Power grid data processing method based on dynamic mode decomposition

Abstract

The invention discloses a power grid data processing method based on dynamic mode decomposition and belongs to the technical field of power system data analysis. In order to solve the technical problem that the accuracy and the real-time performance of power grid data monitoring are poor due to the fact that data sources are diversified and data densities are different in an existing data monitoring system, so that the data are difficult to effectively compare and analyze in time sequence, the invention provides a new data processing method. By processing the power grid operation data from different data sources and different sampling frequencies, the time sequence registration of the data is realized, so that the accuracy and the real-time performance of power grid data monitoring are improved, the automation level and the intelligent degree of a power system are further improved, powerful support is provided for safe and stable operation of the power system, and the power grid data monitoring method is suitable for popularization and application. And the method plays an important role in guiding actual operation.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system data analysis, and particularly relates to a power grid data processing method based on dynamic mode decomposition. Background Art

[0002] With the rapid development of social economy and the deep integration of power systems, the scale of the power grid is constantly expanding. At the same time, the access of new energy sources such as wind power and photovoltaic power generation is also increasing day by day. The characteristics of intermittency and volatility on the new energy power generation side are obvious. Coupled with the influence of factors such as extreme weather, the difficulty of predicting the peak load of the power grid is constantly increasing. In this context, the real-time monitoring and analysis of power grid operation data become particularly important to ensure the stability and reliability of the power system. Existing data monitoring systems face many challenges, such as diverse data sources and inconsistent data densities. These problems make it difficult to effectively compare and analyze data in time series, resulting in poor accuracy and real-time performance of power grid data monitoring.

[0003] Therefore, there is an urgent need for a new technical solution in the existing technology to solve this problem. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: to provide a power grid data processing method based on dynamic mode decomposition to solve the technical problem that existing data monitoring systems are difficult to effectively compare and analyze data in time series due to diverse data sources and inconsistent data densities, resulting in poor accuracy and real-time performance of power grid data monitoring.

[0005] The power grid data processing method based on dynamic mode decomposition includes the following steps, and the following steps are carried out sequentially:

[0006] Step 1: Perform data preprocessing on power grid data

[0007] 1) Take data packets composed of data from different data sources during power grid operation. Among them, data from the same data source at the same collection point is stored in the same column, and data from each collection point of the same measurement unit is stored in the same row, and is used as the original data; the data sources include the Energy Management System (EMS), the Electric Energy Meter (Meter), and the Distribution Automation System (DA);

[0008] Perform detrending processing on the original data through the data detrending formula;

[0009] 2) Reconstructing data packets using the detrended data. In the data packet reconstruction, if there is no missing data after detrending, the data packet is directly formed. If there is missing data after detrending, the data packet is formed after the data is filled. The method used for filling the data is: using linear interpolation and historical average values, and using the optimized weighted average data filling method to perform weighted average to obtain the optimized weighted average interpolation result as the value of the pre-filled data;

[0010] 3) filtering the formed data packets through a bandpass filter;

[0011] Step 2: Use the filtered data packets to construct a data matrix and perform dynamic mode decomposition (DMD);

[0012] 1) Use the n measured values ​​of the m acquisition points in the data packet to establish two matrices X and X', respectively. Both matrices X and X' have n rows and m-1 columns, as shown below:

[0013]

[0014] In the formula, x m represents the measurement vector of the mth column;

[0015] 2) Calculate and obtain the modal matrix B;

[0016] 3) Perform singular value x m break down;

[0017] 4) construct a low-rank approximation of the modal matrix B;

[0018] 5) Using the DMD block enhancement formula of the augmented coordinate set established by considering the time delay coordinate, by multiplying the time delay coordinate s by the measurement vector x m And superimpose the constructed matrix to construct a (n×s)×(m-s+1) matrix H:

[0019] Apply dynamic mode decomposition (DMD) to the matrices Y and Y' derived from H:

[0020]

[0021] If the matrices Y and Y' derived from H are low-dimensional, then increase the number of delay coordinates s until they are full rank;

[0022] Step 3: Modality Identification and Ranking

[0023] Extract the elements of the modal matrix B in step 2) through dynamic modal decomposition (DMD), and calculate and obtain the energy of each mode:

[0024]

[0025] where: b ij is an element in the modal matrix B, representing the j-th coefficient of the i-th mode;

[0026] Arrange the energy values of each mode in descending order from large to small. The mode with the largest energy is the dominant mode, and obtain the modal index Rank(E) arranged in descending order of energy:

[0027] Rank(E) = argsort(E j , decending);

[0028] where: Rank(E) is the modal index arranged in descending order according to the energy E j ;

[0029] Step Four: Data Registration

[0030] 1) Perform data interpolation to convert data from different data sources into multi-dimensional time series data with a unified data granularity. The data interpolation formula is as follows:

[0031] x interp (t) = x(t - k·Δt);

[0032] where: Δt is the target data granularity; k is an integer, representing the k-th time point at the target data granularity Δt; x is the original data; x interp (t) is the interpolated data;

[0033] 2) Form time validity labels:

[0034]

[0035] where: label(t) represents the validity of the data at time t; mod 2 is a modulo 2 operation used to calculate the remainder of a number divided by 2, and thus mark whether the data point is an integer multiple of every other time interval;

[0036] Store the multi-dimensional time series data with the unified data granularity and the corresponding time validity labels in a new data structure;

[0037] Step Five: Bad Data Removal

[0038] Calculate the error of the multi-dimensional time series data after unifying the data granularity:

[0039]

[0040] where: x 基准 is the reference data, selected from the historical stable operation data; x 被评价 is the data to be evaluated; the full scale value is the maximum value of all the data to be evaluated;

[0041] Set the threshold and remove bad data:

[0042] Threshold = α + β · Time validity label;

[0043] Where: α and β are preset parameters;

[0044] If the data error exceeds the set threshold, it is regarded as bad data and removed;

[0045] For the vacant data positions after removal, use step 2) in step one to fill the data;

[0046] Step Six: Data synthesis

[0047] Synthesize the data of each data source after being processed in step five through the formula for synthesizing multi-dimensional data, and use the synthesized multi-dimensional data matrix as the data for subsequent power grid data monitoring.

[0048] The data detrending formula in step one 1) is:

[0049] x detrend = x - (a 0 + a 1 t + a 2 t 2 + a 3 t 3 );

[0050] Where: x is the original data, x detrend is the data after detrending, a 0 , a 1 , a 2 , a 3 are the trend parameters obtained by least squares fitting, and t is the time variable.

[0051] The formula for linear interpolation in step one 2) is:

[0052]

[0053] Where: α is the weight parameter, w interp (t) is the result of linear interpolation at time t, tprev is the time of the nearest known data point before the current time point t for which interpolation is required, t next is the time of the nearest known data point after the current time point t for which interpolation is required, w prev (t) is the known data at time t, w prev is the known data at time t prev and w next is the known data at time t next ;

[0054] The formula for the historical average is as follows:

[0055]

[0056] where: w hist is the historical average, N is the number of historical data points used to calculate the average, and w hist,i is the historical value of the i-th data point;

[0057] Perform weighted average as follows:

[0058] w owa = αw interp + (1 - α)w hist ;

[0059] where: w owa is the optimized weighted average interpolation result, α is the weight parameter used to balance the weights of the linear interpolation and historical average methods.

[0060] Optimize the weight parameter α using the global optimal weight parameter optimization method:

[0061]

[0062] where: α* is the optimal weight parameter, w owa,i is the interpolation value of the i-th data point, w true,i is the true value of the i-th data point, and N is the number of training data points used for optimization.

[0063] The data of each data source in step six includes data from the energy management system EMS, the electric energy meter Meter, and the distribution automation system DA. The synthesized multi-dimensional data matrix is:

[0064]

[0065] where: X multi is the synthesized multi-dimensional data matrix.

[0066] Through the above design scheme, the present invention can bring the following beneficial effects:

[0067] The present invention provides a new data processing method, which can process grid operation data from different data sources and different sampling frequencies, realize time series registration of data, improve the accuracy and real-time performance of grid data monitoring, and further enhance the automation level and intelligence degree of the power system, providing strong support for the safe and stable operation of the power system and playing an important role in guiding actual operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The following further describes the present invention in conjunction with the drawings and specific embodiments:

[0069] Figure 1 It is the original active power data diagram of the energy management system in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention;

[0070] Figure 2 It is the original active power data diagram of the electric energy meter in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention;

[0071] Figure 3 It is the original active power data diagram of the distribution automation system in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention;

[0072] Figure 4 It is the active power data diagram of the energy management system after data processing in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention;

[0073] Figure 5 It is the original active power data diagram of the electric energy meter after data processing in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention;

[0074] Figure 6 It is the original active power data of the distribution automation system after data processing in the embodiment of the power grid data processing method based on dynamic mode decomposition of the present invention. Specific implementation manners

[0075] The power grid data processing method based on dynamic mode decomposition disclosed by the present invention includes the following steps:

[0076] 1. Data preprocessing:

[0077] Perform detrending processing, data packet reconstruction and data filtering on the data from different data sources during the operation of the power grid to reduce the influence of noise and missing values. The data sources include the energy management system (EMS), electric energy meter (Meter) and distribution automation system (DA), etc. The data of each data source at the same collection point is stored in the same column, and the data of each collection point from the same measurement unit is stored in the same row and used as the original data;

[0078] 1) Perform detrending processing, and the formula is as follows:

[0079] x detrend = x - (a 0 + a 1 t + a 2 t 2 + a 3 t 3 ) (1);

[0080] In the formula: x is the original data, x detrend is the data after detrending, a 0 , a1 , a 2 , a 3 is the trend parameter obtained by least squares fitting, and t is the time variable.

[0081] 2) Data packet reconstruction, the formula is as follows:

[0082]

[0083] Where: is the missing data reconstructed by interpolation.

[0084] In data packet reconstruction, if there is no missing data after detrending, a data packet is directly formed. If there is missing data after detrending, a data packet is formed after filling in the data. The data filling adopts the optimized weighted average data filling method, which combines the advantages of linear interpolation and historical averaging, and balances the two methods through a weight parameter.

[0085] Among them, when the linear interpolation method is used to reconstruct the data packet, the linear interpolation inference estimates a nearest available value from the front and back, and the formula of the linear interpolation is as follows:

[0086]

[0087] Where: α is the weight parameter, w interp (t) is the linear interpolation result at time t, tprev is the time of the most recent known data point before the current time point t that needs to be interpolated, and t next is the time of the most recent known data point after the current time point t that needs to be interpolated, w prev (t) is the known data at time t, w prev is at time t prev The known data at w next is at time t next Known data at

[0088] The historical average formula is as follows:

[0089]

[0090] Where: w hist is the historical average, N is the number of historical data points used to calculate the average, and w hist,i is the historical value of the ith data point.

[0091] Take a weighted average:

[0092] w owa =αw interp +(1-α)whist (5);

[0093] where: w owa To optimize the weighted average interpolation result, α is the weight parameter, which is used to balance the weights of the linear interpolation and historical average methods.

[0094] The weight parameter α is optimized by the global optimal weight parameter optimization method:

[0095]

[0096] where: α* is the optimal weight parameter, w owa,i is the interpolation value of the i-th data point, w true,i is the true value of the i-th data point, and N is the number of training data points used for optimization.

[0097] 3) Data filtering:

[0098] x filtered = H(f)·x (7);

[0099] where: H(f) is the Hilbert band-pass filter, x is the original data, and x filtered is the filtered data.

[0100] 2. Dynamic mode decomposition: The exact DMD combined with the block enhancement formula is used to perform mode decomposition on the preprocessed data, providing a source for subsequent extraction of the dominant modes of the power grid dynamics.

[0101] 1) Construct a data matrix. Arrange n measurement values among m acquisition points into two matrices X and X', and perform exact dynamic mode decomposition. The measurement results are stored in matrix X, with data from the same acquisition point stored in the same column, and data from the same measurement unit stored in the same row. X' has the same structure, and the data time shift is t.

[0102]

[0103] where: x is the state vector representing the system at time t, and X and X' are data matrices.

[0104] 2) Calculate the mode matrix B

[0105] B = X'X + (9);

[0106] where: X + represents the Moore-Penrose pseudoinverse of X;

[0107] 3) Singular value decomposition

[0108] X ≈ UΣV T (10);

[0109] where: U and V are orthogonal matrices, and Σ is a diagonal matrix containing the singular values of X.

[0110] 4) Construct the low-rank approximation

[0111]

[0112] where: U r , Σ r and V r are the first r singular values and the corresponding singular vectors selected from U, Σ, and V. The (n×s)×(m-s+1) matrix H is constructed by using the DMD block enhancement formula for the augmented coordinate set established considering the time-delay coordinates, and constructing the matrix by multiplying the time-delay coordinate s by the measurement vector x and superimposing:

[0113] 5) The block enhancement formula involves applying the exact DMD to the matrices Y and Y' derived from H:

[0114]

[0115] If the matrices Y and Y' derived from H are of low dimension, more time-delay coordinates can be used to increase the rank of H. The number of time-delay coordinates s can be increased until the system reaches full rank.

[0116] 3. Mode identification and ranking: Extract the elements in the mode matrix B in step 2) through the dynamic mode decomposition DMD, calculate and obtain the energy of each mode: Rank the modes to identify the dominant modes.

[0117] Calculate the energy of each mode, and the formula is as follows:

[0118]

[0119] where: b ij is an element in the mode matrix B, representing b ij the j-th coefficient of the i-th mode.

[0120] Obtain the mode index Rank(E) arranged in descending order of energy:

[0121] Rank(E) = argsort(E j , decending) (15);

[0122] where: Rank(E) is the mode index arranged in descending order according to the energy E j The descending order is based on the energy magnitude of each mode. The modes with large energy are ranked in the front, and the modes with small energy are ranked in the back. The modes with large energy are the dominant modes.

[0123] 4. Data registration: Convert the data from different data sources into multi-dimensional time series data with a unified data granularity, and attach a time validity label to the data.

[0124] Data interpolation:

[0125] x interp (t) = x(t - k·Δt) (16);

[0126] In the formula: Δt is the target data granularity (e.g., 15 seconds); k is an integer representing the k-th time point under the target data granularity Δt; x is the original data; x interp (t) is the interpolated data.

[0127] Time validity label:

[0128]

[0129] In the formula: label(t) represents the validity of the data at time t, mod 2 is the modulo 2 operation used to calculate the remainder of a number divided by 2, and is used to mark whether a data point is an integer multiple of every other time interval (e.g., 15 seconds) to generate the time validity label.

[0130] The interpolated data and the time validity label are stored in a new data structure. The time validity label applies not only to the interpolated data but also to the original data, ensuring that the validity of the data can be tracked at each stage of data processing.

[0131] 5. Removal of bad data

[0132] Calculate the data error:

[0133]

[0134] In the formula: x 基准 is the reference data, x 被评价 is the data to be evaluated, and the full-scale value is the maximum value of all the data to be evaluated.

[0135] Set the threshold:

[0136] Threshold = α + β·time validity label (19);

[0137] In the formula: α and β are preset parameters.

[0138] If the data error exceeds the set threshold, it is regarded as bad data and removed.

[0139] For the vacant data positions after removal, use the method in step 1) 2) to fill and process the data.

[0140] 6. Data synthesis

[0141] Synthesized multi-dimensional data:

[0142]

[0143] Where: X multi is the synthesized multi-dimensional data matrix, containing data from the Energy Management System (EMS), Electric Energy Meter (Meter), and Distribution Automation System (DA).

[0144] Example:

[0145] The present invention will be further described below with reference to the accompanying drawings and specific examples.

[0146] The following data are from:

[0147] As Figure 1 shown, the active power data of the Energy Management System, 15-second sampling period: 21, 22, 22, 22.5, 22.9, 23.3, 23.7, 24.1, 24.5, 24.9, 25.3, 25.7, 26.1, 26.5, 26.9, 27.3, 27.7, 28.1, 28.5, 28.9, 29.3.

[0148] As Figure 2 shown, the active power data of the Electric Energy Meter, 1-minute sampling period: 23, 20, 25, 20, 24, 25.

[0149] As Figure 3 shown, the active power data of the Distribution Automation System, 1-minute sampling period: 15, 17, 16, 14, 15, 18.

[0150] According to the method of the present invention, the above data are processed to obtain the following results. It is required to obtain the results as shown in the appendix at a 15-second period. Since the sampling period of the Energy Management data is already 15 seconds, no interpolation processing is required, maintaining the integrity and accuracy of the original data. The Electric Energy Meter data and the Distribution System Automation data are successfully matched to a 15-second sampling period through the method of the present invention. The interpolated data not only has the same time scale as the Energy Management data, but also generates a smooth and naturally variable data curve by using the DMD method. This processing method effectively avoids the data monotonicity that may be caused by simple linear interpolation, making the data more truly reflect the actual dynamic changes and providing a higher-quality data basis for subsequent data analysis and applications. Figures 4 - 6 shown. Since the sampling period of the Energy Management data is already 15 seconds, no interpolation processing is required, maintaining the integrity and accuracy of the original data. The Electric Energy Meter data and the Distribution System Automation data are successfully matched to a 15-second sampling period through the method of the present invention. The interpolated data not only has the same time scale as the Energy Management data, but also generates a smooth and naturally variable data curve by using the DMD method. This processing method effectively avoids the data monotonicity that may be caused by simple linear interpolation, making the data more truly reflect the actual dynamic changes and providing a higher-quality data basis for subsequent data analysis and applications.

[0151] The DMD-based grid data registration and analysis method provided by the present invention can effectively process and analyze the grid operation data from different data sources, and has important practical application value.

Claims

1. A power grid data processing method based on dynamic mode decomposition is characterized by: The method comprises the following steps, and the following steps are performed in sequence: Step 1: Preprocess the power grid data 1) A data packet consisting of data from different data sources in the operation of the power grid, wherein the data from the same collection point of each data source is stored in the same column, and the data from each collection point of the same measurement unit is stored in the same row, and is used as the original data; the data sources include the energy management system EMS, the electric energy meter Meter and the distribution automation system DA; The original data is detrended by using the data detrending formula; 2) Reconstructing data packets using the detrended data. In the data packet reconstruction, if there is no missing data after detrending, the data packet is directly formed. If there is missing data after detrending, the data packet is formed after the data is filled. The method used for filling the data is: using linear interpolation and historical average values, and using the optimized weighted average data filling method to perform weighted average to obtain the optimized weighted average interpolation result as the value of the pre-filled data; 3) filtering the formed data packets through a bandpass filter; Step 2: Use the filtered data packets to construct a data matrix and perform dynamic mode decomposition (DMD); 1) Use the n measured values ​​of the m acquisition points in the data packet to establish two matrices X and X', respectively. Both matrices X and X' have n rows and m-1 columns, as shown below: In the formula, x m represents the measurement vector of the mth column; 2) Calculate and obtain the modal matrix B; 3) Perform singular value x m break down; 4) construct a low-rank approximation of the modal matrix B; 5) Using the DMD block enhancement formula of the augmented coordinate set established by considering the time delay coordinate, by multiplying the time delay coordinate s by the measurement vector x m And superimpose the constructed matrix to construct a (n×s)×(m-s+1) matrix H: Apply dynamic mode decomposition (DMD) to the matrices Y and Y' derived from H: If the matrices Y and Y' derived from H are low-dimensional, then increase the number of delay coordinates s until they are full rank; Step 3: Modality Identification and Ranking Extract the elements of the modal matrix B in step 2) through dynamic modal decomposition (DMD), and calculate and obtain the energy of each mode: Where: b ij is the element in the modal matrix B, representing the jth coefficient of the i-th mode; Arrange the energy values ​​of each mode in descending order from large to small, the one with large energy is the dominant mode, and obtain the modal index Rank (E) in descending order of energy: Rank(E)=argsort(E j ,decending); Where: Rank(E) is based on energy E j modal index in descending order; Step 4: Data Registration 1) Perform data interpolation to convert data from different data sources into multi-dimensional time series data with uniform data granularity. The data interpolation formula is as follows: x interp (t)=x(t-k·Δt); Where: Δt is the target data granularity; k is an integer, representing the kth time point under the target data granularity Δt; x is the original data; x interp (t) is the interpolated data; 2) Forming a time validity label: Where: label(t) indicates the validity of the data at time t; mod 2 is the modulo 2 operation, which is used to calculate the remainder of a number divided by 2, and then mark whether the data point is an integer multiple of every time interval; Store the multi-dimensional time series data with unified data granularity and the corresponding time validity labels in a new data structure; Step 5: Remove bad data Calculate the error of multi-dimensional time series data after unifying the data granularity: Where: x 基准 is the benchmark data, selected from the historical stable operation data; x 被评价 is the data being evaluated; the full value is the maximum value of all the evaluated data; Set thresholds and remove bad data: Threshold = α + β·time validity label; Where: α and β are preset parameters; If the data error exceeds the set threshold, it will be considered as bad data and removed; After removing the vacant data positions, fill the data using step 2) in step 1; Step 6: Data synthesis The data from each data source processed in step five are synthesized using a formula for synthesizing multidimensional data, and the synthesized multidimensional data matrix is ​​used as data for subsequent power grid data monitoring.

2. The power grid data processing method based on dynamic mode decomposition according to claim 1 is characterized in that: The data detrending formula in step 1) is: x detrend =x-(a0+a1t+a2t 2 +a3t 3 ); Where: x is the original data, x detrend is the data after detrending, a0, a1, a2, a3 are the trend parameters obtained by least squares fitting, and t is the time variable.

3. The power grid data processing method based on dynamic mode decomposition according to claim 1 is characterized in that: The linear interpolation formula in step 1) is: Where: α is the weight parameter, w interp (t) is the linear interpolation result at time t, tprev is the time of the most recent known data point before the current time point t that needs to be interpolated, and t next is the time of the most recent known data point after the current time point t that needs to be interpolated, w prev (t) is the known data at time t, w prev is at time t prev The known data at w next is at time t next Known data at The formula for the historical average is: Where: w hist is the historical average, N is the number of historical data points used to calculate the average, and w hist,i is the historical value of the i-th data point; The weighted average is performed as follows: w owa =αw interp +(1-a)w hist ; Where: w owa To optimize the weighted average interpolation results, α is a weight parameter used to balance the weights of the linear interpolation and historical averaging methods.

4. The power grid data processing method based on dynamic mode decomposition according to claim 3 is characterized in that: The weight parameter α is optimized using the global optimal weight parameter optimization method: Where: α* is the optimal weight parameter, w owa,i is the interpolation value of the ith data point, w true,i is the true value of the ith data point, and N is the number of training data points used for optimization.

5. The power grid data processing method based on dynamic mode decomposition according to claim 1 is characterized in that: The data sources in step 6 include data from the energy management system EMS, the electric energy meter Meter and the distribution automation system DA, and the synthesized multidimensional data matrix is: Where: X multi is a synthetic multidimensional data matrix.

Images