Abnormal operation data identification method and device of offshore wind plant, terminal equipment and storage medium

Abstract

The invention discloses an abnormal operation data identification method and device of an offshore wind power plant, terminal equipment and a storage medium, and belongs to the field of offshore wind power. The method comprises the following steps: constructing a low-altitude time sequence matrix according to low-altitude environment data and generated power data of the offshore wind power plant in a plurality of time periods to extract a first seasonal feature; and according to the high-altitude environment data and the generated power data, a high-altitude time sequence matrix is constructed to extract the second seasonal feature, so that the recognition accuracy of the abnormal generated power data is effectively improved. And according to the low-altitude environment data, the high-altitude environment data and the equipment operation data, constructing an operation environment matrix to extract global features, and in the abnormal data identification process, through combination of the first seasonal features and the second seasonal features, the abnormal equipment operation data can be effectively identified. The problem that abnormal offshore wind power abnormal operation data is difficult to correctly recognize at present is solved.

Description

Technical Field

[0001] The present invention relates to a field, and in particular to a method, device, terminal equipment and storage medium for identifying abnormal operation data of an offshore wind farm. Background Art

[0002] With the increasing global demand for clean energy, offshore wind power, as a highly efficient and clean energy source, is rapidly developing. Offshore wind power boasts 20%-40% higher efficiency than onshore wind power and is a key measure for reducing carbon emissions. Current offshore wind power forecasting, power generation planning, and equipment fault identification all rely on historical power generation and equipment operating parameters. Therefore, offshore wind power data quality requirements are extremely high. However, current data anomaly identification only identifies obvious anomalies, such as null or missing values. However, subtle anomalies can also occur, such as those caused by electromagnetic interference, communication delays, or wind farm curtailment. These outliers appear consistent with normal values, but do not conform to the expected overall distribution and correspondence between power generation and equipment operating parameters. Using this data directly without processing will severely impact subsequent tasks such as power forecasting, power generation planning, and fault identification.

[0003] However, since the power generation of offshore wind power is not only affected by wind speed, but is also closely related to environmental factors such as season, temperature and humidity, the power generation of offshore wind power is highly random and uncertain. These characteristics make it difficult to correctly identify abnormal offshore wind power operation data at present. Summary of the Invention

[0004] The present invention provides a method, device, terminal equipment and storage medium for identifying abnormal operation data of an offshore wind farm. The method can solve the current problem of difficulty in correctly identifying abnormal offshore wind farm operation data.

[0005] An embodiment of the present invention provides a method for identifying abnormal operation data of an offshore wind farm, comprising:

[0006] Obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of offshore wind farms in several time periods;

[0007] Constructing a low-altitude time series matrix based on the low-altitude environment data and the power generation data, constructing a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and constructing an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data;

[0008] Extracting, based on the low-altitude time series matrix, first seasonal features associated with the generated power data and low-altitude environmental data of the offshore wind farm; extracting, based on the high-altitude time series matrix, second seasonal features associated with the generated power data and high-altitude environmental data of the offshore wind farm; and extracting, based on the operating environment matrix, global features associated with the equipment operating data and environmental data of the offshore wind farm;

[0009] The first seasonal feature, the second seasonal feature and the global feature are input into a preset abnormal data identification model, so that the abnormal data identification model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series change pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

[0010] Furthermore, extracting a first seasonal feature related to the power generation data and low-altitude environment data of the offshore wind farm according to the low-altitude time series matrix includes:

[0011] performing differential processing on the low-altitude time series matrix to generate a stationary transformation sequence;

[0012] According to the stable change sequence, the autocorrelation diagram analysis method and the partial autocorrelation diagram analysis method are used to identify non-seasonal parameters and seasonal parameters;

[0013] The first seasonal feature is constructed according to the non-seasonal parameter and the seasonal parameter.

[0014] Furthermore, extracting a second seasonal feature related to the power generation data and the high-altitude environmental data of the offshore wind farm according to the high-altitude time series matrix includes:

[0015] Performing seasonal decomposition on the high-altitude time series matrix to decompose the high-altitude time series matrix into trend features, seasonal features, and residuals;

[0016] The trend feature and the seasonal feature are constructed as the second seasonal feature.

[0017] Furthermore, the abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer;

[0018] The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including:

[0019] Performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature;

[0020] Inputting the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature;

[0021] Concatenate the target fusion feature with the second seasonal feature to generate a third fusion feature, input the third fusion feature into an empty convolutional layer, identify and expand the dependency relationship between different features in the third fusion feature, and output a feature to be identified;

[0022] The features to be identified are input into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

[0023] An embodiment of the present invention further provides a device for identifying abnormal operation data of an offshore wind farm, comprising:

[0024] A data acquisition module is used to obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of the offshore wind farm in several time periods;

[0025] a data processing module, configured to construct a low-altitude time series matrix based on the low-altitude environment data and the power generation data, a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data;

[0026] a feature extraction module, configured to extract, based on the low-altitude time series matrix, a first seasonal feature associated with the generated power data and low-altitude environmental data of the offshore wind farm; extract, based on the high-altitude time series matrix, a second seasonal feature associated with the generated power data and high-altitude environmental data of the offshore wind farm; and extract, based on the operating environment matrix, a global feature associated with the equipment operating data and environmental data of the offshore wind farm;

[0027] The abnormal data identification module is used to input the first seasonal feature, the second seasonal feature and the global feature into a preset abnormal data identification model, so that the abnormal data identification model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series change pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

[0028] Furthermore, the feature extraction module extracts a first seasonal feature related to the power generation data and low-altitude environment data of the offshore wind farm according to the low-altitude time series matrix, including:

[0029] performing differential processing on the low-altitude time series matrix to generate a stationary transformation sequence;

[0030] According to the stable change sequence, the autocorrelation diagram analysis method and the partial autocorrelation diagram analysis method are used to identify non-seasonal parameters and seasonal parameters;

[0031] The first seasonal feature is constructed according to the non-seasonal parameter and the seasonal parameter.

[0032] Furthermore, the feature extraction module extracts a second seasonal feature related to the power generation data and the high-altitude environment data of the offshore wind farm according to the high-altitude time series matrix, including:

[0033] Performing seasonal decomposition on the high-altitude time series matrix to decompose the high-altitude time series matrix into trend features, seasonal features, and residuals;

[0034] The trend feature and the seasonal feature are constructed as the second seasonal feature.

[0035] Furthermore, the abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer;

[0036] The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including:

[0037] Performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature;

[0038] Inputting the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature;

[0039] Concatenate the target fusion feature with the second seasonal feature to generate a third fusion feature, input the third fusion feature into an empty convolutional layer, identify and expand the dependency relationship between different features in the third fusion feature, and output a feature to be identified;

[0040] The features to be identified are input into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

[0041] The present application also provides a terminal device, including:

[0042] one or more processors;

[0043] a memory, coupled to the processor, for storing one or more programs;

[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying abnormal operation data of an offshore wind farm as described in the above-mentioned embodiment of the invention.

[0045] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for identifying abnormal operation data of an offshore wind farm as described in the above-mentioned embodiment of the invention is implemented.

[0046] The following beneficial effects are achieved by implementing the present invention:

[0047] The present invention provides a method, apparatus, terminal device, and storage medium for identifying abnormal operating data of an offshore wind farm. The method constructs a low-altitude time series matrix based on low-altitude environmental data and power generation data of the offshore wind farm in several time periods to extract a first seasonal feature, and constructs a high-altitude time series matrix based on high-altitude environmental data and power generation data to extract a second seasonal feature. By capturing the seasonal characteristics of power generation and environmental factors at two altitudes, two different judgment bases are provided for the subsequent identification of abnormal power generation data, effectively improving the accuracy of identifying abnormal power generation data. Based on the low-altitude environmental data, the high-altitude environmental data, and the equipment operation data, an operating environment matrix is ​​constructed to extract global features for mining the coupling relationship between equipment operating status and the environment. During the abnormal data identification process, by combining the first seasonal feature with the second seasonal feature, the correlation between equipment operation data, environmental data, and power generation data can be effectively identified, thereby identifying abnormal equipment operation data. This method solves the current problem of difficulty in correctly identifying abnormal offshore wind power operation data. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0049] Figure 1 This is a flow chart of a method for identifying abnormal operation data of an offshore wind farm provided in one embodiment of the present application;

[0050] Figure 2 This is a structural diagram of an abnormal operation data identification device for an offshore wind farm provided by a certain embodiment of the present application;

[0051] Figure 3 This is a schematic diagram of the structure of a terminal device provided in a certain embodiment of the present application. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned figure descriptions are intended to cover non-exclusive inclusions.

[0054] In the description of the embodiments of this application, the technical terms "first" and "second" are used only to distinguish different objects and should not be understood to indicate or imply relative importance or implicitly specify the quantity, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, the meaning of "plurality" is more than two, unless otherwise clearly and specifically defined.

[0055] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0056] In the description of the embodiments of this application, the term "and / or" is simply a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent the following three situations: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0057] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two). Similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0058] In the description of the embodiments of the present application, unless otherwise expressly specified or limited, technical terms such as "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they can refer to fixed connections, detachable connections, or integration; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; internal connections between two components or interactions between two components. Those skilled in the art can understand the specific meanings of the above terms in the embodiments of the present application based on specific circumstances.

[0059] See also Figure 1 To solve the problems in the prior art, an embodiment of the present invention provides a method for identifying abnormal operation data of an offshore wind farm, comprising:

[0060] S1. Obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of the offshore wind farm in several time periods;

[0061] In a preferred embodiment of the present invention, low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data to be identified are collected from the monitoring system of the offshore wind farm. Specifically, the power generation data is the total amount of electricity generated by the offshore wind farm in each time period. The low-altitude environmental data (15 meters altitude) and the high-altitude environmental data (100 meters altitude) both include: wind speed, wind direction, temperature, humidity, and air pressure. The equipment operation data includes: the operation data of the wind rotor, generator, and gearbox. Furthermore,

[0062] A time period may be one day, one week, one month, etc. In this embodiment, it is set to obtain data in ten time periods.

[0063] S2. Constructing a low-altitude time series matrix based on the low-altitude environment data and the power generation data, constructing a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and constructing an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data;

[0064] In a preferred embodiment of the present invention, to improve the accuracy of subsequent abnormal data recognition, it is necessary to remove obvious abnormal values ​​in the data, such as noise and missing values, before constructing the matrix. Furthermore, the low-altitude time series matrix is ​​constructed by low-altitude environmental data and power generation data for 10 consecutive time periods at an altitude of 100 meters. The high-altitude time series matrix is ​​constructed by high-altitude environmental data and power generation data for 10 consecutive time periods at an altitude of 15 meters. The operating environment matrix is ​​constructed by the environmental change data matrix generated by subtracting the low-altitude environmental data from the high-altitude data for 10 time periods and the equipment operation data.

[0065] S3. Extracting, based on the low-altitude time series matrix, a first seasonal feature related to the generated power data and low-altitude environmental data of the offshore wind farm; extracting, based on the high-altitude time series matrix, a second seasonal feature related to the generated power data and high-altitude environmental data of the offshore wind farm; and extracting, based on the operating environment matrix, a global feature related to the equipment operating data and environmental data of the offshore wind farm;

[0066] Preferably, extracting the first seasonal feature related to the power generation data and low-altitude environment data of the offshore wind farm according to the low-altitude time series matrix includes:

[0067] S31, performing differential processing on the low-altitude time series matrix to generate a stable transformation sequence;

[0068] S32. Identify non-seasonal parameters and seasonal parameters using autocorrelation analysis and partial autocorrelation analysis based on the stationary variation sequence;

[0069] S33: Construct the first seasonal feature according to the non-seasonal parameter and the seasonal parameter.

[0070] In a preferred embodiment of the present invention, a seasonal autoregressive moving average (SARIMA) model is used to model and analyze the low-altitude time series matrix. Specifically, the data is first stabilized, and the trend and seasonality are eliminated by differencing (checking the ADF test). Furthermore, the parameters (p, d, q) (P, D, Q)s are determined by autocorrelogram (ACF) and partial autocorrelogram (PACF) analysis. The optimal parameters are selected using the AIC / BIC criteria. Finally, the parameters are fitted using methods such as maximum likelihood estimation.

[0071] The mathematical representation of the SARIMA model is as follows:

[0072] φ p (B)Φ P (B s )(1-B s ) d (1-B s )D y t =θ q (B)Θ Q (B s )∈t;

[0073] Among them, φ p (B) and θ q (B) is a polynomial of non-seasonal autoregression and moving average. Φ P (B s ) and Θ Q (B s ) is a polynomial of seasonal autoregression and seasonal moving average. t is the observed value. ∈ t is white noise. B is the lag operator, t is the specific time period, for the non-seasonal part: p is the order of the autoregressive (AR) term, which indicates the linear relationship between the current observation and the previous p observations. d is the number of differences (I), which means that the time series is differentiated d times to eliminate the trend. q is the order of the moving average (MA) term, which indicates the linear relationship between the current observation and the previous q forecast errors. For the seasonal part, P is the order of the seasonal autoregressive (SAR) term, which indicates the linear relationship between the current observation and the previous P seasonal observations. D is the number of seasonal differences (I), which means that the time series is differentiated D times to eliminate the seasonal trend. Q is the order of the seasonal moving average (SMA) term, which indicates the linear relationship between the current observation and the previous Q seasonal forecast errors. s is the total number of time periods. In this embodiment, s=10, for monthly data s=12, and for quarterly data s=4;

[0074] Preferably, extracting the second seasonal feature related to the power generation data and high-altitude environmental data of the offshore wind farm according to the high-altitude time series matrix includes:

[0075] S34, performing seasonal decomposition on the high-altitude time series matrix, decomposing the high-altitude time series matrix into trend characteristics, seasonal characteristics, and residuals;

[0076] S35: Construct the trend feature and the seasonal feature into the second seasonal feature.

[0077] In a preferred embodiment of the present invention, a seasonal decomposition model (STL) is used to model the second seasonal feature, and decompose it into three parts: trend, seasonality and residual.

[0078] Specifically, the STL decomposition process includes the following main steps: Initial decomposition: Smoothing the time series data to obtain preliminary estimates of the trend and seasonal components. Seasonal smoothing: Subtracting the initial trend from the time series to obtain a detrended series. Each seasonal cycle position is smoothed separately, and LOESS is used to estimate the mean of each seasonal position to form a preliminary seasonal component. Trend smoothing: Subtracting the initial seasonal component from the original series to obtain a deseasonalized series. LOESS smoothing is performed on the deseasonalized series to estimate a new trend component, such as different levels. Residual calculation: Subtracting the newly estimated trend and seasonal components from the original time series to obtain the residual. The STL model is suitable for datasets with a large number of outliers or noise, and can handle noisy or irregular time series data. This is especially true when the data volume is small and computing resources are limited. Altitudes closer to the surface are more susceptible to external influences, resulting in more anomalies and noise data. Therefore, STL is selected as the input branch model for the third branch. The STL model uses the LOESS (locally weighted regression) smoothing method, which can adaptively estimate trends and seasonal components, such as temperature, humidity, wind speed and other indicators that are more affected by seasons.

[0079] Furthermore, the STL model converts the high altitude time series matrix Y t It is broken down into three main parts:

[0080] Y t =0.3T t +0.4S t +0.4R t ;

[0081] Where: T t It is the trend component, which indicates the long-term trend of data. t It is a seasonal component, indicating periodic fluctuations with a fixed cycle length. Since seasonal factors have a greater impact on whether the power generation data is specific data, the weight of this part will be set higher, to 0.4. t is the residual component, which represents the random fluctuation or noise part.

[0082] Furthermore, a deep convolutional model (DNN) is used to extract global features related to the equipment operation data and environmental data of the offshore wind farm. It should be noted that the structure of the deep convolutional model is: the first convolutional layer, the first pooling layer, the hidden layer, the second convolutional layer, and the second pooling layer. The convolution kernel of the first convolutional layer is 3, the step size is 3, the padding is 2, and the activation function is rule. The convolution kernel of the second convolutional layer is 3, the step size is 5, the padding is 2, and the activation function is rule. Specifically, the first convolutional layer is used for local feature extraction and preliminary abstraction. With a window size of 3 and a movement of 3 units each time, the local correlation features of the environmental changes and equipment operation data within the time period are extracted. The first pooling layer is used for feature dimensionality reduction and local feature aggregation. The pooling operation is used to compress the local features output by the first convolutional layer, extract the key features in the local area, form an overall summary of the local time period, and lay the foundation for global features. The hidden layer is used for cross-layer feature fusion and global pattern capture. Through full connections or convolution operations, the hidden layer further fuses the locally aggregated features of the first pooling layer, capturing long-term correlations across multiple local regions to identify the cumulative impact of environmental change trends over multiple consecutive time periods on the device's operating status. The periodic fluctuations of the air pressure difference are matched with the periodic patterns of the device's operating parameters. The second convolutional layer is used for secondary abstraction of the global features, spanning more time periods with a larger step size of 5 to extract global correlation features of environmental operating data across long time periods. This larger step size allows the model to capture jump-like correlations between non-continuous time periods. The second pooling layer is used for the final aggregation of global features. Through pooling, the long-term features of the second convolutional layer are compressed into a global feature vector, representing the overall pattern of environmental operating data for the entire time series (10 time periods). This determines the global correlation between the overall trend of environmental change (such as the increasing or decreasing low-altitude-high-altitude temperature difference) and the device's operating status, i.e., the global feature.

[0083] S4. Input the first seasonal feature, the second seasonal feature and the global feature into a preset abnormal data recognition model, so that the abnormal data recognition model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

[0084] It should be noted that the seasonal autoregressive moving average model can capture the seasonality and autocorrelation of the collected data time series. The deep convolutional model can extract the global characteristics of the collected data time series. The seasonal decomposition model can separate the trend, seasonality, and residual components of the collected data time series.

[0085] By concatenating and convolving the first seasonal feature, the second seasonal feature, and the global feature, the advantages of different features can be combined, fully leveraging the seasonality, global characteristics, and trend information of the time series. The concatenation operation achieves feature enhancement. Concatenating the outputs of different models increases the feature dimension, enabling the model to learn richer feature representations. Convolution operations are then performed on the concatenated features to further extract relationships between features and enhance their expressive power.

[0086] By combining the strengths of multiple models, we can more comprehensively capture the characteristics of time series, thereby improving forecast accuracy. Different models fit data differently. Fusion of the outputs of multiple models can reduce overfitting of a single model and enhance its generalization. Concatenation and convolution operations increase the dimensionality and complexity of features, enabling the model to learn richer feature representations and better understand the inherent structure of the data. Time series data often exhibits complex patterns, including seasonality, trends, and global features. By fusing the outputs of multiple models, we can better adapt to these complex patterns and improve the robustness of the model.

[0087] Preferably, the abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer;

[0088] The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including:

[0089] S41, performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature;

[0090] S42. Input the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature.

[0091] S43, concatenating the target fusion feature with the second seasonal feature to generate a third fusion feature, inputting the third fusion feature into an empty convolutional layer, identifying and expanding the dependency relationship between different features in the third fusion feature, and outputting a feature to be identified;

[0092] S44. Input the features to be identified into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

[0093] In a preferred embodiment of the present invention, point-by-point convolution, typically a 1x1 convolution operation, is primarily used to extract channel features. It performs convolution on the input feature matrix to integrate information across channels and reduce its dimensionality. Point-by-point convolution is advantageous when integrating multi-channel data such as wind speed and power output because it aggregates information from different feature channels, thereby extracting more abstract feature representations.

[0094] The feature matrix represents the characteristic information of the input data at different levels. Through the convolution operation, the convolution kernel can extract data features such as wind speed, temperature, power, and altitude from the input data. Information compression: The size of the feature map is usually smaller than the size of the input data, which allows the feature matrix to compress the input data to a certain extent. Through the convolution operation, redundant information in the input data is removed, retaining more useful feature information, thereby improving the computational efficiency and generalization ability of the model.

[0095] Dilated convolution increases the receptive field of the convolution kernel by introducing dilations into the standard convolution operation without increasing the number of parameters. This convolution method helps capture a wider range of contextual information. When analyzing features such as temperature and humidity that may have large-scale dependencies, dilated convolution can help the model identify and exploit these dependencies. When processing highly uncertain and random data such as wind power, this model can more accurately capture the inherent patterns in the data, thereby better identifying anomalies.

[0096] Furthermore, during the training process of the abnormal data recognition model, by using many years of historical operating data and high and low altitude environmental data as training data, the model parameters are continuously updated and iterated until the loss function converges or reaches the specified number of iterations. The specific loss function is as follows:

[0097] Loss=0.6*FocalLoss+0.4*SME;

[0098] Focal Loss (P t )=-α t (1-P t)γ log(P t );

[0099] Focal Loss can solve the imbalance problem between specific data and normal data.

[0100] Where: p t Is the model's predicted probability for the sample. For positive samples, p t is the probability of predicting a positive result; for negative samples, p t is the complement of the probability of predicting negative, α tis a parameter that adjusts the weight of positive and negative samples. γ is a parameter that adjusts the weight of difficult and easy samples, also known as the focusing parameter. Dynamic scaling factor (1-P t ) γ It is the core of Focal Loss, which dynamically adjusts the loss value according to the predicted probability of the sample: when P t When it is close to 1 (i.e. the sample is easy to classify), (1-P t ) γ approaches 0, thus reducing the loss contribution of the sample. t When it is small (i.e. the sample is difficult to classify), (1-P t ) γ Close to 1, the loss contribution is relatively large. t Complete the weight adjustment of positive and negative samples, adjusting the weight ratio between positive and negative samples to alleviate the problem of positive and negative sample imbalance. Through dynamic scaling factors, Focal Loss can significantly reduce the loss contribution of easy-to-classify samples, allowing the model to pay more attention to difficult-to-classify samples, significantly improving the model's detection accuracy and recall rate.

[0101] You can also use the SME loss function, which calculates the average of the squares of the differences between the predicted value and the actual value. For each training sample, the loss is defined as:

[0102]

[0103] The SME loss function is highly adaptable to data distributions and can effectively handle non-normal distributions in wind power data. Wind power data is often affected by multiple factors, such as wind speed, wind direction, and equipment status, resulting in a complex data distribution. The SME loss function, through its squared error function, is well-suited to this complex data distribution, enabling better identification of outliers during model training. Furthermore, because the loss function uses a squared term, large errors are amplified, making the SME highly sensitive to outliers.

[0104] In summary, this embodiment provides a method for identifying abnormal operating data of an offshore wind farm. By constructing a low-altitude time series matrix based on the low-altitude environmental data and power generation data of the offshore wind farm in several time periods to extract the first seasonal feature, and constructing a high-altitude time series matrix based on the high-altitude environmental data and power generation data to extract the second seasonal feature, by capturing the seasonal characteristics of power generation and the two high-altitude environmental factors, two different judgment bases are provided for the subsequent identification of abnormal power generation data, effectively improving the accuracy of identifying abnormal power generation data. Based on the low-altitude environmental data, the high-altitude environmental data and the equipment operation data, an operating environment matrix is ​​constructed to extract global features for mining the coupling relationship between the equipment operation status and the environment. In the abnormal data identification process, by combining the first seasonal feature and the second seasonal feature, the correlation between the equipment operation data and the environmental data and the power generation data can be effectively identified, thereby identifying abnormal equipment operation data. This solves the problem that it is difficult to correctly identify abnormal offshore wind power abnormal operation data.

[0105] See Figure 2 , is an abnormal operation data identification device for an offshore wind farm provided by one embodiment of the present invention, comprising:

[0106] A data acquisition module is used to obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of the offshore wind farm in several time periods;

[0107] a data processing module, configured to construct a low-altitude time series matrix based on the low-altitude environment data and the power generation data, a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data;

[0108] a feature extraction module, configured to extract, based on the low-altitude time series matrix, a first seasonal feature associated with the generated power data and low-altitude environmental data of the offshore wind farm; extract, based on the high-altitude time series matrix, a second seasonal feature associated with the generated power data and high-altitude environmental data of the offshore wind farm; and extract, based on the operating environment matrix, a global feature associated with the equipment operating data and environmental data of the offshore wind farm;

[0109] The abnormal data identification module is used to input the first seasonal feature, the second seasonal feature and the global feature into a preset abnormal data identification model, so that the abnormal data identification model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series change pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

[0110] Furthermore, the feature extraction module extracts a first seasonal feature related to the power generation data and low-altitude environment data of the offshore wind farm according to the low-altitude time series matrix, including:

[0111] performing differential processing on the low-altitude time series matrix to generate a stationary transformation sequence;

[0112] According to the stable change sequence, the autocorrelation diagram analysis method and the partial autocorrelation diagram analysis method are used to identify non-seasonal parameters and seasonal parameters;

[0113] The first seasonal feature is constructed according to the non-seasonal parameter and the seasonal parameter.

[0114] Furthermore, the feature extraction module extracts a second seasonal feature related to the power generation data and the high-altitude environment data of the offshore wind farm according to the high-altitude time series matrix, including:

[0115] Performing seasonal decomposition on the high-altitude time series matrix to decompose the high-altitude time series matrix into trend features, seasonal features, and residuals;

[0116] The trend feature and the seasonal feature are constructed as the second seasonal feature.

[0117] Furthermore, the abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer;

[0118] The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including:

[0119] Performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature;

[0120] Inputting the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature;

[0121] Concatenate the target fusion feature with the second seasonal feature to generate a third fusion feature, input the third fusion feature into an empty convolutional layer, identify and expand the dependency relationship between different features in the third fusion feature, and output a feature to be identified;

[0122] The features to be identified are input into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

[0123] See also Figure 3 , an embodiment of the present application further provides a terminal device, including:

[0124] one or more processors;

[0125] a memory, coupled to the processor, for storing one or more programs;

[0126] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for identifying abnormal operation data of an offshore wind farm.

[0127] The processor is used to control the overall operation of the terminal device to complete all or part of the steps of the above-mentioned abnormal operation data identification method of the offshore wind farm. The memory is used to store various types of data to support the operation of the terminal device. These data may include, for example, instructions for any application or method used to operate on the terminal device, as well as application-related data. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0128] In an exemplary embodiment, the terminal device can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components, and is used to execute the method for identifying abnormal operation data of an offshore wind farm as described in any of the above embodiments, and achieve the same technical effect as the above method.

[0129] In another exemplary embodiment, a computer-readable storage medium including a computer program is further provided. When executed by a processor, the computer program implements the steps of the method for identifying abnormal operating data of an offshore wind farm as described in any of the aforementioned embodiments. For example, the computer-readable storage medium may be the aforementioned memory including the computer program. The computer program may be executed by a processor of a terminal device to implement the method for identifying abnormal operating data of an offshore wind farm as described in any of the aforementioned embodiments, thereby achieving the same technical effects as the aforementioned methods.

[0130] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying abnormal operation data of an offshore wind farm, characterized in that: include: Obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of offshore wind farms in several time periods; Constructing a low-altitude time series matrix based on the low-altitude environment data and the power generation data, constructing a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and constructing an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data; Extracting, based on the low-altitude time series matrix, first seasonal features associated with the generated power data and low-altitude environmental data of the offshore wind farm; extracting, based on the high-altitude time series matrix, second seasonal features associated with the generated power data and high-altitude environmental data of the offshore wind farm; and extracting, based on the operating environment matrix, global features associated with the equipment operating data and environmental data of the offshore wind farm; The first seasonal feature, the second seasonal feature and the global feature are input into a preset abnormal data identification model, so that the abnormal data identification model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series change pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

2. The method for identifying abnormal operation data of an offshore wind farm according to claim 1, characterized in that: The extracting, based on the low-altitude time series matrix, a first seasonal feature related to the generated power data and the low-altitude environment data of the offshore wind farm comprises: performing differential processing on the low-altitude time series matrix to generate a stationary transformation sequence; According to the stable change sequence, the autocorrelation diagram analysis method and the partial autocorrelation diagram analysis method are used to identify non-seasonal parameters and seasonal parameters; The first seasonal feature is constructed according to the non-seasonal parameter and the seasonal parameter.

3. The method for identifying abnormal operation data of an offshore wind farm according to claim 2, characterized in that: The extracting, based on the high-altitude time series matrix, a second seasonal feature related to the generated power data and the high-altitude environmental data of the offshore wind farm comprises: Performing seasonal decomposition on the high-altitude time series matrix to decompose the high-altitude time series matrix into trend features, seasonal features, and residuals; The trend feature and the seasonal feature are constructed as the second seasonal feature.

4. The method for identifying abnormal operation data of an offshore wind farm according to claim 3, characterized in that: The abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer; The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including: Performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature; Inputting the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature; Concatenate the target fusion feature with the second seasonal feature to generate a third fusion feature, input the third fusion feature into an empty convolutional layer, identify and expand the dependency relationship between different features in the third fusion feature, and output a feature to be identified; The features to be identified are input into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

5. A device for identifying abnormal operation data of an offshore wind farm, characterized in that: include: A data acquisition module is used to obtain low-altitude environmental data, high-altitude environmental data, equipment operation data, and power generation data of the offshore wind farm in several time periods; a data processing module, configured to construct a low-altitude time series matrix based on the low-altitude environment data and the power generation data, a high-altitude time series matrix based on the high-altitude environment data and the power generation data, and an operating environment matrix based on the low-altitude environment data, the high-altitude environment data, and the equipment operating data; a feature extraction module, configured to extract, based on the low-altitude time series matrix, a first seasonal feature associated with the generated power data and low-altitude environmental data of the offshore wind farm; extract, based on the high-altitude time series matrix, a second seasonal feature associated with the generated power data and high-altitude environmental data of the offshore wind farm; and extract, based on the operating environment matrix, a global feature associated with the equipment operating data and environmental data of the offshore wind farm; The abnormal data identification module is used to input the first seasonal feature, the second seasonal feature and the global feature into a preset abnormal data identification model, so that the abnormal data identification model integrates the first seasonal feature, the second seasonal feature and the global feature, identifies the time series change pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data.

6. The abnormal operation data identification device for an offshore wind farm according to claim 5, characterized in that: The feature extraction module extracts a first seasonal feature related to the power generation data and low-altitude environment data of the offshore wind farm according to the low-altitude time series matrix, including: performing differential processing on the low-altitude time series matrix to generate a stationary transformation sequence; According to the stable change sequence, the autocorrelation diagram analysis method and the partial autocorrelation diagram analysis method are used to identify non-seasonal parameters and seasonal parameters; The first seasonal feature is constructed according to the non-seasonal parameter and the seasonal parameter.

7. The abnormal operation data identification device for an offshore wind farm according to claim 6, characterized in that: The feature extraction module extracts a second seasonal feature related to the power generation data and the high-altitude environment data of the offshore wind farm according to the high-altitude time series matrix, including: Performing seasonal decomposition on the high-altitude time series matrix to decompose the high-altitude time series matrix into trend features, seasonal features, and residuals; The trend feature and the seasonal feature are constructed as the second seasonal feature.

8. The abnormal operation data identification device for an offshore wind farm according to claim 7, characterized in that: The abnormal data recognition model includes: a point-by-point convolution layer, a void convolution layer, and a fully connected layer; The abnormal data identification model integrates the first seasonal feature, the second seasonal feature, and the global feature to identify the temporal variation pattern of the equipment operation data and the power generation data of the offshore wind farm, and then outputs the abnormal equipment operation data and the abnormal power generation data, including: Performing weighted fusion on the first seasonal feature and the global feature to generate a first fused feature; Inputting the first fused feature into a point-by-point convolution layer, so that the point-by-point convolution layer convolves the first fused feature through a plurality of feature channels, identifies and removes redundant information in the first fused feature, and outputs a target fused feature; Concatenate the target fusion feature with the second seasonal feature to generate a third fusion feature, input the third fusion feature into an empty convolutional layer, identify and expand the dependency relationship between different features in the third fusion feature, and output a feature to be identified; The features to be identified are input into a fully connected layer, so that the fully connected layer identifies and outputs abnormal equipment operation data and abnormal power generation data according to the features to be identified.

9. A terminal device, characterized in that: include: one or more processors; a memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the abnormal operation data identification method of the offshore wind farm according to any one of claims 1 to 4.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for identifying abnormal operation data of an offshore wind farm according to any one of claims 1 to 4 is implemented.

Images