Automatic weather station wind direction abnormity identification method, system and equipment based on machine learning and storage medium

Abstract

The invention discloses an automatic weather station wind direction anomaly identification method, system and device based on machine learning and a storage medium. The identification method comprises the steps of selecting typical weather process time periods; constructing a parameter model suitable for wind direction anomaly classification; performing wind direction anomaly detection, and identifying wind direction anomaly data of the automatic weather station by using the constructed parameter model suitable for wind direction anomaly classification; according to the method, indirect abnormal characterization monitoring can be carried out under the condition that hardware of an observation instrument of the automatic weather station is not detected, wrong wind direction data with high concealment can be effectively detected, and the accuracy of the abnormal wind direction identification is improved. The wind direction data quality of the automatic weather station is improved, and an accurate wind direction data set support is provided for real-time weather monitoring, future weather change prediction and climate change trend and rule knowing.

Description

Technical Field

[0001] The present invention belongs to the field of meteorological technology, and in particular relates to a method, system, device and storage medium for identifying wind direction anomalies in an automatic weather station based on machine learning. Background Art

[0002] Wind measurement data plays a vital role in meteorological monitoring and early warning, meteorological services, climate statistical analysis, urban ventilation corridor planning, environmental pollution meteorological conditions analysis, port gale navigation control and other industry scenarios. Since automatic weather stations are built in diverse geographical environments such as urban green spaces, rooftops of high-rise buildings, mountainous areas, port terminals, islands, and highways, their distribution points are numerous and wide, and all are unmanned stations, which brings considerable challenges to the maintenance of instruments, affecting the quality of observation data of automatic weather stations. Some stations may have abnormal or wrong wind directions. If these abnormal values ​​are not detected and processed in time, it will cause misjudgment and misleading of meteorological early warning monitoring and climate statistics, and thus have an adverse impact on meteorological service decision-making and climate statistics.

[0003] Wind direction anomaly detection methods are mainly divided into two categories: observation instrument hardware detection method and indirect anomaly characterization analysis method based on observation data. For the observation instrument hardware detection method, the significant disadvantage of this method is that it requires on-site verification of equipment, which is not only time-consuming and labor-intensive, but also impossible to conduct on-site verification at remote or difficult-to-reach sites; the indirect anomaly characterization analysis method is to identify anomalies by analyzing the abnormal performance characteristics of wind direction data. Its disadvantage is that it cannot effectively detect erroneous wind direction data with high concealment. Summary of the invention

[0004] The purpose of the present invention is to provide a method, system, device and storage medium for identifying wind direction anomalies in an automatic weather station based on machine learning, so as to solve the problems that the prior art cannot perform on-site verification at remote or difficult-to-reach sites and cannot effectively detect erroneous wind direction data with high concealment.

[0005] The embodiment of the present application is implemented as follows: a method for identifying abnormal wind direction in an automatic weather station based on machine learning, comprising:

[0006] S1. Selection of typical weather process periods;

[0007] S2, construct a parameter model suitable for wind direction anomaly classification;

[0008] S3, performing wind direction anomaly detection, using the constructed parameter model suitable for wind direction anomaly classification to identify wind direction anomaly data of the automatic weather station;

[0009] S4. Statistical analysis of abnormal wind direction data, dividing the abnormal warning level according to the percentage range of abnormal wind direction data, and forming a wind direction abnormality identification result.

[0010] Optionally, in some embodiments of the present application, the typical weather process includes a typhoon weather process and / or a cold air weather process.

[0011] Optionally, in some embodiments of the present application, the typical weather process period selection includes:

[0012] S11. The 10-minute average wind direction distribution of the automatic weather station is divided into 72 distribution areas according to the 5° distance interval. The specific interval divisions are as follows:

[0013] 0°≤W1<5°,5°≤W2<10°,…,350°≤W 71 <355°,355°≤W 72 <360°;

[0014] S12. Taking the dominant wind direction as the center, count the data samples of the ten-minute average wind direction distribution of the automatic weather station to obtain the sector size ST where the sample ratio is greater than or equal to 80% i ;

[0015] S13. If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector less than or equal to 120° for 12 consecutive hours, the 12th hour in the past is selected as the start time of the wind direction anomaly detection of the typical weather process, and the conditions for starting the wind direction anomaly detection are:

[0016] ST i ≤120°, i=1,2,…,11,12;

[0017] In the formula, ST i It indicates the size of the sector where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour;

[0018] If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector greater than 120° for 12 consecutive hours, the 12th hour in the past is selected as the end time of the wind direction anomaly detection of the typical weather process. The end condition of the anomaly detection is:

[0019] ST i >120°, i=1,2,…,11,12;

[0020] In the formula, ST i It indicates the sector size where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour.

[0021] Optionally, in some embodiments of the present application, a method for constructing a parameter model suitable for wind direction anomaly classification includes:

[0022] S21, using the ten-minute average wind direction data of the automatic weather station in a typical weather process to adjust the two parameters of the DBSCAN algorithm, radius ε and the minimum number of points in the neighborhood MinPts;

[0023] S22. According to prior experience, the initial values ​​of ε and MinPts parameters of the DBSCAN algorithm are set, and the ε and MinPts parameters are adjusted by the method of increasing the initial values ​​of the ε and MinPts parameters step by step, and the clustering results of the ten-minute average wind direction data under different parameter conditions are obtained, and the clustering results of the ten-minute average wind direction data are visualized, and the parameters ε0 and MinPts0 when the clustering results of the ten-minute average wind direction data are most consistent with the ground dominant wind direction are manually judged and obtained, that is, the optimal parameters ε0 and the optimal parameters MinPts0, and saved;

[0024] S23. According to the optimal parameter ε0 and the optimal parameter MinPts0, a DBSCAN clustering wind direction anomaly classification model is obtained, that is, a parameter model suitable for wind direction anomaly classification.

[0025] Optionally, in some embodiments of the present application, wind direction anomaly detection is performed on historical wind direction data of the automatic weather station and / or real-time wind direction data of the automatic weather station.

[0026] Optionally, in some embodiments of the present application, the method for detecting wind direction anomaly using historical wind direction data of an automatic weather station includes:

[0027] S31. Count the typical weather process cases and number the typical weather process cases;

[0028] S32, select typical weather process cases and influence time periods hourly T0~T i The historical ten-minute average wind direction data of the automatic weather station;

[0029] S33, from T0 to T i The parameter model suitable for wind direction anomaly classification is used to detect anomalies in the historical ten-minute average wind direction dataset of all times of the automatic weather station during the period of typical weather process case influence. The historical ten-minute average wind direction dataset of the automatic weather station at a certain time is:

[0030] W his_t_i =(x h0 , x h1 , …, x hn-1 , x hn )

[0031] In the formula, x hn Represents the historical ten-minute average wind direction data of the nth automatic weather station at a certain time;

[0032] S34, randomly select unprocessed wind direction data x hn , find the hn The historical ten-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0033] S35. If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x hn Form a cluster C with data set C j , and x hn Marked as processed;

[0034] s36, recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j data samples until cluster C j All data samples are marked as processed;

[0035] S37. If the number of samples in data set C is less than MinPts0, then the current wind direction data x hn Labeled as noise data;

[0036] s38, repeat steps S34 to S37 until W his_t_i All wind direction data are grouped into cluster C j Or marked as noise data, and the classification results are stored in the first database.

[0037] The method of detecting wind direction anomaly using the real-time wind direction data of the automatic weather station includes:

[0038] S301, determine the start condition of wind direction abnormality detection, perform start condition detection, number the wind direction abnormality detection cases, and mark the start time as T0;

[0039] S302: After wind direction anomaly detection begins, a parameter model suitable for wind direction anomaly classification is used to classify T i The real-time ten-minute average wind direction data of the automatic weather station at that time is used for anomaly detection. The real-time ten-minute average wind direction data set of a certain automatic weather station at that time is expressed as:

[0040] W living_t =(x l0 , x l1 , …, x ln-1 , x ln );

[0041] In the formula, x ln It represents the real-time 10-minute average wind direction data of the nth automatic weather station at a certain time;

[0042] S303, randomly select unprocessed wind direction data x ln , find theln The real-time 10-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0043] S304: If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x ln Form a cluster C with data set C j , and x ln Marked as processed;

[0044] S305, recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j Data sample C j , until cluster C j All samples were marked as processed;

[0045] S306. If the number of samples in data set C is less than MinPts0, then the current wind direction data x ln Labeled as noise data;

[0046] S307, repeat steps S303 to S306 until W living_t All wind direction data are grouped into cluster C j or marked as noise data, and storing the classification results in a second database;

[0047] S308, determine T i+1 Whether the data at this time meets the abnormal detection end conditions, if not, continue to execute steps S302 to S307; if the detection end conditions are met, stop the abnormal detection of the real-time ten-minute average wind direction data of the automatic weather station.

[0048] Optionally, in some embodiments of the present application, the statistical analysis of wind direction abnormality data and the method of dividing the abnormality warning level according to the percentage range of wind direction abnormality data include:

[0049] S41. Total number of statistical data of typical weather processes N id ;

[0050] S42. Statistical abnormal data of typical weather process N times id_noise ;

[0051] S43. Percentage of abnormal wind direction data in typical weather processes according to different automatic weather stations P id_noise , which is expressed as:

[0052]

[0053] S44, P id_noise≥95%, first-level abnormality, there is a high possibility of wind direction data error or wind direction sensor operating status failure, and on-site equipment operating status verification is required;

[0054] 80%≤P id_noise <95%, secondary abnormality, the possibility of wind direction data error or wind direction sensor operating status failure is high, strengthen monitoring efforts, and check the operating status of equipment on site if necessary;

[0055] P id_noise <80%, level 3 abnormality, appropriate attention;

[0056] S45. For the first-level anomaly and the second-level anomaly, a wind direction anomaly identification product is generated and stored in the third database.

[0057] Accordingly, the embodiment of the present application also provides an automatic weather station wind direction anomaly recognition system based on machine learning, including: a typical weather process period selection module, used for typical weather process period selection;

[0058] Constructing a parameter model module suitable for wind direction anomaly classification, used to construct a parameter model suitable for wind direction anomaly classification;

[0059] A wind direction anomaly detection module is used to detect wind direction anomalies and identify wind direction anomaly data of an automatic weather station using a constructed parameter model suitable for wind direction anomaly classification;

[0060] The statistical analysis module of wind direction abnormal data is used for statistical analysis of wind direction abnormal data, and divides the abnormal warning level according to the percentage range of wind direction abnormal data to form wind direction abnormality identification results.

[0061] Accordingly, an embodiment of the present application also provides a computer device, including a storage device and a processor, wherein the storage device stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0062] Accordingly, an embodiment of the present application also provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor executes the steps of the above method.

[0063] In this application:

[0064] DBSCAN is a density-based spatial clustering algorithm for noise applications.

[0065] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0066] The present invention can perform indirect abnormal characterization monitoring without detecting the hardware of the automatic weather station observation instrument. It can effectively detect erroneous wind direction data with high concealment, improve the quality of wind direction data of the automatic weather station, and enhance the ability to monitor the operating status of the wind direction instrument of the automatic weather station. It provides accurate wind direction data set support for real-time weather monitoring, prediction of future weather changes, and understanding of the trends and laws of climate change. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flow chart of the method for identifying abnormal wind direction in an automatic weather station based on machine learning of the present invention;

[0068] Figure 2 A ten-minute average wind direction distribution diagram of an automatic weather station provided by an embodiment of the present invention;

[0069] Figure 3 The real-time wind direction data anomaly detection result of the automatic weather station provided in the embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0071] The technical solution of this application is as follows:

[0072] First, see Figure 1 The embodiment of the present application provides a method for identifying abnormal wind direction in an automatic weather station based on machine learning, comprising:

[0073] S1. Selection of typical weather process periods;

[0074] S2, construct a parameter model suitable for wind direction anomaly classification;

[0075] S3, performing wind direction anomaly detection, using the constructed parameter model suitable for wind direction anomaly classification to identify wind direction anomaly data of the automatic weather station;

[0076] S4. Statistical analysis of abnormal wind direction data, dividing the abnormal warning level according to the percentage range of abnormal wind direction data, and forming a wind direction abnormality identification result.

[0077] The technical solution of the present application can perform indirect anomaly characterization monitoring without detecting the hardware of the automatic weather station observation instrument. It can effectively detect erroneous wind direction data with high concealment, improve the quality of wind direction data of the automatic weather station, and provide accurate wind direction data set support for real-time weather monitoring, prediction of future weather changes, and understanding of climate change trends and laws.

[0078] In S1:

[0079] In some embodiments, typical weather processes include typhoon weather processes and / or cold air weather processes.

[0080] Furthermore, cold air weather processes include cold wave weather processes.

[0081] In some embodiments, the typical weather process period selection includes:

[0082] S11. The 10-minute average wind direction distribution of the automatic weather station is divided into 72 distribution areas according to the 5° distance interval. The specific interval divisions are as follows:

[0083] 0°≤W1<5°,5°≤W2<10°,…,350°≤W 71 <355°,355°≤W 72 <360°;

[0084] S12. Taking the dominant wind direction as the center, count the data samples of the ten-minute average wind direction distribution of the automatic weather station to obtain the sector size ST where the sample ratio is greater than or equal to 80% i ;

[0085] S13. If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector less than or equal to 120° for 12 consecutive hours, the 12th hour in the past is selected as the start time of the wind direction anomaly detection of the typical weather process, and the conditions for starting the wind direction anomaly detection are:

[0086] ST i ≤120°, i=1,2,…,11,12;

[0087] In the formula, ST i It indicates the size of the sector where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour;

[0088] If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector greater than 120° for 12 consecutive hours, the 12th hour in the past is selected as the end time of the wind direction anomaly detection of the typical weather process. The end condition of the anomaly detection is:

[0089] ST i >120°, i=1,2,…,11,12;

[0090] In the formula, ST i It indicates the sector size where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour.

[0091] In S2:

[0092] In some embodiments, a method for constructing a parameter model suitable for wind direction anomaly classification includes:

[0093] S21, using the ten-minute average wind direction data of the automatic weather station in a typical weather process to adjust the two parameters of the DBSCAN algorithm, radius ε and the minimum number of points in the neighborhood MinPts;

[0094] S22. According to prior experience, the initial values ​​of ε and MinPts parameters of the DBSCAN algorithm are set, and the ε and MinPts parameters are adjusted by the method of increasing the initial values ​​of the ε and MinPts parameters step by step, and the clustering results of the ten-minute average wind direction data under different parameter conditions are obtained, and the clustering results of the ten-minute average wind direction data are visualized, and the parameters ε0 and MinPts0 when the clustering results of the ten-minute average wind direction data are most consistent with the ground dominant wind direction are manually judged and obtained, that is, the optimal parameters ε0 and the optimal parameters MinPts0, and saved;

[0095] S23. According to the optimal parameter ε0 and the optimal parameter MinPts0, a DBSCAN clustering wind direction anomaly classification model is obtained, that is, a parameter model suitable for wind direction anomaly classification.

[0096] In the S3:

[0097] In some embodiments, wind direction anomaly detection is performed on historical wind direction data of the automatic weather station and / or real-time wind direction data of the automatic weather station.

[0098] Furthermore, the method for detecting wind direction anomaly using historical wind direction data of an automatic weather station includes:

[0099] S31. Count the typical weather process cases and number the typical weather process cases;

[0100] S32, select typical weather process cases and influence time periods hourly T0~T i The historical ten-minute average wind direction data of the automatic weather station;

[0101] S33, from T0 to T i The parameter model suitable for wind direction anomaly classification is used to detect anomalies in the historical ten-minute average wind direction dataset of all times of the automatic weather station during the period of typical weather process case influence. The historical ten-minute average wind direction dataset of the automatic weather station at a certain time is:

[0102] W his_t_i =(x h0 , x h1 , …, x hn-1 , x hn )

[0103] In the formula, x hnRepresents the historical ten-minute average wind direction data of the nth automatic weather station at a certain time;

[0104] S34, randomly select unprocessed wind direction data x hn , find the hn The historical ten-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0105] S35. If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x hn Form a cluster C with data set C j , and x hn Marked as processed;

[0106] S36. Recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j data samples until cluster C j All data samples are marked as processed;

[0107] S37. If the number of samples in data set C is less than MinPts0, then the current wind direction data x hn Labeled as noise data;

[0108] S38, repeat steps S34 to S37 until W his_t_i All wind direction data are grouped into cluster C j Or marked as noise data, and the classification results are stored in the first database.

[0109] Furthermore, the method for detecting wind direction anomaly using the real-time wind direction data of the automatic weather station includes:

[0110] S301, determine the start condition of wind direction abnormality detection, perform start condition detection, number the wind direction abnormality detection cases, and mark the start time as T0;

[0111] S302, after the wind direction anomaly detection starts, the real-time 10-minute average wind direction data of the automatic weather station is used to perform anomaly detection starting from T0 time using a parameter model suitable for wind direction anomaly classification. The real-time 10-minute average wind direction data set of the automatic weather station at a certain time is expressed as:

[0112] W living_t =(x l0 , x l1 , …, x ln-1 , x ln );

[0113] In the formula, x ln It represents the real-time 10-minute average wind direction data of the nth automatic weather station at a certain time;

[0114] S303, randomly select unprocessed wind direction data x ln , find the ln The real-time 10-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0115] S304: If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x ln Form a cluster C with data set C j , and x ln Marked as processed;

[0116] S305, recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j Data sample C j , until cluster C j All samples were marked as processed;

[0117] S306. If the number of samples in data set C is less than MinPts0, then the current wind direction data x ln Labeled as noise data;

[0118] S307, repeat steps S303 to S306 until W living_t All wind direction data are grouped into cluster C j or marked as noise data, and storing the classification results in a second database;

[0119] S308, determine T i+1 Whether the data at this time meets the abnormal detection end conditions, if not, continue to execute steps S302 to S307; if the detection end conditions are met, stop the abnormal detection of the real-time ten-minute average wind direction data of the automatic weather station.

[0120] In some embodiments, a computer program is deployed to automatically determine the conditions for starting wind direction anomaly detection.

[0121] In some embodiments, under the start conditions of wind direction anomaly detection, a computer program is used to automatically determine whether the wind direction anomaly detection has reached the end conditions. If the end conditions are reached, the start condition detection is executed cyclically (the purpose is to achieve real-time automatic detection of wind direction anomalies); if the wind direction anomaly detection has not reached the end conditions, the above steps are recursively performed timed times until the end conditions are reached.

[0122] It can be understood that the above cluster C j In the above figure, j represents different clusters, and its value is 1, 2, 3, etc.

[0123] In S4:

[0124] In some embodiments, the statistical analysis of wind direction abnormality data and the method of dividing abnormality warning levels according to the percentage range of wind direction abnormality data include:

[0125] S41. Total number of statistical data of typical weather processes N id ;

[0126] S42. Statistical abnormal data of typical weather process N times id_noise ;

[0127] S43. Percentage of abnormal wind direction data in typical weather processes according to different automatic weather stations P id_noise , which is expressed as:

[0128]

[0129] S44, P id_noise ≥95%, first-level abnormality, there is a high possibility of wind direction data error or wind direction sensor operating status failure, and on-site equipment operating status verification is required;

[0130] 80%≤P id_noise <95%, secondary abnormality, the possibility of wind direction data error or wind direction sensor operating status failure is high, strengthen monitoring efforts, and check the operating status of equipment on site if necessary;

[0131] P id_noise <80%, level 3 abnormality, appropriate attention;

[0132] S45. For the first-level anomaly and the second-level anomaly, a wind direction anomaly identification product is generated and stored in the third database.

[0133] In said S41:

[0134] In some embodiments, the data of typical weather processes include data of typical weather processes of historical ten-minute average wind direction data of an automatic weather station or data of typical weather processes of real-time ten-minute average wind direction data of an automatic weather station.

[0135] In some embodiments, an interface service for forming a wind direction anomaly identification product is provided. It can be understood that the wind direction anomaly data is sorted from large to small according to the percentage, and a JSON format product access service for secondary anomalies and primary anomalies is provided through HTTP.

[0136] In a second aspect, an embodiment of the present application provides an automatic weather station wind direction anomaly recognition system based on machine learning, comprising:

[0137] Typical weather process period selection module, used for typical weather process period selection;

[0138] Constructing a parameter model module suitable for wind direction anomaly classification, used to construct a parameter model suitable for wind direction anomaly classification;

[0139] A wind direction anomaly detection module is used to detect wind direction anomalies and identify wind direction anomaly data of an automatic weather station using a constructed parameter model suitable for wind direction anomaly classification;

[0140] The statistical analysis module of wind direction abnormal data is used for statistical analysis of wind direction abnormal data, and divides the abnormal warning level according to the percentage range of wind direction abnormal data to form wind direction abnormality identification results.

[0141] In the typical weather process period selection module:

[0142] In some embodiments, typical weather processes include typhoon weather processes and / or cold air weather processes.

[0143] Furthermore, typical weather processes include typhoon weather processes and / or cold air weather processes.

[0144] In some embodiments, the typical weather process period selection includes:

[0145] The 10-minute average wind direction distribution of the automatic weather station is divided into 72 distribution areas according to the 5° distance interval. The specific interval divisions are as follows:

[0146] 0°≤W1<5°,5°≤W2<10°,…,350°≤W 71 <355°,355°≤W 72 <360°;

[0147] Centered on the dominant wind direction, the data samples of the ten-minute average wind direction distribution of the automatic weather station are counted to obtain the sector size ST with a sample ratio greater than or equal to 80% i ;

[0148] If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector less than or equal to 120° for 12 consecutive hours, the 12th hour in the past is selected as the start time of the typical weather process wind direction anomaly detection. The conditions for starting wind direction anomaly detection are:

[0149] ST i ≤120°, i=1,2,…,11,12;

[0150] In the formula, ST i It indicates the size of the sector where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour;

[0151] If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector greater than 120° for 12 consecutive hours, the 12th hour in the past is selected as the end time of the wind direction anomaly detection of the typical weather process. The end condition of the anomaly detection is:

[0152] ST i >120°, i=1,2,…,11,12;

[0153] In the formula, ST i It indicates the sector size where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour.

[0154] Construct a parameter model module suitable for wind direction anomaly classification:

[0155] In some embodiments, a method for constructing a parameter model suitable for wind direction anomaly classification includes:

[0156] The ten-minute average wind direction data of the automatic weather station in typical weather processes are used to adjust the two parameters of the DBSCAN algorithm: radius ε and the minimum number of points in the neighborhood MinPts;

[0157] The initial values ​​of ε and MinPts parameters of the DBSCAN algorithm are set according to prior experience. The ε and MinPts parameters are adjusted by increasing their initial values ​​step by step. The clustering results of the ten-minute average wind direction data under different parameter conditions are obtained. The clustering results of the ten-minute average wind direction data are visualized. The parameters ε0 and MinPts0 when the clustering results of the ten-minute average wind direction data are most consistent with the dominant wind direction on the ground are manually judged and obtained, which are the optimal parameters ε0 and the optimal parameters MinPts0, and are saved.

[0158] According to the optimal parameters ε0 and MinPts0, the DBSCAN clustering wind direction anomaly classification model is obtained, that is, the parameter model suitable for wind direction anomaly classification.

[0159] In the wind direction anomaly detection module:

[0160] In some embodiments, wind direction anomaly detection is performed on historical wind direction data of the automatic weather station and / or real-time wind direction data of the automatic weather station.

[0161] Furthermore, the method for detecting wind direction anomaly using historical wind direction data of an automatic weather station includes:

[0162] Count the typical weather process cases and number them;

[0163] Select typical weather process cases and influence the period from hour to hour T0 to T i The historical ten-minute average wind direction data of the automatic weather station;

[0164] From T0 to T i The parameter model suitable for wind direction anomaly classification is used to detect anomalies in the historical ten-minute average wind direction dataset of all times of the automatic weather station during the period of typical weather process case influence. The historical ten-minute average wind direction dataset of the automatic weather station at a certain time is:

[0165] W his_t_i =(x h0 , x h1 , …, x hn-1 , x hn )

[0166] In the formula, x hn Represents the historical ten-minute average wind direction data of the nth automatic weather station at a certain time;

[0167] Randomly select unprocessed wind direction data x hn , find the hn The historical ten-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0168] If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x hn Form a cluster C with data set C j , and x hn Marked as processed;

[0169] Recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j data samples until cluster C j All data samples are marked as processed;

[0170] If the number of samples in data set C is less than MinPts0, then the current wind direction data x hn Labeled as noise data;

[0171] Repeat the above steps until W his_t_i All wind direction data are grouped into cluster C j Or marked as noise data, and the classification results are stored in the first database.

[0172] Furthermore, the method for detecting wind direction anomaly using the real-time wind direction data of the automatic weather station includes:

[0173] Determine the start condition of wind direction anomaly detection, perform start condition detection, number the wind direction anomaly detection cases, and mark the start time as T0;

[0174] After wind direction anomaly detection begins, a parameter model suitable for wind direction anomaly classification is used to classify Ti The real-time ten-minute average wind direction data of the automatic weather station at that time is used for anomaly detection. The real-time ten-minute average wind direction data set of a certain automatic weather station at that time is expressed as:

[0175] W living_t =(x l0 , x l1 , …, x ln-1 , x ln );

[0176] In the formula, x ln It represents the real-time 10-minute average wind direction data of the nth automatic weather station at a certain time;

[0177] Randomly select unprocessed wind direction data x ln , find the ln The real-time 10-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0;

[0178] If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x ln Form a cluster C with data set C j , and x ln Marked as processed;

[0179] Recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j Data sample C j , until cluster C j All samples were marked as processed;

[0180] If the number of samples in data set C is less than MinPts0, then the current wind direction data x ln Labeled as noise data;

[0181] Repeat the above steps until W living_t All wind direction data are grouped into cluster C j or marked as noise data, and storing the classification results in a second database;

[0182] Judge T i+1 Whether the data at this time meets the abnormal detection end conditions, if not, continue to execute the above steps; if the detection end conditions are met, stop the abnormal detection of the real-time ten-minute average wind direction data of the automatic weather station.

[0183] In the statistical analysis module of abnormal wind direction data:

[0184] In some embodiments, the statistical analysis of wind direction abnormality data and the method of dividing abnormality warning levels according to the percentage range of wind direction abnormality data include:

[0185] The total number of statistical data of typical weather processes N id ;

[0186] Statistical abnormal data of typical weather process times N id_noise ;

[0187] The percentage of abnormal wind direction data P in typical weather processes according to different automatic weather stations id_noise , which is expressed as:

[0188]

[0189] P id_noise ≥95%, first-level abnormality, there is a high possibility of wind direction data error or wind direction sensor operating status failure, and on-site equipment operating status verification is required;

[0190] 80%≤P id_noise <95%, secondary abnormality, wind direction data error or wind direction sensor operating status failure is likely, strengthen monitoring efforts, and conduct on-site equipment operating status verification if necessary;

[0191] P id_noise <80%, level 3 abnormality, appropriate attention;

[0192] For the first-level anomalies and the second-level anomalies, wind direction anomaly identification products are generated and stored in the third database.

[0193] In a third aspect, the present application provides a computer device comprising a storage device and a processor, wherein the storage device stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above-mentioned method for identifying wind direction anomaly in an automatic weather station based on machine learning.

[0194] The computer device may be a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with the user through a keyboard, a mouse, a remote control, a touch pad, or a voice control device.

[0195] The memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or D interface display memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory can be an internal storage unit of the computer device, such as a hard disk or memory of the computer device. In other embodiments, the memory can also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory can also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory is often used to store the operating system and various application software installed on the computer device, such as the program code of the automatic weather station wind direction anomaly recognition method based on machine learning, etc. In addition, the memory can also be used to temporarily store various types of data that have been output or are to be output.

[0196] The processor may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor is generally used to control the overall operation of the computer device. In this embodiment, the processor is used to run the program code stored in the memory or process data, such as running the program code of the automatic weather station wind direction anomaly identification method based on machine learning.

[0197] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the steps of the automatic weather station wind direction anomaly identification method based on machine learning as described above.

[0198] Among them, the computer-readable storage medium stores an interface display program, and the interface display program can be executed by at least one processor so that the at least one processor performs the steps of the automatic weather station wind direction anomaly identification method based on machine learning as mentioned above.

[0199] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for a terminal device (which can be a mobile phone, computer, server or network device, etc.) to execute the automatic weather station wind direction anomaly recognition method based on machine learning described in the embodiment of the present application.

[0200] Test Example 1

[0201] The test example selected the weather process of Typhoon “Sura” No. 2309 as the application case.

[0202] like Figure 2 As shown, the ten-minute average wind direction distribution of the automatic weather station at 03:00 on September 1, 2023, the dominant wind direction is northerly, and the sample ratio is greater than or equal to 80% of the clustered sector size ST i is 90°.

[0203] Starting from 03:00 on September 1, 2023, there will be 12 consecutive hours (5 minutes each) in which more than 80% of the ten-minute average wind direction data samples of the automatic weather station are concentrated in an arc sector less than or equal to 120°. Therefore, 03:00 on September 1, 2023 is used as the condition for starting wind direction anomaly detection, and the program automatically performs wind direction anomaly detection hour by hour.

[0204] Starting from 09:05 on September 2, 2023, more than 80% of the ten-minute average wind direction data of the automatic weather stations will have an arc sector greater than 120° for 12 consecutive hours. Therefore, the wind direction anomaly detection will end from 09:05 on September 2, 2023.

[0205] Methods for constructing parameter models suitable for wind direction anomaly classification include:

[0206] Step 1: Use the 10-minute average wind direction data of the automatic weather station during the cold air weather process from January 24 to 26, 2023 and the typhoon weather process from July 16 to 17, 2023 to adjust the two parameters of the DBSCAN algorithm: radius ε and the minimum number of points in the neighborhood MinPts;

[0207] Step 2, according to prior experience, the initial values ​​of ε and MinPts parameters of the DBSCAN algorithm are set to 5 and 1° respectively, 1°≤ε≤16° is used as the first-level cycle parameter, and the ε parameter is adjusted by increasing 1° step by step according to the initial value, and 5≤MinPts≤35 is used as the second-level cycle parameter, and the MinPts parameter is adjusted by increasing 2 step by step according to the initial value, and the clustering results of the ten-minute average wind direction data under different parameter conditions are obtained. The clustering results of the ten-minute average wind direction data are visualized, and the parameters ε0 and MinPts0 when the clustering results of the ten-minute average wind direction data are most consistent with the ground dominant wind direction are manually judged and obtained, that is, the optimal parameters ε0 and the optimal parameters MinPts0, and finally the parameter MinPts0 is determined to be 15 and the parameter ε0 is 8°;

[0208] Step 3: Use the final determined parameter MinPts0 of 15 and parameter ε0 of 8° to obtain the DBSCAN clustering wind direction anomaly classification model, that is, the parameter model suitable for wind direction anomaly classification.

[0209] Step 4: The ten-minute average wind direction distribution of the automatic weather station at 03:00 on September 1, 2023, with the dominant wind direction being northerly and the sample ratio being greater than or equal to 80% of the clustered sector size ST i is 90°, such as Figure 3 As shown. Since 03:00 on September 1, 2023, there have been 12 consecutive hours (5 minutes each) in which more than 80% of the automatic weather station's ten-minute average wind direction data samples have an arc sector less than or equal to 120°. Therefore, 03:00 on September 1, 2023 is used as the start condition for wind direction anomaly detection. The wind direction anomaly detection cases are numbered as 2023090103, 2023090109, 2023090115, 2023090121, 2023090203, and 2023090209. The starting time is T0, which is 03:00 on September 1, 2023. The real-time ten-minute average wind direction data set of the automatic weather station at T0 is expressed as:

[0210] {6,26,341,349,359,359,8,3,326,271,336,343,350,357,357,348,334,338,331,351,4,1,19,39,294,48,335,345,308,337,326,15,238,23,318,339,344,353,351,358,4,16,8,9,356,352,75,5,349,227,321,4,280,0,358,146,299,355,273,342,338,8,317,15,293,354,358,343,358,345,23,192,8,56,96,4,33,10,25,21,323,9,337,320,213,11,24,0,2,327,356,335,266,278,14,3,17,342,356,18,358,356,358,338,8,3,339,286,40,304,40,335,344,348,331,357,0,348,333,3,354,300,330,319,353,319,336,341,317,338,250,0,26,321,345,337,233,347,346,338,339,18,317,352,283,354,313,297,14,311,44,344,335,338,344,330,178,323,345,337,327,335,285,354,164,8,103,328,83,184,0,320,313,216,172,6,334,285,359,325,114,334,320,3,307,348,246,351,343,5,79,122,11,4,128,321,339,352,14,32,333,351,203,293,128,355,359,315,11,93,23,321,353,225,352,270,3,241,354,357,56,164,332,319,150,26,12,343,349,166,104,297,3,315,345,84,282,309,14,330,345,20,88,160,312,93,343,308,239,301,219,82,341,358,99,350,324,211,333,360,348,2,40,28,90,337,30,339,295,341,19,10,299,332,346,323,146,155,355,342,139,148,22,301,308,49,9,5,341,343,354,355,342,343,198,320,271,56,346,329,13,271,334,357,319,293,349,334,294,315,359,357,340,342,6,315,342,110,269,352, 349,318,329,344,5,272,347,343,336,327,335,22,169,16,15,359,358,47,231,345,336,321,8,295,6,339,12,0,330,23,42,11,8,329,340,356,325,138,3,321,337,324,345,299,14, 12,19,5,9,348,287,325,7,9,351,346,331,250,10,59,6,107,65,308,302,336,292,71,33,342,21,302,91,310,318,354,150,77,320,358,348,355,332,312,18,316,335,18,286,29,3 40,233,337,313,67,287,45,327,13,173,341,255,315,297,80,19,344,0,5,352,350,356,328,344,337,13,294,55,21,225,356,298,26,341,183,338,282,299,323,139,263,80,159};,

[0211] Step 5: Randomly select unprocessed wind direction data x ln is 6, then x ln The real-time ten-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0 is expressed as {6,8,3,4,1,4,8,9,5,4,0,8,8,4,10,9,11,0,2,14,3,8,3,0,3,0,14,8,0,6,3,5,11,4,14,11,3,12,3,14,2,10,9,5,13,6,5,8,6,12,0,11,8,3,14,12,5,9,7,9,10,6,13,0,5,13};

[0212] Step 6: The number of samples in data set C is 66, which is greater than 15, forming a cluster C1. ln =6 Mark as processed;

[0213] Step 7. In cluster C1, recursively traverse samples 8,3,4,1,4,8,9,5,4,0,8,8,4,10,9,11,0,2,14,3,8,3,0,3,0,14,8,0,6,3,5,11,4,14,11,3,12,3,14,2,10,9,5,13,6,5,8,6,12,0,11,8,3,14,12,5,9,7,9,10,6,13,0,5,13, ​​and cluster C1 is expanded to {6,8,3,4,1,19,15,4,16 ,8,9,5,4,0,8,15,8,4,10,21,9,11,0,2,14,3,17,18,8,3,0,3,0,18,14,8,0,6,3,5,11,4,14,11,3,12,3,14,20,2,19,10,22,9,5,13,6,5,22,16,15,8,6,12,0,11,8,3,14,12,19,5,9,7,9,10,6,21,18,18,13,19,0,5,13,21};

[0214] Step 8: Unprocessed wind direction data x ln is 178, then x ln The real-time 10-minute average wind direction data set C of all automatic weather stations with a distance less than or equal to ε0 is represented as {178, 164, 184, 172, 192}. The number of samples in data set C is 5, which is less than 15. Therefore, the current wind direction data x ln 178 are marked as noise data;

[0215] Step 9. Repeat steps 4 to 8. All wind direction data are divided into clusters C1 and C2, as well as noise data. The data samples of C1 are {6, 26, 8, 3, 4, 1, 19, 39, 15, 23, 4, 16, 8, 9, 5, 4, 0, 8, 15, 23, 8, 4, 33, 10, 25, 21, 9, 11, 24, 0, 2, 14, 3, 17, 18, 8, 3, 40, 40, 0, 3, 0, 26, 18, 14, 8,0,6,3,5,11,4,14,32,11,23,3,26,12,3,14,20,2,40,28,30,19,10,22,9,5,13,6,5,22,16,15,8,6,12,0,23,11,8,3,14,12,19,5,9,7,9,10,6,33,21,18,18,29,13,19,0,5,13,21,26}

[0216] The data sample of C2 is

[0217] {341,349,359,359,326,271,336,343,350,357,357,348,334,338,331,351,294,48,335,345,308,337,326,238,318,339,344,353,351,358,356,352,75,349,227,321,280,358,146,299,355,273,342,338,317,293,354,358,343,358,345,192,56,96,323,337,320,213,327,356,335,266,278,342,356,358,356,358,338,339,286,304,335,344,348,331,357,348,333,354,300,330,319,353,319,336,341,317,338,250,321,345,337,233,347,346,338,339,317,352,283,354,313,297,311,44,344,335,338,344,330,178,323,345,337,327,335,285,354,164,103,328,83,184,320,313,216,172,334,285,359,325,114,334,320,307,348,246,351,343,79,122,128,321,339,352,333,351,203,293,128,355,359,315,93,321,353,225,352,270,241,354,357,56,164,332,319,150,343,349,166,104,297,315,345,84,282,309,330,345,88,160,312,93,343,308,239,301,219,82,341,358,99,350,324,211,333,360,348,90,337,339,295,341,299,332,346,323,146,155,355,342,139,148,301,308,49,341,343,354,355,342,343,198,320,271,56,346,329,271,334,357,319,293,349,334,294,315,359,357,340,342,315,342,110,269,352,349,318,329,344,272,347,343,336,327,335,169,359,358,47,231,345,336,321,295,339,330,42,329,340,356,325,138,321,337,324,345,299,348,287,325,351,346,331,250,59,107,65,308,302,336,292,71,342,302,91,310,318,354,15 0,77,320,358,348,355,332,312,316,335,286,340,233,337,313,67,287,45,327,173,341,255,315,297,80,344,352,350,356,328,344,337,294,55,225,356,298,341,183,338,282,299,323,139,263,80,159},

[0218] The noise data samples are:

[0219] 48,238,75,227,146,192,56,96,213,266,250,233,44,178,164,103,83,184,216,172,114,246,79,122,128,203,128,93,225,241,56,164,150,166,104,84,88,160,93,2 39,219,82,99,211,90,146,155,139,148,49,198,56,110,269,169,47,231,42,138,250,59,107,65,71,91,150,77,233,67,45,173,255,80,55,225,183,139,263,80,159

[0220] For steps 4 to 9 above, please refer to Figure 3 (a) Figure 3 (a), (b), (c), (d), (e), and (f) are visualization distribution diagrams of wind direction data anomaly detection results at 03:00 on September 1, 2023, 09:00 on September 1, 2023, 15:00 on September 1, 2023, 21:00 on September 1, 2023, and 03:00 on September 2, 2023, and 09:00 on September 2, 2023. The detection method divides the wind direction sequence into 2 or 3 clusters, among which the red sample points represent abnormal wind direction points.

[0221] Experimental Example 2

[0222] Statistical analysis of abnormal wind direction data, and methods for dividing abnormal warning levels according to the percentage range of abnormal wind direction data include:

[0223] Step 1: Count the total number of times N the weather process of Typhoon "Sura" No. 2309 id ;

[0224] Step 2: Count the abnormal data of the weather process of Typhoon "Sura" No. 2309, time N id_noise ;

[0225] Step 3: Statistical analysis of abnormal wind direction data percentage P of different automatic weather stations during the typhoon "Sura" No. 2309 weather process id_noise , which is expressed as:

[0226]

[0227] Step 4: P id_noise ≥95%, first-level abnormality, there is a high possibility of wind direction data error or wind direction sensor operating status failure, and on-site equipment operating status verification is required;

[0228] 80%≤P id_noise <95%, secondary abnormality, the possibility of wind direction data error or wind direction sensor operating status failure is high, strengthen monitoring efforts, and check the operating status of equipment on site if necessary;

[0229] P id_noise <80%, level 3 abnormality, appropriate attention;

[0230] Step 5: For the first-level anomaly and the second-level anomaly, a wind direction anomaly identification product is generated and stored in the third database.

[0231] The statistical analysis results of the wind direction anomaly data of the 2309 typhoon "Sura" weather process are shown in Table 1. From Table 1, we can see that the P id_noise The values ​​of are all greater than 95%, which are all first-level abnormalities. There is a high possibility that the wind direction data is wrong or the wind direction sensor is in a faulty state. The on-site equipment operation status needs to be checked. The P id_noise The values ​​are all less than 95% and greater than 80%, which are both level 2 abnormalities. There is a high possibility that the wind direction data is wrong or the wind direction sensor is in an operating state fault. Strengthen monitoring efforts and check the operating state of the equipment on site if necessary.

[0232] Table 1 Statistical analysis results of wind direction anomaly data

[0233]

[0234] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying abnormal wind direction in an automatic weather station based on machine learning, characterized in that: include: S1. Selection of typical weather process periods; S2, construct a parameter model suitable for wind direction anomaly classification; S3, performing wind direction anomaly detection, using the constructed parameter model suitable for wind direction anomaly classification to identify wind direction anomaly data of the automatic weather station; S4. Statistical analysis of abnormal wind direction data, dividing the abnormal warning level according to the percentage range of abnormal wind direction data, and forming a wind direction abnormality identification result.

2. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 1, characterized in that: Typical weather processes include typhoon weather processes and / or cold air weather processes.

3. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 1 or 2, characterized in that: Typical weather process periods include: S11. The 10-minute average wind direction distribution of the automatic weather station is divided into 72 distribution areas according to the 5° distance interval. The specific interval divisions are as follows: 0°≤W1<5°,5°≤W2<10°,…,350°≤W 71 <355°,355°≤W 72 <360°; S12. Taking the dominant wind direction as the center, count the data samples of the ten-minute average wind direction distribution of the automatic weather station to obtain the sector size ST where the sample ratio is greater than or equal to 80% i ; S13. If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector less than or equal to 120° for 12 consecutive hours, the 12th hour in the past is selected as the start time of the wind direction anomaly detection of the typical weather process, and the conditions for starting the wind direction anomaly detection are: ST i ≤120°,i=1,2,…,11,12; In the formula, ST i It indicates the size of the sector where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour; If more than 80% of the 10-minute average wind direction data samples of the automatic weather station have an arc sector greater than 120° for 12 consecutive hours, the 12th hour in the past is selected as the end time of the wind direction anomaly detection of the typical weather process. The end condition of the anomaly detection is: ST i >120°,i=1,2,…,11,12; In the formula, ST i It indicates the sector size where the ten-minute average wind direction data samples of the automatic weather station are gathered, and i indicates the hour.

4. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 1 or 2, characterized in that: Methods for constructing parameter models suitable for wind direction anomaly classification include: S21, using the ten-minute average wind direction data of the automatic weather station in a typical weather process to adjust the two parameters of the DBSCAN algorithm, radius ε and the minimum number of points in the neighborhood MinPts; S22. According to prior experience, the initial values ​​of ε and MinPts parameters of the DBSCAN algorithm are set, and the ε and MinPts parameters are adjusted by the method of increasing the initial values ​​of the ε and MinPts parameters step by step, and the clustering results of the ten-minute average wind direction data under different parameter conditions are obtained, and the clustering results of the ten-minute average wind direction data are visualized, and the parameters ε0 and MinPts0 when the clustering results of the ten-minute average wind direction data are most consistent with the ground dominant wind direction are manually judged and obtained, that is, the optimal parameters ε0 and the optimal parameters MinPts0, and saved; S23. According to the optimal parameter ε0 and the optimal parameter MinPts0, a DBSCAN clustering wind direction anomaly classification model is obtained, that is, a parameter model suitable for wind direction anomaly classification.

5. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 1 or 2, characterized in that: Wind direction anomaly detection is performed on historical wind direction data of the automatic weather station and / or real-time wind direction data of the automatic weather station.

6. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 5, characterized in that: The methods for detecting wind direction anomalies using historical wind direction data from automatic weather stations include: S31. Count the typical weather process cases and number the typical weather process cases; S32, select typical weather process cases and influence time periods hourly T0~T i The historical ten-minute average wind direction data of the automatic weather station; S33, from T0 to T i The parameter model suitable for wind direction anomaly classification is used to detect anomalies in the historical ten-minute average wind direction dataset of the automatic weather station at all times during the period affected by typical weather processes. The historical ten-minute average wind direction dataset of the automatic weather station at a certain time is: W his_t_i =(x h0 ,x h1 ,…,x hn-1 ,x hn ) In the formula, x hn Represents the historical ten-minute average wind direction data of the nth automatic weather station at a certain time; S34, randomly select unprocessed wind direction data x hn , find the hn The historical ten-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0; S35. If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x hn Form a cluster C with data set C j , and x hn Marked as processed; S36. Recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j data samples until cluster C j All data samples are marked as processed; S37. If the number of samples in data set C is less than MinPts0, then the current wind direction data x hn Labeled as noise data; S38, repeat steps S34 to S37 until W his_t_i All wind direction data are grouped into cluster C j or marked as noise data, and storing the classification result in the first database; and / or The method of detecting wind direction anomaly using the real-time wind direction data of the automatic weather station includes: S301, determine the start condition of wind direction abnormality detection, perform start condition detection, number the wind direction abnormality detection cases, and mark the start time as T0; S302: After wind direction anomaly detection begins, a parameter model suitable for wind direction anomaly classification is used to classify T i The real-time ten-minute average wind direction data of the automatic weather station at that time is used for anomaly detection. The real-time ten-minute average wind direction data set of a certain automatic weather station at that time is expressed as: W living_t =(x l0 ,X L1 ,…,x ln-1 ,x ln ); In the formula, x ln It represents the real-time ten-minute average wind direction data of the nth automatic weather station at a certain time; S303, randomly select unprocessed wind direction data x ln , find the ln The real-time 10-minute average wind direction dataset C of all automatic weather stations with a distance less than or equal to ε0; S304: If the number of samples in data set C is greater than or equal to MinPts0, then the current wind direction data x ln Form a cluster C with data set C j , and x ln Marked as processed; S305, recursively traverse cluster C j The wind direction data that is not marked as processed in the cluster C is expanded j Data sample C j , until cluster C j All samples were marked as processed; S306. If the number of samples in data set C is less than MinPts0, then the current wind direction data x ln Labeled as noise data; S307, repeat steps S303 to S306 until W living_t All wind direction data are grouped into cluster C j or marked as noise data, and storing the classification results in a second database; S308, determine T i+1 Whether the data at this time meets the abnormal detection end conditions, if not, continue to execute steps S302 to S307; if the detection end conditions are met, stop the abnormal detection of the real-time ten-minute average wind direction data of the automatic weather station.

7. The method for identifying abnormal wind direction in an automatic weather station based on machine learning according to claim 1 or 2, characterized in that: Statistical analysis of abnormal wind direction data, and methods for dividing abnormal warning levels according to the percentage range of abnormal wind direction data include: S41. Total number of statistical data of typical weather processes N id ; S42. Statistical abnormal data of typical weather process N times id_noise ; S43. Percentage of abnormal wind direction data in typical weather processes according to different automatic weather stations P id_noise , which is expressed as: S44, P id_noise ≥95%, first-level abnormality, there is a high possibility of wind direction data error or wind direction sensor operating status failure, and on-site equipment operating status verification is required; 80%≤P id_noise <95%, secondary abnormality, wind direction data error or wind direction sensor operating status failure is likely, strengthen monitoring efforts, and conduct on-site equipment operating status verification if necessary; P id_noise <80%, level 3 abnormality, appropriate attention; S45. For the first-level anomaly and the second-level anomaly, a wind direction anomaly identification product is generated and stored in the third database.

8. The automatic weather station wind direction anomaly recognition system based on machine learning is characterized by: include: Typical weather process period selection module, used for typical weather process period selection; Constructing a parameter model module suitable for wind direction anomaly classification, used to construct a parameter model suitable for wind direction anomaly classification; A wind direction anomaly detection module is used to detect wind direction anomalies and identify wind direction anomaly data of an automatic weather station using a constructed parameter model suitable for wind direction anomaly classification; The statistical analysis module of wind direction abnormal data is used for statistical analysis of wind direction abnormal data, and divides the abnormal warning level according to the percentage range of wind direction abnormal data to form wind direction abnormality identification results.

9. Computer device, characterized in that The invention comprises a storage and a processor, wherein the storage stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 7.

Images