A method for identifying observation data anomaly based on SIP-LOF

Abstract

A kind of observation data anomaly identification method based on SIP-LOF, comprising the following steps: first, the observation data to be detected is extracted by PLR-SIP method All data points, then two data points of starting and ending are used as segmentation point, the point of maximum y direction distance between two segmentation points is calculated, then the point of maximum y direction distance between adjacent two segmentation points is calculated in turn, obtain multiple segmentation points, and sub-sequence is formed between adjacent two segmentation points;Extract multiple characteristic values of sub-sequence, construct multidimensional space according to all characteristic values, then all characteristic values are mapped in multidimensional space, and the distance between each characteristic value is calculated;After arranging distance value according to size, the k distance of each sample and the first k neighbors are found, then the reachable distance of each sample to k neighbors, local reachable density are calculated, then local outlier factor LOF is calculated, and LOF value is obtained.The present application can quickly and automatically locate abnormal data in mass data.

Description

Technical Field

[0001] This invention relates to an improvement in earthquake anomaly data identification technology, belonging to the field of data processing, and particularly to a method for identifying anomalies in observation data based on SIP-LOF. Background Technology

[0002] The geophysical observation network consists of three major disciplines: topographic deformation, geomagnetism, and subsurface fluids. Among them, topographic deformation measurement is an indispensable and important means in earthquake monitoring. Crustal deformation instruments conduct scientific observations from multiple perspectives, such as earth tilt, stress, strain, gravity, spatial large-scale topographic deformation measurement, and fault deformation measurement. Common topographic deformation instruments include, but are not limited to, borehole strain gauges, water pipe inclinometers, vertical pendulum inclinometers, extensometers, and volumetric strain gauges.

[0003] With the dense deployment of digital seismic monitoring networks, the number of topographic deformation instruments used has also increased, leading to an exponential growth in the total amount of observation data. While this provides more substantial and abundant data samples for earthquake prediction research, it also greatly increases the difficulty of identifying anomalous data. However, during the observation process of topographic deformation instruments, various factors inevitably affect the data, such as solid tides, natural environment, human interference, observation systems, or earthquake precursors. These factors can cause abnormal dynamics in the data samples based on the original periodic changes, such as data spikes, large jumps, step transitions, and missing data. Under these circumstances, traditional processing methods are no longer able to quickly and automatically locate anomalies in massive amounts of observation data, which in turn affects the quality of the observation data.

[0004] Chinese patent application CN202310937865.X, filed on July 28, 2023, discloses a method and system for constructing a geomagnetic anomaly dataset. The method includes: extracting and removing linear trend components from minute-by-minute data of geomagnetic vertical components from each station to obtain a residual geomagnetic sequence containing periodic fluctuations and local anomalous fluctuations of the geomagnetic signal; constructing a reference signal for the geomagnetic data of each station based on the residual geomagnetic sequence; calculating the window-weighted correlation between the daily geomagnetic data of each station and the reference signal; extracting geomagnetic anomaly waveforms based on the magnitude of the window-weighted correlation, and forming a geomagnetic anomaly dataset. This method can improve the accuracy and sensitivity of geomagnetic anomaly waveform detection, so as to detect possible geomagnetic anomaly signals in a timely manner. However, the above solution does not solve the problem of being unable to quickly and automatically locate anomalous data in massive amounts of data.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of this patent application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to overcome the problem in the prior art that it is impossible to quickly and automatically locate abnormal data in massive data, and to provide a SIP-LOF-based method for identifying anomalies in observation data that can quickly and automatically locate abnormal data in massive data.

[0007] To achieve the above objectives, the technical solution of the present invention is: a method for identifying anomalies in observation data based on SIP-LOF, wherein the method for identifying anomalies in observation data based on SIP-LOF includes the following steps:

[0008] Step 1: First, extract all data points from the observation data to be detected using the PLR-SIP method. Then, treat all data points as a time series. Use the start and end points as first-generation segmentation points and connect them to form a first-generation baseline. Calculate the y-direction distance between the remaining data points and the first-generation baseline, and identify the point with the largest y-direction distance, which is the important point. If the distance is less than a threshold, the segmentation point calculation ends; if it is greater than the threshold, the important point is the second-generation segmentation point. Then, connect the second-generation segmentation point with the start and end points respectively to form two second-generation baselines. Calculate the new y-direction distance between the remaining data points and the two second-generation baselines, and compare the point with the largest y-direction distance with the threshold. If it is greater than the threshold, a third-generation segmentation point is obtained. Connect the first-generation, second-generation, and third-generation segmentation points to obtain three third-generation baselines. Then, calculate the y-direction distance between the remaining data points and the third-generation baselines using the same method. Repeat this process to form segmentation points until the largest y-direction distance is less than the threshold, at which point the segmentation point calculation ends. At this point, a subsequence is formed between two adjacent segmentation points.

[0009] Step 2: Extract multiple feature values ​​from the subsequence, construct a multidimensional space based on all feature values, then map all feature values ​​into the multidimensional space, calculate the distance between each feature value, and obtain the corresponding distance value;

[0010] Step 3: After sorting the distance values ​​by size, find the k-th distance and the k nearest neighbors of each sample. Then calculate the reachability distance and local reachability density of each sample to the k nearest neighbors to obtain the corresponding distance value and density value. Then calculate the Local Outlier Factor (LOF) based on all the distance values ​​and density values ​​to obtain all the LOF values.

[0011] Step 4: If the LOF value is close to 1, the data point is considered a normal point; if the LOF value is far from 1, the data point is an outlier.

[0012] In step one, if the distance is less than the threshold, the calculation of the dividing point ends. Specifically, when the distance is less than the threshold, the point cannot become a new dividing point. The point is merged into the line segment formed by the previous dividing point, and the dividing process ends.

[0013] In step two, the various eigenvalues ​​include length, extreme value difference, standard deviation, and mean. These four eigenvalues ​​are the four elements of the feature space.

[0014] The length represents the trend of subsequence variation, the extreme value difference is the difference between the maximum and minimum values, the standard deviation is the dispersion of the sequence data, and the mean is the average trend.

[0015] The threshold is 0.001 to 0.005 times the standard deviation.

[0016] In step two, multiple feature values ​​of the subsequence are extracted, a multidimensional space is constructed based on all feature values, and then all feature values ​​are mapped in the multidimensional space. The distance between each feature value is calculated to obtain the corresponding distance value. Specifically:

[0017] The time series is mapped to this feature space, and each subsequence corresponds to a point in the feature space. The position of each subsequence in the feature space is represented by a four-element structure composed of feature values. Based on this, the distance between each pattern is calculated to obtain the corresponding distance value.

[0018] Since the value ranges of the four feature quantities differ significantly, and different magnitudes can affect the anomaly detection results of the observation data, normalization processing is performed on these four sets of feature values ​​respectively, specifically as follows:

[0019] Let x1 = (x 11 x 12 x 13 , ..., x 1n Let x be one of the subsequences. max With x min These represent the maximum and minimum values ​​of each eigenvalue;

[0020]

[0021] In step three, after sorting the distance values ​​by size, the k-th distance and the k nearest neighbors of each sample are found. Then, the reachability distance and local reachability density from each sample to its k nearest neighbors are calculated to obtain the corresponding density values. Then, the Local Outlier Factor (LOF) is calculated based on all density values ​​to obtain all LOF values. Specifically, the density-based Local Outlier Factor algorithm is used for anomaly pattern detection, and the specific steps are as follows:

[0022] A. Calculate the Euclidean distance between any two subsequences;

[0023] B. After arranging the Euclidean distances, find the k-th distance and the k nearest neighbors of each sample;

[0024] C. Calculate the reachability distance and local reachability density of each sample to its k nearest neighbors.

[0025] The reachable distance k is calculated as follows:

[0026] reach-dist(p, q)=max(k-distance(q), d(p, q));

[0027] Where point p is a data point in the sample, point q is one of the k nearest neighbors of point p, d(p,q) represents the actual distance between point p and point q, and k-distance(q) is the distance from point q to its k-th nearest neighbor.

[0028] The locally reachable density is calculated as follows:

[0029]

[0030] In step three, the Local Outlier Factor (LOF) is calculated based on all distance and density values ​​to obtain all LOF values, specifically:

[0031]

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0033] 1. In this invention, a method for identifying anomalies in observation data based on SIP-LOF, all data points are first extracted from the observation data to be detected using the PLR-SIP method. Then, all data points are treated as a time series. The first and last data points are used as segmentation points. The point with the largest y-distance between two segmentation points is calculated. If the distance is less than a threshold, the two ends are merged; if it is greater than the threshold, this important point is considered a segmentation point. This process is repeated until all data is segmented. At this point, a subsequence is formed between adjacent segmentation points. In application, time series analysis techniques from data mining are introduced. Observation data of typical events such as equipment failure, site environment, and earthquake precursors are treated as time series. The series is further segmented, and subsequence features are extracted. Finally, the LOF value of the subsequence is calculated. This method can accurately and effectively identify typical interference events and earthquake precursor information, thereby quickly and automatically locating abnormal data in massive datasets, improving instrument data availability, and increasing the efficiency of fault diagnosis for maintenance personnel. Therefore, this invention can quickly and automatically locate abnormal data in massive datasets.

[0034] 2. In this invention, a method for identifying anomalies in observational data based on SIP-LOF, the SIP-LOF algorithm preprocesses time-series data using statistical interpolation to remove long-term trends and seasonal fluctuations, making the data more stable. This helps the algorithm more accurately distinguish between genuine anomalous signals and periodic changes caused by natural variations, reducing false alarms and missed alarms. Furthermore, the detrended data better reflects short-term anomalous changes, which the SIP-LOF algorithm can capture, enabling timely detection of potential anomalies even when the overall data trend is stable. Therefore, this invention reduces false alarms and missed alarms and can detect anomalies even under stable conditions.

[0035] 3. In this invention, a method for identifying anomalies in observation data based on SIP-LOF, the SIP-LOF method does not rely on assumptions about data distribution and can handle anomalies of various shapes and sizes. It is applicable to various complex data structures and monitoring scenarios, improving the method's universality and adaptability. Although the LOF method has high computational complexity when processing large-scale datasets, the SIP preprocessing step simplifies the subsequent anomaly detection process to some extent, thereby improving efficiency, which is particularly important in real-time monitoring systems. By combining statistical interpolation and local anomaly factor algorithms, the ability to distinguish noise and outliers is enhanced. Therefore, this invention has strong adaptability and high detection efficiency. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the data processing of the present invention.

[0037] Figure 2 This is a schematic diagram of the detection results of typical events at the Mudanjiang Earthquake Station in this invention.

[0038] Figure 3 This is a schematic diagram of the detection results of typical events at the Jiaxiang Seismic Station in this invention.

[0039] Figure 4 This is a schematic diagram of the detection results of typical events at Zhongshan Station in this invention.

[0040] Figure 5 This is a schematic diagram of the detection results of typical events at the Kuancheng Seismic Station in this invention.

[0041] Figure 6 This is a schematic diagram of the detection results of typical events at the Nanshan Seismic Station in this invention.

[0042] Figure 7 This is a schematic diagram of the detection results of typical events at the Yixian Earthquake Station in this invention. Detailed Implementation

[0043] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] See Figures 1 to 7 A method for identifying anomalies in observation data based on SIP-LOF, comprising the following steps:

[0045] Step 1: First, extract all data points from the observation data to be detected using the PLR-SIP method. Then, treat all data points as a time series. Use the start and end points as first-generation segmentation points and connect them to form a first-generation baseline. Calculate the y-direction distance between the remaining data points and the first-generation baseline, and identify the point with the largest y-direction distance, which is the important point. If the distance is less than a threshold, the segmentation point calculation ends; if it is greater than the threshold, the important point is the second-generation segmentation point. Then, connect the second-generation segmentation point with the start and end points respectively to form two second-generation baselines. Calculate the new y-direction distance between the remaining data points and the two second-generation baselines, and compare the point with the largest y-direction distance with the threshold. If it is greater than the threshold, a third-generation segmentation point is obtained. Connect the first-generation, second-generation, and third-generation segmentation points to obtain three third-generation baselines. Then, calculate the y-direction distance between the remaining data points and the third-generation baselines using the same method. Repeat this process to form segmentation points until the largest y-direction distance is less than the threshold, at which point the segmentation point calculation ends. At this point, a subsequence is formed between two adjacent segmentation points.

[0046] Step 2: Extract multiple feature values ​​from the subsequence, construct a multidimensional space based on all feature values, then map all feature values ​​into the multidimensional space, calculate the distance between each feature value, and obtain the corresponding distance value;

[0047] Step 3: After sorting the distance values ​​by size, find the k-th distance and the k nearest neighbors of each sample. Then calculate the reachability distance and local reachability density of each sample to the k nearest neighbors to obtain the corresponding distance value and density value. Then calculate the Local Outlier Factor (LOF) based on all the distance values ​​and density values ​​to obtain all the LOF values.

[0048] Step 4: If the LOF value is close to 1, the data point is considered a normal point; if the LOF value is far from 1, the data point is an outlier.

[0049] In step one, if the value is less than the threshold, the two ends are merged. Specifically, when the value is less than the threshold, the point cannot become a new dividing point, and the point is merged into the line segment formed by the previous dividing point.

[0050] In step two, the various eigenvalues ​​include length, extreme value difference, standard deviation, and mean. These four eigenvalues ​​are the four elements of the feature space.

[0051] The length represents the trend of subsequence variation, the extreme value difference is the difference between the maximum and minimum values, the standard deviation is the dispersion of the sequence data, and the mean is the average trend.

[0052] The threshold is 0.001 to 0.005 times the standard deviation.

[0053] In step two, multiple feature values ​​of the subsequence are extracted, a multidimensional space is constructed based on all feature values, and then all feature values ​​are mapped in the multidimensional space. The distance between each feature value is calculated to obtain the corresponding distance value. Specifically:

[0054] The time series is mapped to this feature space, and each subsequence corresponds to a point in the feature space. The position of each subsequence in the feature space is represented by a four-element structure composed of feature values. Based on this, the distance between each pattern is calculated to obtain the corresponding distance value.

[0055] Since the value ranges of the four feature quantities differ significantly, and different magnitudes can affect the anomaly detection results of the observation data, normalization processing is performed on these four sets of feature values ​​respectively, specifically as follows:

[0056] Let x1 = (x 11 x 12 x 13 , ..., x 1n Let x be one of the subsequences. max With x min These represent the maximum and minimum values ​​of each eigenvalue;

[0057]

[0058] In step three, after sorting the distance values ​​by size, the k-th distance and the k nearest neighbors of each sample are found. Then, the reachability distance and local reachability density from each sample to its k nearest neighbors are calculated to obtain the corresponding density values. Then, the Local Outlier Factor (LOF) is calculated based on all density values ​​to obtain all LOF values. Specifically, the density-based Local Outlier Factor algorithm is used for anomaly pattern detection, and the specific steps are as follows:

[0059] A. Calculate the Euclidean distance between any two subsequences;

[0060] B. After arranging the Euclidean distances, find the k-th distance and the k nearest neighbors of each sample;

[0061] C. Calculate the reachability distance and local reachability density of each sample to its k nearest neighbors.

[0062] The reachable distance k is calculated as follows:

[0063] reach-dist(p, q)=max(k-distance(q), d(p, q));

[0064] Where point p is a data point in the sample, point q is one of the k nearest neighbors of point p, d(p,q) represents the actual distance between point p and point q, and k-distance(q) is the distance from point q to its k-th nearest neighbor.

[0065] The locally reachable density is calculated as follows:

[0066]

[0067] In step three, the Local Outlier Factor (LOF) is calculated based on all distance and density values ​​to obtain all LOF values, specifically:

[0068]

[0069] The following are supplementary descriptions of the present invention:

[0070] The geophysical network time series observation data used has a high dimension and a wide range of data volume, and the curve shapes are diverse. Directly applying data mining techniques to the original data point by point is often inefficient. Therefore, using pattern representation to replace the original time series and reduce the dimensionality of the time series can compress the data while maintaining the basic shape of the time series.

[0071] Example 1:

[0072] A method for identifying anomalies in observation data based on SIP-LOF, comprising the following steps:

[0073] Step 1: First, extract all data points from the observation data to be detected using the PLR-SIP method. Then, treat all data points as a time series. Use the start and end points as first-generation segmentation points, connecting them to form a first-generation baseline. Calculate the y-distance between the remaining data points and the first-generation baseline, identifying the point with the largest y-distance as the important point. If the distance is less than a threshold, the segmentation calculation ends; if it is greater than the threshold, the important point is designated as a second-generation segmentation point. Then, connect the second-generation segmentation points to the start and end points respectively to form two second-generation baselines. Finally, calculate the remaining data points... The remaining data points are then compared with the new y-direction distances between the two second-generation baselines. The point with the largest y-direction distance is compared with a threshold. If the distance is greater than the threshold, a third-generation segmentation point is obtained. The first-generation, second-generation, and third-generation segmentation points are then connected to form three third-generation baselines. The y-direction distances between the remaining data points and the third-generation baselines are calculated using the same method. This process is repeated to form segmentation points until the largest y-direction distance is less than the threshold, at which point the segmentation point calculation ends. The resulting compressed segmented time series accurately reflects the overall trend and characteristics of the original data. At this point, a subsequence is formed between two adjacent segmentation points.

[0074] Step 2: Extract multiple feature values ​​from the subsequence, construct a multidimensional space based on all feature values, then map all feature values ​​into the multidimensional space, calculate the distance between each feature value, and obtain the corresponding distance value;

[0075] Step 3: After sorting the distance values ​​by size, find the k-th distance and the k nearest neighbors of each sample. Then calculate the reachability distance and local reachability density of each sample to the k nearest neighbors to obtain the corresponding distance value and density value. Then calculate the Local Outlier Factor (LOF) based on all the distance values ​​and density values ​​to obtain all the LOF values.

[0076] Step 4: If the LOF value is close to 1, the data point is considered normal; if the LOF value is far from 1, the data point is an outlier.

[0077] Example 2:

[0078] Example 2 is basically the same as Example 1, except that:

[0079] In step two, multiple feature values ​​of the subsequences are extracted, a multidimensional space is constructed based on all feature values, and then all feature values ​​are mapped into the multidimensional space. The distance between each feature value is calculated to obtain the corresponding distance value. Specifically, the time series is mapped to this feature space, each subsequence corresponds to a point in this feature space, and the position of each subsequence in the feature space is represented by a four-element representation of the feature values. Based on this, the distance between each pattern is calculated to obtain the corresponding distance value. Since the value ranges of the four feature quantities differ greatly, different magnitudes can affect the anomaly detection results of the observation data. Therefore, these four sets of feature values ​​are normalized separately, specifically as follows:

[0080] Let x1 = (x 11 x 12 x 13 , ..., x 1n Let x be one of the subsequences. max With x min These represent the maximum and minimum values ​​of each eigenvalue;

[0081]

[0082] Example 3:

[0083] Example 3 is basically the same as Example 1, except that:

[0084] In step three, after sorting the distance values ​​by size, the k-th distance and the k nearest neighbors of each sample are found. Then, the reachability distance and local reachability density from each sample to its k nearest neighbors are calculated to obtain the corresponding density values. Then, the Local Outlier Factor (LOF) is calculated based on all density values ​​to obtain all LOF values. Specifically, the density-based Local Outlier Factor algorithm is used for anomaly pattern detection, and the specific steps are as follows:

[0085] A. Calculate the Euclidean distance between any two subsequences;

[0086] B. After arranging the Euclidean distances, find the k-th distance and the k nearest neighbors of each sample;

[0087] C. Calculate the reachability distance and local reachability density of each sample to its k nearest neighbors.

[0088] The reachable distance k is calculated as follows:

[0089] reach-dist(p, q)=max(k-distance(q), d(p, q));

[0090] Where point p is a data point in the sample, point q is one of the k nearest neighbors of point p, d(p,q) represents the actual distance between point p and point q, and k-distance(q) is the distance from point q to its k-th nearest neighbor.

[0091] The locally reachable density is calculated as follows:

[0092]

[0093] In step three, the Local Outlier Factor (LOF) is calculated based on all distance and density values ​​to obtain all LOF values, specifically:

[0094]

[0095] Example 4:

[0096] Example 4 is basically the same as Example 1, except that:

[0097] Raw observation data from topographic deformation instruments at multiple seismic stations were selected, and six typical events with obvious curve characteristics were used as templates to illustrate and analyze the anomaly detection research content of typical events such as site environment, equipment failure, and earthquakes. All data used came from the precursor observation database of the Hubei Provincial Earthquake Bureau's Subject Center; the time sampling rate of the observation data was minute sampling.

[0098] Data source and measurement item information

[0099]

[0100] Directly performing anomaly detection on the raw data will greatly reduce the accuracy of the algorithm. First, the data must be preprocessed. The moving average method is used to transform non-stationary sequences into stationary sequences. The results show that the trend and periodic effects of most precursor observation data can be eliminated by the moving average method.

[0101] First, select an appropriate window length, calculate the moving average trend, and continuously slide the window forward while taking the average, such as... Figure 1 As shown, the black dashed line represents the moving average trend of the curve. The moving average trend is subtracted from the original data, and the threshold is set to twice the standard deviation of the data after removing the trend. If the detrended data is greater than twice the standard deviation, it is judged as an anomaly. At the same time, interpolation is used to replace the outlier values, thereby smoothing the short-term fluctuation data and preserving the long-term trend, forming preprocessed data, which helps to improve the quality and reliability of the precursor data and better serve the subsequent anomaly detection.

[0102] The SIP-LOP algorithm is applied to the preprocessed data, and the LOF values ​​of the subsequences and the detection curves are output. Figures 2-7The detection results of various typical events are presented. The upper part of the figure is the preprocessed observation curve, and the lower part is the anomaly detection result curve. The specific values ​​of the LOF detection values ​​reflect the degree of anomaly in each part of the original observation curve.

[0103] Figure 2 and Figure 4 Step-like phenomena appeared at the red marked areas, indicating significant data abrupt changes. The anomaly detection results showed a significant increase in the LOF value. Among them, Figure 2 The step phenomenon only shows a small vertical upward trend, proving that the method in this paper can also easily detect small-amplitude abrupt jumps. Figure 3 The trend reversal of the curve was confirmed to be caused by the instability of the pendulum's movement, and the corresponding LOF curve also changed significantly.

[0104] Figure 5 and Figure 6 Similarly, V-shaped and irregular distortions appeared on the originally smooth and stable periodic solid tide curves, which were determined to be caused by air pressure interference and surrounding infrastructure construction interference, respectively. The algorithm in this paper also successfully detected these two events. Figure 7 The water pipe instrument in Yi County recorded relatively clear coseismic changes, which showed an oscillating pattern. After the earthquake, the area returned to calm, and the LOF detection value also showed a continuous increase during the oscillation period.

[0105] Observing the results of the six tests above, it can be seen that when the observed curve is in a stable and regularly changing solid tide pattern, its corresponding LOF detection value always hovers around 1. When the observed curve exhibits distortions caused by typical events such as steps, sudden jumps, or trend reversals, its corresponding LOF detection value will increase significantly to between 2 and 10. To facilitate anomaly detection in actual observation work, this scheme selects topographic deformation instrument observation data from the Yixian Seismic Station in 2022 to further quantify anomaly detection, that is, to set a limit for the LOF detection value. When the LOF detection value exceeds this limit, it is determined that an anomaly has been detected.

[0106] The selection of the LOF limit is crucial for determining whether the observed data has reached anomalies. If the LOF limit is too small, normal fluctuations that do not belong to anomalies will be misjudged as anomalies. If the LOF limit is too large, it will cause the omission of local minor distortions caused by events such as microseismic events, site environment, or natural disturbances. Therefore, this paper, based on a comprehensive consideration of recall and precision, initially selects the lower limit of the range 2-10 mentioned above as the limit. Statistical experiments were conducted around the value of 2. The statistical results are shown in the table below. The average accuracy of the algorithm for the four measurement items reaches 90.1%. When the LOF limit is 2.5, the average anomaly detection accuracy is the highest.

[0107] Anomaly detection statistics

[0108]

[0109] The above description is only a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. Any equivalent modifications or changes made by those skilled in the art based on the content disclosed in the present invention should be included within the scope of protection set forth in the claims.

Claims

1. A SIP-LOF-based observation data anomaly identification method, characterized in that: The SIP-LOF-based observation data anomaly identification method includes the following steps: Step 1: First, select the raw observation data from the seismic station's topographic deformation instrument. Extract all data points from the raw observation data using the PLR-SIP method. Then, treat all data points as a time series. Use the starting and ending data points as first-generation segmentation points. Connect the two first-generation segmentation points to form the first-generation baseline. Calculate the y-direction distance between the remaining data points and the first-generation baseline, and identify the point with the largest y-direction distance, which is the important point. If the distance is less than a threshold, the segmentation point calculation ends; if it is greater than the threshold, the important point is the second-generation segmentation point. Then, divide the second-generation segmentation points... Connect the remaining data points to the starting and ending points to form two second-generation baselines. Then, calculate the new y-distance between the remaining data points and the two second-generation baselines. Find the point with the largest y-distance and compare it with a threshold. If it is greater than the threshold, a third-generation segmentation point is obtained. Then, connect the first-generation, second-generation, and third-generation segmentation points to obtain three third-generation baselines. Then, calculate the y-distance between the remaining data points and the third-generation baselines. Repeat this process to form segmentation points until the largest y-distance is less than the threshold. At this point, a subsequence is formed between two adjacent segmentation points. Step 2: Extract multiple feature values ​​from the subsequence, construct a multidimensional space based on all feature values, then map all feature values ​​into the multidimensional space, calculate the distance between each feature value, and obtain the corresponding distance value; Step 3: After sorting the distance values ​​by size, find the k-th distance and the k nearest neighbors of each sample. Then calculate the reachability distance and local reachability density of each sample to the k nearest neighbors to obtain the corresponding distance value and density value. Then calculate the Local Outlier Factor (LOF) based on all the distance values ​​and density values ​​to obtain all the LOF values. Step 4: If the LOF value is close to 1, the data point is considered normal and is therefore judged as normal. If the LOF value is far from 1, the data point is considered an outlier and is therefore judged as abnormal. This method can accurately and effectively identify typical events, including equipment failure, natural disturbances, site environment, and geophysical events.

2. The method for identifying anomalies in observation data based on SIP-LOF according to claim 1, characterized in that: In step one, if the distance is less than the threshold, the calculation of the dividing point ends. Specifically, when the distance is less than the threshold, the point cannot become a new dividing point. The point is merged into the line segment formed by the previous dividing point, and the dividing process ends.

3. The method for identifying anomalies in observation data based on SIP-LOF according to claim 2, characterized in that: The threshold is 0.001 to 0.005 times the standard deviation.

4. The method for identifying anomalies in observation data based on SIP-LOF according to claim 1, characterized in that: In step two, the various eigenvalues ​​include length, extreme value difference, standard deviation, and mean. These four eigenvalues ​​are the four elements of the feature space. The length represents the trend of subsequence variation, the extreme value difference is the difference between the maximum and minimum values, and the standard deviation is the sequence data. The degree of dispersion, with the mean being the average trend.

5. The method for identifying anomalies in observation data based on SIP-LOF according to claim 4, characterized in that: In step two, multiple feature values ​​of the subsequence are extracted, a multidimensional space is constructed based on all feature values, and then all feature values ​​are mapped in the multidimensional space. The distance between each feature value is calculated to obtain the corresponding distance value. Specifically: The time series is mapped to this feature space, and each subsequence corresponds to a point in this feature space. The position of each subsequence in the feature space is represented by a four-element structure composed of feature values. Based on this, the distance between each pattern is calculated to obtain the corresponding distance value.

6. The method for identifying anomalies in observation data based on SIP-LOF according to claim 5, characterized in that: Because the value ranges of the four feature quantities differ significantly, and different magnitudes can affect the anomaly detection results of the observed data, these four sets of feature values ​​are normalized separately, as follows: set up For one of the subsequences, and These represent the maximum and minimum values ​​of each eigenvalue; 。 7. The method for identifying anomalies in observation data based on SIP-LOF according to claim 1, characterized in that: In step three, after sorting the distance values ​​by size, the k-th distance and the k nearest neighbors of each sample are found. Then, the reachability distance and local reachability density from each sample to its k nearest neighbors are calculated to obtain the corresponding density values. Then, the Local Outlier Factor (LOF) is calculated based on all density values ​​to obtain all LOF values. Specifically, the density-based Local Outlier Factor algorithm is used for anomaly pattern detection, and the specific steps are as follows: A. Calculate the Euclidean distance between any two subsequences; B. After arranging the Euclidean distances, find the k-th distance and the k nearest neighbors of each sample; C. Calculate the reachability distance and local reachability density of each sample to its k nearest neighbors.

8. The method for identifying anomalies in observation data based on SIP-LOF according to claim 7, characterized in that: The reachable distance k is calculated as follows: ； Among them, point It is a data point in the sample. It is a point of One of the neighbors, Point With point The actual distance between them It is a point The distance to its k-th nearest neighbor.

9. The method for identifying anomalies in observation data based on SIP-LOF according to claim 8, characterized in that: The locally reachable density is calculated as follows: 。 10. The method for identifying anomalies in observation data based on SIP-LOF according to claim 9, characterized in that: In step three, the Local Outlier Factor (LOF) is calculated based on all distance and density values ​​to obtain all LOF values, specifically: 。

Images