Offshore wind farm electrical abnormal state discrimination method based on multi-dimensional matrix profile

Abstract

The application discloses a method for distinguishing abnormal electrical states of offshore wind farms based on a multi-dimensional matrix profile, takes common fault states and disturbance phenomena of offshore wind farms as the invention object, and proves the effectiveness of the algorithm through theoretical analysis, simulation calculation and actual data verification.The offshore wind farm electrical abnormal state feature extraction strategy based on EEMD effectively decomposes the abnormal waveform into signals with their own time characteristic scales, and provides a basis for offshore wind farm abnormal state identification.The multi-dimensional matrix profile sub-sequence similarity measurement method based on the improved DTW algorithm reflects the change trend and frequency difference change of the time sequence fragment, and the embedding of the heartbeat packet mechanism effectively reduces the calculation burden in the actual detection process, and does not cause the missed detection, and has wide engineering application value.

Description

Technical Field

[0001] This invention relates to the field of electrical technology, and more specifically to a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles. Background Technology

[0002] Building a new power system is a crucial means to achieve the "dual carbon" target, and renewable energy sources, represented by wind power and photovoltaics, are an important component of this new power system. In recent years, offshore wind power technology has experienced rapid development in my country due to its advantages of being clean, renewable, and possessing abundant wind resources. According to research, by the end of 2021, China's cumulative installed capacity of offshore wind power had reached 23.4 GW, surpassing the UK and Germany to become the world's largest offshore wind power market. Compared to the limitations of onshore wind power, such as significant noise pollution and large land area requirements, offshore wind power has the potential for large-scale development due to its abundant and stable wind resources.

[0003] However, offshore wind farms are far from the coast, and the environment and operating conditions of offshore wind turbines are harsh and complex. The turbines are prone to failure and disturbances, and the maintenance costs are higher and more difficult than those onshore. The economic losses caused by a shutdown are enormous.

[0004] In addition, since offshore wind farms are generally not limited by region and are generally large, this means that there are more wind turbines. It also means that there is a huge amount of status data in the wind farm database that needs to be monitored. Therefore, it is urgent to develop relevant technologies for identifying abnormal conditions in offshore wind farms, monitor the health status of offshore wind turbines, reduce the operation and maintenance costs of offshore wind farms, and improve the efficiency of wind farms. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles. Taking common fault states and disturbance phenomena in offshore wind farms as the subject of the invention, the effectiveness of the algorithm has been proven through theoretical analysis, simulation calculation and actual data verification, which solves the problem of insufficient identification accuracy due to the scarcity of offshore wind farm state data samples.

[0006] To achieve the above objectives, this invention provides a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles, the method comprising:

[0007] Acquire unidentified data for offshore wind farms;

[0008] The data to be identified is decomposed to obtain IMF components;

[0009] Abnormal state identification is performed in the constructed abnormal state feature sample library, and the corresponding abnormal state features are matched.

[0010] Optionally, the decomposition of the data to be identified specifically includes:

[0011] Taking advantage of the characteristic that Gaussian white noise cancels out noise during multiple averaging processes, the data to be identified is decomposed using the EEMD algorithm.

[0012] Optionally, the step of identifying abnormal states in the constructed abnormal state feature sample library specifically includes:

[0013] The MDMP algorithm is used to identify abnormal states in a constructed abnormal state feature sample library. The DTW distance metric is used to determine the shape difference between the sequence to be identified and the abnormal sample feature library sequence, and it has the characteristic of adaptive window size.

[0014] Optionally, the method further includes: constructing the abnormal state feature sample library.

[0015] Optionally, constructing the abnormal state feature sample library specifically includes:

[0016] Obtain the time series of abnormal states of offshore wind farms;

[0017] Remove noise from the time series of the abnormal states;

[0018] The denoised abnormal state time series is decomposed into IMF components, and a multidimensional time series is reconstructed based on the IMF components to construct an abnormal state feature sample library.

[0019] Optionally, the denoised abnormal state time series is decomposed into IMF components using the EEMD algorithm. Specifically, this includes adding Gaussian white noise of equal amplitude multiple times during the IMF decomposition process, utilizing the statistical characteristic of Gaussian white noise having a uniform frequency distribution to cancel out the noise during multiple averaging processes.

[0020] Optionally, the method further includes: determining the wind field detection status based on a heartbeat mechanism.

[0021] Optionally, the heartbeat mechanism sends a specific signal to the server at fixed intervals. Upon receiving the specific signal, the server replies with a response signal, and the server determines the current communication status of the client based on the response signal. The heartbeat mechanism has a triggering mechanism, which is as follows:

[0022]

[0023] Where t1 is the time when the last heartbeat signal was sent, t2 is the time when the current heartbeat signal is sent, and v b and v m These are the baseline value and the maximum value of wind speed fluctuation, respectively, U b and fb These are the reference values ​​for voltage and frequency fluctuations, respectively.

[0024] If the s output is 1, it indicates that the heartbeat packet signal received by the computer is normal. At this time, the abnormal state identification system can be in low-frequency detection state. If the s signal received by the computer is 0, it proves that a fault may occur in the future. The proposed identification strategy needs to be activated. After a period of high-frequency detection, if no abnormality is displayed and the received heartbeat packet signal is 1, the system enters low-frequency detection state.

[0025] Optionally, the method further includes:

[0026] The distance judgment is performed based on the most similar subsequence identified by the multidimensional matrix contour algorithm. The identification mechanism is as follows:

[0027]

[0028] in, The average distance between the most similar subsequences, {d1,d2,…,d N} represents the average distance of the most similar subsequences under N real abnormal state events. It can be seen that only when the output is "1" is the current object to be identified an anomaly type matched by the anomaly feature library; otherwise, it indicates that the current offshore wind farm is operating normally.

[0029] Optionally, the method further includes:

[0030] Determine whether the matched abnormal state features meet the requirements;

[0031] If the matched abnormal state features meet the requirements, then the identified abnormal state is determined to be correct.

[0032] Optionally, the method further includes: if the determined abnormal state characteristics do not meet the requirements, then the identified abnormal state is determined to be incorrect, and the data to be identified is not abnormal.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] This invention proposes a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles. Taking common fault states and disturbance phenomena in offshore wind farms as the subject of the invention, the effectiveness of the algorithm is proved through theoretical analysis, simulation calculation and actual data verification, which solves the problem of insufficient identification accuracy due to the scarcity of offshore wind farm state data samples. Attached Figure Description

[0035] Figure 1 This is a flowchart of the EEMD signal decomposition of voltage sequence data of a wind turbine in an offshore wind farm provided in one or more embodiments of the present invention;

[0036] Figure 2 This is a flowchart of an electrical anomaly identification process for offshore wind farms based on MDMP, provided in one or more embodiments of the present invention.

[0037] Figure 3 This refers to an offshore wind farm model provided in one or more embodiments of the present invention;

[0038] Figure 4 The figures show the results of different signal processing methods under single-phase ground fault waveforms provided in one or more embodiments of the present invention.

[0039] Figure 5 This is a fault EEMD decomposition diagram provided in one or more embodiments of the present invention;

[0040] Figure 6 This is a feature library of electrical anomaly states of offshore wind farms provided in one or more embodiments of the present invention;

[0041] Figure 7 This invention provides an abnormal state identification of a wind turbine in an offshore wind farm based on Matlab simulation, as described in one or more embodiments of the present invention.

[0042] Figure 8 This invention provides an electrical anomaly identification method for a wind turbine in an offshore wind farm, as described in one or more embodiments of the present invention. Detailed Implementation

[0043] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0044] The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular forms “the” and “the” as used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0045] It should be understood that although the terms "first," "second," "third," etc., may be used to describe structures / information in this invention, these structures / information should not be limited to these terms. These terms are only used to distinguish structures / information of the same type from one another. For example, without departing from the scope of this invention, the first structure / information may also be referred to as the second structure / information, and similarly, the second structure / information may also be referred to as the first structure / information.

[0046] In addition, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0047] This invention proposes a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles. Taking common fault states and disturbance phenomena in offshore wind farms as the subject of the invention, the effectiveness of the algorithm is proved through theoretical analysis, simulation calculation and actual data verification, which solves the problem of insufficient identification accuracy due to the scarcity of offshore wind farm state data samples.

[0048] This invention uses EEMD as a feature extraction method to address the frequent mode confusion phenomenon that occurs during EMD decomposition of intrinsic mode functions (IMFs). Signal confusion under different modes will affect the accuracy of feature extraction. This algorithm adds Gaussian white noise of equal amplitude multiple times during IMF decomposition, utilizing the statistical characteristic of Gaussian white noise's uniform frequency distribution to cancel out the noise during multiple averaging processes. Due to the harsh working environment and fluctuating wind speeds in offshore wind farms, EEMD can not only improve feature extraction accuracy for abnormal state feature extraction but also effectively reduce signal noise and separate IMF components representative of abnormal states.

[0049] EEMD decomposes time-series data such as voltage and current under abnormal offshore wind field conditions into single-component signals containing only a single instantaneous frequency, that is, composed of a finite number of IMF components and a residual component.

[0050] Given a voltage sequence X(t) of a wind turbine in an offshore wind farm, the algorithm flowchart is as follows: Figure 1 As shown, the specific decomposition steps are as follows:

[0051] 1) Define the overall mean frequency F;

[0052] 2) Add Gaussian white noise n with a standard normal distribution to the original data X(t). i (t) represents the noise signal added in the i-th experiment, resulting in a new original time series data x. i (t), as shown in formula (1):

[0053] x i (t)=X(t)+n i (t) (1)

[0054] 3) Use the EMD algorithm to determine x i For all local maxima and minima in (t), the i-th mean envelope function X is obtained by fitting all local extrema points using cubic spline interpolation. mean(i) (t). x i (t) Remove mean envelope function X mean(i) (t) obtains the intermediate signal h i (t), calculated as shown in formula (2):

[0055] h i (t)=x i (t)-x mean(i) (t) (2)

[0056] 4) Determine h i (t) Whether the IMF conditions are met: ①h i (t) has the same number of maxima and minimums or differs by 1; ②h i The mean of the envelopes defined by local maxima and local minima in (t) is zero. If the above conditions are not met, repeat steps 3) to 4); if they are met, h i (t) is the IMF score C i (t) quantity, and the data x i (t) Separate the IMF component to obtain the remaining signal m i (t), as shown in equation (3):

[0057] m i (t)=x i (t)-C i (t) (3)

[0058] 5) m i (t) Repeat steps 2) to 4) to obtain all IMF components and a residual r(t), as shown in equation (4):

[0059]

[0060] 6) To eliminate the white noise added during the above decomposition process, the IMF is averaged to obtain the final IMF component C. i (t), as shown in equation (5):

[0061]

[0062] The EEMD algorithm is applied to the decomposition of abnormal state signals in offshore wind farms. Because multiple additions of white noise can completely preserve the frequency components under abnormal conditions, the decomposed IMF reflects the characteristics of different time scales under abnormal conditions. This not only helps to obtain a complete sample library of abnormal state features but also provides a practical basis for abnormal state identification. Since real-world offshore wind farm anomaly data is relatively scarce in this invention, Matlab / Simulink is used to simulate offshore wind farm anomalies to obtain various abnormal states of wind turbines. Based on this, the features of the abnormal states after EEMD decomposition are extracted and combined to construct a sample library of abnormal state features of offshore wind farms.

[0063] Please see Figure 1This invention proposes an anomaly identification technique based on the multi-dimensional matrix profile (MDMP) algorithm using dynamic time warping (DTW). This method improves the MDMP algorithm by replacing the original Euclidean distance with a DTW distance metric. This not only effectively addresses the issue of neglecting the temporal differences in the identified sequence trend during Euclidean distance calculation but also enables adaptive window size adjustment within the algorithm.

[0064] The algorithm takes as input the reconstructed multidimensional time series T after EEMD decomposition of the anomalous state, and outputs as a multidimensional matrix contour and contour index M. list The multidimensional matrix contour M represents the distance between the multidimensional subsequences in T and L, and is arranged in ascending order, meaning the similarity is from high to low; the multidimensional matrix contour index M... list It represents the position of the most similar subsequence that T matches in L.

[0065] Specifically, firstly, the index matrix storing the most similar subsequences in the multidimensional matrix contour is initialized; then, the distance between the current query sequence and the subsequences is calculated and stored in the distance matrix D. When calculating the distance between two sequences, this invention abandons the original Euclidean distance and adopts the DTW algorithm. This algorithm can effectively capture the trend mapping relationship between two time series by utilizing dynamic regularization. It not only effectively solves the adaptive problem of sliding window, but also reflects the trend and shape similarity of the dynamic change amplitude of the time series, making up for the problem that Euclidean distance is insufficient in characterizing the temporal differences in identifying the trend of the sequence. Its principle is briefly described as follows:

[0066] Given two time series x(i), i = 1, 2, ..., m and x(j), j = 1, 2, ..., n, solving the DTW problem becomes an optimization problem. Under the constraints of boundary conditions, continuity, and monotonicity, the regularized path must be from d(1,1) to d(m,n), where d is the Euclidean distance metric. Then, the cumulative distance r(i,j) of the optimal regularized path is:

[0067]

[0068] DTW=min{r(m,n)} (2)

[0069] Therefore, the distance matrix D between the current window and the query is obtained according to the DTW algorithm. Then, D is sorted in ascending order by column, and the resulting index D0 is obtained. list Finally, the D values, sorted in ascending column order, are summed and averaged to obtain the multidimensional matrix outline M and index M. listBy determining the minimum distance and index, the most similar multidimensional subsequence and its location are obtained, thus achieving the matching of similar subsequences. The multidimensional matrix contour algorithm based on DTW distance is as follows: Figure 2 As shown.

[0070] Furthermore, it is worth noting that, by Figure 2 It can be seen that the algorithm will always obtain the minimum distance and index of the most similar subsequence, and the matching result is not the most similar subsequence in the true sense. In order to reduce the false judgment rate of anomaly identification, this invention applies a threshold judgment algorithm, which judges the most similar subsequence identified by the multidimensional matrix contour algorithm by a threshold using "0" and "1" logic. The judgment criteria are shown in equation (3):

[0071]

[0072] In the formula, Let ε be the average distance between the most similar subsequences, and let ε be a set threshold determined through data analysis. The data analysis process is as follows: Assuming the total number of abnormal state identifications is V, and the number of actual abnormal states is N, then calculate the average distance between the most similar subsequences under N abnormal state identification events, denoted as D. Then D = {d1, d2, ..., d...} n The threshold is set as ε = maxD, which is the minimum value of the average distance between the most similar subsequences of N real abnormal states. Therefore, it can be seen that only subsequence segments with output "1" in j indicate that the current working condition is abnormal, and a precise match with the abnormal state type in the sample library has been successfully achieved, thus identifying the abnormal state type; otherwise, it indicates that no abnormality has occurred in the current observation data.

[0073] Since the time spent in abnormal states at wind farms is much shorter than the time spent in normal states, and in reality, wind farms are in a normal state most of the time, if computers or related equipment are constantly in a high-frequency inspection state, it will consume a lot of useless computing power. Therefore, we consider designing a heartbeat mechanism to ensure that computing power can be allocated according to the current state of the offshore wind farm, thereby reducing the computational cost of low-probability, useless identification areas.

[0074] Generally, a heartbeat mechanism sends a specific signal to the server at fixed intervals. If the server receives the signal, it replies with a response signal. The main purpose is to inform the server of the current communication status of the client. In designing the heartbeat mechanism, this invention considers the actual situation of offshore wind farms. Compared with standard heartbeat signals, the computer does not determine the status of the wind farm by whether it receives a heartbeat signal, but by whether an anomaly occurs when it receives a heartbeat signal. Therefore, this invention sets a triggering mechanism for the heartbeat signal, and the triggering condition for the anomaly is shown in equation (4).

[0075] • Sudden changes in wind speed or extremely high wind speeds: The wind speed fluctuates greatly, or there is a prolonged period of high wind speed.

[0076] • Sudden voltage amplitude fluctuation: The voltage amplitude fluctuates significantly;

[0077] • Voltage frequency change: A sudden increase or decrease in the frequency of the voltage;

[0078] • Sudden change in phase current amplitude: The phase current amplitude fluctuates significantly;

[0079] • Weather warning: Weather forecasts indicate that severe weather is expected in the near future.

[0080]

[0081] Where t1 is the time when the last heartbeat signal was sent, t2 is the time when the current heartbeat signal is sent, and v b and v m U represents the baseline value and maximum value of wind speed fluctuation, respectively. b and f b These represent the baseline values ​​for voltage and frequency fluctuations, respectively. If the s output is 1, it indicates that the heartbeat signal received by the computer is normal, and the abnormal state identification system can be in low-frequency detection mode. If the s signal received by the computer is 0, it indicates that a fault may occur in the future, and the proposed identification strategy needs to be activated. After a period of high-frequency detection, if no abnormality is displayed and the received heartbeat signal is 1, the system enters low-frequency detection mode.

[0082] This invention uses the simulation software Matlab / Simulink to establish a simulation model of an offshore wind farm in Shanghai with a frequency of 50Hz. The wind farm consists of 16 1.5MW doubly-fed asynchronous turbines. The output voltage of the turbine terminals is 575V, which is stepped up to 25kV by a transformer and transmitted via a 10km medium-voltage submarine cable. After being stepped up to 120kV by an offshore substation, the voltage is transmitted via a 15km high-voltage submarine cable and then connected to the onshore power grid. Figure 3 As shown in Table 1, to obtain electrical anomaly data for offshore wind farms, this invention conducted extensive simulation experiments, simulating common electrical faults in offshore wind farms under different transition resistances (1-10Ω) at locations L1-L4, including single-phase grounding, two-phase grounding, three-phase grounding, and phase-to-phase short circuits. Disturbances were introduced at location L5: transient pulses, transient oscillations, voltage drops, and current surges, as shown in Table 1, totaling 52 abnormal operating conditions with different parameters. The duration of the fault and disturbance was set to 0.1s (5 cycles; the operating time of relay protection devices is generally 4-6 cycles), and voltage was the data acquisition target, with a sampling frequency of 5kHz, i.e., 100 points were sampled per cycle.

[0083] This invention uses the EEMD algorithm to extract the characteristic frequencies characterized by the time-frequency response of abnormal waveforms. Figure 4 The results of processing the single-phase ground fault voltage signal at the wind turbine terminal using Fourier transform, EMD, and EEMD are shown, with the fault voltage waveform at the wind turbine terminal shown as the red curve. Due to the selection of the sliding window in the Fourier transform, its adaptability to multi-scale signal processes or abrupt changes is poor; therefore, the decomposed spectrum cannot yield representative fault frequency characteristics. Comparing EMD and EEMD, both decompose the original signal into IMF components represented by high and low frequencies. As shown in the figure, compared to EMD, EEMD can capture more frequency components, refining the fault characteristics. This indicates that EEMD better decomposes the abnormal waveform into signals with their own time-scale characteristics, solving the mode mixing problem of EMD and demonstrating the superiority of EEMD in extracting electrical anomaly characteristics of offshore wind farms.

[0084] This invention uses wind turbine terminal faults and onshore grid connection point disturbances as examples to demonstrate the electrical anomaly features extracted by the EEMD algorithm, such as... Figure 5 As shown. Figure 5 The EEMD decomposition diagrams for unidirectional and three-phase grounding faults at the terminal of offshore wind turbines are presented. The resulting IMFs (Integrated Frequency Components) 1-7 are composed of high and low frequency components, arranged in order from high to low frequency. Each IMF component contains characteristic information of the voltage waveform in each frequency band under abnormal electrical conditions. Subsequently, the IMFs obtained from the EEMD decomposition of the abnormal signal are used as fault features under abnormal conditions and reconstructed into multidimensional time series T1-T7. For this purpose, an offshore wind farm electrical abnormal state feature sample library is created as follows: Figure 6 As shown.

[0085] Table 1 Types of Electrical Anomalies in Offshore Wind Farms

[0086]

[0087] The electrical anomaly state feature library established based on simulation data is used to identify the state of the measured data. Figure 7 The figure shows a partial feature sample library of abnormal states, including common fault and disturbance types in offshore wind farms. The red curve in the figure shows the measured abnormal data. The identification method based on MDMP realizes the matching of the identified object with the IMF feature components in the feature sample library.

[0088] Depend on Figure 7 As shown, the MDMP identification method was used to identify the attributes of the best dimension subsequence pairs, and the high and low frequency IMF feature components of fault type 1 in the sample library were successfully matched with the tested object, thus determining the fault type of the current tested data and realizing the identification of electrical anomalies in offshore wind farms.

[0089] Furthermore, considering that in the actual operation and maintenance of offshore wind farms, the identification of abnormal states is affected by harsh environmental factors such as extreme weather, which can cause the high-frequency components of the IMF decomposed by EEMD to be interfered with by noise and thus fail to accurately characterize the state characteristics. To verify the effectiveness of the abnormal state identification strategy based on MDMP in various scenarios, this invention analyzes the high-frequency signals of abnormal states under fault location, fault type, different transition resistance magnitudes, and lightning strike effects, and realizes the identification of electrical abnormal states in offshore wind farms based on the IMF combination results selected by MDMP, as shown in Table 2.

[0090] Table 2. Results of IMF combination selected by MDMP in various scenarios

[0091]

[0092]

[0093] As can be seen from Table 2, under high-frequency noise interference, the combination of IMF components selected by MDMP for various fault conditions in the electrical anomaly state of offshore wind farms is different. The selected IMF components take into account both high-frequency and low-frequency components, indicating that the MDMP identification method comprehensively considers the high, medium and low frequency components of the IMF and achieves anomaly identification. This fully demonstrates the adaptive capability and robustness of the proposed method in anomaly identification.

[0094] To further demonstrate the identification accuracy and effectiveness of the proposed method, it was compared with two other identification strategies: VMD and LS-SVM, and WT and Logistic Regression. The results are shown in Table 3.

[0095] Table 3 shows the accuracy comparison results between the proposed method and other methods.

[0096]

[0097] Because the number of samples used in the modeling process is small, the two algorithms compared did not achieve satisfactory recognition accuracy. On the contrary, the recognition method proposed in this invention does not require a large number of samples to train the recognition model. It is not only time-efficient but also achieves excellent recognition performance, which demonstrates the effectiveness and superiority of the method.

[0098] The identification strategy proposed in this invention can be used for real-time monitoring of both large-scale steady-state SCADA data and high-precision real-time power quality data in the actual electrical anomaly monitoring of offshore wind power. The engineering applicability of the MDMP-based identification method is verified using SCADA monitoring data as an example. SCADA data from an offshore wind farm in Shanghai under actual abnormal operating conditions of its turbines were selected. This data was sampled from a 13MW turbine with a ten-minute average wind speed of 11.2m / s, at a sampling frequency of 30s, totaling 20,160 data points.

[0099] Results of identifying real anomaly sample data of offshore wind fields as follows Figure 8 As shown, the MDMP identification method was used to achieve the matching of high-frequency features T1, low-frequency features T4 and T7. The red curve in the figure indicates that the identified object and the feature sample library were successfully matched, which shows the effectiveness of the method proposed in this invention for processing real data.

[0100] Furthermore, the MDMP algorithm proposed in this invention utilizes DTW as a similarity measure between multidimensional time series. Compared to traditional Euclidean distance, DTW not only reflects the trend and shape similarity of the dynamic changes in time series, but also compensates for the window adaptability problem in the algorithm. The results of MDMP calculation of multidimensional subsequence distances are shown in Table 4, where the blue highlighted parts represent the distance values ​​of the closest subsequences in different dimensions.

[0101] Table 4 shows the distance calculation of multidimensional subsequences based on the MDMP algorithm.

[0102]

[0103] To verify the effectiveness of the heartbeat mechanism designed in this invention, this invention simulates the application of the proposed identification strategy to the actual abnormal state detection process using real SCADA data, setting v b and v m Set to 2m / s and 25m / s respectively, U b and f b Set them to 0.05U respectively. 额 and 0.05f 额 Table 5 compares the anomaly identification results with and without the embedded heartbeat mechanism. It can be seen that the detection count and time of the identification strategy using the heartbeat mechanism are significantly reduced compared to the detection process without the heartbeat mechanism, specifically by 75.89%. In addition, although the identification frequency is significantly reduced, the identification rate using the heartbeat mechanism is still 100%, which indicates that there is no problem of missed detection of abnormal states due to the embedding of the heartbeat mechanism, thus verifying the effectiveness of the heartbeat mechanism in anomaly identification.

[0104] Table 5. Anomaly identification results for embedded and non-embedded heartbeat mechanisms.

[0105]

[0106] This invention proposes a method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix contours, addressing the problem of insufficient identification accuracy due to the scarcity of offshore wind farm state data samples. Taking common fault states and disturbance phenomena in offshore wind farms as the subject of the invention, the effectiveness of the algorithm is demonstrated through theoretical analysis, simulation calculations, and verification with actual data. The EEMD-based feature extraction strategy for electrical anomalies in offshore wind farms effectively decomposes abnormal waveforms into signals with their own time-dependent scales, providing a basis for identifying anomalies in offshore wind farms. The similarity measurement method between multidimensional matrix contour subsequences based on the improved DTW algorithm reflects the changing trends and frequency differences of time series segments; the embedding of the heartbeat mechanism effectively reduces the computational burden in the actual detection process and avoids missed detections, demonstrating broad engineering application value.

[0107] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles, characterized in that, The method includes: Acquire unidentified data for offshore wind farms; The data to be identified is decomposed using the EEMD algorithm to obtain IMF components; An abnormal state feature sample library is constructed. An abnormal state is identified within this library using a multidimensional matrix contour algorithm. The shape difference between the sequence to be identified and the sequence in the abnormal sample feature library is determined using the DTW distance metric, and corresponding abnormal state features are matched. Specifically, this includes: Initialize the index matrix in the multidimensional matrix contour to store the most similar subsequences; The distance between the current query sequence and its subsequences is calculated using the DTW algorithm and stored in the distance matrix D. Sort matrix D in ascending order by column and obtain the index of the sorted matrix. Accumulate the sorted matrix D by column and average the results to obtain the multidimensional matrix outline and index. Determine the most similar multidimensional subsequence and its position by using the minimum distance and index to achieve matching of similar subsequences. The distance judgment is performed based on the most similar subsequence identified by the multidimensional matrix contour algorithm. The identification mechanism is as follows: in, The average distance between the most similar subsequences, {d1,d2,…,d N } represents the average distance of the most similar subsequences under N real abnormal state events. When the output is "1", the current object to be identified is the abnormal type matched by the abnormal feature library; otherwise, the current offshore wind farm is operating normally.

2. The method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles as described in claim 1, characterized in that, The construction of the abnormal state feature sample library specifically includes: Obtain the time series of abnormal states of offshore wind farms; Remove noise from the time series of the abnormal states; The denoised abnormal state time series is decomposed into IMF components, and a multidimensional time series is reconstructed based on the IMF components to construct an abnormal state feature sample library.

3. The method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles as described in claim 1, characterized in that, The method further includes: The wind field detection status is determined based on the heartbeat mechanism.

4. The method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles as described in claim 3, characterized in that, The heartbeat mechanism sends a specific signal to the server at fixed intervals. Upon receiving this signal, the server replies with a response signal, which determines the current communication status of the client. The heartbeat mechanism has a triggering mechanism, which is as follows: in, t 1 represents the time when the heartbeat signal was last sent. t 2 represents the current time when the heartbeat signal is sent. v b and v m These are the baseline value and the maximum wind speed, respectively, for wind speed fluctuations. U b and f b These are the reference values ​​for voltage and frequency fluctuations, respectively. If the output of s is 1, it indicates that the heartbeat packet signal received by the computer is normal. At this time, the abnormal state identification system is in a low-frequency detection state. When the s signal received by the computer is 0, the proposed identification strategy is started. After a period of high-frequency detection, if no abnormality is displayed and the received heartbeat packet signal is 1, the system enters the low-frequency detection state.

5. The method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles as described in claim 1, characterized in that, The method further includes: Determine whether the matched abnormal state features meet the requirements; If the matched abnormal state features meet the requirements, then the identified abnormal state is determined to be correct.

6. The method for identifying electrical anomalies in offshore wind farms based on multidimensional matrix profiles as described in claim 5, characterized in that, The method further includes: If the identified abnormal state characteristics do not meet the requirements, then the identified abnormal state is determined to be incorrect, and the data to be identified is not abnormal.

Images