the Industrial Internet communication standard, for example, using a nested key-value pair structure. Then, it writes the standardized relative anomaly score calculated in the preceding steps into the core result field of this structured data container as the primary index for triggering graded alarms. Next, it completely embeds a local operating condition summary, including the local unique ID, precise timestamp, ambient wind speed, and power, into the context description field of the container to solidify the physical scenario at the time of data generation. Simultaneously, it serializes the list of all member unit IDs constituting the final peer group and their corresponding intrinsic similarity scores and fills them into the reference basis field, clearly recording the dynamic benchmark source for this comparison. Finally, it performs serialization encoding and checksum calculation on the assembled structured data container to generate a compact preprocessed peer comparison data packet, which is then sent to the wind farm control center or uploaded to the cloud database via the on-site data bus, completing the final delivery from edge computing results to high-value operation and maintenance knowledge.

[0043] In summary, the edge computing-based wind power PLC data preprocessing method according to embodiments of the present invention is explained. First, the original PLC data stream of the wind turbine generator is parsed at the edge side of the generator, extracting key parameters to generate a summary of the generator's operating conditions. Next, to address the problem of weak fault characteristics being masked by field-wide aggregation, a hybrid similarity scoring method is used to dynamically select candidate generators highly consistent with the generator's current operating state from within the wind farm, constructing a final peer group and establishing a precise local comparison benchmark. To address the interference of wake effects on similarity assessment, an aerodynamic model is introduced to quantify the wind speed loss generated by upstream generators, and similarity is calculated based on the intrinsic performance after wake compensation, thereby eliminating spurious biases caused by differences in the physical environment. Finally, the relative anomaly score of the generator is calculated based on the statistical characteristics of the high-purity peer group, i.e., the final peer group, and then packaged and output. This achieves a shift from coarse macroscopic statistics to refined intrinsic performance comparison, improving the sensitivity and reliability of early fault detection in wind turbines.

[0044] The present invention also provides a wind power PLC data preprocessing system based on edge computing.

[0045] Figure 6 This is a block diagram of a wind power PLC data preprocessing system based on edge computing according to an embodiment of the present invention. Figure 6 As shown, the wind power PLC data preprocessing system 100 based on edge computing according to an embodiment of the present invention includes: a raw PLC data stream acquisition module 110, used to acquire the local raw PLC data stream of the wind turbine generator; a local operating condition summary acquisition module 120, used to extract operating condition dimension parameters and monitoring target parameters from the local raw PLC data stream to obtain a local operating condition summary; a dynamic peer group construction module 130, used to construct a dynamic peer group based on a hybrid similarity score from the local operating condition summary and the candidate peer list to obtain a final peer group; a relative anomaly score calculation module 140, used to calculate the relative anomaly score between the final peer group and the local operating condition summary; and a data packet encapsulation module 150, used to encapsulate the relative anomaly score, the final peer group, and the local operating condition summary into a data packet to obtain a preprocessed peer comparison data packet.

[0046] For example, the local operating condition summary acquisition module 120 includes: The raw data stream parsing unit is used to parse the raw PLC data stream based on the parameter configuration table to obtain the instant value buffer and time series buffer mapping table. The sliding time window parameter aggregation unit is used to perform sliding time window parameter aggregation calculations on the original PLC data stream based on the query aggregation function and window size in the parameter configuration table to obtain an aggregated value mapping table. The operating condition summary encapsulation unit is used to encapsulate the instant value buffer, aggregate value mapping table, local identity identifier and current timestamp into an operating condition summary to obtain the local operating condition summary.

[0047] For example, the dynamic peer group building module 130 includes: The candidate peer wake field influence quantification unit is used to quantify the wake field influence of the candidate peers based on the wind turbine layout database and wake model parameters to obtain the wind speed loss mapping table. The intrinsic similarity mapping table acquisition unit is used to calculate the wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list based on the wind speed loss mapping table and the wind turbine layout database to obtain the intrinsic similarity mapping table. The weighted sorting and peer group final determination unit is used to perform weighted sorting and peer group final determination on the intrinsic similarity mapping table to obtain the final peer group.

[0048] For example, the relative anomaly score calculation module 140 includes: The peer group statistical baseline model construction unit is used to construct the peer group statistical baseline model based on the target parameter name, and to obtain the peer group statistical characteristics; The standardized relative anomaly score calculation unit is used to calculate the relative anomaly score for each monitored target value in the local machine operating condition summary based on the statistical characteristics of peer groups.

[0049] The specific implementation method of the wind power PLC data preprocessing system 100 based on edge computing provided in this embodiment of the invention can be found in the wind power PLC data preprocessing method based on edge computing provided in this embodiment of the invention, and will not be repeated here.

[0050] The edge computing-based wind power PLC data preprocessing system 100 according to embodiments of the present invention can be implemented in various types of computing devices or control units. For example, it can be a high-performance industrial personal computer deployed in the nacelle or tower base control cabinet of a wind turbine generator, an embedded controller serving as an edge computing gateway, or a programmable automation controller integrated into the wind turbine main control system. In one possible implementation, the edge computing-based wind power PLC data preprocessing system 100 according to embodiments of the present invention can be integrated into the computing device as a software module and / or a hardware module. For example, the edge computing-based wind power PLC data preprocessing system 100 can be a resident data processing service in the operating system of the computing device. This software module is configured to perform real-time parsing and operating condition summary generation of the native raw PLC data stream, dynamic peer group construction based on wake compensation model and hybrid similarity scoring, relative anomaly score calculation in response to peer group statistical characteristics, and preprocessing peer comparison data packet encapsulation containing context information. Alternatively, it can be a dedicated wind turbine edge-side intelligent monitoring and fault early warning algorithm program developed for the computing device. Of course, the wind power PLC data preprocessing system 100 based on edge computing can also be one of the many hardware modules of the computing device or control unit, or it can be embedded in the field programmable gate array circuit to accelerate the solution of the aerodynamic wake model and the large-scale peer similarity matching process in parallel, or it can be an industrial Internet of Things edge data analysis integrated circuit for a specific application.

[0051] Those skilled in the art will understand that the above embodiments are specific implementations of the present invention, and in practical applications, various changes can be made in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A wind power PLC data preprocessing method based on edge computing, characterized in that, include: Acquire the original PLC data stream of the wind turbine generator set; Extract operating condition dimension parameters and monitoring target parameters from the original PLC data stream to obtain a summary of the operating condition of the machine; Dynamic peer grouping based on hybrid similarity scoring is performed on the local operating condition summary and candidate peer list to obtain the final peer group; Calculate the relative anomaly score between the final peer group and the machine condition summary; The relative anomaly score, the final peer group, and the local condition summary are encapsulated into a preprocessed peer comparison data packet.

2. The wind power PLC data preprocessing method based on edge computing according to claim 1, characterized in that, Extract operating condition dimension parameters and monitoring target parameters from the original PLC data stream to obtain a local operating condition summary, including: Based on the parameter configuration table, the original PLC data stream is parsed to obtain the mapping table of instant value buffer and time series buffer; Based on the query aggregation function and window size in the parameter configuration table, a sliding time window parameter aggregation calculation is performed on the native raw PLC data stream in the time series buffer mapping table to obtain the aggregated value mapping table. The instant value buffer, aggregate value mapping table, local identity identifier and current timestamp are encapsulated into a working condition summary to obtain the local working condition summary.

3. The wind power PLC data preprocessing method based on edge computing according to claim 1, characterized in that, The local operating condition summary and candidate peer list are dynamically constructed based on a hybrid similarity score to obtain the final peer group, including: Based on the wind turbine layout database and wake model parameters, the influence of the wake field of the candidate peers is quantified in the local operating condition summary and candidate peer list to obtain the wind speed loss mapping table. Based on the wind speed loss mapping table and the wind turbine layout database, the wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list is calculated to obtain the intrinsic similarity mapping table. The intrinsic similarity mapping table is weighted and sorted, and the peer groups are finally determined to obtain the final peer groups.

4. The wind power PLC data preprocessing method based on edge computing according to claim 3, characterized in that, Based on the wind turbine layout database and wake model parameters, the influence of the wake field of candidate peers is quantified from the local operating condition summary and candidate peer list to obtain a wind speed loss mapping table, including: The impact of the wake field of the candidate equivalents is quantified using the following formula for the summary of the machine's operating conditions and the list of candidate equivalents: ； in, Representative candidate wind turbine Total wind speed loss rate Representatives are located in the candidate wind turbines The collection of wind turbines upstream that have a wake effect on them. It is an upstream wind turbine The real-time thrust coefficient, It is an upstream wind turbine rotor radius, It is the wake expansion coefficient. It is an upstream wind turbine With candidate wind turbines Projected distance in the wind direction.

5. The wind power PLC data preprocessing method based on edge computing according to claim 4, characterized in that, Based on the wind speed loss mapping table and the wind turbine layout database, the wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list is calculated to obtain the intrinsic similarity mapping table, including: The wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list is calculated using the following formula: ； in, For this machine With candidate wind turbines The intrinsic similarity score, This is the local machine. With candidate wind turbines The physical distance between them It is the spatial attenuation coefficient. and These are the local machines. Measured power and candidate wind turbines The measured power express function.

6. The wind power PLC data preprocessing method based on edge computing according to claim 1, characterized in that, Calculate the relative anomaly score between the final peer group and the machine condition summary, including: Based on the target parameter name, a peer group statistical baseline model is constructed for the final peer group and the local operating condition summary to obtain the peer group statistical characteristics. Based on the statistical characteristics of peer groups, the relative anomaly score is calculated by standardizing the relative anomaly score of each monitored target value in the local machine operating condition summary.

7. The wind power PLC data preprocessing method based on edge computing according to claim 6, characterized in that, Based on the statistical characteristics of peer groups, a standardized relative anomaly score is calculated for each monitored target value in the local machine operating condition summary to obtain a relative anomaly score, including: The standardized relative anomaly score for each monitored target value in the machine's operating condition summary is calculated using the following formula: ； ； in, For relative outlier scores, and These are the mean and standard deviation, respectively, of the statistical characteristics of the equivalent groups. The target value to be monitored on this machine. The minimum standard deviation threshold, For the effective standard deviation, express function.

8. A wind power PLC data preprocessing system based on edge computing, characterized in that, include: The raw PLC data stream acquisition module is used to acquire the original PLC data stream of the wind turbine generator set. The local operating condition summary acquisition module is used to extract operating condition dimension parameters and monitoring target parameters from the local PLC raw data stream to obtain a local operating condition summary. The dynamic peer group construction module is used to construct dynamic peer groups based on hybrid similarity scoring from the local operating condition summary and the candidate peer list to obtain the final peer group; The relative anomaly score calculation module is used to calculate the relative anomaly score between the final peer group and the local operating condition summary; The packet encapsulation module is used to encapsulate the relative anomaly score, the final peer group, and the local condition summary into preprocessed peer comparison packets.

9. The wind power PLC data preprocessing system based on edge computing according to claim 8, characterized in that, The local machine operating condition summary acquisition module includes: The raw data stream parsing unit is used to parse the raw PLC data stream based on the parameter configuration table to obtain the instant value buffer and time series buffer mapping table. The sliding time window parameter aggregation unit is used to perform sliding time window parameter aggregation calculations on the original PLC data stream based on the query aggregation function and window size in the parameter configuration table to obtain an aggregated value mapping table. The operating condition summary encapsulation unit is used to encapsulate the instant value buffer, aggregate value mapping table, local identity identifier and current timestamp into an operating condition summary to obtain the local operating condition summary.

10. The wind power PLC data preprocessing system based on edge computing according to claim 8, characterized in that, The dynamic peer group building module includes: The candidate peer wake field influence quantification unit is used to quantify the wake field influence of the candidate peers based on the wind turbine layout database and wake model parameters to obtain the wind speed loss mapping table. The intrinsic similarity mapping table acquisition unit is used to calculate the wake-compensated intrinsic performance similarity between the local operating condition summary and the candidate peer list based on the wind speed loss mapping table and the wind turbine layout database to obtain the intrinsic similarity mapping table. The weighted sorting and peer group final determination unit is used to perform weighted sorting and peer group final determination on the intrinsic similarity mapping table to obtain the final peer group.

Images