A method for detecting abnormal PMU frequency data

Abstract

A detection method for abnormal frequency data of a phasor measurement unit (PMU). The method includes: generating respective frequency sequence data according to the sampling points monitored by the reference station and the measuring station; extracting the maximum value sequence and the minimum value sequence composed of time serial numbers according to the frequency values ​​of the elements in the frequency sequence data; adapting the maximum value sequence and the minimum value sequence of the measuring station according to the maximum value sequence and the minimum value sequence of the reference station; merging the maximum value sequence and the minimum value sequence of the reference station and the measuring station respectively to generate the key feature point sequence of the reference station and the measuring station; subtracting the elements in the key feature point sequence of the reference station from the corresponding elements in the key feature point sequence of the measuring station to obtain a difference sequence; if the absolute value of any element of the difference sequence exceeds the delay threshold, the output abnormality is that there is a frequency delay. The present disclosure can detect various abnormal situations in a refined manner.

Description

Technical Field

[0001] The present invention belongs to the field of WAMS (Wide Area Measurement System) / PMU (Phasor Measurement Unit) data quality evaluation, and more specifically, relates to a method for detecting abnormal PMU frequency data. Background Art

[0002] The real-time dynamic monitoring system of the power system, also known as WAMS, mainly uses the high-precision dynamic real-time data with a unified time scale collected and uploaded by the synchronous PMU to accurately monitor and analyze the dynamic process of the power system over a wide area. Since PMU data has the characteristics of high collection accuracy, high transmission density, and high time scale accuracy, it is of great significance for the dynamic analysis of the power system and subsequent accident handling. However, after the PMU device is put into operation, the corresponding PMU data in the WAMS system is affected by factors such as the inherent design defects of the PMU device, equipment aging, operating conditions, and external equipment interference, resulting in varying degrees of quality problems in the acquired measurement data, which is then characterized by different types of data deviation anomalies. At present, the problem analysis and detection of PMU abnormal data is still at the stage of manual regular inspections, and there are problems such as cumbersome on-site detection processes and difficult detection, so the problems that cause abnormal data cannot be discovered in time.

[0003] The PMU device can obtain synchronous sampling data such as the amplitude and phase angle of three-phase voltage and current, and calculate power, frequency, etc. Among them, since the frequencies between nodes in the regional power system at the same time have a certain mutual reference value, through information mining and analysis of its data representation, PMU data channels with data quality problems can be found, which helps the daily manual inspection work of the WAMS system.

[0004] At present, the methods for detecting abnormal data of PMU are mainly divided into two categories: 1. Methods based on system state estimation (SE); 2. Data-driven methods. If the system-wide PMU measurements are redundant for calculating network power flow, SE can be effectively performed to detect abnormal data of PMU. However, accurate network parameters and topology information are prerequisites for ensuring the effectiveness of SE methods. In actual engineering applications, for example, line operating parameters may change due to uncertain factors such as the external environment, so the performance of such methods is often not well guaranteed. The data-driven method realizes abnormality detection based on the potential correlation between normal measurement data at each point in the system under the constraints of the power system model (such as the dynamic similarity of frequency data of each node in the regional system). It does not need to explicitly model the more complex power system model, so it does not have the limitations of traditional SE methods. Summary of the invention

[0005] In order to address the deficiencies in the prior art, the purpose of the present invention is to construct detection rules for different PMU frequency data anomaly types to perform refined detection, thereby quickly locating the PMU data channel with quality problems and efficiently realizing problem detection and defect elimination processing for the PMU data channel.

[0006] The present invention adopts the following technical solution.

[0007] A method for detecting abnormal PMU frequency data, wherein the phasor measurement unit includes a reference station and a measurement station, and the method includes:

[0008] Step 1, generating frequency series data of the reference station and the measurement station respectively according to the sampling points monitored by the reference station and the measurement station, wherein the elements of the frequency series data include: frequency value and time serial number;

[0009] Step 2, extracting a maximum value sequence and a minimum value sequence respectively according to the frequency values ​​of the elements in the frequency sequence data, wherein each element in the maximum value sequence and the minimum value sequence is composed of a time sequence number;

[0010] Step 3, adapting the maximum value sequence and the minimum value sequence of the measurement station according to the maximum value sequence and the minimum value sequence of the reference station;

[0011] Step 4, merging the maximum value sequence and the minimum value sequence of the reference station to generate a key feature point sequence of the reference station;

[0012] Step 5, merging the maximum value sequence and the minimum value sequence of the measurement station to generate a key feature point sequence of the measurement station;

[0013] Step 6, subtracting the elements in the key feature point sequence of the reference station from the corresponding elements in the key feature point sequence of the measurement station to obtain a difference sequence;

[0014] Step 7, determining whether the absolute value of any element of the difference sequence exceeds a delay threshold;

[0015] Step 8: In response to the absolute value of any element of the difference sequence exceeding a preset threshold, outputting an abnormality that there is a frequency delay.

[0016] Furthermore, the step 1 also includes:

[0017] Perform filtering and denoising on each element in the frequency series data: Among them, x(i) is the frequency value of the element with time number i in the frequency series data before filtering and denoising, y(i) is the frequency value of the element with time number i in the frequency series data after filtering and denoising, and k is the length of the sliding window.

[0018] Furthermore, the length of the sliding window is 600.

[0019] Furthermore, the step 3 specifically includes:

[0020] In the reference station, the first element R of the maximum value sequence is obtained respectively max (1) and the first element R of the minimum sequence min (1);

[0021] In the measuring station, the first element E of the maximum value sequence is obtained respectively max (1) and the first element E of the minimum sequence min (1);

[0022] If R max (1) Smaller than R min (1), and E max (1) Greater than E min (1), further determine R max (1) With E min (1) If R max (1) Less than E min (1), then delete the first element of the maximum sequence of the reference station, if R max (1) Greater than E min (1), then delete the first element of the minimum value sequence of the measuring station;

[0023] If R max (1) Greater than R min (1), and E max (1) Less than E min (1), further determine R min (1) With E min (1) If R min (1) Less than E max (1), then delete the first element of the minimum sequence of the reference station, if R min (1) Greater than E max (1), then delete the first element of the maximum value sequence of the measuring station;

[0024] In the reference station, the last element R of the maximum value sequence is obtained respectively max (-1) and the last element R of the minimum sequence min (-1);

[0025] In the measuring station, the last element E of the maximum value sequence is obtained respectively max (-1) and the last element E of the minimum sequence min (-1);

[0026] If R max (-1) is less than R min (-1), and E max (-1) greater than E min (-1), further determine R min (-1) and E max (-1), if R min (-1) greater than E max (-1), then delete the last element of the minimum sequence of the reference station. If R min (-1) is less than E max (-1), then delete the last element of the maximum value sequence of the measuring station;

[0027] If R max (-1) greater than R min (-1), and E max (-1) is less than E min (-1), further determine R max (-1) and E min (-1), if R max (-1) greater than E min (-1), then delete the last element of the maximum sequence of the reference station. If R max (-1) is less than E min (-1), then delete the last element of the minimum value sequence of the measuring station.

[0028] Furthermore, the delay threshold is 50.

[0029] Furthermore, the step 1 also includes:

[0030] If the frequency value of an element in the frequency sequence data is less than 49.5 Hz or greater than 50.5 Hz, the output abnormality is that the frequency has an obvious outlier value.

[0031] If the frequency values ​​of more than 80% of the elements in the frequency sequence data are the same value, the output abnormality is that the frequency does not refresh error.

[0032] Furthermore, the step 5 also includes:

[0033] If the number of elements in the key feature point sequence of the reference station is inconsistent with the number of elements in the key feature point sequence of the measurement station, the output abnormality is that the frequency cannot follow the reference station.

[0034] Furthermore, the method further comprises:

[0035] Obtaining time delay information of the measurement station relative to the reference station according to the difference sequence;

[0036] In combination with the time delay information, subtract the frequency value of the element in the frequency series data of the reference station from the frequency value of the corresponding element in the frequency series data of the measurement station to obtain a frequency difference sequence;

[0037] If half of the elements in the frequency difference sequence are greater than +0.001 Hz, the output abnormality is that the overall frequency is too high;

[0038] If half of the elements in the frequency difference sequence are less than -0.001 Hz, the output abnormality is that the overall frequency is too low.

[0039] Furthermore, the method further comprises:

[0040] Sorting the elements in the frequency difference sequence from small to large;

[0041] Calculate the quartile difference Q = Q of the sorted frequency difference sequence 3n / 4 -Q n / 4 , where n is the total number of elements in the frequency difference sequence, Q i is the i-th element of the sorted frequency difference sequence;

[0042] If Q is greater than 0.0025Hz, the output abnormality is that the overall frequency fluctuates significantly.

[0043] A refined detection system for PMU frequency data anomaly, comprising: a monitoring module, a data processing module, a logic judgment module, a calculation module and an output module, characterized in that:

[0044] The monitoring module is used to monitor the sampling points to generate frequency series data of the reference station and the measurement station;

[0045] The data processing module is used to extract the maximum value sequence and the minimum value sequence respectively according to the frequency values ​​of the elements in the frequency sequence data, and merge the maximum value sequence and the minimum value sequence of the reference station to generate the key feature point sequence of the reference station, and merge the maximum value sequence and the minimum value sequence of the measurement station to generate the key feature point sequence of the measurement station;

[0046] The logic judgment module is used to adapt the maximum value sequence and the minimum value sequence of the measurement station according to the maximum value sequence and the minimum value sequence of the reference station, and to judge whether the absolute value of any element of the difference value sequence exceeds the delay threshold;

[0047] The calculation module is used to subtract the elements in the key feature point sequence of the reference station from the corresponding elements in the key feature point sequence of the measurement station to obtain a difference sequence;

[0048] The output module is used to output abnormal results.

[0049] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0050] (1) The present invention takes into account the frequency response characteristics of regional power system nodes under small system disturbances such as load changes and transformer tap switching, and constructs detection rules according to the data anomaly types. It can realize the refined detection and identification of 7 types of common anomalies, namely: 1) obvious outliers appear in the frequency; 2) the frequency has non-refresh errors; 3) the frequency cannot follow the reference station; 4) there is a time delay in the frequency; 5) the overall frequency is high; 6) the overall frequency is low; 7) the overall frequency fluctuates significantly.

[0051] (2) The present invention takes into account the time synchronization of PMU sampling data and the frequency characteristics of regional power grid nodes, and combines manual experience to introduce reference stations for frequency data comparison, effectively quantifying the difference between the frequency data of the measurement station and the frequency data of the reference station. Further, an abnormal feature mining method based on key feature points of the sequence is proposed, which extracts the effective features of the PMU frequency sequence data delay and follow-up anomalies, which helps to deeply analyze the abnormal type, so as to achieve the purpose of refined detection.

[0052] (3) The present invention has strong industrial application potential and helps to identify PMU frequency data anomalies caused by various problems after the PMU device is officially put into operation and which are difficult to detect using expert experience, thereby quickly locating the PMU data channel with quality problems and efficiently realizing problem detection and defect elimination processing of the PMU data channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The invention is a flow chart of a method for detecting abnormal PMU frequency data. DETAILED DESCRIPTION

[0054] The present application is further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present application.

[0055] The main purpose of the present invention is to perform refined detection and identification of PMU frequency data anomalies by considering the frequency response characteristics of regional power system nodes under small system disturbances such as load changes and transformer tap switching. To achieve the above purpose, the technical solution adopted by the present invention mainly includes the following steps:

[0056] (1) Obtain the measured frequency data of the regional power grid PMU measurement points collected by the WAMS system of a province. The specific steps are as follows:

[0057] (1.1) Considering the overall grid structure and tidal current transmission characteristics of the province, a plant station was selected as the reference station based on the on-site engineering practice experience. The measurement data of the reference station is of high quality and there is no data deviation abnormality.

[0058] (1.2) There may be multiple PMU measurement points in the same plant. For example, a new energy plant may deploy PMU measurement points at the main transformer, busbar, and outgoing line. At the same time, the PMU measurement points in a plant are mostly from the same manufacturer. Therefore, the PMU data in the same plant are highly similar, and the slight differences between the PMU data will not affect the subsequent refined detection and identification. Therefore, consider selecting a representative measurement point to represent all the measurement point data of the plant where it is located to simplify the subsequent detection and identification work.

[0059] (2) According to the actual situation of the acquired frequency data, corresponding processing and detection and identification are performed. The specific steps are as follows:

[0060] (2.1) The sampling frequency of the acquired frequency data is 50 Hz, so the frequency of the measuring station within 5 minutes is represented by a time series containing 5×60×50=15000 sampling points. Within the time scale of 5 minutes, the frequency series data of the measuring station to be detected and analyzed are traversed to perform the following types of data anomaly detection. It should be noted that each element in the frequency series data includes: frequency value and time sequence number.

[0061] (2.1.1) According to the national grid frequency standards, the normal grid frequency is 50 Hz. For grids with a capacity of 3 million kilowatts or more, the deviation shall not exceed ±0.2 Hz; for grids with a capacity of less than 3 million kilowatts, the deviation shall not exceed ±0.5 Hz.

[0062] Therefore, considering the differences in the grid capacity of power systems in different regions, 49.5-50.5Hz is set as the normal range of sampling point frequency of the measuring station within the time scale. If it meets the normal range requirements, the next step (2.1.2) detection and analysis is carried out, otherwise it is identified as the abnormal type of "obvious outliers in frequency" and the detection is terminated.

[0063] (2.1.2) The power system is always under the influence of external disturbances. When a large active power imbalance suddenly occurs due to the disturbance, the system frequency will undergo a relatively obvious transition process, which is manifested as: the frequency of each node in the regional system fluctuates around the normal value of 50Hz, and under normal circumstances, the overall frequency transition process of each measuring station and the reference station remains consistent.

[0064] Therefore, the sampling frequency of the measuring station within the time scale is detected to see if it is a fixed value. By obtaining the distribution of elements in the frequency series data, the element with the largest number of occurrences is found, and it is determined whether its occurrence number exceeds 80% of the sampling points. If the number of occurrences of the element is less than 80% of the sampling points, the next step (3) detection and analysis is performed. Otherwise, it is identified as an abnormal type of "frequency occurrence non-refresh error" and the detection is terminated.

[0065] (3) At the same time, considering the response order and delay of the frequency of each node in the regional system, that is, under normal circumstances, the closer the location is to the disturbance point of the power system, the earlier the frequency responds and then begins to change. Therefore, under normal circumstances, the frequency sequence data of different measurement stations have a certain lead or lag relationship compared with the reference station. After multiple tests and analyses, it was found that the delay was relatively subtle, while the frequency data obtained from the PMU data channel with data quality problems had a more obvious delay, and there was a certain difference between the two.

[0066] In order to further detect and analyze the abnormal types, a method for mining the abnormal features of frequency data based on key feature points of the sequence is proposed. The implementation ideas are as follows:

[0067] (3.1) Since the original frequency sequence data is measured data, a large amount of environmental noise and random errors are inevitably introduced in the sampling process, which is characterized by the presence of a large number of burrs and spikes in the drawn frequency sequence data waveform, making it difficult to extract key feature points from it.

[0068] Therefore, considering using a high-smoothness sliding average filtering technology on the frequency series data of the reference station and the measurement station within a time scale of 5 minutes, a completely consistent filtering and denoising process is performed to obtain the corresponding smooth curve reflecting the overall transition process of the original sequence data.

[0069] The basic idea of ​​the sliding average filtering technique is to set a sliding window of fixed length, which slides along the time series, and takes the arithmetic mean of the data in the window as the output value, which constitutes the filtered time series. Let k be the length of the sliding window, then The input and output are x(n) and y(n) respectively. In order to reduce the workload of subsequent analysis, combined with the transition process characteristics of the measured frequency sequence data, the present invention sets the sliding window length of the sliding average filter to about 1200 sampling points, that is, k=600.

[0070] (3.2) The sampling point position sequences of the minimum and maximum values ​​in the smooth curves of the reference station and the measurement station are obtained, respectively: are all characterized as one-dimensional sequences. min , R max is the minimum and maximum value point sequence of the reference station, E min、E max is the minimum and maximum value point sequence of the measuring station, N R_min 、N R_max is the number of minimum and maximum value points of the reference station, N E_min 、N E_max is the number of minimum and maximum value points of the measuring station. It should be noted that The elements in consist only of time serial numbers.

[0071] (3.3) According to step (3), the frequency series data of different measuring stations have a certain lead or lag relationship. In the detection and analysis process, since only the frequency series data within the 5-minute time scale is intercepted for observation and analysis, the extreme value conditions at the edge of each measuring station may be different. Therefore, the 4 extreme point sequences obtained in step (3.2) are corrected to eliminate the edge effect and obtain the key feature points of the frequency series data. The specific steps are as follows:

[0072] (3.3.1) Obtain the endpoint sequence of the smooth curve of the reference station R and the measurement station E, denoted as P.

[0073] (3.3.2) If the elements in the P sequence are in the sequence If it appears in , it will be eliminated and a new sequence will be obtained.

[0074] (3.3.3) Take the sequence respectively The first element in is denoted by R max (1) R min (1). Calculate the head direction R of the reference station dir_start , R dir_start =R min (1)-R max (1), the positive or negative value represents the direction. Similarly, the head direction E of the measuring station is obtained dir_start .

[0075] (3.3.4) If the head directions of the measuring station and the reference station are inconsistent, the head end edge effect elimination processing is performed: Get R max (1) R min (1) E max (1) E min (1) and in the corresponding sequence Otherwise, the next step is to eliminate the edge effect at the end.

[0076] (3.3.5) Similarly, calculate the terminal direction R of the reference station and the measurement station dir_end 、E dir_end, and make a direction consistency judgment. If the end directions of the measuring station and the reference station are inconsistent, the edge effect elimination processing of the end is performed: Get R max (-1), R min (-1), E max (-1), E min (-1) and in the corresponding sequence Otherwise, proceed to the next step.

[0077] (3.3.6) Merge the data processed by steps (3.3.1)-(3.3.5) Get the key feature point sequence R of the reference station frequency series data K 1×N Similarly, we can get the key feature point sequence E of the measuring station K 1×N , are represented as one-dimensional sequences, and N is the number of key feature points.

[0078] (3.4) Considering step (2.1.2), under normal circumstances, the overall frequency transition process of each measurement station is consistent with that of the reference station. Check whether the number of key feature points in the time scale of the measurement station is consistent with that of the reference station. If it meets the consistency requirements, proceed to the next step (3.5) detection and analysis, otherwise it will be identified as the abnormal type of "frequency cannot follow the reference station" and the detection is terminated.

[0079] (3.5) Since the frequency series data of the reference station and the measurement station have been subjected to completely consistent filtering and denoising in step (3.1), and the smoothed frequency series data reflects the key transition characteristics of the original frequency series data. The difference sequence of the key feature points of the measurement station and the reference station is D K 1×N , D Ki =E Ki -R Ki , where E K 1×N , R K 1×N are the key feature point sequences of the measurement station and the reference station respectively. By analyzing the difference sequence, the time delay of the measurement station compared with the reference station is detected. Ki If the absolute value exceeds 50 sampling points (i.e., the delay is ±1 second), it is identified as the abnormal type of "frequency has delay" and the delay is

[0080] (4) Within the 5-minute time scale, further intercept a period of time in which the frequency data shows the same change trend as the observation time window. The method is to select the first maximum sampling point and the minimum sampling point after the correction processing in step (3.3) as the two ends of the observation time window. At the same time, if there is a delay in the measurement station, the delay of the measurement station relative to the reference station obtained in step (3.5) is eliminated, where the (overall) delay of the measurement station relative to the reference station is the average value of the delay of each element. Then compare the difference in the original frequency series data between the measurement station and the reference station in the observation time window, and the frequency difference sequence is Among them, M i =E 0i -R 0i , N m is the length of the observation time window, E0 and R0 are the original frequency sequence data in the observation time window.

[0081] (4.1) In the observation time window, traverse the frequency difference sequence The following types of data anomaly detection are performed, where the detection thresholds are set after actual manual verification:

[0082] (4.1.1) If If more than 50% of the sampling points have values ​​higher than +0.001Hz, it is identified as an abnormal type of "overall frequency is too high".

[0083] (4.1.2) If If more than 50% of the sampling points have values ​​lower than -0.001 Hz, it is identified as an "overall low frequency" anomaly type.

[0084] (4.1.3) If If the interquartile difference is higher than 0.0025Hz, it is identified as an abnormal type of "obvious overall frequency fluctuation". 3n / 4 -Q n / 4 , n is the total number of elements in the frequency difference sequence. After the values ​​of each element in the sequence are arranged in ascending order, Q n / 4 (First quartile) indicates that 25% of the element values ​​in the sequence data are less than or equal to Q n / 4 , Q 3n / 4 (Third quartile) indicates that 75% of the element values ​​in the sequence data are less than or equal to Q 3n / 4 The interquartile range reflects the degree of dispersion of the middle 50% of the data in the sequence. The larger the value, the more dispersed the data in the middle. Therefore, the interquartile range is used to measure the overall fluctuation of the sequence.

[0085] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal PMU frequency data, wherein: The PMU includes a reference station and a measurement station, and is characterized in that it includes the following steps: Step 1, generating frequency series data of the reference station and the measurement station respectively according to the sampling points monitored by the reference station and the measurement station, wherein the elements of the frequency series data include: frequency value and time serial number; Step 2, extracting a maximum value sequence and a minimum value sequence respectively according to the frequency values ​​of the elements in the frequency sequence data, wherein each element in the maximum value sequence and the minimum value sequence is composed of a time sequence number; Step 3, adapting the maximum value sequence and the minimum value sequence of the measurement station according to the maximum value sequence and the minimum value sequence of the reference station; Specifically, the first element R of the maximum value sequence is obtained in the reference station. max (1) and the first element R of the minimum sequence min (1); In the measuring station, the first element E of the maximum value sequence is obtained respectively max (1) and the first element E of the minimum sequence min (1); If R max (1) Smaller than R min (1), and E max (1) Greater than E min (1), further determine R max (1) With E min (1) If R max (1) Less than E min (1), then delete the first element of the maximum sequence of the reference station, if R max (1) Greater than E min (1), then delete the first element of the minimum value sequence of the measuring station; If R max (1) Greater than R min (1), and E max (1) Less than E min (1), further determine R min (1) With E min (1) If R min (1) Less than E max (1), then delete the first element of the minimum sequence of the reference station, if R min (1) Greater than E max (1), then delete the first element of the maximum value sequence of the measuring station; In the reference station, the last element R of the maximum value sequence is obtained respectively max (-1) and the last element R of the minimum sequence min (-1); In the measuring station, the last element E of the maximum value sequence is obtained respectively max (-1) and the last element E of the minimum sequence min (-1); If R max (-1) is less than R min (-1), and E max (-1) greater than E min (-1), further determine R min (-1) and E max (-1), if R min (-1) greater than E max (-1), then delete the last element of the minimum sequence of the reference station. If R min (-1) is less than E max (-1), then delete the last element of the maximum value sequence of the measuring station; If R max (-1) greater than R min (-1), and E max (-1) is less than E min (-1), further determine R max (-1) and E min (-1), if R max (-1) greater than E min (-1), then delete the last element of the maximum sequence of the reference station. If R max (-1) is less than E min (-1), then delete the last element of the minimum value sequence of the measuring station; Step 4, merging the maximum value sequence and the minimum value sequence of the reference station to generate a key feature point sequence of the reference station; Step 5, merging the maximum value sequence and the minimum value sequence of the measurement station to generate a key feature point sequence of the measurement station; Step 6, subtracting the elements in the key feature point sequence of the reference station from the corresponding elements in the key feature point sequence of the measurement station to obtain a difference sequence; Step 7, determining whether the absolute value of any element of the difference sequence exceeds a delay threshold; Step 8: In response to the absolute value of any element of the difference sequence exceeding a preset threshold, outputting an abnormality that there is a frequency delay.

2. A method for detecting abnormal PMU frequency data according to claim 1, characterized in that: The step 1 also includes: Perform filtering and denoising on each element in the frequency series data: Among them, x(i) is the frequency value of the element with time number i in the frequency series data before filtering and denoising, y(i) is the frequency value of the element with time number i in the frequency series data after filtering and denoising, and k is the length of the sliding window.

3. A method for detecting abnormal PMU frequency data according to claim 2, characterized in that: The length of the sliding window is 600.

4. A method for detecting abnormal PMU frequency data according to claim 1, characterized in that: The delay threshold is 50.

5. A method for detecting abnormal PMU frequency data according to claim 1, characterized in that: The step 1 also includes: If the frequency value of an element in the frequency sequence data is less than 49.5 Hz or greater than 50.5 Hz, the output abnormality is that the frequency has obvious outliers; If the frequency values ​​of more than 80% of the elements in the frequency sequence data are the same value, the output abnormality is that the frequency does not refresh error.

6. A method for detecting abnormal PMU frequency data according to claim 1, characterized in that: The step 5 also includes: If the number of elements in the key feature point sequence of the reference station is inconsistent with the number of elements in the key feature point sequence of the measurement station, the output abnormality is that the frequency cannot follow the reference station.

7. A method for detecting abnormal PMU frequency data according to claim 1, characterized in that: The method further comprises: Obtaining time delay information of the measurement station relative to the reference station according to the difference sequence; In combination with the time delay information, subtract the frequency value of the element in the frequency series data of the reference station from the frequency value of the corresponding element in the frequency series data of the measurement station to obtain a frequency difference sequence; If half of the elements in the frequency difference sequence are greater than +0.001 Hz, the output abnormality is that the overall frequency is too high; If half of the elements in the frequency difference sequence are less than -0.001 Hz, the output abnormality is that the overall frequency is too low.

8. A method for detecting abnormal PMU frequency data according to claim 7, characterized in that: The method further comprises: Sorting the elements in the frequency difference sequence from small to large; Calculate the quartile difference Q = Q of the sorted frequency difference sequence 3n / 4 -Q n / 4 , where n is the total number of elements in the frequency difference sequence, Q i is the i-th element of the sorted frequency difference sequence; If Q is greater than 0.0025Hz, the output abnormality is that the overall frequency fluctuates significantly.

9. A refined detection system for PMU frequency data anomaly, running the method according to any one of claims 1 to 8, comprising: The monitoring module, data processing module, logic judgment module, calculation module and output module are characterized by: The monitoring module is used to monitor the sampling points to generate frequency series data of the reference station and the measurement station; The data processing module is used to extract the maximum value sequence and the minimum value sequence respectively according to the frequency values ​​of the elements in the frequency sequence data, and merge the maximum value sequence and the minimum value sequence of the reference station to generate the key feature point sequence of the reference station, and merge the maximum value sequence and the minimum value sequence of the measurement station to generate the key feature point sequence of the measurement station; The logic judgment module is used to adapt the maximum value sequence and the minimum value sequence of the measurement station according to the maximum value sequence and the minimum value sequence of the reference station, and to judge whether the absolute value of any element of the difference value sequence exceeds the delay threshold; The calculation module is used to subtract the elements in the key feature point sequence of the reference station from the corresponding elements in the key feature point sequence of the measurement station to obtain a difference sequence; The output module is used to output abnormal results.

Images