An iceemdan-lrtc-based power load completion method and system

Abstract

The application discloses an ICCEEMDAN-LRTC-based power load completion method and system and belongs to the field of power equipment state monitoring. The improved adaptive noise complete ensemble empirical mode decomposition is performed on the power load time series data containing defects, and the non-stationary signal of the load is decomposed into different IMF sub-sequences; a limited number of IMF sub-sequences are reconstructed into a tensor array according to a time period, and a low-rank tensor completion algorithm is used to complete the missing values of the power load tensor containing defects. The method can fully consider the internal relationship of multidimensional data, so that the missing data can be more comprehensively completed. Finally, the completed IMF sub-sequences are reconstructed to obtain complete load time series data.

Description

TECHNICAL FIELD

[0001] The application relates to an ICEEMDAN-LRTC-based power load completion method and system and belongs to the field of power equipment state monitoring. BACKGROUND

[0002] Power load data is of great significance to the stable operation of a power system, and is related to key fields such as power plant power generation planning, substation load control and power grid dispatching control. With the development and application of smart power in China, a large number of smart meters are installed on the user side, which directly leads to an explosive growth of measurement data in the power system. Effective use of the measurement data is helpful to the safe and stable operation of the power system, but the access of a large number of smart devices also brings challenges to the power system. During the measurement data collection process, due to measurement device failure, communication device failure, poor physical environment and data transmission process failure, the monitoring data is missing.

[0003] The missing data will lead to incomplete transmission of information of the monitoring system, and will cause deviation in the data analysis process of the power system control terminal, so that accurate power system operation information cannot be obtained. Therefore, the completion of missing data of the power load is studied, the completion of the missing data is completed by performing characteristic division on the load data, so that accurate power load data is obtained, which is helpful to the safe and stable operation of the power system, and can also provide data support for the planning and dispatching of the power system. SUMMARY

[0004] The application provides an ICEEMDAN-LRTC-based power load completion method and system, which performs empirical mode decomposition on a power load time sequence, performs structural reorganization on a limited number of IMF subsequences obtained through decomposition to obtain a tensor array, uses an LRTC algorithm to complete missing values of the tensor containing the missing values, and reconstructs the completed IMF subsequences to obtain a complete tensor load time sequence.

[0005] The technical scheme of the application is as follows: an ICEEMDAN-LRTC-based power load completion method, S1, collecting power load time sequence data containing missing values;

[0006] S2, decomposing the power load time sequence data containing missing values by using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a limited number of power load intrinsic mode function (IMF) sequences and a residual term;

[0007] S3, performing normalization processing on the limited number of power load intrinsic mode function (IMF) sequences to obtain normalized IMF sequences;

[0008] S4, determine a time period, restructure the normalized IMF sequence, and obtain a tensor array containing missing power load data;

[0009] S5, use a low-rank tensor completion method to complete the data of the tensor array containing missing data, and obtain a completed tensor array;

[0010] S6, perform inverse normalization processing on the completed tensor array, and obtain the restored IMF sequence;

[0011] S7, reconstruct the IMF sequence and the residual term to obtain complete power load data.

[0012] The power load time series data containing missing data satisfies that the proportion of missing samples in the total sampling samples should be less than or equal to 40%.

[0013] The S2 comprises:

[0014] Let the power load time series data containing missing data be x(t), assign a "0" value to the missing position of x(t); add a Gaussian noise signal to the load time series, and calculate the first-order residual γ1 of the load time series after adding the noise;

[0015] Difference between x(t) and the first-order residual γ1 to obtain the IMF1 sequence;

[0016] Continue to add noise to the first-order residual γ1 to obtain the second-order residual, difference between the first-order residual and the second-order residual to obtain the IMF2 sequence; repeat the step until the power load time series reaches the maximum iteration number or the residual term cannot be further decomposed, and all IMF sequences and residual terms are obtained.

[0017] The S4 comprises: converting the IMF sequence into a matrix form according to a determined time period, and then constructing a three-dimensional tensor array according to the corresponding time sequence. , l1 represents a sampling period, l2 represents a time period, and l3 represents the sum of the number of IMF sequences and the number of residual terms.

[0018] The S5 comprises:

[0019] The rank of the low-rank tensor completion defined tensor is , and the low-rank tensor completion objective function is established as follows:

[0020]

[0021] Wherein, rank() is the rank of the tensor, is the tensor array containing missing data to be completed, is the collected tensor array, and Ω is the known load data index;

[0022] The above formula is converted by replacing the tensor rank minimum with the tensor core norm minimum as follows:

[0023]

[0024] wherein, is the tensor core norm, ; alpha i is a weight factor, X (i) is a matrix of the tensor unfolded along the ith mode;

[0025] The objective function is solved by using an alternating multiplier direction method to obtain the completed tensor.

[0026] According to another aspect of the present application, there is provided an ICEEMDAN-LRTC-based power load completion system, comprising:

[0027] A collection module is configured to collect incomplete power load time series data.

[0028] A first obtaining module is configured to decompose the incomplete power load time series data by using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a finite number of power load intrinsic mode function (IMF) sequences and a residual term.

[0029] A second obtaining module is configured to normalize the finite number of power load intrinsic mode function (IMF) sequences to obtain normalized IMF sequences.

[0030] A third obtaining module is configured to determine a time period, restructure the normalized IMF sequences, and obtain a power load data incomplete tensor array.

[0031] A fourth obtaining module is configured to complete the data of the incomplete tensor array by using a low-rank tensor completion method to obtain a completed tensor array.

[0032] A fifth obtaining module is configured to perform inverse normalization on the completed tensor array to obtain recovered IMF sequences.

[0033] A sixth obtaining module is configured to reconstruct the IMF sequences and the residual term to obtain complete power load data.

[0034] According to another aspect of the present application, there is provided a processor configured to run a program, wherein the program performs the ICEEMDAN-LRTC-based power load completion method according to any one of the above aspects when running.

[0035] The beneficial effects of the present application are: the improved adaptive noise complete ensemble empirical mode decomposition is performed on the power load time series data containing missing values, the non-stationary signal of the load is decomposed into different IMF sub-sequences; a limited number of IMF sub-sequences are reconstructed into a tensor array according to a time period, and a low-rank tensor completion algorithm is used to complete the missing values of the power load tensor containing missing values, the method can fully consider the internal relationship of multi-dimensional data, so that the missing data is more fully completed; finally, the completed IMF sub-sequences are reconstructed to obtain complete load time series data. BRIEF DESCRIPTION OF DRAWINGS

[0036] Fig. 1 is a flowchart of the present application;

[0037] Fig. 2 is each sub-sequence obtained by decomposing the original power load time series by the ICEEMDAN method;

[0038] Fig. 3 is a time sequence diagram of the true value and the completed value of part of the load data. DETAILED DESCRIPTION

[0039] The application will be further described below in conjunction with the drawings and examples, but the content of the application is not limited to the scope described.

[0040] Example 1: as shown, according to an aspect of an embodiment of the present application, a power load completion method based on ICEEMDAN-LRTC is provided, comprising: Figs. 1-3

[0041] S1, collecting power load time series data containing missing values;

[0042] S2, decomposing the power load time series data containing missing values by using an improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) method, obtaining a limited number of power load intrinsic mode function (IMF) sequences and a residual term;

[0043] S3, normalizing the limited number of power load intrinsic mode function (IMFI) sequences to obtain normalized IMF sequences;

[0044] S4, determining a time period, and restructuring the normalized IMF sequences to obtain a power load data missing tensor array;

[0045] ​S5. Use the Low Rank Tensor Completion (LRTC) method to complete the missing tensor array and obtain the completed tensor array.

[0046] S6. Perform inverse normalization on the completed tensor array to obtain the restored IMF sequence;

[0047] S7. Reconstruct the IMF sequence and residual terms to obtain complete power load data.

[0048] For S1, the missing data in general power load data includes: completely random missing, random missing, and non-random missing. The missing power load time series data should satisfy the condition that the proportion of missing samples in the total sample should be less than or equal to 40%. Based on this design, it is possible to select an appropriate time scale for load data collection according to different missing patterns, thereby ensuring the effectiveness of data recovery and completion.

[0049] In step S2, the power load data is decomposed according to the following process:

[0050] (2a) Let the power load time series be x(t), and assign a value of "0" to the load values ​​at the missing positions in x; E k (·) represents the k-th order mode component generated by empirical mode decomposition, and N(·) represents the local mean of the generated signal;

[0051] (2b) Add the first group of Gaussian white noise signal η to the original load time series. i Construct a noisy signal x i (t)=x(t)+β1E1(η i (t)), where i represents the number of times noise is added, and β1 represents the expected signal-to-noise ratio of the first decomposition iteration;

[0052] (2c) Calculate the first-order residual γ1, and subtract the first-order residual from the original power load time series to obtain the first-order IMF component;

[0053]

[0054]

[0055] in, The symbol indicates that the mean of the whole is taken;

[0056] (2d) Continue to add Gaussian white noise to the first-order IMF component, calculate the second-order residual, and obtain the second-order IMF2 subsequence;

[0057]

[0058]

[0059] (2e) According to the above calculation process, the k-th order mode is derived as:

[0060]

[0061]

[0062] Repeat the above steps until the power load time series reaches the maximum number of iterations or the residual term cannot be further decomposed;

[0063] The normalization processing formula of the power load IMF sub-sequence in step S3 is as follows:

[0064]

[0065] In step S4, the IMF sub-sequences are restructured. The IMF sub-sequences are converted into matrix form according to the determined time period, and then each IMF matrix is constructed into a three-dimensional tensor array according to the corresponding time sequence , l1 represents the sampling period, l2 represents the time period, l3 represents the sum of the number of IMF sequences and the number of residual terms, and the time period represents the quotient of the number of missing power load time series data points and the sampling period; In this embodiment, the time period is selected as 24h, i.e. 1 day. The selection of this value conforms to the variation law of the power load data, so that the constructed tensor is more consistent with the actual situation.

[0066] In step S5, the low-rank tensor completion defines the tensor The rank of the tensor is established, and the low-rank tensor completion objective function is:

[0067]

[0068] Wherein, rank() is the rank of the tensor, is the missing tensor array to be completed, is the tensor array collected, and Ω is the known load data index;

[0069] The tensor nuclear norm minimum is used to replace the tensor rank minimum, and the above formula is converted into

[0070]

[0071] Wherein, is the tensor nuclear norm, ; α i is the weight factor, X (i) is the matrix unfolded along the i-th mode of the tensor, and n is the dimension of the tensor; In this embodiment, the dimension of the tensor is 3.

[0072] The target function is solved by using an alternating direction method of multipliers (ADMM) to obtain a completed tensor.

[0073] The completed tensor array is subjected to inverse normalization processing, all IMF subsequences and residual terms are added to reconstruct a completed power load time sequence.

[0074] In the embodiment, the public data provided by the Global Energy Forecasting Competition 2012 (GEFCom2012) is used to simulate the load data, the data of Zone_1 from January 1, 2004 to December 31, 2004 is used as a completely random missing power load sampling data set, the missing rate is 40%, the sampling frequency is 1h, and the sampling period is 366 days, so the embodiment is a time sequence containing 8784 data.

[0075] Reference Fig. 2 The original power load time sequence is decomposed by the ICEEMDAN to obtain each IMF subsequence and a residual term.

[0076] Each IMF subsequence is subjected to normalization processing, and the normalized data is reconstructed into a tensor array The missing tensor is completed by using a low-rank tensor completion method, the complete tensor array is subjected to inverse normalization, and all IMF subsequences and residual terms are added to obtain a complete load time sequence, and Table 1 is a power load completion effect.

[0077] Table 1 Power load completion effect

[0078]

[0079] It can be seen from the results that the method has strong randomness and large fluctuation characteristics of the power load, the ICEEMDAN method is used to obtain time sequence components under each mode, internal information of the load time sequence is extracted, the nonlinearity and non-stationarity of the power load time sequence data have obvious inhibitory effect. More modal characteristics are provided for the LRTC method to complete the missing data, and the data completion precision is improved.

[0080] Fig. 3 For the completely random missing of the load time sequence in the embodiment with a missing rate of 40%, after the missing values are completed by using the load data completion method, the comparison between 168 sampling samples in the time sequence and the completed samples can directly show the effect of the load data completion method.

[0081] According to another aspect of the embodiments of the present application, there is provided an ICEEMDAN-LRTC-based power load completion system, comprising:

[0082] a collection module configured to collect power load time series data with missing values;

[0083] a first obtaining module configured to decompose the power load time series data with missing values by using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a finite number of intrinsic mode function (IMF) sequences of power load and a residual term;

[0084] a second obtaining module configured to normalize the finite number of intrinsic mode function (IMF) sequences of power load to obtain normalized IMF sequences;

[0085] a third obtaining module configured to determine a time period and restructure the normalized IMF sequences to obtain a power load data missing value tensor array;

[0086] a fourth obtaining module configured to complete data of the missing value tensor array by using a low-rank tensor completion method to obtain a completed tensor array;

[0087] a fifth obtaining module configured to perform inverse normalization on the completed tensor array to obtain recovered IMF sequences;

[0088] a sixth obtaining module configured to reconstruct the IMF sequences and the residual term to obtain complete power load data.

[0089] According to another aspect of the embodiments of the present application, there is provided a processor configured to run a program, wherein the program performs the ICEEMDAN-LRTC-based power load completion method according to any one of the above embodiments when running.

[0090] The specific embodiments of the present application are described in detail above with reference to the accompanying drawings, but the present application is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the spirit of the present application.

Claims

1. A power load completion method based on ICEEMDAN-LRTC, characterized in that, The method comprises the following steps: S1, collecting power load time series data with missing values; S2, decomposing the power load time series data with missing values by using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a plurality of power load intrinsic mode function (IMF) sequences and a residual term; S3, normalizing the plurality of power load intrinsic mode function (IMF) sequences to obtain normalized IMF sequences; S4, determining a time period, and restructuring the normalized IMF sequences to obtain a power load data missing value tensor array; S5, completing data of the tensor array with missing values by using a low-rank tensor completion method to obtain a completed tensor array; S6, performing reverse normalization on the completed tensor array to obtain recovered IMF sequences; S7, reconstructing the IMF sequences and the residual term to obtain complete power load data. The S2 comprises the following steps: The power load time series data with missing values is denoted as x(t), and the load values at the missing positions in x(t) are assigned a value of "0"; a Gaussian noise signal is added to the load time series, and a first-order residual γ1 is calculated for the load time series after the noise is added; The x(t) is subtracted from the first-order residual γ1 to obtain an IMF1 sequence; Noise is continuously added to the first-order residual γ1 to obtain a second-order residual, and the first-order residual and the second-order residual are subtracted to obtain an IMF2 sequence; the step is repeated until the power load time series reaches a maximum number of iterations or the residual term cannot be decomposed any more, and all IMF sequences and residual terms are obtained; The S5 comprises the following steps: Defining a tensor of rank , a low-rank tensor completion objective function is established:​ ； wherein rank() is the rank of the tensor, is the incomplete tensor array to be completed, is the acquired tensor array, Ω is the known load data index; l1 represents the sampling period, l2 represents the time period, and l3 represents the sum of the number of IMF sequences and the number of residual terms. The tensor nuclear norm minimum is used to replace the tensor rank minimum, and the above formula is converted into: ； wherein, is the tensor nuclear norm, ; a i is a weight factor, X (i) is the matrix of the tensor unfolded along the i-th mode; The completed tensor is obtained by solving the objective function by using an alternating multiplier direction method.

2. The ICEEMDAN-LRTC-based power load completion method according to claim 1, wherein, The power load time series data with missing values satisfies that the proportion of missing samples in the total sampling samples is less than or equal to 40%. 3.The ICEEMDAN-LRTC-based power load completion method according to claim 1, wherein, The S4 comprises: converting the IMF sequence into a matrix form according to a determined time period, and then forming a three-dimensional tensor array according to the corresponding time sequence of each IMF matrix .

4. An ICEEMDAN-LRTC-based power load completion system, characterized in that, The method comprises the following steps: A collection module is configured to collect power load time series data with missing values; A first obtaining module is configured to decompose the power load time series data with missing values by using an improved adaptive noise complete ensemble empirical mode decomposition method to obtain a plurality of power load intrinsic mode function (IMF) sequences and a residual term; A second obtaining module is configured to normalize the plurality of power load intrinsic mode function (IMF) sequences to obtain normalized IMF sequences; A third obtaining module is configured to determine a time period, and restructure the normalized IMF sequences to obtain a power load data missing value tensor array; A fourth obtaining module is configured to complete data of the tensor array with missing values by using a low-rank tensor completion method to obtain a completed tensor array; A fifth obtaining module is configured to perform reverse normalization on the completed tensor array to obtain recovered IMF sequences; A sixth obtaining module is configured to reconstruct the IMF sequences and the residual term to obtain complete power load data. The first obtaining module comprises the following steps: The power load time series data with missing values is denoted as x(t), and the load values at the missing positions in x(t) are assigned a value of "0"; a Gaussian noise signal is added to the load time series, and a first-order residual γ1 is calculated for the load time series after the noise is added; Subtracting x(t) from the first order residual γ1, the IMF1 sequence is obtained; Continue to add noise to the first order residual γ1 to obtain the second order residual, subtract the first order residual from the second order residual to obtain the IMF2 sequence; repeat the step until the power load time sequence reaches the maximum iteration number or the residual term cannot continue to decompose, and all IMF sequences and residual terms are obtained; The fourth obtaining module comprises: Defining a tensor of rank , a low-rank tensor completion objective function is established:​ ； wherein rank() is the rank of the tensor, is the incomplete tensor array to be completed, is the acquired tensor array, Ω is the known load data index; l1 represents the sampling period, l2 represents the time period, and l3 represents the sum of the number of IMF sequences and the number of residual terms. The above formula is converted into: ； wherein, is the tensor nuclear norm, ; a i is a weight factor, X (i) is the matrix of the tensor unfolded along the ith mode; The tensor core norm minimum is used to replace the tensor rank minimum, and the above formula is converted into:

5. A processor, comprising: The processor is configured to run a program, and the program performs the power load completion method based on the ICEEMDAN-LRTC when running. The processor is configured to run a program, and the program performs the power load completion method based on the ICEEMDAN-LRTC when running.

Images