Digital-twin-oriented multi-source streaming data alignment and enhancement method for wind farm

Abstract

The application provides a wind farm multi-source stream data alignment and enhancement method for digital twinning, including: constructing a static physical topology graph based on the physical connection relationship between the collection devices in the target wind farm; obtaining multi-source stream data collected by the collection devices; dividing the multi-source stream data into multiple data units, adding a feature label to each data unit, and obtaining a feature data packet stream; inputting the feature data packet stream and the static physical topology graph into a pre-trained fusion alignment model, performing spatio-temporal fusion alignment in a sliding time window, and obtaining a global alignment state matrix on a unified time sequence; inputting the global alignment state matrix into a pre-trained enhancement model, performing online enhancement and repair of abnormal or missing data, and outputting enhanced synchronous data stream; and encoding the synchronous data stream to generate a serialized data packet stream for direct analysis by a digital twinning simulation system. The application can provide data support for a wind farm digital twinning system.

Description

Technical Field

[0001] This application relates to the field of wind farm digital twin technology, and in particular to a method, apparatus, equipment and storage medium for aligning and enhancing multi-source streaming data of wind farms for digital twins. Background Technology

[0002] The core challenge in building a high-fidelity digital twin system for wind farms lies in converting multi-source streaming data from heterogeneous acquisition devices such as wind turbines, transformer substations, and booster stations into a high-quality synchronous data stream that can directly drive the simulation model. Due to differences in sampling frequency, communication protocols, and time bases among various devices, the original data streams generally suffer from problems such as time asynchrony, data loss, and transient noise, which restrict the accuracy of subsequent simulation analysis and decision-making.

[0003] Existing multi-source data alignment methods often employ fixed-frequency resampling combined with interpolation to force data streams to align to a unified time grid. They then use fixed threshold detection, historical averages, or neighboring values ​​to fill in outliers and missing data. However, these methods only consider temporal proximity, ignoring electrical connections and causal constraints between devices. This results in a lack of physical consistency in synchronized data, easily leading to logically contradictory simulation inputs. Furthermore, they struggle to distinguish between real fault transients and sensor noise, and cannot generate data consistent with physical coupling principles when data is missing for extended periods. Therefore, their application in strongly coupled wind farm systems presents significant limitations. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Firstly, this application proposes a method for aligning and enhancing multi-source streaming data in wind farms for digital twins. The method includes: constructing a static physical topology map based on the physical connection relationships between acquisition devices within a target wind farm; acquiring multi-source streaming data acquired by the acquisition devices; dividing the multi-source streaming data into multiple data units and adding feature labels to each data unit to obtain a feature-defined data packet stream; inputting the feature-defined data packet stream and the static physical topology map into a pre-trained fusion alignment model, performing spatiotemporal fusion alignment within a sliding time window to obtain a global alignment state matrix on a unified time series; inputting the global alignment state matrix into a pre-trained enhancement model to perform online enhancement and repair of abnormal or missing data, and outputting an enhanced synchronous data stream; encoding the synchronous data stream to generate a serialized data packet stream that can be directly parsed by a digital twin simulation system.

[0006] In one implementation, the fusion alignment model sequentially includes an input layer, a temporal encoding layer, a spatial graph attention layer, a temporal fusion layer, and an output layer. The input layer processes the characteristic data packet stream and the static physical topology graph, outputting a sequence of feature vectors for each aligned data packet. The temporal encoding layer encodes the event timestamp of each data packet into a time-position embedding vector, adds it to the feature vector, and outputs a node feature sequence with temporal information. The spatial graph attention layer calculates the attention weights between the node and its neighboring nodes at each target time based on the static physical topology graph for each node in the node feature sequence, and outputs a spatial feature vector aggregating neighbor information based on the attention weights. The temporal fusion layer concatenates and fuses the spatial feature vector corresponding to each node with its initial estimate obtained through Hermite interpolation, outputting a corresponding alignment state vector. The output layer arranges the alignment state vectors corresponding to all nodes in a unified time sequence and outputs the global alignment state matrix.

[0007] In one alternative implementation, the time coding layer uses sinusoidal position coding to map event timestamps into fixed-dimensional vectors, with the coding formula as follows:

[0008]

[0009]

[0010] in, This represents the relative position number of the event timestamp within the sliding time window. For dimension index and , The total dimension of the encoded vector. This represents the value of the 2d dimension in the position encoding vector. Indicates the position encoding vector of the first The value of the dimension.

[0011] In one alternative implementation, the attention weights are calculated using the following formula:

[0012] in, The attention weights, Represents a node The set of all neighboring nodes, This is an index variable for traversing neighboring nodes. For nodes At time 10:00, its neighboring nodes Attention score; The formula for calculating the attention score is:

[0013] in, The attention score is... For nodes At any moment eigenvectors, Neighboring nodes At its most recent valid data time eigenvectors, Encode the edge relationship vector. This is a time difference encoding vector. For learnable query transformation matrix, The key transformation matrix is ​​a learnable matrix. For learnable weight vectors, This represents a vector concatenation operation; The formula for calculating the spatial feature vector is as follows:

[0014] in, For nodes The corresponding spatial feature vector, is a learnable value transformation matrix.

[0015] The step of concatenating and fusing the spatial feature vector corresponding to the node with its preliminary estimate obtained through Hermite interpolation to output the corresponding alignment state vector includes: estimating the data stream corresponding to the node using Hermite interpolation to obtain a preliminary interpolation result; concatenating the preliminary interpolation result corresponding to the node with the aggregated neighbor feature vector output by the spatial graph attention layer, and calculating the alignment state vector through a fully connected layer; the concatenation process is represented as follows:

[0016] in, Let be the alignment state vector. For activation function, The output transformation matrix is ​​learnable. This represents a vector concatenation operation. It includes the corrected synchronized data values ​​and their corresponding confidence levels. The confidence levels are calculated based on the number of valid data points in the node's own data stream within the sliding time window, the reliability level in the data quality label, and the variance of the attention weight.

[0017] In one implementation, the enhancement model sequentially includes a conditional coding layer, a generator network, and a discriminator network in the order of data processing; wherein, the conditional coding layer takes a globally aligned state matrix as input, extracts the context information of each data segment as a conditional vector, and outputs a conditional code; the generator network takes a random noise vector as input. and condition vector The system generates a synchronization data segment that matches the conditions, using the real data segment or the synchronization data segment and the condition vector as input, and outputs the discrimination result.

[0018] In one implementation, encoding the synchronous data stream to generate a serialized data packet stream for direct parsing by the digital twin simulation system includes: performing lossy compression on continuously changing quantities in the synchronous data stream using differential encoding, and performing lossless compression on discrete state quantities using dictionary encoding to obtain encoded state vectors at each time point; encapsulating the encoded state vector at each time point into a corresponding custom UDP frame, each custom UDP frame containing a frame header synchronization word, a timestamp, a data body length, an encoded state vector, and a cyclic redundancy check code; and arranging all custom UDP frames in chronological order to form the serialized data packet stream.

[0019] Secondly, this application proposes a multi-source streaming data alignment and enhancement system for wind farms oriented towards digital twins. The system includes: a physical topology map construction module for constructing a static physical topology map based on the physical connection relationships between acquisition devices within the target wind farm; a data acquisition module for acquiring multi-source streaming data acquired by the acquisition devices; a data packet stream generation module for dividing the multi-source streaming data into multiple data units and adding feature labels to each data unit to obtain a feature-defined data packet stream; a state matrix generation module for inputting the feature-defined data packet stream and the static physical topology map into a pre-trained fusion alignment model, performing spatiotemporal fusion alignment within a sliding time window to obtain a global alignment state matrix on a unified time series; a data stream generation module for inputting the global alignment state matrix into a pre-trained enhancement model for online enhancement and repair of abnormal or missing data, outputting an enhanced synchronous data stream; and a data packet stream generation module for encoding the synchronous data stream to generate a serialized data packet stream for direct parsing by the digital twin simulation system.

[0020] In one implementation, the fusion alignment model sequentially includes an input layer, a temporal encoding layer, a spatial graph attention layer, a temporal fusion layer, and an output layer. The input layer processes the characteristic data packet stream and the static physical topology graph, outputting a sequence of feature vectors for each aligned data packet. The temporal encoding layer encodes the event timestamp of each data packet into a time-position embedding vector, adds it to the feature vector, and outputs a node feature sequence with temporal information. The spatial graph attention layer calculates the attention weights between the node and its neighboring nodes at each target time based on the static physical topology graph for each node in the node feature sequence, and outputs a spatial feature vector aggregating neighbor information based on the attention weights. The temporal fusion layer concatenates and fuses the spatial feature vector corresponding to each node with its initial estimate obtained through Hermite interpolation, outputting a corresponding alignment state vector. The output layer arranges the alignment state vectors corresponding to all nodes in a unified time sequence and outputs the global alignment state matrix.

[0021] In one alternative implementation, the time coding layer uses sinusoidal position coding to map event timestamps into fixed-dimensional vectors, with the coding formula as follows:

[0022]

[0023] in, This represents the relative position number of the event timestamp within the sliding time window. For dimension index and , The total dimension of the encoded vector. This represents the value of the 2d dimension in the position encoding vector. Indicates the position encoding vector of the first The value of the dimension.

[0024] In one alternative implementation, the attention weights are calculated using the following formula:

[0025] in, The attention weights, Represents a node The set of all neighboring nodes, This is an index variable for traversing neighboring nodes. For nodes At time 10:00, its neighboring nodes Attention score; The formula for calculating the attention score is:

[0026] in, The attention score is... For nodes At any moment eigenvectors, Neighboring nodes At its most recent valid data time eigenvectors, Encode the edge relationship vector. This is a time difference encoding vector. For learnable query transformation matrix, The key transformation matrix is ​​a learnable matrix. For learnable weight vectors, This indicates a vector concatenation operation.

[0027] The formula for calculating the spatial eigenvectors mentioned above can be expressed as:

[0028] in, For nodes The corresponding spatial feature vector, is a learnable value transformation matrix.

[0029] The step of concatenating and fusing the spatial feature vector corresponding to the node with its preliminary estimate obtained through Hermite interpolation to output the corresponding alignment state vector includes: estimating the data stream corresponding to the node using Hermite interpolation to obtain a preliminary interpolation result; concatenating the preliminary interpolation result corresponding to the node with the aggregated neighbor feature vector output by the spatial graph attention layer, and calculating the alignment state vector through a fully connected layer; the concatenation process is represented as follows:

[0030] in, Let be the alignment state vector. For activation function, The output transformation matrix is ​​learnable. This represents a vector concatenation operation. It includes the corrected synchronized data values ​​and their corresponding confidence levels. The confidence levels are calculated based on the number of valid data points in the node's own data stream within the sliding time window, the reliability level in the data quality label, and the variance of the attention weight.

[0031] In one implementation, the enhancement model sequentially includes a conditional coding layer, a generator network, and a discriminator network in the order of data processing; wherein, the conditional coding layer takes a globally aligned state matrix as input, extracts the context information of each data segment as a conditional vector, and outputs a conditional code; the generator network takes a random noise vector as input. and condition vector The system generates a synchronization data segment that matches the conditions, using the real data segment or the synchronization data segment and the condition vector as input, and outputs the discrimination result.

[0032] In one implementation, the data packet stream generation module can be used to: perform lossy compression on continuously changing quantities in the synchronous data stream using differential encoding, and perform lossless compression on discrete state quantities using dictionary encoding to obtain encoded state vectors at each time point; encapsulate the encoded state vector at each time point into a corresponding custom UDP frame, each custom UDP frame containing a frame header synchronization word, timestamp, data body length, encoded state vector, and cyclic redundancy check code; and arrange all custom UDP frames in chronological order to form the serialized data packet stream.

[0033] Thirdly, this application provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described in the first aspect.

[0034] Fourthly, this application proposes a storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the method described in the first aspect.

[0035] Fifthly, this application proposes a program product comprising at least one of a program and instructions, wherein when the program or instructions are executed by an electronic device, they implement the steps of the method described in the first aspect.

[0036] This application provides a method, apparatus, equipment, and storage medium for aligning and enhancing multi-source streaming data in wind farms for digital twins. It constructs a static physical topology map based on the physical connections between data acquisition devices such as wind turbines, transformer substations, and booster stations within the target wind farm. Multi-source streaming data from each device is simultaneously acquired and divided into independent data units. Each data unit is appended with a timestamp, source device identifier, semantic tag, and data quality tag to form a characteristic data packet stream. The characteristic data packet stream and the static physical topology map are then input into a pre-trained fusion alignment model. Spatiotemporal fusion alignment is completed within a sliding time window, resulting in a global alignment state matrix under a unified time series. An enhanced model then performs online enhancement and repair on abnormal and missing data segments, outputting an enhanced synchronous data stream. Finally, the synchronous data stream is compressed and encapsulated into a custom UDP frame sequence using a differentiated encoding strategy, generating a serialized data packet stream that can be directly parsed by the digital twin simulation system. This solution ensures the physical consistency of synchronous data, improves robustness in scenarios with data anomalies and missing data, and provides high-precision, high-reliability data support for wind farm digital twin systems.

[0037] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0038] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for aligning and enhancing multi-source streaming data in a wind farm for digital twins, provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a wind farm multi-source streaming data alignment and enhancement system for digital twins provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0039] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0040] The following describes, with reference to the accompanying drawings, a method and system for aligning and enhancing multi-source streaming data of wind farms for digital twins, according to embodiments of this application.

[0041] Figure 1This is a flowchart illustrating a method for aligning and enhancing multi-source streaming data in wind farms for digital twins, provided in an embodiment of this application. Figure 1 As shown, the method may include, but is not limited to, the following steps: S1: Construct a static physical topology map based on the physical connection relationship between the data acquisition devices in the target wind farm.

[0042] In the embodiments of this application, the above-mentioned data acquisition equipment includes at least one of the following: a fan, a transformer substation, and a booster station; the physical connection relationship between the data acquisition equipment includes at least one of the following: the electrical connection relationship between the fan and the corresponding transformer substation, the power transmission connection relationship between the transformer substation and the booster station, and the electrical topology relationship between the various devices inside the booster station.

[0043] For example, physical entities with data acquisition capabilities, such as wind turbines, transformer substations, and booster stations within the target wind farm, are used as topology nodes. Each node is associated with rated parameters such as equipment type, rated voltage, capacity, and power, as well as the spatial location information of the equipment. Based on the electrical connection relationship between the wind turbine and the corresponding transformer substation, the power transmission connection relationship between the transformer substation and the booster station, and the electrical topology relationship between the various devices within the booster station, topology edges are established between nodes with physical connections. A connection type label is added to each edge to identify whether it is a direct electrical connection, an indirect electrical connection, or a non-electrical connection, thus constructing a static physical topology map that reflects the physical connection relationship between the data acquisition devices in the wind farm.

[0044] S2: Acquire multi-source streaming data collected by the acquisition device.

[0045] For example, the data collected from the wind turbine includes some or all of the following: wind speed, wind direction, impeller speed, generator speed, vibration acceleration, nacelle temperature, gearbox oil temperature, generator winding temperature, stator voltage, stator current, active power, reactive power, wind turbine operating status code, and fault code.

[0046] The data collected from the transformer substation includes some or all of the following: high-voltage side line voltage, high-voltage side phase current, high-voltage side active power, high-voltage side reactive power, low-voltage side line voltage, low-voltage side phase current, low-voltage side active power, low-voltage side reactive power, oil temperature, oil level, circuit breaker opening and closing position signals, protection action event types, and protection action times.

[0047] The data collected from the substation includes circuit breaker opening and closing position signals, disconnector opening and closing position signals, bus voltage, line current, active power, reactive power, protection system SOE (Sequence of Events) records, and some or all of the voltage amplitude, voltage phase angle, current amplitude, current phase angle, and frequency from the synchronous phasor measurement data.

[0048] S3: Divide the multi-source streaming data into multiple data units and add feature labels to each data unit to obtain a feature-based data packet stream.

[0049] For example, based on multiple sets of continuous measurement data collected by each acquisition device, each set of measurement data and its corresponding acquisition time are divided into an independent data unit. Then, a corresponding event timestamp and a globally unique device identifier are added to each data unit. Semantic tags containing physical quantity names, dimensions and subsystems are matched from a preset semantic tag library according to the measurement type and acquisition device type. Data quality tags containing data validity identifiers, data source reliability levels and timestamp confidence are also labeled. Finally, all tagged data units are summarized in the stream form to obtain a characteristic data packet stream.

[0050] S4: Input the characteristic data packet stream and static physical topology graph into the pre-trained fusion alignment model, and perform spatiotemporal fusion alignment within the sliding time window to obtain a global alignment state matrix on a unified time series.

[0051] In some embodiments, the fusion alignment model sequentially includes an input layer, a temporal encoding layer, a spatial graph attention layer, a temporal fusion layer, and an output layer. The input layer processes the characteristic data packet stream and the static physical topology graph, outputting a sequence of feature vectors for each aligned data packet. The temporal encoding layer encodes the event timestamp of each data packet into a time-position embedding vector, adds it to the feature vector, and outputs a node feature sequence with temporal information. The spatial graph attention layer calculates the attention weights between the node and its neighboring nodes at each target time based on the static physical topology graph for each node in the node feature sequence, and outputs a spatial feature vector aggregating neighbor information based on the attention weights. The temporal fusion layer concatenates and fuses the spatial feature vector corresponding to each node with its initial estimate obtained through Hermite interpolation, outputting the corresponding alignment state vector. The output layer arranges the alignment state vectors corresponding to all nodes in a unified time series, outputting a global alignment state matrix.

[0052] In some embodiments, the time coding layer uses sinusoidal position coding to map event timestamps into fixed-dimensional vectors, with the coding formula as follows:

[0053]

[0054] in, This represents the relative position number of the event timestamp within the sliding time window. For dimension index and , The total dimension of the encoded vector. This represents the value of the 2d dimension in the position encoding vector. Indicates the position encoding vector of the first The value of the dimension.

[0055] The attention weights of the temporal coding layer can be calculated using the following formula:

[0056] in, For attention weights, Represents a node The set of all neighboring nodes, This is an index variable for traversing neighboring nodes. For nodes At time 10:00, its neighboring nodes The attention score. The formula for calculating the attention score can be expressed as:

[0057] in, For attention score, For nodes At any moment eigenvectors, Neighboring nodes At its most recent valid data time eigenvectors, Encode the edge relationship vector. This is a time difference encoding vector. For learnable query transformation matrix, The key transformation matrix is ​​a learnable matrix. For learnable weight vectors, This represents a vector concatenation operation; The formula for calculating the spatial eigenvectors mentioned above can be expressed as:

[0058] in, For nodes The corresponding spatial feature vector, is a learnable value transformation matrix.

[0059] In some embodiments, the above-mentioned concatenation and fusion of the spatial feature vector corresponding to the node with its own preliminary estimate obtained by Hermite interpolation to output the corresponding alignment state vector includes: estimating the data stream corresponding to the node using Hermite interpolation to obtain a preliminary interpolation result; concatenating the preliminary interpolation result corresponding to the node with the aggregated neighbor feature vector output by the spatial graph attention layer, and calculating the alignment state vector through a fully connected layer.

[0060] For example, the above splicing process can be represented as follows:

[0061] in, To align the state vector, For activation function, The output transformation matrix is ​​learnable. This represents a vector concatenation operation. It includes the corrected synchronized data values ​​and their corresponding confidence levels. The confidence level is calculated based on the number of valid data points in the node's own data stream within the sliding time window, the reliability level in the data quality label, and the variance of the attention weight.

[0062] As an example, the formula for calculating the confidence level above can be expressed as:

[0063] in, The confidence score is... The number of valid data points. For the reliability level in the data quality label, Attention weights The variance is used to calculate the confidence score. , The preset first weighting coefficient, The second weighting coefficient is preset. This is the preset third weighting coefficient.

[0064] In some embodiments, the fusion alignment model is pre-trained using a self-supervised approach. The specific training process can be as follows: acquire the historical multi-source streaming data of the target wind farm, which includes the original measurement values ​​and corresponding acquisition timestamps of each acquisition device within a continuous historical time interval; randomly select some time points from the historical data stream of each acquisition device as masking points, remove the original measurement values ​​at the masking points, and retain the time location information and source device identifiers of the masking points to construct training samples; input the masked historical multi-source streaming data and the static physical topology map into the fusion alignment model, and have the model output the state prediction values ​​corresponding to each masking point under a unified time series; use a joint loss function composed of a weighted average squared error loss and a physical consistency loss to calculate the total loss between the predicted value and the original true measurement value of each masking point, perform backpropagation based on the total loss, and iteratively update the model network parameters using the gradient descent algorithm until the loss converges, thus obtaining the pre-trained fusion alignment model.

[0065] In some embodiments, the joint loss function described above can be expressed as:

[0066] in, For joint losses, For mean square error loss, For physical consistency loss, This is the balance coefficient; By fixing the model parameters, a pre-trained fusion alignment model is obtained.

[0067] In some specific embodiments, physical consistency loss Based on the power balance equation, the formula can be expressed as:

[0068] in, It is the set of edges in a static physical topology graph that have direct electrical connections. This represents the total number of edges in a static physical topology graph that have direct electrical connections. and The two ends of the edge are respectively and nodes At any moment The power value is extracted from the aligned state vector.

[0069] S5: Input the global alignment state matrix into the pre-trained augmentation model to perform online augmentation and repair of abnormal or missing data, and output the augmented synchronous data stream.

[0070] In some embodiments, the enhancement model includes a conditional coding layer, a generator network, and a discriminator network in the order of data processing. The conditional coding layer takes the globally aligned state matrix as input, extracts the context information of each data segment as a conditional vector, and outputs the conditional code. The generator network takes a random noise vector and a conditional vector as input and generates a synchronized data segment that matches the condition. The discriminator network takes a real data segment or a synchronized data segment and a conditional vector as input and outputs the discrimination result.

[0071] In some specific embodiments, during the pre-training process of the augmented model, the generator network and discriminator network The adversarial loss function is:

[0072]

[0073] in, For real data samples extracted from the globally aligned state matrix, This is the conditional vector output by the conditional coding layer. It is a random noise vector. For the true data distribution, For noise distribution, This indicates that the discriminator network is under conditional Make a judgment The probability of real data. This indicates that the generator network is under conditional Below is noise The generated data sample.

[0074] In some specific embodiments, after the augmentation model is pre-trained, the weights of the generator network are fixed. During online application, the augmentation model receives a global alignment state matrix. For data segments marked as anomalous or missing in the global alignment state matrix, the augmentation model inputs the context information of that data segment into the generator network to obtain the corresponding generated data segment. Here, data segments marked as anomalous or missing refer to data segments whose validity is invalid in the data quality label or whose timestamp confidence is below a preset threshold.

[0075] S6: Encode the synchronous data stream to generate a serialized data packet stream that can be directly parsed by the digital twin simulation system.

[0076] In one implementation, the synchronous data stream is encoded to generate a serialized data packet stream that can be directly parsed by the digital twin simulation system. This includes: performing lossy compression on continuously changing quantities in the synchronous data stream using differential encoding, and performing lossless compression on discrete state quantities using dictionary encoding to obtain encoded state vectors at each time point; encapsulating the encoded state vector at each time point into a corresponding custom UDP (User Datagram Protocol) frame, each custom UDP frame containing a frame header synchronization word, timestamp, data body length, encoded state vector, and cyclic redundancy check code; and arranging all custom UDP frames in chronological order to form a serialized data packet stream.

[0077] Specifically, in the process of generating serialized data packet streams, a differentiated encoding and compression strategy is adopted for different data attributes of the enhanced post-synchronization data stream: for analog data such as wind speed, active power, and winding temperature that change continuously over time, differential encoding is used for lossy compression. The difference between adjacent time points is used to replace the original absolute value, and the time correlation of continuous quantities is used to compress the data volume to reduce the transmission load; for discrete state quantities such as circuit breaker opening and closing positions, wind turbine operating status codes, and fault codes, a preset state mapping dictionary is used to map discrete state values ​​into shorter codes, reducing the data length while fully preserving the state information; after encoding processing, the encoded state vectors corresponding to each time point are obtained. Subsequently, the encoded state vector at each time point is encapsulated into a custom UDP frame structure. Each frame sequentially includes a frame header synchronization word, a timestamp, a data body length, an encoded state vector, and a cyclic redundancy check (CRC) code. The frame header synchronization word is used by the receiver to quickly identify the start boundary of the data frame. The timestamp corresponds to the unified alignment time of the data in that frame. The data body length indicates the byte length of the encoded state vector to support correct unpacking by the receiver. The CRC code is used by the receiver to verify the integrity of data transmission and avoid transmission errors from interfering with simulation accuracy. Finally, all the encapsulated custom UDP frames are arranged in chronological order according to their timestamps to obtain a serialized data packet stream that can be directly parsed and called by the digital twin simulation system with low latency.

[0078] By implementing the embodiments of this application, a static physical topology map can be constructed based on the physical connection relationships between data acquisition devices such as wind turbines, transformer substations, and booster stations within the target wind farm. Multi-source streaming data from each device is synchronously acquired and divided into independent data units. Each data unit is appended with a timestamp, source device identifier, semantic tag, and data quality tag to form a characteristic data packet stream. Subsequently, the characteristic data packet stream and the static physical topology map are input into a pre-trained fusion alignment model. Spatiotemporal fusion alignment is completed within a sliding time window, resulting in a global alignment state matrix under a unified time series. Then, a pre-trained enhancement model performs online enhancement and repair on abnormal and missing data segments, outputting an enhanced synchronous data stream. Finally, the synchronous data stream is compressed and encapsulated into a custom UDP frame sequence using a differentiated encoding strategy, generating a serialized data packet stream that can be directly parsed by a digital twin simulation system. This scheme ensures the physical consistency of synchronous data, improves robustness in scenarios with data anomalies and missing data, and provides high-precision, high-reliability data support for wind farm digital twin systems.

[0079] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a multi-source streaming data alignment and enhancement system for wind farms oriented towards digital twins, provided in an embodiment of this application. Figure 2As shown, the system 200 includes: a physical topology map construction module 201, used to construct a static physical topology map based on the physical connection relationship between acquisition devices within the target wind farm; a data acquisition module 202, used to acquire multi-source streaming data collected by the acquisition devices; a data packet stream generation module 203, used to divide the multi-source streaming data into multiple data units and add feature labels to each data unit to obtain a featured data packet stream; a state matrix generation module 204, used to input the featured data packet stream and the static physical topology map into a pre-trained fusion alignment model, and perform spatiotemporal fusion alignment within a sliding time window to obtain a global alignment state matrix on a unified time series; a data stream generation module 205, used to input the global alignment state matrix into a pre-trained enhancement model, perform online enhancement and repair of abnormal or missing data, and output an enhanced synchronous data stream; and a data packet stream generation module 206, used to encode the synchronous data stream to generate a serialized data packet stream for direct parsing by the digital twin simulation system.

[0080] In one implementation, the fusion alignment model sequentially comprises an input layer, a temporal encoding layer, a spatial graph attention layer, a temporal fusion layer, and an output layer. The input layer processes the characteristic data packet stream and the static physical topology graph, outputting a sequence of feature vectors for each aligned data packet. The temporal encoding layer encodes the event timestamps of each data packet into time-position embedding vectors, adding them to the feature vectors to output a node feature sequence with temporal information. The spatial graph attention layer calculates the attention weights between each node and its neighboring nodes at each target time based on the static physical topology graph, and outputs a spatial feature vector aggregating neighbor information based on these attention weights. The temporal fusion layer concatenates and fuses the spatial feature vector corresponding to each node with its initial estimate obtained through Hermite interpolation, outputting the corresponding alignment state vector. The output layer arranges the alignment state vectors corresponding to all nodes in a unified time sequence, outputting a global alignment state matrix.

[0081] In one alternative implementation, the time coding layer uses sinusoidal position coding to map event timestamps into fixed-dimensional vectors, with the coding formula as follows:

[0082]

[0083] in, This represents the relative position number of the event timestamp within the sliding time window. For dimension index and , The total dimension of the encoded vector. This represents the value of the 2d dimension in the position encoding vector. Indicates the position encoding vector of the first The value of the dimension.

[0084] In one alternative implementation, the attention weights are calculated using the following formula:

[0085] in, For attention weights, Represents a node The set of all neighboring nodes, This is an index variable for traversing neighboring nodes. For nodes At time 10:00, its neighboring nodes Attention score; The formula for calculating attention score can be expressed as:

[0086] in, For attention score, For nodes At any moment eigenvectors, Neighboring nodes At its most recent valid data time eigenvectors, Encode the edge relationship vector. This is a time difference encoding vector. For learnable query transformation matrix, The key transformation matrix is ​​a learnable matrix. For learnable weight vectors, This represents a vector concatenation operation; The formula for calculating spatial eigenvectors is:

[0087] in, For nodes The corresponding spatial feature vector, is a learnable value transformation matrix.

[0088] The spatial feature vector corresponding to a node is concatenated and fused with its initial estimate obtained through Hermite interpolation to output the corresponding alignment state vector. This process includes: estimating the data stream corresponding to the node using Hermite interpolation to obtain an initial interpolation result; concatenating the initial interpolation result corresponding to the node with the aggregated neighbor feature vector output by the spatial graph attention layer; and calculating the alignment state vector through a fully connected layer. The concatenation process is represented as follows:

[0089] in, To align the state vector, For activation function, The output transformation matrix is ​​learnable. This represents a vector concatenation operation. It includes the corrected synchronized data values ​​and their corresponding confidence levels. The confidence level is calculated based on the number of valid data points in the node's own data stream within the sliding time window, the reliability level in the data quality label, and the variance of the attention weight.

[0090] In one implementation, the augmentation model includes a conditional coding layer, a generator network, and a discriminator network in the order of data processing. The conditional coding layer takes the globally aligned state matrix as input, extracts the contextual information of each data segment as a conditional vector, and outputs the conditional code. The generator network takes a random noise vector as input. and condition vector The system takes the input of either the real data segment or the synchronized data segment and the condition vector as input, and outputs the discrimination result.

[0091] In one implementation, the data packet stream generation module can be used to: perform lossy compression on continuously changing quantities in the synchronous data stream using differential encoding, and perform lossless compression on discrete state quantities using dictionary encoding to obtain encoded state vectors at each time point; encapsulate the encoded state vector at each time point into a corresponding custom UDP frame, each custom UDP frame containing a frame header synchronization word, timestamp, data body length, encoded state vector, and cyclic redundancy check code; and arrange all custom UDP frames in chronological order to form a serialized data packet stream.

[0092] It should be noted that the foregoing explanation of the embodiment of the wind farm multi-source streaming data alignment and enhancement method for digital twins also applies to the wind farm multi-source streaming data alignment and enhancement system for digital twins in this embodiment, and will not be repeated here.

[0093] To implement the above embodiments, this application also proposes an electronic device. Please see [link to relevant documentation]. Figure 3 , Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of this application. For example... Figure 3 As shown, the electronic device 300 includes: a processor 301 and a memory 302 communicatively connected to the processor 301; the memory 302 stores computer execution instructions; the processor 301 executes the computer execution instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0094] To implement the above embodiments, this application also proposes a storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the methods provided in the foregoing embodiments.

[0095] To implement the above embodiments, this application also proposes a program product, including at least one of a program and instructions, wherein when the program and instructions are executed by an electronic device, they implement the steps of the method provided in the foregoing embodiments.

[0096] It should be noted that the acquisition, transmission, storage, use, and processing of data in this application comply with the relevant provisions of national laws and regulations and do not violate public order and good morals.

[0097] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, data stored, data displayed, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0098] It is worth noting that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0099] In the description of this application, unless otherwise stated, " / " means "or", for example, A / B can mean A or B; "and / or" in this document is merely a description of the relationship between related objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone.

[0100] In the foregoing descriptions of the embodiments, 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 this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0101] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0102] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0103] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0104] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0105] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0107] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for aligning and enhancing multi-source streaming data of wind farms for digital twins, characterized in that, include: A static physical topology map is constructed based on the physical connection relationships between the data acquisition devices within the target wind farm. Acquire the multi-source streaming data collected by the acquisition device; The multi-source streaming data is divided into multiple data units, and a feature label is added to each data unit to obtain a feature-based data packet stream; The characterized data packet stream and the static physical topology graph are input into a pre-trained fusion alignment model, and spatiotemporal fusion alignment is performed within a sliding time window to obtain a global alignment state matrix on a unified time series. The global alignment state matrix is ​​input into the pre-trained augmentation model to perform online augmentation and repair of abnormal or missing data, and the augmented synchronous data stream is output. The synchronous data stream is encoded to generate a serialized data packet stream that can be directly parsed by the digital twin simulation system.

2. The method according to claim 1, characterized in that, The fusion alignment model comprises, in sequence, an input layer, a temporal encoding layer, a spatial graph attention layer, a temporal fusion layer, and an output layer; The input layer is used to process the characteristic data packet stream and the static physical topology graph, and outputs an aligned feature vector sequence for each data packet. The time coding layer is used to encode the event timestamp of each data packet into a time location embedding vector, and after adding it to the feature vector, output a node feature sequence with time sequence information; The spatial graph attention layer is used to calculate the attention weights of the node with its neighboring nodes at each target time based on the static physical topology graph for each node in the node feature sequence, and output a spatial feature vector after aggregating neighbor information based on the attention weights. The temporal fusion layer is used to concatenate and fuse the spatial feature vector corresponding to the node with its own preliminary estimate obtained by Hermite interpolation for each node, and output the corresponding alignment state vector. The output layer arranges the alignment state vectors corresponding to all the nodes according to a unified time series and outputs the global alignment state matrix.

3. The method according to claim 2, characterized in that, The time coding layer uses sinusoidal position coding to map event timestamps into fixed-dimensional vectors. The coding formula is as follows: in, This represents the relative position number of the event timestamp within the sliding time window. For dimension index and , The total dimension of the encoded vector. This represents the value of the 2d dimension in the position encoding vector. Indicates the position encoding vector of the first The value of the dimension.

4. The method according to claim 2, characterized in that, The formula for calculating the attention weight is: in, The attention weights, Represents a node The set of all neighboring nodes, This is an index variable for traversing neighboring nodes. For nodes At time 10:00, its neighboring nodes Attention score; The formula for calculating the attention score is: in, The attention score is... For nodes At any moment eigenvectors, Neighboring nodes At its most recent valid data time eigenvectors, Encode the edge relationship vector. This is a time difference encoding vector. For learnable query transformation matrix, The key transformation matrix is ​​a learnable matrix. For learnable weight vectors, This represents a vector concatenation operation; The formula for calculating the spatial feature vector is as follows: in, For nodes The corresponding spatial feature vector, is a learnable value transformation matrix.

5. The method as described in claim 3, characterized in that, The step of concatenating and fusing the spatial feature vector corresponding to the node with its initial estimate obtained through Hermite interpolation to output the corresponding alignment state vector includes: Hermite interpolation is used to estimate the data stream corresponding to the node to obtain preliminary interpolation results; The preliminary interpolation result corresponding to the node is concatenated with the aggregated neighbor feature vector output by the spatial graph attention layer, and the alignment state vector is calculated through a fully connected layer; the concatenation process is represented as follows: in, Let be the alignment state vector. For activation function, The output transformation matrix is ​​learnable. This represents a vector concatenation operation. It includes the corrected synchronized data values ​​and their corresponding confidence levels. The confidence levels are calculated based on the number of valid data points in the node's own data stream within the sliding time window, the reliability level in the data quality label, and the variance of the attention weight.

6. The method as described in claim 1, characterized in that, The enhancement model, in the order of data processing, includes a conditional coding layer, a generator network, and a discriminator network. The conditional coding layer takes the global alignment state matrix as input, extracts the context information of each data segment as a conditional vector, and outputs the conditional code. The generator network uses a random noise vector and condition vector As input, generate a synchronized data segment that matches the conditions; The discriminator network takes the real data segment or the synchronous data segment and the condition vector as input and outputs the discrimination result.

7. The method as described in claim 1, characterized in that, The process of encoding the synchronous data stream to generate a serialized data packet stream that can be directly parsed by the digital twin simulation system includes: The continuously changing quantities in the synchronous data stream are compressed using differential coding with lossy compression, and the discrete state quantities are compressed using dictionary coding with lossless compression, to obtain the encoded state vectors at each time point. The encoded state vector at each time point is encapsulated into the corresponding custom UDP frame. Each custom UDP frame contains a frame header synchronization word, timestamp, data body length, encoded state vector, and cyclic redundancy check code. All custom UDP frames are arranged in chronological order to form the serialized data packet stream.

8. A wind farm multi-source streaming data alignment and enhancement system for digital twins, characterized in that, include: The physical topology map construction module is used to construct a static physical topology map based on the physical connection relationships between data acquisition devices within the target wind farm. The data acquisition module is used to acquire multi-source streaming data collected by the acquisition device; A data packet stream generation module is used to divide the multi-source streaming data into multiple data units and add feature tags to each data unit to obtain a feature-based data packet stream. The state matrix generation module is used to input the featured data packet stream and the static physical topology map into the pre-trained fusion alignment model, and perform spatiotemporal fusion alignment within a sliding time window to obtain a global alignment state matrix on a unified time series. The data stream generation module is used to input the global alignment state matrix into the pre-trained augmentation model, perform online augmentation and repair of abnormal or missing data, and output the augmented synchronous data stream. The data packet stream generation module is used to encode the synchronous data stream to generate a serialized data packet stream that can be directly parsed by the digital twin simulation system.

9. A storage medium storing instructions, characterized in that, When the instructions are executed on an electronic device, the electronic device causes the electronic device to perform the method of any one of claims 1 to 7.

10. A program product comprising at least one of a program and instructions, characterized in that, When at least one of the program or instructions is executed by an electronic device, it implements the steps of the method according to any one of claims 1 to 7.

Images