Abnormality detection method, processing method, device and system for transformer oil chromatography time sequence data

Abstract

The invention discloses an anomaly detection method, an anomaly processing method, an anomaly processing device and an anomaly processing system for oil chromatography time sequence data of a transformer. The method comprises the following steps: acquiring time sequence data of a transformer oil chromatogram; preprocessing the time series data and then decomposing the time series data to obtain components of seasons, trends and residual errors; detecting the position of an abnormal value according to the residual component, marking the value of the position of the time series data as the abnormal value, processing the abnormal value, and extracting a trend component and a season component from the processed time series data; obtaining a trend component prediction value by using the trend component prediction model, and obtaining a seasonal component prediction value by using the seasonal component prediction model; combining the trend component prediction value with the season component prediction value to form integral prediction time sequence data; and finally, processing the transformer oil chromatography time sequence data of which the detection result is an abnormal detection result. According to the method, the transformer oil chromatography time sequence abnormal data can be effectively and accurately identified in a dynamic environment based on the trend change of the time sequence data.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformers, and in particular relates to an abnormality detection method, a processing method, a device and a system for transformer oil chromatogram time series data. Background Art

[0002] Transformer oil chromatographic data is easily affected by the operating status and measurement accuracy of the sensor during transmission, which may lead to missing or abnormal data. At present, the anomaly detection method for oil chromatographic data mainly judges abnormal points by analyzing the distribution characteristics of the data. Commonly used methods include Z-score and box plot method. In addition, there are distance and density-based methods. However, these methods often fail to pay attention to the trend changes of data, and it is difficult to accurately detect abnormal points, resulting in a high probability of misdiagnosis and missed diagnosis, and failing to be close to the actual data. Summary of the invention

[0003] The technical problem to be solved by the present invention is to address the above-mentioned deficiencies in the prior art and propose an anomaly detection method, processing method, device and system for transformer oil chromatogram time series data. The anomaly detection method can effectively and accurately identify transformer oil chromatogram time series abnormal data in a dynamic environment based on the trend changes of the time series data.

[0004] In a first aspect, the present invention provides a method for detecting abnormality of transformer oil chromatogram time series data, the method comprising the following steps:

[0005] Obtain time series data of transformer oil chromatogram;

[0006] Decomposing the time series data to obtain residual components of the time series data;

[0007] According to the residual component, it is determined whether the time series data is abnormal data, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

[0008] Furthermore, the decomposing the time series data to obtain the residual components of the time series data is specifically: decomposing the time series data according to the STL method to obtain the residual components of the time series data;

[0009] Decomposing the time series data can also obtain the seasonal component and trend component of the time series data, and the specific steps are: decomposing the time series data according to the STL method to obtain the seasonal component of the time series data; and decomposing the time series data according to the STL method to obtain the trend component of the time series data;

[0010] The specific calculation formula for decomposing the time series data according to the STL method is as follows:

[0011] Y v =Tv +s v +r v ;

[0012] in,

[0013] Y v Represents time series data;

[0014] T v represents the trend component;

[0015] S v Indicates seasonal component;

[0016] R v represents the residual component;

[0017] v is the time index, v = 1,…,N;

[0018] N is the total number of time series;

[0019] The determining, based on the residual component, whether the time series data is abnormal data specifically includes: when the residual component exceeds a threshold range, determining that the time series data is abnormal data.

[0020] Furthermore, the decomposition of the time series data according to the STL method to obtain the trend component of the time series data is combined with the inner loop iteration to process the time series data to obtain the trend component, which specifically includes the following steps:

[0021] Step A1: Remove the trend component from time series data

[0022] The method of removing the trend component in the time series data specifically includes the following steps:

[0023] Step A11: Initialization assignment;

[0024] The initialization assignment specifically includes: setting K is the number of iterations;

[0025] Step A12: Calculate the Kth trend component according to the following formula (1):

[0026] in, is the trend value in the Kth iteration;

[0027] Step A13: Calculate the time series data with trend components removed according to the following formula (2):

[0028]

[0029] Step A2: Perform subsequence smoothing on the time series data with trend components removed;

[0030] The step A2 specifically includes:

[0031] Step A21: Set the time series period to n p , then we get n p periodic subsequences;

[0032] Step A22: Perform the LOESS process on each periodic subsequence, and extend each periodic subsequence forward and backward by one period, and combine them to obtain a length of N+2×n p Time Series N is the total number of time series;

[0033] Step A3: Perform 3 times with length n p , n p , 3 sliding averages, and one LOESS process is performed to obtain a time series of length N

[0034] Step A4: Based on the time series of length N Find the seasonal component;

[0035] The calculation formula of the seasonal component is as follows:

[0036]

[0037] Step A5: Remove the seasonal component from the time series data and then perform LOESS processing to obtain the trend component

[0038] The trend component The calculation formula is as follows:

[0039]

[0040] Express Perform LOESS process.

[0041] Furthermore, in the process of decomposing the time series data, the adjustment of the robust weight is used as an outer loop for processing the time series data. After the inner loop outputs the STL decomposition result of the current iteration, the robust weight required for the next iteration is obtained according to the STL decomposition result of the current iteration and the calculation formula of the robust weight.

[0042] Among them, the robustness weight ρ v The calculation formula is as follows:

[0043] ρv =B*(R v | / h); (4)

[0044] Where:

[0045] R v Represents the remainder at time v, whose value is equal to Y v -T v -S v , that is, the residual component;

[0046] The B function is a bisquare weight function, that is, a bisquare function, and its specific calculation process is as shown in formula (5):

[0047]

[0048] h is an intermediate variable, and its obtaining process is as shown in formula (6):

[0049] h=6*median(R v ); (6)

[0050] u is the independent variable of the bisquare function;

[0051] median() represents the median function.

[0052] Furthermore, when the residual component exceeds a threshold range, the time series data is determined to be abnormal data, which specifically includes the following steps:

[0053] Calculate the mean and standard deviation of the residual components;

[0054] Calculate the absolute value of the difference between the residual component value and the mean, and determine whether the absolute value is greater than twice the standard deviation:

[0055] If the absolute value is greater than twice the standard deviation, the time series data corresponding to the residual component value is determined to be abnormal data; if the absolute value is not greater than twice the standard deviation, the time series data corresponding to the residual component value is determined to be not abnormal data.

[0056] Furthermore, before detecting the transformer oil chromatogram time series data, the method further comprises the steps of: preprocessing the transformer oil chromatogram time series data;

[0057] The preprocessing of the transformer oil chromatogram time series data specifically includes the following steps:

[0058] Step B1: Check whether the transformer oil chromatogram time series data is missing:

[0059] If the transformer oil chromatogram time series data does not have any missing, the transformer oil chromatogram time series data is not preprocessed; if the transformer oil chromatogram time series data has any missing, proceed to step B2;

[0060] Step B2: Fill missing values ​​or remove data according to the missing rate and the proportion of consecutive missing values, which specifically includes the following steps:

[0061] Step B21: Determine whether the missing rate meets the first condition or the second condition. The first condition is that the missing rate does not exceed the first threshold value. The second condition is that the missing rate is between the first threshold value and the second threshold value, and the continuous missing does not exceed the third threshold value of the total missing value. Then continue to determine whether the missing item is the first value in the transformer oil chromatogram time series data. Otherwise, proceed to step B22:

[0062] If the missing item is the first value in the transformer oil chromatogram time series data, the next valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing; otherwise, the previous valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing;

[0063] Step B22: Eliminate the transformer oil chromatogram time series data;

[0064] The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

[0065] In a second aspect, the present invention provides a method for processing abnormalities in transformer oil chromatogram time series data, the method comprising the following steps:

[0066] The abnormality detection method for transformer oil chromatogram time series data described in the first aspect is adopted to obtain abnormality detection results of transformer oil chromatogram time series data;

[0067] Extract trend and seasonal components;

[0068] Taking the trend component as input, and using the trend component prediction model to obtain a predicted value of the trend component; and taking the season component as input, and using the season component prediction model to obtain a predicted value of the season component;

[0069] Combine the forecast value of the trend component with the forecast value of the seasonal component to form the overall forecast time series data;

[0070] According to the predicted time series data, the transformer oil chromatogram time series data with abnormal detection results is processed.

[0071] In a third aspect, the present invention provides a device for detecting abnormality of transformer oil chromatogram time series data, the device comprising:

[0072] An acquisition unit, used for acquiring time series data of transformer oil chromatogram;

[0073] a decomposition unit, connected to the acquisition unit, and configured to decompose the time series data to obtain a residual component of the time series data;

[0074] The determination unit is connected to the decomposition unit and is used to determine whether the time series data is abnormal data according to the residual component, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

[0075] Further, the decomposition unit includes a third decomposition unit;

[0076] The third decomposition unit is connected to the acquisition unit, and is used to decompose the time series data according to the STL method to obtain the residual component of the time series data;

[0077] The decomposition unit further comprises a first decomposition unit and a second decomposition unit;

[0078] The first decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the seasonal component of the time series data;

[0079] The second decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the trend component of the time series data;

[0080] The decomposition unit stores the following calculation formula for decomposition:

[0081] Y v =T v +S v +R v ;

[0082] in,

[0083] Y v Represents time series data;

[0084] T v represents the trend component;

[0085] S v Indicates seasonal component;

[0086] R v represents the residual component;

[0087] v is the time index, v = 1,…,N;

[0088] N is the total number of time series;

[0089] The determination unit includes a first determination unit, which is connected to the decomposition unit and is used to determine that the time series data is abnormal data when the residual component exceeds a preset threshold range therein.

[0090] Furthermore, the device also includes a pre-processing unit,

[0091] The preprocessing unit is connected to the acquisition unit and is used to preprocess the transformer oil chromatogram time series data;

[0092] The pre-processing unit specifically comprises the following steps:

[0093] A checking unit, used for checking whether the transformer oil chromatogram time series data is missing;

[0094] A preprocessing unit connected to the inspection unit, and used for preprocessing the transformer oil chromatogram time series data when the inspection unit determines that the transformer oil chromatogram time series data is missing;

[0095] The pre-processing unit comprises:

[0096] A first determination module is used to determine whether the missing rate satisfies a first condition or a second condition, wherein the first condition is that the missing rate does not exceed a first threshold, and the second condition is that the missing rate is between the first threshold and the second threshold, and the number of consecutive missing does not exceed a third threshold of the total number of missing data;

[0097] A second determination module, connected to the first determination module, is used to determine whether the missing item is the first value in the transformer oil chromatogram time series data when the first determination module determines that the missing rate meets the first condition or the second condition;

[0098] A filling module, connected to the second determination module, is used to fill the missing item with the next valid value of the missing item when the second determination module determines that the missing item is the first value in the transformer oil chromatogram time series data; and is also used to fill the missing item with the previous valid value of the missing item when the second determination module determines that the missing item is not the first value in the transformer oil chromatogram time series data;

[0099] A rejection module, connected to the first determination module, for rejecting the transformer oil chromatogram time series data when the first determination module determines that the missing rate cannot meet the first condition or the second condition;

[0100] The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

[0101] In a fourth aspect, the present invention provides an abnormality processing system for transformer oil chromatogram time series data, the system comprising:

[0102] The abnormality detection device for transformer oil chromatogram time series data of the third aspect; the abnormality detection device is used to obtain the abnormality detection result of the transformer oil chromatogram time series data;

[0103] an extraction unit, connected to the anomaly detection device, for extracting a trend component and a seasonal component from the anomaly detection result;

[0104] a first prediction unit, connected to the extraction unit, configured to take the trend component as input and obtain a predicted value of the trend component using a trend component prediction model;

[0105] a second prediction unit, connected to the extraction unit, for taking the seasonal component as input and obtaining a predicted value of the seasonal component using a seasonal component prediction model;

[0106] a combining unit, connected to the first prediction unit and the second prediction unit respectively, for combining the predicted value of the trend component with the predicted value of the season component to form overall predicted time series data;

[0107] The processing unit is connected to the combining unit and is used to process the transformer oil chromatogram time series data whose detection result is an abnormal detection result.

[0108] Beneficial effects of the present invention:

[0109] 1. The present invention focuses on the trend change of time series data, so that the abnormal time series data of transformer oil chromatogram can be more effectively identified in a dynamic environment. At the same time, it can identify the dynamic change of data trend and provide more sensitive abnormal detection results.

[0110] 2. The present invention can improve the detection accuracy of transformer oil chromatographic time series data.

[0111] 3. The present invention can enhance robustness and has stronger adaptability.

[0112] 4. The present invention enhances the ability to understand complex data and identify anomalies by combining trend analysis and machine learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0113] Figure 1 Schematic diagram of abnormal processing flow of transformer oil chromatogram time series data in an embodiment of the present invention;

[0114] Figure 2 Schematic diagram of an abnormality detection method for transformer oil chromatogram time series data in an embodiment of the present invention;

[0115] Figure 3 This is a diagram showing the abnormal processing effect of transformer oil chromatogram time series data in an embodiment of the present invention;

[0116] Figure 4 Schematic diagram of an abnormality detection device for transformer oil chromatogram time series data in an embodiment of the present invention;

[0117] Among them, the figure numerals are: 10, acquisition unit, 20, decomposition unit, 30, determination unit. DETAILED DESCRIPTION

[0118] In order to enable those skilled in the art to better understand the technical solution of the present invention, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0119] It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention rather than to limit the present invention.

[0120] It can be understood that, in the absence of conflict, the various embodiments of the present invention and the various features in the embodiments can be combined with each other.

[0121] It can be understood that, for the convenience of description, the drawings of the present invention only show the parts related to the present invention, while the parts irrelevant to the present invention are not shown in the drawings.

[0122] It can be understood that each unit and module involved in the embodiments of the present invention may correspond to only one physical structure, or may be composed of multiple physical structures, or multiple units and modules may be integrated into one physical structure.

[0123] It can be understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present invention may occur in an order different from that marked in the drawings.

[0124] It is understood that the flowcharts and block diagrams of the present invention illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present invention. Each box in the flowchart or block diagram may represent a unit, module, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based system that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0125] It can be understood that the units and modules involved in the embodiments of the present invention can be implemented by software or hardware. For example, the units and modules can be located in a processor.

[0126] Embodiment 1:

[0127] like Figure 1 and Figure 2As shown, this embodiment provides a method for detecting abnormality of transformer oil chromatogram time series data, which is used in the field of transformer production. The method comprises the following steps:

[0128] Step S1: Acquire time series data of transformer oil chromatogram.

[0129] The time series data of transformer oil chromatogram usually includes the following specific data:

[0130] 1. Basic time series data

[0131] Timestamp: records the specific time when data is collected, usually expressed in the format of date and time.

[0132] Measurement period: the frequency of data collection (such as hourly, daily, etc.).

[0133] 2. Oil chromatographic composition data

[0134] Solvent component concentration: includes the concentration values ​​of different gases (such as hydrogen, hydrocarbons, oxygen, carbon dioxide, etc.), usually expressed in ppm (parts per million) or mg / L (milligrams per liter).

[0135] Elemental composition: such as the content of impurities such as water, ether, alcohol, etc.

[0136] 3. Electrical parameters

[0137] Dielectric strength of oil: the voltage strength before dielectric breakdown.

[0138] Oil temperature: The temperature of the transformer oil, usually expressed in degrees Celsius (℃).

[0139] 4. Physical parameters

[0140] Oil density: The density of oil, usually in kg / m 3 express.

[0141] Oil viscosity: The viscosity value of oil reflects its fluidity.

[0142] 5. Environmental parameters

[0143] Ambient temperature and humidity: External environmental factors that affect transformer and oil chromatographic data.

[0144] Air pressure: Ambient air pressure may affect oil chromatography measurements.

[0145] 6. Sensor status data

[0146] Sensor operating status: Indicates whether the sensor is working properly, such as online / offline status.

[0147] Sensor calibration records: including sensor calibration date and calibration results, etc.

[0148] 7. Anomaly detection and alert logging

[0149] Outlier marking: Identification and recording of abnormal data.

[0150] Alarm event records: time, type and handling measures of the alarm, etc.

[0151] 8. Historical data snapshot

[0152] Past data records: Historical oil chromatographic composition data for easy trend analysis and comparison.

[0153] Step S2: preprocessing the transformer oil chromatogram time series data;

[0154] The preprocessing of the transformer oil chromatogram time series data specifically includes the following steps:

[0155] Step B1: Check whether the transformer oil chromatogram time series data is missing:

[0156] If the transformer oil chromatogram time series data does not have any missing, the transformer oil chromatogram time series data is not preprocessed; if the transformer oil chromatogram time series data has any missing, proceed to step B2;

[0157] Step B2: Fill missing values ​​or remove data according to the missing rate and the proportion of consecutive missing values, which specifically includes the following steps:

[0158] Step B21: Determine whether the missing rate meets the first condition or the second condition. The first condition is that the missing rate does not exceed the first threshold value. The second condition is that the missing rate is between the first threshold value and the second threshold value, and the continuous missing does not exceed the third threshold value of the total missing value. Then continue to determine whether the missing item is the first value in the transformer oil chromatogram time series data. Otherwise, proceed to step B22:

[0159] If the missing item is the first value in the transformer oil chromatogram time series data, the next valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing; otherwise, the previous valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing;

[0160] Step B22: Eliminate the transformer oil chromatogram time series data;

[0161] The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

[0162] According to practical applications, the missing values ​​of oil chromatography time series data can be divided into two situations: scattered missing values, that is, there are a small number of isolated missing values ​​in the data set, which may be due to accidental errors in the data collection or transmission process; continuous missing values, that is, there are continuous missing values ​​at multiple time points in the data set, which may be caused by equipment failure, sensor deactivation or other systematic problems.

[0163] Based on the above description of missing data, for scattered missing data (i.e., the missing rate does not exceed 10%), the missing value is filled with the previous valid value. If the first value of the time series data is missing, it is filled with the next valid value. For time series data with a missing rate exceeding 40%, it is directly eliminated. If it does not exceed 40% and the consecutive missing values ​​do not exceed 50% of the total missing values, the missing value is filled with the previous valid value (if the first value of the time series data is missing, it is filled with the next valid value); otherwise, it is directly eliminated.

[0164] Step S3: Decompose the time series data to obtain residual components of the time series data.

[0165] As a specific implementation manner, the decomposing the time series data to obtain the residual components of the time series data is specifically: decomposing the time series data according to the STL method to obtain the residual components of the time series data.

[0166] By decomposing the time series data, the seasonal component and trend component of the time series data can also be obtained. The specific steps are: decomposing the time series data according to the STL method to obtain the seasonal component of the time series data; and, decomposing the time series data according to the STL method to obtain the trend component of the time series data.

[0167] In time series data analysis, in addition to the STL (Seasonal-Trend decomposition using Loess) method, there are many other methods to decompose time series data. The following are some common alternative methods:

[0168] 1. Addition and multiplication models

[0169] Depending on the characteristics of the data, an additive or multiplicative model can be used for decomposition.

[0170] Additive model: suitable for situations where the seasonal variation is relatively constant.

[0171] Multiplicative model: Suitable for situations where the magnitude of seasonal changes changes as the trend changes.

[0172] 2. X11 / X12 decomposition

[0173] X11 and X12 are procedures developed by the U.S. Census Bureau for seasonal adjustment and trend analysis of time series data. These methods can handle seasonal effects very effectively.

[0174] 3. Moving Average

[0175] The trend component is extracted by smoothing the data by calculating the moving average of a certain window size. This method is often used to eliminate short-term fluctuations while observing long-term trends.

[0176] 4. Hodrick-Prescott (HP) filtering

[0177] HP filtering is a commonly used economic method to separate the trend component and the cycle component in time series data. By adjusting the smoothing parameter, the smoothing degree of the trend component can be controlled.

[0178] 5. Sarima Model

[0179] The Seasonal Autoregressive Integrated Moving Average (SARIMA) model can be used to analyze and model time series data, which takes into account seasonal and trend components. The SARIMA method can not only predict but also describe the seasonal and cyclical characteristics of data.

[0180] 6. FFT (Fast Fourier Transform)

[0181] Fourier transform can convert time series data from the time domain to the frequency domain, thereby analyzing the frequency components in the data. This method is often used to identify periodic behavior.

[0182] 7. Machine Learning Methods

[0183] LSTM (Long Short-Term Memory Network): A deep learning model suitable for processing temporal dependencies in long time series.

[0184] Random Forest: It can also be used for time series forecasting, although it is not designed specifically for time series data in nature.

[0185] 8.ETS (Exponential Smoothing State Space Model)

[0186] The ETS model uses exponential smoothing methods to model time series data, taking into account trends (Exponentialsmoothing Trend), seasonality (Seasonal) and periodicity (Error).

[0187] In fact, when judging whether the time series data is abnormal data, only the residual component is needed, and the implementation process is as follows:

[0188] Calculate the mean and standard deviation of the residual components;

[0189] Calculate the absolute value of the difference between the residual component value and the mean, and determine whether the absolute value is greater than twice the standard deviation:

[0190] If the absolute value is greater than twice the standard deviation, the time series data corresponding to the residual component value is determined to be abnormal data; if the absolute value is not greater than twice the standard deviation, the time series data corresponding to the residual component value is determined not to be abnormal data.

[0191] Theoretically, 3σ can also be used, but the 3σ (three times the standard deviation) principle is stricter than the 2σ (twice the standard deviation) principle. Only a few data points will be considered outliers. In contrast, the 2σ principle will include more data points and will more sensitively identify possible outliers. Therefore, the obtained residual component is used to perform anomaly detection using the 2σ principle.

[0192] The specific calculation formula for decomposing the time series data according to the STL method is as follows:

[0193] y v =T v +S v +R v ;

[0194] in,

[0195] Y v Represents time series data;

[0196] T v represents the trend component;

[0197] S v Indicates seasonal component;

[0198] R v represents the residual component;

[0199] v is the time index, v = 1,…,N;

[0200] N is the total number of time series.

[0201] As a more specific implementation, the decomposition of the time series data according to the STL method to obtain the trend component of the time series data is combined with an inner loop iteration to process the time series data to obtain the trend component, which specifically includes the following steps:

[0202] Step A1: Remove the trend component from the time series data

[0203] The step of removing the trend component from the time series data specifically includes the following steps:

[0204] Step A11: Initialization assignment;

[0205] The initialization assignment specifically includes: setting K=0,

[0206] Step A12: Calculate the Kth trend component according to the following formula

[0207] in, is the trend value in the Kth iteration;

[0208] Step A13: Calculate the time series data with trend components removed according to the following formula:

[0209]

[0210] Step A2: Perform subsequence smoothing on the time series data with trend components removed;

[0211] The step A2 specifically includes:

[0212] Step A21: Set the time series period to n p , then we get n p periodic subsequences;

[0213] Step A22: Perform the LOESS process on each periodic subsequence, and extend each periodic subsequence forward and backward by one time point, and combine them to obtain a length of N+2×n p Time Series

[0214] Step A3: Perform 3 times with length n p , n p , 3 sliding averages, and one LOESS process is performed to obtain a time series of length N

[0215] Step A4: Based on the time series of length N Find the seasonal component;

[0216] The calculation formula of the seasonal component is as follows:

[0217]

[0218] Step A5: Remove the seasonal component from the time series data and then perform LOESS processing to obtain the trend component

[0219] The trend component The calculation formula is as follows:

[0220]

[0221] Express Perform LOESS process.

[0222] As a specific implementation method, in the process of decomposing time series data, the adjustment of robust weights is used as an outer loop for processing the time series data. After the inner loop outputs the STL decomposition result of the current iteration, the robust weights required for the next iteration are obtained according to the STL decomposition result of the current iteration and the calculation formula of the robust weights;

[0223] Among them, the robustness weight ρ v The calculation formula is as follows:

[0224] ρ v =B*(|R v | / h);

[0225] Where:

[0226] R v Represents the remainder at time v, whose value is equal to Y v -T v -S v , that is, the residual component;

[0227] The B function is a bisquare weight function, i.e., a bisquare function;

[0228]

[0229] h is the intermediate variable; h=6*median(|R v |);

[0230] u is the independent variable of the bisquare function;

[0231] median() represents the median function.

[0232] Step S4: judging whether the time series data is abnormal data according to the residual component:

[0233] If the residual component exceeds the threshold range, the time series data is determined to be abnormal data, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

[0234] Embodiment 2:

[0235] like Figure 1 As shown, this embodiment provides a method for handling abnormalities in transformer oil chromatogram time series data, the method comprising the following steps:

[0236] The abnormality detection method for transformer oil chromatogram time series data described in Example 1 is used to obtain abnormality detection results of transformer oil chromatogram time series data;

[0237] Extract trend and seasonal components;

[0238] Taking the trend component as input, and using the trend component prediction model to obtain a predicted value of the trend component; and taking the season component as input, and using the season component prediction model to obtain a predicted value of the season component;

[0239] Combine the forecast value of the trend component with the forecast value of the seasonal component to form the overall forecast time series data;

[0240] According to the predicted time series data, the transformer oil chromatogram time series data with abnormal detection results is processed.

[0241] Among them, the trend component prediction model and the seasonal component prediction model are both trained based on historical data.

[0242] The following combination Figure 1 , a preferred specific embodiment is further described.

[0243] Oil chromatogram gas data mainly include hydrogen (H2), methane (C2H4), ethane (C2H6), ethylene (C2H4), acetylene (C2H2), carbon monoxide (CO), and carbon dioxide (CO2). Collect oil chromatogram gas time series data of multiple transformers in monthly units.

[0244] After data preprocessing, these oil chromatogram gas time series data are preprocessed. If there is no missing value, directly enter the data anomaly detection. If there is a missing value, determine whether it is a scattered missing value. If it is, it needs to be filled with the previous valid value of the missing value. If the first value of the time series data is missing, it is filled with the next valid value; if it is not a scattered missing value, further determine whether the data is available. If it is not available, directly eliminate it. Otherwise, fill it in the above method.

[0245] After the data anomaly detection, the preprocessed time series data is subjected to STL time series decomposition. Python development can directly import STL from statsmodels.tsa.seasonal, with a quarter of 3 months as the cycle, set period to 3, robust to True, and other values ​​​​can be defaulted. Since the residual component is obtained by decomposing the original data to remove the trend and seasonal components through STL, in this process, if the abnormal point has a large impact on the trend and seasonal components of the data, its performance in the residual component will become less obvious. Although theoretically 3σ can also be used, the 3σ principle is stricter than the 2σ principle, and only a very small number of data points will be regarded as outliers. In contrast, the 2σ principle will include more data points and will more sensitively identify possible outliers. Therefore, the obtained residual component is used to adopt the 2σ principle for anomaly detection.

[0246] After the abnormal data processing, the sliding average method is used to preliminarily process the abnormal points, and then the parameters used in the STL method in data anomaly detection are used to fit the trend component and seasonal component of the new time series data, and the two are combined as the time series data after abnormal processing. The effect is as follows Figure 3 As shown in the figure, the data will be much smoother after exception processing.

[0247] This example uses the H2 time series data of a 750kV transformer oil chromatogram as an example to verify the abnormal detection data and abnormal processing data (as shown in the following table):

[0248]

[0249] After the STL method is applied to the original data, the residual component is obtained. The distribution of the residual component conforms to the normal distribution. In order to avoid missing outliers, the 2σ principle can be used to detect abnormal points. The abnormal detection method of the abnormal detection device can accurately detect the calibrated abnormal points, and then use the abnormal processing module to correct the abnormal points. The corrected values ​​are: 22.494ppm and 157.401ppm.

[0250] This embodiment is based on the embodiment 1, firstly, the time series data of the transformer oil chromatogram is obtained; then, the time series data is preprocessed and then decomposed to obtain the season, trend and residual components; then, according to the residual component, the position of the abnormal value is detected, the value of the time series data at this position is marked as an abnormal value and processed, and then the trend component and the season component are extracted from the processed time series data; then, the trend component prediction model is used to obtain the trend component prediction value, and the season component prediction model is used to obtain the season component prediction value; then, the trend component prediction value and the season component prediction value are combined to form the overall predicted time series data; finally, the transformer oil chromatogram time series data whose detection result is an abnormal detection result is processed. This embodiment can effectively and accurately identify the abnormal transformer oil chromatogram time series data in a dynamic environment based on the trend change of the time series data.

[0251] Embodiment 3:

[0252] like Figure 4 As shown, this embodiment provides a device for detecting abnormality of transformer oil chromatogram time series data, the device comprising:

[0253] An acquisition unit 10 is used to acquire time series data of transformer oil chromatogram;

[0254] A decomposition unit 20, connected to the acquisition unit 10, for decomposing the time series data to obtain residual components of the time series data;

[0255] The determination unit 30 is connected to the decomposition unit 20 and is used to determine whether the time series data is abnormal data according to the residual component, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

[0256] As a specific implementation, the decomposition unit 20 includes a third decomposition unit;

[0257] The third decomposition unit is connected to the acquisition unit, and is used to decompose the time series data according to the STL method to obtain the residual component of the time series data;

[0258] The decomposition unit 20 also includes a first decomposition unit and a second decomposition unit;

[0259] The first decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the seasonal component of the time series data;

[0260] The second decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the trend component of the time series data;

[0261] The third decomposition unit is connected to the acquisition unit, and is used to decompose the time series data according to the STL method to obtain the residual component of the time series data;

[0262] The decomposition unit stores the following calculation formula for decomposition:

[0263] Y v =T v +S v +R v ;

[0264] in,

[0265] Y v Represents time series data;

[0266] T v represents the trend component;

[0267] S v Indicates seasonal component;

[0268] R v represents the residual component;

[0269] v is the time index, v = 1,…,N;

[0270] N is the total number of time series.

[0271] The determination unit 30 includes a first determination unit, which is connected to the decomposition unit and is used to determine that the time series data is abnormal data when the residual component exceeds a threshold range.

[0272] As a specific implementation, the device further includes a preprocessing unit.

[0273] The preprocessing unit is connected to the acquisition unit and is used to preprocess the transformer oil chromatogram time series data;

[0274] The pre-processing unit specifically comprises the following steps:

[0275] A checking unit, used for checking whether the transformer oil chromatogram time series data is missing;

[0276] A preprocessing unit connected to the inspection unit, and used for preprocessing the transformer oil chromatogram time series data when the inspection unit determines that the transformer oil chromatogram time series data is missing;

[0277] The pre-processing unit comprises:

[0278] A first determination module is used to determine whether the missing rate satisfies a first condition or a second condition, wherein the first condition is that the missing rate does not exceed a first threshold, and the second condition is that the missing rate is between the first threshold and the second threshold, and the number of consecutive missing does not exceed a third threshold of the total number of missing data;

[0279] A second determination module, connected to the first determination module, is used to determine whether the missing item is the first value in the transformer oil chromatogram time series data when the first determination module determines that the missing rate meets the first condition or the second condition;

[0280] A filling module, connected to the second determination module, is used to fill the missing item with the next valid value of the missing item when the second determination module determines that the missing item is the first value in the transformer oil chromatogram time series data; and is also used to fill the missing item with the previous valid value of the missing item when the second determination module determines that the missing item is not the first value in the transformer oil chromatogram time series data;

[0281] A rejection module, connected to the first determination module, for rejecting the transformer oil chromatogram time series data when the first determination module determines that the missing rate cannot meet the first condition or the second condition;

[0282] The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

[0283] Embodiment 4:

[0284] This embodiment provides a system for handling abnormalities in transformer oil chromatogram time series data, the system comprising:

[0285] The abnormality detection device for transformer oil chromatogram time series data described in Example 3; the abnormality detection device is used to obtain the abnormality detection result of the transformer oil chromatogram time series data;

[0286] an extraction unit, connected to the anomaly detection device, for extracting a trend component and a seasonal component from the anomaly detection result;

[0287] a first prediction unit, connected to the extraction unit, configured to take the trend component as input and obtain a predicted value of the trend component using a trend component prediction model;

[0288] a second prediction unit, connected to the extraction unit, for taking the seasonal component as input and obtaining a predicted value of the seasonal component using a seasonal component prediction model;

[0289] a combining unit, connected to the first prediction unit and the second prediction unit respectively, for combining the predicted value of the trend component with the predicted value of the season component to form overall predicted time series data;

[0290] The processing unit is connected to the combining unit and is used to process the transformer oil chromatogram time series data whose detection result is an abnormal detection result.

[0291] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for detecting abnormality of transformer oil chromatographic time series data, characterized in that: The method comprises the following steps: Obtain time series data of transformer oil chromatogram; Decomposing the time series data to obtain residual components of the time series data; According to the residual component, it is determined whether the time series data is abnormal data, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

2. The method for detecting abnormality of transformer oil chromatogram time series data according to claim 1, characterized in that: Decomposing the time series data to obtain the residual components of the time series data specifically comprises: decomposing the time series data according to the STL method to obtain the residual components of the time series data; By decomposing the time series data, the seasonal component and trend component of the time series data can also be obtained. The specific steps are: decomposing the time series data according to the STL method to obtain the seasonal component of the time series data; And, decomposing the time series data according to the STL method to obtain the trend component of the time series data; The specific calculation formula for decomposing the time series data according to the STL method is as follows: Y v =T v +S v +R v ; in, Y v Represents time series data; T v represents the trend component; S v Indicates seasonal component; R v represents the residual component; v is the time index, v = 1,…,N; N is the total number of time series; The determining, based on the residual component, whether the time series data is abnormal data specifically includes: when the residual component exceeds a threshold range, determining that the time series data is abnormal data.

3. The abnormality detection method for transformer oil chromatogram time series data according to claim 2 is characterized in that: Decomposing the time series data according to the STL method to obtain the trend component of the time series data is to iteratively process the time series data in combination with an inner loop to obtain the trend component, which specifically includes the following steps: Step A1: Remove the trend component from time series data The method of removing the trend component in the time series data specifically includes the following steps: Step A11: Initialization assignment; The initialization assignment specifically includes: setting K is the number of iterations; Step A12: Calculate the Kth trend component according to the following formula (1): in, is the trend value in the Kth iteration; Step A13: Calculate the time series data with trend components removed according to the following formula (2): Step A2: Perform subsequence smoothing on the time series data with trend components removed; The step A2 specifically includes: Step A21: Set the time series period to n p , then we get n p periodic subsequences; Step A22: Perform the LOESS process on each periodic subsequence, and extend each periodic subsequence forward and backward by one period, and combine them to obtain a length of N+2×n p Time Series N is the total number of time series; Step A3: Perform 3 times with length n p , n p , 3 sliding averages, and one LOESS process is performed to obtain a time series of length N Step A4: Based on the time series of length N Find the seasonal component; The calculation formula of the seasonal component is as follows: Step A5: Remove the seasonal component from the time series data and then perform LOESS processing to obtain the trend component The trend component The calculation formula is as follows: Express Perform LOESS process.

4. The method for detecting abnormality of transformer oil chromatogram time series data according to claim 3, characterized in that: In the process of decomposing the time series data, the adjustment of the robust weight is used as the outer loop for processing the time series data. After the inner loop outputs the STL decomposition result of the current iteration, the robust weight required for the next iteration is obtained according to the STL decomposition result of the current iteration and the calculation formula of the robust weight. Among them, the robustness weight ρ v The calculation formula is as follows: r v =B*(R v | / h); (4) Where: R v Represents the remainder at time v, whose value is equal to Y v -T v -S v , that is, the residual component; The B function is a bisquare weight function, that is, a bisquare function, and its specific calculation process is as shown in formula (5): h is an intermediate variable, and its obtaining process is as shown in formula (6): h=6*median(R v ); (6) u is the independent variable of the bisquare function; median() represents the median function.

5. The method for detecting abnormality of transformer oil chromatogram time series data according to claim 2, characterized in that: When the residual component exceeds the threshold range, the time series data is determined to be abnormal data, which specifically includes the following steps: Calculate the mean and standard deviation of the residual components; Calculate the absolute value of the difference between the residual component value and the mean, and determine whether the absolute value is greater than twice the standard deviation: If the absolute value is greater than twice the standard deviation, the time series data corresponding to the residual component value is determined to be abnormal data; if the absolute value is not greater than twice the standard deviation, the time series data corresponding to the residual component value is determined to be not abnormal data.

6. The method for detecting abnormality of transformer oil chromatogram time series data according to any one of claims 1 to 5, characterized in that: Before detecting the transformer oil chromatogram time series data, the method further comprises the steps of: preprocessing the transformer oil chromatogram time series data; The preprocessing of the transformer oil chromatogram time series data specifically includes the following steps: Step B1: Check whether the transformer oil chromatogram time series data is missing: If the transformer oil chromatogram time series data does not have any missing, then the transformer oil chromatogram time series data is not preprocessed; If the transformer oil chromatogram time series data is missing, proceed to step B2; Step B2: Fill missing values ​​or remove data according to the missing rate and the proportion of consecutive missing values, which specifically includes the following steps: Step B21: Determine whether the missing rate meets the first condition or the second condition. The first condition is that the missing rate does not exceed the first threshold value. The second condition is that the missing rate is between the first threshold value and the second threshold value, and the continuous missing does not exceed the third threshold value of the total missing value. Then continue to determine whether the missing item is the first value in the transformer oil chromatogram time series data. Otherwise, proceed to step B22: If the missing item is the first value in the transformer oil chromatogram time series data, the next valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing; otherwise, the previous valid value of the missing item is used to fill the missing item, and the process ends to complete the preprocessing; Step B22: Eliminate the transformer oil chromatogram time series data; The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

7. A method for handling abnormalities in transformer oil chromatographic time series data, characterized in that: The method comprises the following steps: Using the abnormality detection method for transformer oil chromatogram time series data according to any one of claims 2 to 6, obtaining abnormality detection results of transformer oil chromatogram time series data; Extract trend and seasonal components; Taking the trend component as input, the trend component prediction model is used to obtain the predicted value of the trend component; and, taking the seasonal component as input, obtaining the predicted value of the seasonal component using the seasonal component prediction model; Combine the forecast value of the trend component with the forecast value of the seasonal component to form the overall forecast time series data; According to the predicted time series data, the transformer oil chromatogram time series data with abnormal detection results is processed.

8. A device for detecting abnormality of transformer oil chromatographic time series data, characterized in that: The device comprises: An acquisition unit, used for acquiring time series data of transformer oil chromatogram; a decomposition unit, connected to the acquisition unit, and configured to decompose the time series data to obtain a residual component of the time series data; The determination unit is connected to the decomposition unit and is used to determine whether the time series data is abnormal data according to the residual component, so as to complete the abnormality detection of the transformer oil chromatogram time series data.

9. The abnormality detection device for transformer oil chromatogram time series data according to claim 8, characterized in that: The device comprises: The decomposition unit includes a third decomposition unit; The third decomposition unit is connected to the acquisition unit, and is used to decompose the time series data according to the STL method to obtain the residual component of the time series data; The decomposition unit further comprises a first decomposition unit and a second decomposition unit; The first decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the seasonal component of the time series data; The second decomposition unit is connected to the acquisition unit and is used to decompose the time series data according to the STL method to obtain the trend component of the time series data; The third decomposition unit is connected to the acquisition unit, and is used to decompose the time series data according to the STL method to obtain the residual component of the time series data; The decomposition unit stores the following calculation formula for decomposition: Y v =T v +S v +R v ; in, Y v Represents time series data; T v represents the trend component; S v Indicates seasonal component; R v represents the residual component; v is the time index, v = 1,…,N; N is the total number of time series; The determination unit includes a first determination unit, which is connected to the decomposition unit and is used to determine that the time series data is abnormal data when the residual component exceeds a preset threshold range therein.

10. The abnormality detection device for transformer oil chromatogram time series data according to claim 9, characterized in that: The device comprises: The first determination unit comprises: A first calculation module, used for calculating the mean and standard deviation of the residual component; A second calculation module, connected to the first calculation module, for calculating the absolute value of the difference between the residual component value and the mean value; A determination module, connected to the second calculation module, is used to determine that the time series data corresponding to the residual component value is abnormal data when the absolute value is greater than twice the standard deviation; and is also used to determine that the time series data corresponding to the residual component value is not abnormal data when the absolute value is not greater than twice the standard deviation.

11. The abnormality detection device for transformer oil chromatogram time series data according to any one of claims 8 to 10, characterized in that: The device also includes a pre-processing unit, The preprocessing unit is connected to the acquisition unit and is used to preprocess the transformer oil chromatogram time series data; The pre-processing unit specifically comprises the following steps: A checking unit, used for checking whether the transformer oil chromatogram time series data is missing; A preprocessing unit connected to the inspection unit, and used for preprocessing the transformer oil chromatogram time series data when the inspection unit determines that the transformer oil chromatogram time series data is missing; The pre-processing unit comprises: A first determination module is used to determine whether the missing rate satisfies a first condition or a second condition, wherein the first condition is that the missing rate does not exceed a first threshold, and the second condition is that the missing rate is between the first threshold and the second threshold, and the consecutive missing does not exceed a third threshold of the total number of missings; A second determination module, connected to the first determination module, is used to determine whether the missing item is the first value in the transformer oil chromatogram time series data when the first determination module determines that the missing rate meets the first condition or the second condition; A filling module, connected to the second determination module, is used to fill the missing item with the next valid value of the missing item when the second determination module determines that the missing item is the first value in the transformer oil chromatogram time series data; and is also used to fill the missing item with the previous valid value of the missing item when the second determination module determines that the missing item is not the first value in the transformer oil chromatogram time series data; A rejection module, connected to the first determination module, for rejecting the transformer oil chromatogram time series data when the first determination module determines that the missing rate cannot meet the first condition or the second condition; The first threshold, the second threshold, and the third threshold are all preset values, and the first threshold is smaller than the second threshold.

12. A transformer oil chromatogram time series data abnormality processing system, characterized in that: The system comprises: The abnormality detection device for transformer oil chromatogram time series data according to any one of claims 8 to 11; the abnormality detection device is used to obtain the abnormality detection result of the transformer oil chromatogram time series data; an extraction unit, connected to the anomaly detection device, for extracting a trend component and a seasonal component from the anomaly detection result; a first prediction unit, connected to the extraction unit, configured to take the trend component as input and obtain a predicted value of the trend component using a trend component prediction model; a second prediction unit, connected to the extraction unit, for taking the seasonal component as input and obtaining a predicted value of the seasonal component using a seasonal component prediction model; a combining unit, connected to the first prediction unit and the second prediction unit respectively, for combining the predicted value of the trend component with the predicted value of the season component to form overall predicted time series data; The processing unit is connected to the combining unit and is used to process the transformer oil chromatogram time series data whose detection result is an abnormal detection result.

Images