Method and device for cleaning transformer monitoring data based on multiple models

Abstract

The invention relates to the technical field of data processing, and discloses a transformer monitoring data cleaning method and device based on multiple models. The method comprises the steps of performing anomaly detection on a time sequence of transformer monitoring data based on a time sequence model, and determining a residual sequence; determining whether the monitoring data of each detection point belongs to abnormal data or not based on the residual error of each detection point in the residual error sequence and the absolute deviation mean value of the monitoring data; if it is determined that the monitoring data belongs to the abnormal data, dirty data to be cleaned in the abnormal data is determined; and performing data cleaning on the dirty data based on a bidirectional long-short-term memory network model. According to the method, the abnormal data can be more accurately determined, and dirty data in the abnormal data can be effectively cleaned.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing, and in particular to a method and device for cleaning transformer monitoring data based on multiple models. Background Art

[0002] As an important part of the power system, the state of the transformer will greatly affect the safety and stability of the power system. With the vigorous development of information technology and sensor technology, the field of transformer monitoring has ushered in an explosive growth in data volume. This trend provides a data basis for the prediction and status evaluation of transformer state parameters. However, due to the existence of external interference and short-term failure of communication ports during data transmission, there are some abnormal data in the collected data, which are mainly missing values ​​and noise values. In order to improve the accuracy of subsequent transformer state parameter prediction and status evaluation, it is very necessary to clean the collected transformer monitoring data. Summary of the invention

[0003] In view of the above situation, an embodiment of the present application provides a method and device for cleaning transformer monitoring data based on multiple models, aiming to solve the above problem or at least partially solve the above problem.

[0004] In a first aspect, an embodiment of the present application provides a method for cleaning transformer monitoring data based on multiple models, the method comprising: performing anomaly detection on the time series of transformer monitoring data based on a time series model to determine a residual sequence; determining whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the absolute deviation mean of the monitoring data; if it is determined that the monitoring data is abnormal data, determining the dirty data to be cleaned in the abnormal data; and performing data cleaning on the dirty data based on a bidirectional long short-term memory network model.

[0005] In some embodiments, the determining whether the monitoring data of each detection point is abnormal data based on the absolute deviation mean of the residual and the monitoring data of each detection point in the residual sequence includes: determining the additive outlier AO test statistic and the innovative outlier IO test statistic of each detection point based on the absolute deviation mean of the residual and the monitoring data of each detection point; determining the maximum absolute value of the AO test statistic and the IO test statistic of each detection point; if the maximum absolute value is greater than a preset threshold, determining that the monitoring data of the detection point is abnormal data.

[0006] In some embodiments, determining the dirty data to be cleaned in the abnormal data includes: determining the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point; and taking the abnormal data whose abnormality type is sensor abnormality as the dirty data to be cleaned.

[0007] In some embodiments, the method of determining the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point includes: determining the abnormal point of each detection point based on the AO test statistic and the IO test statistic of each detection point; if abnormal points appear at consecutive moments, determining that the abnormality type of the abnormal data is a sensor abnormality; if abnormal points do not appear at consecutive moments, determining that the abnormality type of the abnormal data is a device abnormality.

[0008] In some embodiments, the outlier point of each detection point is determined based on the AO test statistic and the IO test statistic of each detection point, including: if the AO test statistic is greater than the IO test statistic, determining that the detection point is an AO outlier point; if the AO test statistic is less than the IO test statistic, determining that the detection point is an IO outlier point.

[0009] In some embodiments, the AO test statistic is determined based on the following formula:

[0010]

[0011] The IO test statistic is determined based on the following formula:

[0012]

[0013] in, represents the AO test statistic, n represents the total number of samples in the time series, F represents the forward shift operator, represents the IO test statistic, a T represents the residual of the detection point T, Represents the mean absolute deviation.

[0014] In some embodiments, the mean absolute deviation is determined based on the total number of samples of the time series, the value of the time series, and the median of the time series.

[0015] In a second aspect, an embodiment of the present application also provides a cleaning device for transformer monitoring data based on multiple models, the device comprising: a detection module, used to perform anomaly detection on the time series of transformer monitoring data based on a time series model, and determine a residual sequence; a processing module, used to determine whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the absolute deviation mean of the monitoring data; if it is determined that the monitoring data is abnormal data, determine the dirty data to be cleaned in the abnormal data; a cleaning module, used to perform data cleaning on the dirty data based on a bidirectional long short-term memory network model.

[0016] In some embodiments, the processing module is specifically used to determine the additive outlier AO test statistic and the innovative outlier IO test statistic of each detection point based on the residual of each detection point and the absolute deviation mean of the monitoring data; determine the maximum absolute value of the AO test statistic and the IO test statistic of each detection point; if the maximum absolute value is greater than a preset threshold, determine that the monitoring data of the detection point is abnormal data.

[0017] In some embodiments, the processing module is specifically used to determine the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point; and use the abnormal data with the abnormality type of sensor abnormality as the dirty data to be cleaned.

[0018] In some embodiments, the processing module is specifically used to determine the abnormal point of each detection point based on the AO test statistic and the IO test statistic of each detection point; if the abnormal point appears at consecutive moments, the abnormal type of the abnormal data is determined to be a sensor abnormality; if the abnormal point does not appear at consecutive moments, the abnormal type of the abnormal data is determined to be a device abnormality.

[0019] In some embodiments, the processing module is specifically used to determine that the detection point is an AO abnormal point if the AO test statistic is greater than the IO test statistic; if the AO test statistic is less than the IO test statistic, determine that the detection point is an IO abnormal point.

[0020] In some embodiments, the AO test statistic is determined based on the following formula:

[0021]

[0022] The IO test statistic is determined based on the following formula:

[0023]

[0024] in, represents the AO test statistic, n represents the total number of samples in the time series, F represents the forward shift operator, represents the IO test statistic, a T represents the residual of the detection point T, Represents the mean absolute deviation.

[0025] In some embodiments, the mean absolute deviation is determined based on the total number of samples of the time series, the value of the time series, and the median of the time series.

[0026] In a third aspect, an embodiment of the present application further provides an electronic device, comprising: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to perform the steps of the first aspect described above.

[0027] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes the steps of the first aspect above.

[0028] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0029] The time series model is used to detect anomalies in the time series of transformer monitoring data, determine the residual sequence, and determine whether the monitoring data is abnormal data through the absolute deviation mean of the residual sequence and the time series. The absolute deviation mean can better reflect the overall situation of the residual sequence, so that the abnormal data can be determined more accurately. The dirty data in the abnormal data is further cleaned through the bidirectional long short-term memory network model. The bidirectional long short-term memory network model can learn historical and future information. When performing data cleaning, it can improve the effect of data cleaning and perform effective cleaning. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0031] Figure 1 The LSTM unit structure diagram provided in the embodiment of the present application is shown;

[0032] Figure 2 The structure diagram of the BiLSTM unit provided in the embodiment of the present application is shown;

[0033] Figure 3 A schematic flow chart of a method for cleaning transformer monitoring data based on multiple models provided in an embodiment of the present application is shown;

[0034] Figure 4 A schematic flow chart of a method for cleaning transformer monitoring data based on multiple models provided in another embodiment of the present application is shown;

[0035] Figure 5 A data graph showing the data of dissolved methane in oil when the contamination rate is 4% provided in an embodiment of the present application;

[0036] Figure 6A data graph showing the data of dissolved methane in oil when the contamination rate is 6% provided in an embodiment of the present application;

[0037] Figure 7 A data graph showing the data of dissolved methane in oil when the contamination rate is 8% provided in an embodiment of the present application;

[0038] Figure 8 A data graph showing the data of dissolved methane in oil when the contamination rate is 10% provided in an embodiment of the present application;

[0039] Fig. 9 A comparison chart of data cleaning performance of different models provided in the embodiments of the present application is shown;

[0040] Fig.10 A structural diagram of a cleaning device based on multi-model transformer monitoring data provided by an embodiment of the present application is shown;

[0041] Fig.11 A schematic structural diagram of an electronic device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0043] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that such use is interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the term "including" and its variants are to be interpreted as open-ended terms meaning "including but not limited to".

[0044] As described in the background technology, as an important component of the power system, the state of the transformer will greatly affect the safety and stability of the power system. With the vigorous development of information technology and sensor technology, the field of transformer monitoring has ushered in an explosive growth in data volume. This trend provides a data basis for the state parameter prediction and state evaluation of the transformer. However, due to the existence of factors such as external interference and short-term failure of the communication port during data transmission, there are some abnormal data in the collected data, which are mainly missing values ​​and noise values. In order to improve the accuracy of subsequent transformer state parameter prediction and state evaluation, it is very necessary to clean the collected transformer monitoring data.

[0045] Based on this, an embodiment of the present application provides a method for cleaning transformer monitoring data based on multiple models, which performs anomaly detection on the time series of transformer monitoring data through a time series model, determines the residual sequence, and determines whether the monitoring data is abnormal data through the absolute deviation mean of the residual sequence and the time series. The absolute deviation mean can better reflect the overall situation of the residual sequence, so that the abnormal data can be determined more accurately, and further the dirty data in the abnormal data is cleaned through a bidirectional long short-term memory network model. The bidirectional long short-term memory network model can learn historical and future information. When performing data cleaning, it can improve the effect of data cleaning and perform effective cleaning.

[0046] Before introducing the specific scheme of this application, the model involved in this application is first introduced.

[0047] 1. Autoregressive Moving Average Model

[0048] In some embodiments, the time series model in the present application adopts an autoregressive integrated moving average model (ARIMA).

[0049] ARIMA is a time series prediction model that regards the data sequence formed by the prediction object over time as a random sequence. It assumes that the label value fluctuates around the general trend of time, where the trend is affected by historical labels and the fluctuation is affected by accidental events within a period of time. The general trend itself is not necessarily stable. A certain mathematical model is used to approximate this sequence. Once this sequence is identified, the future value can be predicted through the past and present values ​​of the time series.

[0050] The AR part in the ARIMA model is used to process the autoregressive part in the time series. It takes into account the impact of observations in several past periods on the current value. The I part is used to make the non-stationary time series stable. Through first-order or second-order differential processing, the trend and seasonal factors in the time series are eliminated. The MA part is used to process the moving average part of the time series. It takes into account the impact of past prediction errors on the current value.

[0051] Assume that the time series {z1,z2,z3,…,z t} obeys the ARIMA (p, d, q) model, it can be expressed as:

[0052]

[0053] in, is a time series that reaches the stationary requirement after d times of difference; is the current value of the time series; μ is a constant term; φ1, φ2, φ3, …, φ p are the coefficients of the autoregressive term; {ε1,ε2,ε3,…,ε t} is a white noise sequence; θ1,θ2,θ3,…,θ q is the coefficient of the moving average term.

[0054] If the time series {z t If} is a stationary time series, there is no need to perform differential processing on it, that is, d = 0, then the original ARIMA (p, d, q) model is transformed into an ARMA (p, q) model, and the time series can be expressed as:

[0055] z t =μ+φ1z t-1 +φ2z t-2 +...+φ p z t-p +ε t +θ1ε t-1 +...+θ q ε t-q

[0056] Comparing the representation of time series by ARIMA model and ARMA model, it can be seen that the ARMA model can be regarded as a generalized ARIMA model.

[0057] On this basis, if q = 0, the original ARMA (p, q) model is transformed into an AR (p) model, and the time series can be expressed as:

[0058] z t =μ+φ1z t-1 +φ2z t-2 +…+φ p z t-p+ε t

[0059] If p = 0, the original ARMA (p, q) model is transformed into the MA (q) model, and the time series can be expressed as:

[0060] z t =μ+ε t +θ1ε t-1 +…+θ q ε t-q

[0061] 2. Outlier Test Statistics

[0062] In some embodiments, the abnormal data in the transformer monitoring data are mainly additive outliers (AO) and innovative outliers (IO) in the time series.

[0063] Assume {W t} is a zero-mean time series with no outliers, {U t} is the observed time series containing outliers. Using the ARIMA (p, d, q) model and according to different outlier types, it can be divided into the following two types of outlier models:

[0064] 1) AO outlier model

[0065]

[0066]

[0067] θ(B)=1-θ1B-θ2B 2 -…-θ q B q

[0068] φ(B)=1-φ1B-φ2B 2 -…-φ p B p

[0069] Where: B is a delay operator; θ(B) and φ(B) are stationary and reversible operators without common factors; {ε t} are independent and have the same distribution White noise sequence; is the impulse function; ω is the influence factor of the outlier; is the difference operator.

[0070] AO only affects the observed values ​​at the time T when the anomaly occurs, and the missing values ​​in the abnormal data are classified as AO-type outliers.

[0071] 2) IO outlier model

[0072]

[0073] IO will not only interfere with the time series at the time T when the anomaly occurs, but will also affect the observed values ​​after time T, but the impact will gradually become smaller.

[0074] In order to identify outliers in time series, it is necessary to calculate the fitting residuals of the time series ARIMA model and perform AO and IO outlier test statistics based on the residual sequence to determine whether it is an outlier.

[0075] The traditional AO and IO outlier test statistics are:

[0076]

[0077] π * (F) = 1-π1F-π2F 2 -…-π n-T F n-T

[0078]

[0079]

[0080] Where: AO and λ IO is the traditional AO and IO outlier test statistic; n is the total number of samples in the time series; F is the forward shift operator, and Fa t =a t+1 ; a T is the model fitting residual at time T.

[0081] 3. Long Short Term Memory (LSTM)

[0082] LSTM is a special recurrent neural network (RNN) that can learn long-term dependencies. In order to overcome the gradient vanishing and gradient exploding problems of traditional RNN when processing long sequences, LSTM introduces a gating mechanism, including input gate, forget gate, and output gate. Figure 1 The LSTM unit structure diagram shown in Figure 1. The output gate determines how much of the input information of the current time step can be stored in the memory cell of the LSTM unit, and the forget gate determines how much of the previous information in the memory cell of the LSTM unit will be forgotten. The output gate determines the output of the current LSTM unit, that is, how much of the hidden state of the current time step will be passed to the next time step and as the final output of the LSTM.

[0083] The internal calculation process of each LSTM unit is:

[0084] 1) Calculate the output i of the input gate i

[0085] i i =σ(U i x t +V it-1 +b i )

[0086] Where: σ is the sigmoid function; U i With V i The input gate is the current input x t and the unit state h at the previous moment t-1 Weight matrix; b i is the bias term.

[0087] 2) Calculate the output f of the forget gate t

[0088] f t =σ(U f x t +V ft-1 +b f )

[0089] Where: U f With V f The forget gate is the current input x t and the unit state h at the previous moment t-1 Weight matrix; b f is the bias term.

[0090] 3) Calculate the temporary state of the unit at the current moment

[0091]

[0092] Where: tanh is the hyperbolic tangent function; U c With V c The current unit and the current input x t and the unit state h at the previous moment t-1 Weight matrix; b c is the bias term.

[0093] 4) Calculate the current cell state c t

[0094]

[0095] Where: Represents element-wise matrix multiplication.

[0096] 5) Calculate the output o of the output gate t

[0097] o t =σ(U o x t +V ot-1 +b o )

[0098] Where: U o With V o The output gate is the current input x t and the unit state h at the previous moment t-1 Weight matrix; b c is the bias term.

[0099] 6) Calculation unit output t

[0100] t=o t *tan(c t )

[0101] BiLSTM is an improved network of LSTM, which can capture the forward and backward dependencies in the input sequence, such as Figure 2 As shown in the BiLSTM network structure diagram, a single-layer BiLSTM is composed of two LSTMs, one for forward processing of the input sequence; the other for backward processing of the input sequence. After the processing is completed, the outputs of the two LSTMs are superimposed to obtain the final output.

[0102] In addition to the original LSTM calculation, the forward and backward processes must also be calculated during BiLSTM calculation:

[0103]

[0104] Where: and The outputs of the forward and backward layers at time t respectively; L is the LSTM model; H t is the final output of the BiLSTM network; b y is the bias of the output layer.

[0105] In the embodiment of the present application, the mining of the forward propagation and backward propagation rules of the input sequence enables BiLSTM to learn the historical and future information contained therein, which can improve the data cleaning effect when cleaning abnormal data in transformer monitoring data.

[0106] The present application is described in detail below through specific embodiments.

[0107] Figure 3 The schematic diagram of the process of the cleaning method of transformer monitoring data based on multiple models provided in the embodiment of the present application is shown. Figure 3It can be seen that the present application at least includes steps S101 to S104:

[0108] Step S101: performing anomaly detection on the time series of transformer monitoring data based on a time series model to determine a residual sequence.

[0109] In some embodiments, the time series model is an ARIMA model.

[0110] Furthermore, an ARIMA model is established to fit the transformer monitoring data, and the model parameters (p, d, q) of the ARIMA model are obtained. The fitted ARIMA model is further compared with the transformer monitoring data to calculate the residual sequence.

[0111] Step S102: Determine whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the mean of the absolute deviation of the time series.

[0112] Step S103: If it is determined that the monitoring data is abnormal data, dirty data to be cleaned in the abnormal data is determined.

[0113] Step S104: Clean the dirty data based on the bidirectional long short-term memory network model.

[0114] In some embodiments, historical monitoring data of the transformer is collected in advance, and the historical monitoring data of the transformer is normalized to provide the data repair capability of the model. The normalization formula can be expressed as:

[0115]

[0116] Where: is the normalized data; X is the original data; X min and X max They represent the minimum and maximum values ​​in the original data respectively.

[0117] Furthermore, the hyperparameters of the BiLSTM model are set, the BiLSTM model is trained using the normalized data, and the trained model is used to clean the dirty data determined in step S103.

[0118] from Figure 3It can be seen from the method shown that the present application performs anomaly detection on the time series of transformer monitoring data through a time series model, determines the residual sequence, and determines whether the monitoring data is abnormal data through the absolute deviation mean of the residual sequence and the time series. The absolute deviation mean can better reflect the overall situation of the residual sequence, so that the abnormal data can be determined more accurately. The dirty data in the abnormal data is further cleaned through a bidirectional long short-term memory network model. The bidirectional long short-term memory network model can learn historical and future information. When performing data cleaning, it can improve the effect of data cleaning and perform effective cleaning.

[0119] In some embodiments of the present application, whether the monitoring data of each detection point is abnormal data is determined based on the following method: the AO test statistic and the IO test statistic of each detection point are determined based on the residual of each detection point and the absolute deviation mean of the monitoring data; the maximum absolute value of the AO test statistic and the IO test statistic of each detection point is determined; if the maximum absolute value is greater than a preset threshold, the monitoring data of the detection point is determined to be abnormal data.

[0120] In some embodiments, the maximum absolute value of the AO test statistic and the IO test statistic of each detection point is determined based on the following formula: in, represents the AO test statistic for the t test point, Represents the IO test statistic for the t test point.

[0121] In some embodiments, if t >C, the detection point t is considered as a data abnormal point, otherwise it is considered as a normal point. Wherein, C is a predetermined threshold. Exemplarily, 3<C<4.

[0122] In some embodiments of the present application, dirty data to be cleaned in abnormal data is determined based on the following method, including: determining the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point; and taking abnormal data whose abnormality type is sensor abnormality as the dirty data to be cleaned.

[0123] In some embodiments, the abnormality types of the detection points include equipment abnormality and sensor abnormality.

[0124] Since transformer monitoring data is of various types and data transmission is subject to external interference, the final data will have problems such as noise values ​​and missing values. The abnormal data generated by sensor abnormalities is called dirty data, which needs to be identified in anomaly detection to distinguish it from the abnormal data actually generated by the abnormality of the transformer equipment itself.

[0125] In an embodiment of the present application, data on abnormal equipment status is cleaned through a bidirectional long short-term memory network model, and abnormal sensor data is analyzed by staff for abnormal diagnosis of the sensor.

[0126] In some embodiments, based on the AO test statistic and the IO test statistic of each detection point, the abnormal point of each detection point is determined; if the abnormal point appears at consecutive moments, the abnormal type of the abnormal data is determined to be the transformer state abnormality; if the abnormal point does not appear at consecutive moments, the abnormal type of the abnormal data is determined to be the equipment state abnormality.

[0127] In one implementation, if the AO test statistic is greater than the IO test statistic, the detection point is determined to be an AO outlier point; if the AO test statistic is less than the IO test statistic, the detection point is determined to be an IO outlier point.

[0128] In some embodiments of the present application, the traditional calculation of AO test statistics and IO test statistics will be affected by the residual standard deviation. When the number of outliers in the time series rises to a certain number, the residual standard deviation will not be able to reflect the overall situation of the residual series well, resulting in the occurrence of the "masking" phenomenon, which in turn makes the anomaly detection invalid. Based on this, the embodiment of the present application improves the calculation method of the test statistic and uses the absolute deviation mean for calculation.

[0129] In some embodiments, the mean absolute deviation is determined based on the total number of samples of the time series, the value of the time series, and the median of the time series.

[0130] In one implementation, the mean absolute deviation is determined based on the following formula:

[0131]

[0132] Where N represents the total number of samples in the time series X; x j Represents the value of the jth time series; MED is the median of the time series.

[0133] In some embodiments, the AO test statistic is determined based on the following formula:

[0134]

[0135] The IO test statistic is determined based on the following formula:

[0136]

[0137] in, represents the AO test statistic, n represents the total number of samples in the time series, F represents the forward shift operator, represents the IO test statistic, a2 represents the residual of the detection point T, Represents the mean absolute deviation.

[0138] In the embodiment of the present application, the test statistic is calculated by the absolute deviation mean proposed in the present application, and then based on the test statistic, accurate identification of abnormal points can be achieved when a large number of abnormal points appear in the data.

[0139] In summary, if Figure 4 As shown, the multi-model based transformer monitoring data cleaning method proposed in this application, in the abnormality detection stage, first establishes an ARIMA model, detects the transformer monitoring data through the ARIMA model, calculates the model residual, calculates the AO and IO abnormal value test statistics based on the model residual, and determines the maximum value λ of the absolute value of the AO test statistic and the IO test statistic t , if λ t If it is greater than the threshold C, the data at the detection point is determined to be abnormal data, and the dirty data in the abnormal data is determined. In the data repair stage, the historical transformer monitoring data is first collected and normalized, and then the hyperparameters of the BiLSTM model are set, and the trained BiLSTM model is used to clean and repair the dirty data.

[0140] In order to verify the effectiveness and superiority of the cleaning method proposed in this application, taking the methane data in the dissolved gas in oil as an example, AO and IO anomaly points are added to the methane data that originally does not contain "dirty data" to generate the methane data to be cleaned. When the methane data is subjected to different degrees of contamination rates, the traditional anomaly detection method is compared with the improved anomaly detection method proposed in this article. In addition, the outlier test statistic formed by replacing the original standard deviation with the shrinking standard deviation is also compared at the same time.

[0141] For this purpose, the following four different pollution situations are constructed, with pollution rates of 4%, 6%, 8%, and 10% respectively. The polluted methane data are as follows: Figure 5-8 ,in, Figure 5 The data of dissolved methane in oil when the contamination rate is 4% is shown below. Figure 6 The data of dissolved methane in oil when the contamination rate is 6% is shown below. Figure 7 The data of dissolved methane in oil when the contamination rate is 8% is shown below. Figure 8 The data are for dissolved methane in oil when the contamination rate is 10%.

[0142] On this basis, the traditional anomaly detection method, the improved anomaly detection method with shrunken standard deviation (referred to as anomaly detection method 2 in this paper), and the improved anomaly detection method proposed in this paper are used to detect anomalies in the dissolved methane in oil data with different contamination rates. The detection results of the three anomaly methods are shown in Tables 1 and 2.

[0143] Table 1 Comparison of the total number of abnormal points detected by three anomaly detection methods

[0144]

[0145] Table 2 Comparison of the number of AO and IO anomaly points detected by three anomaly detection methods

[0146]

[0147] According to the detection results of methane data under different pollution rates by the three anomaly detection methods, as the pollution rate increases, the anomaly detection performance of the traditional anomaly detection method and anomaly detection method 2 both decreases, among which the performance of the traditional anomaly detection method decreases more significantly. The anomaly detection method mentioned in this application can accurately identify anomalies in the data under different pollution rates, proving the effectiveness of the method.

[0148] Taking the data of dissolved methane in oil with a contamination rate of 4% as an example, this application compares the data cleaning method based on the BiLSTM model with the data cleaning method based on the LSTM model and the data cleaning method based on the GRU model. Fig. 9 The comparison chart of data cleaning performance of different models is shown in Table 3.

[0149] By comparing the data cleaning methods of the three models in the table, it can be seen that the overall error of the data cleaning method based on the BiLSTM model proposed in this application is 0.75%, which is 0.54% lower than the overall error of the data cleaning method based on LSTM, and 0.7% lower than the overall error of the data cleaning method based on GRU, which proves the effectiveness and superiority of the solution of this application.

[0150] Table 3 Data repair results of methane data with a contamination rate of 4%

[0151]

[0152] In some embodiments of the present application, a cleaning device for transformer monitoring data based on multiple models is provided, and the cleaning device for transformer monitoring data based on multiple models corresponds to the cleaning method for transformer monitoring data based on multiple models in the above embodiments. Fig.10 As shown, the cleaning device for transformer monitoring data based on multiple models includes: a detection module 201, a processing module 202 and a cleaning module 203. The detailed description of each functional module is as follows:

[0153] A detection module 201 is used to perform anomaly detection on the time series of transformer monitoring data based on a time series model to determine a residual sequence;

[0154] Processing module 202, used to determine whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the mean of the absolute deviation of the monitoring data; if it is determined that the monitoring data is abnormal data, determine the dirty data to be cleaned in the abnormal data;

[0155] The cleaning module 203 is used to clean the dirty data based on a bidirectional long short-term memory network model.

[0156] In some embodiments of the present application, in the above-mentioned device, the processing module 202 is specifically used to determine the additive outlier AO test statistic and the innovative outlier IO test statistic of each detection point based on the residual of each detection point and the absolute deviation mean of the monitoring data; determine the maximum absolute value of the AO test statistic and the IO test statistic of each detection point; if the maximum absolute value is greater than a preset threshold, determine that the monitoring data of the detection point is abnormal data.

[0157] In some embodiments of the present application, in the above-mentioned device, the processing module 202 is specifically used to determine the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point; and use the abnormal data with the abnormality type of device status abnormality as the dirty data to be cleaned.

[0158] In some embodiments of the present application, the processing module 202 is specifically used to determine the abnormal point of each detection point based on the AO test statistic and the IO test statistic of each detection point; if the abnormal point appears at consecutive moments, the abnormal type of the abnormal data is determined to be the transformer state abnormality; if the abnormal point does not appear at consecutive moments, the abnormal type of the abnormal data is determined to be the equipment state abnormality.

[0159] In some embodiments of the present application, the processing module 202 is specifically used to determine that the detection point is an AO outlier point if the AO test statistic is greater than the IO test statistic; if the AO test statistic is less than the IO test statistic, determine that the detection point is an IO outlier point.

[0160] In some embodiments of the present application, in the above device, the AO test statistic is determined based on the following formula:

[0161]

[0162] The IO test statistic is determined based on the following formula:

[0163]

[0164] in, represents the AO test statistic, n represents the total number of samples in the time series, F represents the forward shift operator, represents the IO test statistic, a T represents the residual of the detection point T, Represents the mean absolute deviation.

[0165] In some embodiments of the present application, the absolute deviation mean is determined based on the total number of samples of the time series, the value of the time series, and the median of the time series.

[0166] It should be noted that any of the above-mentioned cleaning devices for transformer monitoring data based on multiple models can implement the above-mentioned cleaning method for transformer monitoring data based on multiple models in a one-to-one correspondence, which will not be repeated here.

[0167] Fig.11 FIG. 1 is a schematic diagram showing the structure of an electronic device provided in an embodiment of the present application. Fig.11 As shown, at the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and a memory. Among them, the memory may include a memory, such as a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage, etc. Of course, the electronic device may also include hardware required for other services.

[0168] The processor, network interface and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.11 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0169] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory may include internal memory and non-volatile memory, and provides instructions and data to the processor.

[0170] The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a cleaning device for transformer monitoring data based on multiple models at the logical level. The processor executes the program stored in the memory and is specifically used to execute the above method.

[0171] The processor may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or a combination of hardware and software modules in the decoding processor to be executed. The software module may be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0172] The electronic device can execute the cleaning method of transformer monitoring data based on multiple models provided in multiple embodiments of the present application, and realize a cleaning device for transformer monitoring data based on multiple models. Fig.10 The functions of the illustrated embodiment will not be described in detail in the embodiments of the present application.

[0173] An embodiment of the present application also proposes a computer-readable storage medium, which stores one or more programs, and the one or more programs include instructions. When the instructions are executed by an electronic device including multiple application programs, the electronic device can execute the multi-model based transformer monitoring data cleaning method provided by multiple embodiments of the present application.

[0174] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0175] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0176] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0177] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0178] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0179] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0180] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0181] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity or device including the element.

[0182] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0183] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for cleaning transformer monitoring data based on multiple models, characterized in that: The method comprises: Based on the time series model, anomaly detection is performed on the time series of transformer monitoring data to determine the residual sequence; Determine whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the mean of the absolute deviation of the time series; If it is determined that the monitoring data is abnormal data, determining dirty data to be cleaned in the abnormal data; The dirty data is cleaned based on a bidirectional long short-term memory network model.

2. The method according to claim 1, characterized in that The determining whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the absolute deviation mean of the time series includes: Determine the additive outlier AO test statistic and the innovative outlier IO test statistic of each detection point based on the residual of each detection point and the absolute deviation mean of the monitoring data; Determine the maximum value of the absolute value of the AO test statistic and the IO test statistic of each detection point; If the maximum value of the absolute value is greater than a preset threshold, it is determined that the monitoring data of the detection point is abnormal data.

3. The method according to claim 2, characterized in that The determining of dirty data to be cleaned in the abnormal data includes: Based on the AO test statistic and the IO test statistic of each detection point, determine the abnormal type of each detection point; Abnormal data whose abnormality type is sensor abnormality is used as the dirty data to be cleaned.

4. The method according to claim 3, characterized in that The determining of the abnormality type of each detection point based on the AO test statistic and the IO test statistic of each detection point includes: Based on the AO test statistic and the IO test statistic of each test point, determine the outlier point of each test point; If abnormal points appear at consecutive moments, it is determined that the abnormal type of the abnormal data is a sensor abnormality; If no abnormal point appears at consecutive moments, it is determined that the abnormal type of the abnormal data is a device abnormality.

5. The method according to claim 4, characterized in that The determining of the abnormal point of each detection point based on the AO test statistic and the IO test statistic of each detection point includes: If the AO test statistic is greater than the IO test statistic, the detection point is determined to be an AO outlier point; If the AO test statistic is smaller than the IO test statistic, the detection point is determined to be an IO abnormal point.

6. The method according to any one of claims 2 to 5, characterized in that The AO test statistic is determined based on the following formula: The IO test statistic is determined based on the following formula: in, represents the AO test statistic, n represents the total number of samples in the time series, F represents the forward shift operator, represents the IO test statistic, a T represents the residual of the detection point T, Represents the mean absolute deviation.

7. The method according to claim 1, characterized in that The absolute deviation mean is determined based on the total number of samples of the time series, the value of the time series, and the median of the time series.

8. A cleaning device for transformer monitoring data based on multiple models, characterized in that: The device comprises: A detection module is used to perform anomaly detection on the time series of transformer monitoring data based on a time series model and determine a residual sequence; A processing module, used to determine whether the monitoring data of each detection point is abnormal data based on the residual of each detection point in the residual sequence and the mean of the absolute deviation of the monitoring data; if it is determined that the monitoring data is abnormal data, determine the dirty data to be cleaned in the abnormal data; The cleaning module is used to clean the dirty data based on a bidirectional long short-term memory network model.

9. An electronic device, comprising: processor; as well as A memory arranged to store computer executable instructions, wherein when the executable instructions are executed, the processor executes the steps of the method for cleaning transformer monitoring data based on multiple models as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including a plurality of application programs, enables the electronic device to perform the steps of the multi-model-based transformer monitoring data cleaning method as described in any one of claims 1 to 7.

Images