Electric power system load component identification method based on sparse dictionary learning

Abstract

The invention discloses an electric power system load component identification method based on sparse dictionary learning. Comprising the steps of obtaining standardized historical load data and standardized historical meteorological data of various loads, and establishing a comprehensive historical load data dictionary and a comprehensive historical meteorological data dictionary; acquiring to-be-identified daily load data and to-be-identified daily meteorological data of various loads, and performing load component identification on the to-be-identified daily load data by using an orthogonal matching pursuit algorithm combined with meteorological factors according to the comprehensive historical load data dictionary and the comprehensive historical meteorological data dictionary to obtain a load component identification result, the load component identification result is used for load prediction or scheduling of the power system. According to the method, the load curve of the day to be identified can be decomposed into the load curve of each component, the proportion of each component is calculated, and the method has important value and significance for identifying the generalized load component proportion.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and particularly relates to a method for identifying load components in a power system based on sparse dictionary learning. Background Art

[0002] With the continuous penetration of new energy devices such as distributed photovoltaics, electric vehicles, air conditioners, and small hydropower in the power grid, the accurate extraction and proportion identification of typical load components in the power system have become more complex, posing a severe challenge to the research on the observability of the generalized load at the end. Therefore, it is necessary to study the problem of load component identification.

[0003] The types of generalized load components are numerous and complex. Coupled with inevitable measurement errors and the randomness of the load itself, the decomposition of its component proportions has diversity, making it impossible to accurately characterize various typical loads and difficult to directly select typical load curves for proportion identification. Therefore, it is necessary to explore a method that can accurately characterize the proportion of load components. Summary of the Invention

[0004] To solve the problems in the background art, the present invention provides a method for identifying load components in a power system based on sparse dictionary learning. The present invention first establishes a historical data dictionary according to historical load data and historical meteorological data; then, according to information such as the types of load components, load data, and meteorological data on the day to be identified, models the component decomposition problem as a sparse dictionary model; finally, uses an orthogonal matching pursuit algorithm combined with meteorological factors to decompose the load curve on the day to be identified. The method of the present invention can decompose the load curve on the day to be identified into component load curves and calculate the proportion of each component, which has certain value and significance for the identification of the proportion of generalized load components.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The method of the present invention includes the following steps:

[0007] S1) Collect historical load data and historical meteorological data of multiple types of loads, and obtain standardized historical load data and standardized historical meteorological data of each type of load through preprocessing.

[0008] The multiple types of loads are selected from photovoltaics, wind power, air conditioning loads, small hydropower, charging piles, and basic loads.

[0009] S2) Establish a comprehensive historical load data dictionary and a comprehensive historical meteorological data dictionary according to the standardized historical load data and standardized historical meteorological data of each type of load.

[0010] The step S2 includes the following steps:

[0011] S2.1) Establish historical load matrices for various types of loads based on the standardized historical load data of each type of load, and combine the historical load matrices of all loads to obtain a comprehensive historical load data dictionary; in the historical load matrix of each type of load, the column vector is the load data collected at all sampling points on the same sampling day, and the row vector is the load data collected at the same sampling point on all sampling days, and the sampling point is the sampling time point.

[0012] S2.2) For each type of load, obtain a historical meteorological matrix through correlation analysis of the standardized historical load data and the standardized historical meteorological data; combine the historical meteorological matrices of all loads to obtain a comprehensive historical meteorological data dictionary.

[0013] The specific steps of S2.2 are as follows:

[0014] S2.2.1) For each type of load, unify the resolution of the standardized historical load data and the standardized historical meteorological data;

[0015] S2.2.2) After unifying the resolution, for each type of load, perform correlation analysis on the load data and various meteorological data of each sampling day respectively to obtain the Spearman correlation coefficient between the load data and each type of meteorological data;

[0016] S2.2.3) For each type of load, according to the following formula, based on the various meteorological data of each sampling day and the Spearman correlation coefficient between the load data and each type of meteorological data, obtain the weighted historical meteorological data at the sampling day and each sampling point:

[0017]

[0018] In the formula, is the weighted historical meteorological data of the nth type of load at the dth day and the ith sampling point; is the kth type of meteorological data of the nth type of load at the dth day and the ith sampling point, k = 1 to K n , K n is the total number of meteorological types of the nth type of load, f n is the weighted calculation function of the nth type of load, is the average correlation coefficient between the load data of the nth type of load and the kth type of meteorological data, ρ n,d,k is the correlation coefficient between the load data of the nth type of load at the dth day and the kth type of meteorological data;

[0019] S2.2.4) For each type of load, combine the weighted historical meteorological data of all sampling points on each sampling day to obtain a daily unified historical meteorological vector, use the daily unified historical meteorological vector as the column vector, and arrange the column vectors in the order of the collection date to construct the historical meteorological matrix of the load.

[0020] The historical meteorological matrix of each type of load constructed in step S2.2.4 is specifically as follows:

[0021]

[0022] In the formula, A wn is the historical meteorological matrix of the nth type of load, is the weighted historical meteorological data of the nth type of load at the i-th sampling point on the d-th day, where d is the serial number of the sampling day, d = 1 to D n , D n is the total number of sampling days of the nth type of load, i is the serial number of the sampling point, i = 1 to M t , M t is the total number of sampling points per sampling day after unified resolution.

[0023] S3) Obtain the load data of the day to be identified and the meteorological data of the day to be identified for each type of load. According to the comprehensive historical load data dictionary and the comprehensive historical meteorological data dictionary obtained in step S2, use the orthogonal matching pursuit algorithm combined with meteorological factors to identify the load components of the load data of the day to be identified, and obtain the load component identification result. The load component identification result is used for load forecasting or dispatching of the power system.

[0024] The specific steps of step S3 are as follows:

[0025] S3.1) Take the daily unified historical load vector in the comprehensive historical load data dictionary, that is, the column vector, as an element, and take the minimum number of load data dictionary elements and accurately representing the total load as the solution target, and establish a load component identification model according to the comprehensive historical load data dictionary;

[0026] In step S3.1, the load component identification model is specifically:

[0027] X = arg min||E|| 2 ,

[0028] s.t. Y = AX + E

[0029] ||X|| 0 ≤ L

[0030] In the formula, Y is the load vector of the day to be identified, X is the component proportion vector, A is the comprehensive historical load data dictionary, E is the error vector, ||X|| 0 is the number of non-zero elements in the component proportion vector X, L is the preset upper limit value of the selection of the comprehensive historical load data dictionary elements, ||E|| 2is the 2-norm of the error vector, that is, the square root of the sum of the squares of the errors at all sampling points. s.t. is the constraint condition, and arg min is the value of the variable when this expression reaches the minimum. X = arg min||E|| 2 is the value of the component proportion vector X when the 2-norm of the error vector E reaches the minimum;

[0031] Among them, the component proportion vector X is specifically:

[0032]

[0033] In the formula, X n is the element proportion matrix of the nth type of load in the comprehensive historical load data dictionary, n = 1 to N, where N is the total number of load types, and D n is the total number of sampling days of the nth type of load;

[0034] Among them, the error vector E is specifically:

[0035] E = [ε 1 ... ε i ... ε M T ∈R M×1

[0036] In the formula, ε i is the error at the i-th sampling point, where i = 1 to M, and M is the total number of sampling points on the day to be identified, which is the same as the total number of sampling points on each sampling day.

[0037] S3.2) Obtain the load data on the day to be identified and the meteorological data of each type of load on the day to be identified, and construct them into a load vector on the day to be identified and a meteorological vector on the day to be identified respectively; use the orthogonal matching pursuit algorithm combined with meteorological factors, and according to the load component identification model and the meteorological vector on the day to be identified, perform load component identification on the load vector on the day to be identified to obtain the load component identification result.

[0038] The specific steps of S3.2 are as follows:

[0039] S3.2.1) Collect the load data of M sampling points on the day to be identified and construct them into a load vector on the day to be identified. Collect the meteorological data of M sampling points at the locations of each type of load on the day to be identified to obtain the meteorological data of each type of load on the day to be identified, and construct the meteorological data of each type of load on the day to be identified into a meteorological vector on the day to be identified through the weighted calculation function of each type of load; w In step S3.2.1, according to the following formula, the meteorological data of each type of load on the day to be identified is constructed into a meteorological vector on the day to be identified through the weighted calculation function f

[0040] of each type of load: n Construct the meteorological data of each type of load on the day to be identified into a meteorological vector on the day to be identified:​

[0041]

[0042] Wherein, is the weighted meteorological data of the nth type of load at the ith sampling point, is the original meteorological data of the ith sampling point of the kth type of meteorological data on the day to be identified, Y wn is the meteorological vector of the day to be identified for the nth type of load.

[0043] S3.2.2) Take the inner product of the comprehensive historical meteorological data dictionary and the meteorological vector of the day to be identified to obtain the meteorological inner product matrix, which is used as the meteorological inner product matrix; meanwhile, initialize the index list;

[0044] In the step S3.2.2, take the inner product of the load inner product matrix and the meteorological inner product matrix according to the following formula to obtain the comprehensive inner product matrix:

[0045]

[0046] Wherein, G j is the comprehensive inner product matrix obtained in the jth iteration, is the load inner product matrix obtained in the jth iteration, G w is the meteorological inner product matrix, and ⊙ represents the Hadamard product of matrices.

[0047] S3.2.3) Take the inner product of the comprehensive historical load data dictionary and the load vector of the day to be identified to obtain the load inner product matrix, take the inner product of the load inner product matrix and the meteorological inner product matrix obtained in step S3.2.2 to obtain the comprehensive inner product matrix, and select the index corresponding to the maximum value among the other elements except the elements corresponding to the indexes in the index list from the comprehensive inner product matrix, and add it to the index list;

[0048] S3.2.4) Select the elements corresponding to the indexes from the comprehensive historical load data dictionary according to the index list to construct an index dictionary matrix, use the index dictionary matrix as the comprehensive historical load data dictionary to input into the load component identification model, and obtain the component proportion vector and the residual vector through the least squares method for processing the load component identification model after inputting the index dictionary matrix;

[0049] S3.2.5) Use the residual vector as the load vector of the day to be identified, and obtain the new component proportion vector and the residual vector according to steps S3.2.3 and S3.2.4;

[0050] S3.2.6) Repeat step S3.2.5 until the iteration reaches the preset number of times or the residual converges, to obtain the final index list and the component proportion vector. Stack the same type of elements in the index list according to the component proportion vector to obtain the load component identification result.

[0051] The beneficial effects of the present invention are as follows:

[0052] 1. The present invention takes into account the historical load and meteorological big data, incorporates the load and meteorological data of the day to be identified into the identification system, and improves the model decomposition accuracy;

[0053] 2. The present invention takes into account different types of historical meteorological data, uses correlation analysis and weighted processing of meteorological data, and comprehensively examines the influence of different types of historical meteorological data;

[0054] 3. The present invention uses the sparse dictionary learning method for load component identification, can perform dimensionality reduction representation on large datasets, has good noise reduction ability, and has strong adaptability. Description of the Drawings

[0055] Figure 1 is the flow chart of the load component identification method in the present invention;

[0056] Figure 2 is the load curve graph of the day to be identified before and after applying noise perturbation;

[0057] Figure 3 is the 96-point load curve graph of the load to be identified and its four components;

[0058] Figure 4 is the identification result graph of the load component identification method in the present invention. Detailed Embodiments

[0059] The following further details the present invention in conjunction with the drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, those skilled in the art's various equivalent modifications of the present invention all fall within the scope defined by the appended claims of this application.

[0060] The present invention provides a load component identification method based on sparse dictionary learning, as Figure 1 shown, which includes the following steps:

[0061] The method of the present invention includes the following steps:

[0062] S1) Collect historical load data and historical meteorological data of various types of loads, and obtain the standardized historical load data and standardized historical meteorological data of various types of loads through preprocessing;

[0063] Multiple types of loads are selected from photovoltaic power, wind power, air-conditioning loads, small hydropower, charging piles, and basic loads. In specific implementation, to ensure the identification accuracy, the number and composition of load types in the historical load data and the load data to be identified on the day to be identified can be made consistent.

[0064] The historical meteorological data of each type of load includes multiple meteorological types. In the historical meteorological data of different loads, the number and composition of meteorological types are not necessarily the same. The meteorological types include the weather, temperature, temperature at 2 m above the ground, relative humidity at 2 m above the ground, pressure, wind speed at 10 m above the ground, wind direction at 10 m above the ground, wind speed at 100 m above the ground, wind direction at 100 m above the ground, and radiation at the location of the load.

[0065] S2) Establish a comprehensive historical load data dictionary and a comprehensive historical meteorological data dictionary according to the standardized historical load data and standardized historical meteorological data of each type of load;

[0066] Step S2 includes the following steps:

[0067] S2.1) Establish a historical load matrix for each type of load according to the standardized historical load data of each type of load, and combine the historical load matrices of all loads to obtain a comprehensive historical load data dictionary; in the historical load matrix of each type of load, the column vector is the load data collected at all sampling points on the same sampling day, and the row vector is the load data collected at the same sampling point on all sampling days. The sampling point is the sampling time point;

[0068] Specifically, in step S2.1, the historical load matrix of each type of load is:

[0069]

[0070] In the formula, is the historical load matrix of the nth type of load; n is the serial number of the load type, n = 1 to N, and N is the total number of load types; is the load data of the nth type of load at the dth day and the ith sampling point, d is the serial number of the sampling day, d = 1 to D n , D n is the total number of sampling days of the nth type of load, i is the serial number of the sampling point, i = 1 to M, and M is the total number of sampling points on each sampling day.

[0071] S2.2) For each type of load, obtain a historical meteorological matrix through correlation analysis of the standardized historical load data and standardized historical meteorological data; combine the historical meteorological matrices of all loads to obtain a comprehensive historical meteorological data dictionary.

[0072] Step S2.2 is specifically:

[0073] S2.2.1) For each type of load, unify the resolution of the standardized historical load data and the standardized historical meteorological data; specifically, unifying the resolution means: unifying the number of sampling points and the time interval between two adjacent sampling points; after unifying the resolution, the historical load data and the historical meteorological data have the same number of sampling points and the same sampling point time interval;

[0074] S2.2.2) For each type of load, by performing correlation analysis on the load data and various meteorological data of each sampling day respectively, obtain the Spearman correlation coefficient between the load data and each type of meteorological data;

[0075] S2.2.3) For each type of load, according to the following formula, based on the various meteorological data of each sampling day and the Spearman correlation coefficient between the load data and each type of meteorological data, obtain the weighted historical meteorological data at the sampling day and each sampling point:

[0076]

[0077] In the formula, is the weighted historical meteorological data of the nth type of load at the dth day and the ith sampling point; is the kth type of meteorological data of the nth type of load at the dth day and the ith sampling point, k = 1~K n , K n is the total number of meteorological types of the nth type of load, f n is the weighted calculation function of the nth type of load, is the average correlation coefficient between the load data of the nth type of load and the kth type of meteorological data, ρ n,d,k is the correlation coefficient between the load data of the nth type of load at the dth day and the kth type of meteorological data;

[0078] S2.2.4) For each type of load, combine the weighted historical meteorological data of all sampling points of each sampling day to obtain a daily unified historical meteorological vector, and use the daily unified historical meteorological vector as a column vector, and arrange each column vector in the order of the acquisition date to construct the historical meteorological matrix of the load.

[0079] The historical meteorological matrix of each type of load constructed in step S2.2.4 is specifically:

[0080]

[0081] In the formula, A wn is the historical meteorological matrix of the nth type of load, is the weighted historical meteorological data of the nth type of load at the dth day and the ith sampling point, d is the serial number of the sampling day, d = 1~D n , D nis the total number of sampling days for the nth type of load, i is the serial number of the sampling point, and i = 1 to M t , M t is the total number of sampling points for each sampling day after uniform resolution.

[0082] S3) Obtain the load data of the day to be identified and the meteorological data of the day to be identified for various types of loads. According to the comprehensive historical load data dictionary and the comprehensive historical meteorological data dictionary obtained in step S2, use the orthogonal matching pursuit algorithm combined with meteorological factors to identify the load components of the load data of the day to be identified, and obtain the load component identification result. The load component identification result is used for load forecasting or dispatching of the power system.

[0083] Step S3 is specifically as follows:

[0084] S3.1) Use the daily unified historical load vector in the comprehensive historical load data dictionary, that is, the column vector as an element, and take the minimum number of load data dictionary elements and accurately representing the total load as the solution goal, and establish a load component identification model according to the comprehensive historical load data dictionary;

[0085] In step S3.1, the load component identification model is specifically:

[0086] X = arg min||E|| 2 ,

[0087] s.t. Y = AX + E

[0088] ||X|| 0 ≤ L

[0089] In the formula, Y is the load vector of the day to be identified, X is the component proportion vector, A is the comprehensive historical load data dictionary, E is the error vector, ||X|| 0 is the number of non-zero elements in the component proportion vector X, L is the preset upper limit value of the selection of elements in the comprehensive historical load data dictionary, ||E|| 2 is the 2-norm of the error vector, that is, the square root of the sum of the squares of the errors at all sampling points, s.t. is the constraint condition, arg min is the value of the variable when this formula reaches the minimum value, and X = arg min||E|| 2 is the value of the component proportion vector X when the 2-norm of the error vector E reaches the minimum value;

[0090] Among them, the component proportion vector X is specifically:

[0091]

[0092] In the formula, X n is the element proportion matrix of the nth type of load in the comprehensive historical load data dictionary, n = 1 to N, N is the total number of load types, Dn is the total number of sampling days for the nth type of load;

[0093] Among them, the error vector E is specifically:

[0094] E = [ε 1 ... ε i ... ε M T ∈R M×1

[0095] In the formula, ε i is the error at the i-th sampling point, where i = 1 to M, and M is the total number of sampling points on the day to be identified, which is the same as the total number of sampling points for each sampling day.

[0096] Among them, the load vector Y on the day to be identified is a vector composed of the load data collected at M sampling points on the day to be identified, and is expressed as:

[0097] Y = [y 1 ... y i ... y M T ∈R M×1

[0098] In the formula, y i is the load data collected at the i-th sampling point on the day to be identified.

[0099] S3.2) Obtain the load data on the day to be identified and the meteorological data on the day to be identified for each type of load, and construct them into a load vector and a meteorological vector on the day to be identified respectively; use the orthogonal matching pursuit algorithm combined with meteorological factors, and according to the load component identification model and the meteorological vector on the day to be identified, perform load component identification on the load vector on the day to be identified to obtain the load component identification result.

[0100] Step S3.2 is specifically as follows:

[0101] S3.2.1) Collect the load data at M sampling points on the day to be identified and construct it into a load vector on the day to be identified. Respectively collect the meteorological data at M sampling points on the day to be identified at the locations where each type of load is located to obtain the meteorological data on the day to be identified for each type of load, and construct the meteorological data on the day to be identified for the load into a meteorological vector on the day to be identified through the weighted calculation function for each type of load; w In step S3.2.1, according to the following formula, the meteorological data on the day to be identified for the load is constructed into a meteorological vector on the day to be identified through the weighted calculation function f

[0102] for each type of load: n Construct the meteorological data on the day to be identified for the load into a meteorological vector on the day to be identified:

[0103]

[0104] ​​Wherein, is the weighted meteorological data of the nth type of load at the ith sampling point, is the original meteorological data of the ith sampling point of the kth meteorological data type on the day to be identified, and Y wn is the meteorological vector of the day to be identified for the nth type of load.

[0105] S3.2.2) Take the inner product of the comprehensive historical meteorological data dictionary and the meteorological vector of the day to be identified to obtain the meteorological inner product matrix as the meteorological inner product matrix; meanwhile, initialize the index list;

[0106] Specifically: for the meteorological data of each type of load, use the weighted calculation function of the load to perform weighted summation processing on all types of meteorological data at each sampling point to obtain the unified meteorological data of the sampling point, and combine the unified meteorological data of the M w sampling points on the day to be identified for the load to form the meteorological vector of the day to be identified for the load;

[0107] In step S3.2.2, take the inner product of the load inner product matrix and the meteorological inner product matrix according to the following formula to obtain the comprehensive inner product matrix:

[0108]

[0109] Wherein, G j is the comprehensive inner product matrix obtained in the jth iteration, is the load inner product matrix obtained in the jth iteration, and G w is the meteorological inner product matrix, and ⊙ represents the Hadamard product of the matrices.

[0110] S3.2.3) Take the inner product of the comprehensive historical load data dictionary and the load vector of the day to be identified to obtain the load inner product matrix, take the inner product of the load inner product matrix and the meteorological inner product matrix obtained in step S3.2.2 to obtain the comprehensive inner product matrix, and select, from the comprehensive inner product matrix, the index corresponding to the maximum value among the other elements except the elements corresponding to the indexes in the index list, and add it to the index list;

[0111] S3.2.4) Select the elements corresponding to the indexes in the comprehensive historical load data dictionary according to the index list to construct the index dictionary matrix, use the index dictionary matrix as the comprehensive historical load data dictionary to input into the load component identification model, and obtain the component proportion vector and the residual vector through the least squares method for the load component identification model after inputting the index dictionary matrix;

[0112] S3.2.5) Use the residual vector as the load vector of the day to be identified to obtain the new component proportion vector and the residual vector according to steps S3.2.3 and S3.2.4;

[0113] S3.2.6) Repeat step S3.2.5 until the iteration reaches the preset number of times or the residual converges, obtaining the final index list and the component proportion vector. Stack the same type of elements in the index list according to the component proportion vector to obtain the load component identification result.

[0114] The specific embodiments of the present invention are as follows:

[0115] In this embodiment, the 10kV load data from May to August 2023 in a certain province of China and the matching local meteorological data are used for example analysis. Among them, the load data includes the 10kV medium-voltage outgoing line data of 4 different load types, and the load types include: air conditioner, charging pile, photovoltaic, and basic load. The data sampling frequency is once every 15 minutes, and a total of 96 points are sampled in a day (M = 96). The meteorological data includes the 10kV medium-voltage outgoing line data of 10 different meteorological types, including: weather, temperature, temperature at 2m above the ground, relative humidity at 2m above the ground, pressure, wind speed at 10m above the ground, wind direction at 10m above the ground, wind speed at 100m above the ground, wind direction at 100m above the ground, and radiation. The data sampling frequency is once every 60 minutes, and a total of 24 points are sampled in a day (M w = 24).

[0116] In this embodiment, the four different types of load elements on the same day in the dictionary are stacked, and white noise with a mean of 0 and a variance of 0.005 is applied to simulate the load on the day to be identified.

[0117] This embodiment provides a load component identification method based on sparse dictionary learning, as Figure 1 shown, which includes the following steps:

[0118] S1) Collect the historical load data and historical meteorological data of various loads, and perform data preprocessing to weaken the influence of abnormal sampling points, obtaining the standardized historical load data and standardized historical meteorological data of various loads; Step S1 is specifically:

[0119] In this embodiment, the total number of load types is N (N = 4). Collect the historical load data of N types of loads, perform outlier and missing value processing on the historical load data, and then perform normalization processing to obtain the standardized historical load data of the four types of loads.

[0120] Taking the 96-point load data of one day as an example (i.e., M = 96), perform normalization processing according to the following formula:

[0121]

[0122] Among them, is the normalized load data at the d-th day and the i-th sampling point; is the original load data at the d-th day and the i-th sampling point; d is the serial number of the sampling day, and i is the serial number of the sampling point, with a range of 1 to 96.

[0123] S1.2) After processing the historical meteorological data for outliers and missing values, for the historical meteorological data with different measurement resolutions, where the lowest and highest resolutions are 1h and 15min respectively, perform interval sampling on the historical meteorological data to unify its resolution to the lowest resolution of 1h (i.e., the total number of sampling points M for each sampling day of the historical meteorological data w = 24), to obtain the standardized historical meteorological data for the four types of loads.

[0124] S2) Based on the data obtained in step S1, establish a comprehensive historical load data dictionary, and perform a correlation analysis on the daily historical load data of each type of load and the corresponding daily meteorological data to comprehensively analyze the influence of various meteorological factors on the daily load, and finally establish a comprehensive historical meteorological data dictionary;

[0125] Step S2 specifically includes the following steps:

[0126] S2.1) Establish a historical load matrix for N = 4 types of loads. The historical load matrix for each type of load is as follows:

[0127]

[0128] In the formula, is the historical load matrix of the n-th type of load; n is the serial number of the load type, n = 1 to N, and N is the total number of load types; is the load data of the n-th type of load at the d-th day and the i-th sampling point. d is the serial number of the sampling day, d = 1 to D n , D n is the total number of sampling days of the n-th type of load, and i is the serial number of the sampling point, i = 1 to 96.

[0129] S2.2) Perform a correlation analysis on the daily historical load data of N types of loads and the corresponding daily meteorological data. Taking photovoltaic as an example, denote its load number as n, and its historical meteorological data includes K n types, such as the weather of the day, temperature, temperature at 2m above the ground, relative humidity at 2m above the ground, pressure, wind speed at 10m above the ground, wind direction at 10m above the ground, wind speed at 100m above the ground, wind direction at 100m above the ground, radiation, etc.

[0130] After being processed by step S1, the resolutions of the historical load data and the historical meteorological data of photovoltaic are 15min and 1h respectively. To unify the resolution, first perform interval sampling on the daily historical load data of photovoltaic and then calculate its correlation with the corresponding daily K n types of historical meteorological data (where k is the serial number of the meteorological type, k = 1 to K n ) of the Spearman correlation coefficient, and the calculation formula is as follows:

[0131]

[0132] where ρ n,d,k is the correlation coefficient between the photovoltaic load data on the d-th day and the k-th type of meteorological data; is the load data of the i-th sampling point of the photovoltaic on the d-th day, is the average value of the load data of all sampling points of the photovoltaic on the d-th day; is the k-th type of meteorological data of the i-th sampling point of the photovoltaic on the d-th day, is the average value of the k-th type of meteorological data of all sampling points of the photovoltaic on the d-th day.

[0133] Through correlation analysis, the correlation coefficient ρ between the daily historical load data of the photovoltaic and various types of meteorological data on the corresponding day is obtained n,d,k (d = 1 to D n , k = 1 to K n ). After that, the average correlation coefficient of various types of meteorological data of the photovoltaic is calculated according to the following formula:

[0134]

[0135] In the formula, is the average correlation coefficient between the load data of the photovoltaic and the k-th type of meteorological data, and ρ n,d,k is the correlation coefficient between the photovoltaic load data on the d-th day and the k-th type of meteorological data.

[0136] In order to comprehensively investigate the relationship between various types of meteorological data of the photovoltaic and the daily load data, various types of meteorological data are weighted and calculated according to their average correlation coefficients to obtain the unified daily historical meteorological vector of the photovoltaic, as shown in the following formula:

[0137]

[0138] where is the weighted historical meteorological data of the i-th sampling point of the photovoltaic on the d-th day; f n is the weighted calculation function of the photovoltaic.

[0139] After the above weighted calculation process, a photovoltaic historical meteorological data dictionary A wn is established according to the unified daily historical meteorological vector, specifically as follows:

[0140]

[0141] Similarly, a historical load and meteorological data dictionary for N types of loads is established, and finally, a historical load data dictionary A and a historical meteorological data dictionary A for all types of loads are established. w , specifically as follows:

[0142]

[0143] S3) Based on the comprehensive historical load data dictionary and the comprehensive historical meteorological data dictionary obtained in step S2, use the orthogonal matching pursuit algorithm combined with meteorological factors to identify the load components of the load data on the day to be identified, obtain the load component identification result, and use the load component identification result for load forecasting or dispatching of the power system.

[0144] Step S3 is specifically as follows:

[0145] S3.1) Regard the column vector of the daily historical load data in the historical load data dictionary A as an element in the dictionary, and then determine the solution target as: select the fewest load data dictionary elements to represent the total load as accurately as possible, and then establish a solution model, specifically as follows:

[0146] X = arg min||E|| 2 ,

[0147] s.t. Y = AX + E

[0148] ||X|| 0 ≤L

[0149] where Y is the 96-point load vector to be identified, Y = [y 1 ... y i ...y 96 T ∈R 96×1 , y i is the load data collected at the i-th sampling point on the day to be identified;

[0150] X is the component ratio vector, X n is the element ratio matrix of the n-th type of load in the dictionary,

[0151] E is the 96-point error vector, E = [ε 1 ... ε i ... ε 96 T ∈R 96×1 , ε i is the error at the i-th sampling point on the day to be identified;

[0152] ||X|| 0≤L means that the number of non-zero elements in the component proportion vector needs to be less than the upper limit value L of the selected dictionary elements.

[0153] S3.2) For the above planning problem, an orthogonal matching pursuit algorithm combined with meteorological factors is used for solution. The specific steps are as follows:

[0154] Collect the load data and meteorological data of the day to be identified, preprocess the meteorological data, and construct the meteorological vector of the day to be identified.

[0155] First, collect the load data [y 1 ... y i ... y 96 T and the meteorological data of K n types of meteorological data

[0156] Then, in order to compare the meteorological data of the day to be identified with the historical meteorological data, it is necessary to preprocess the meteorological data of the day to be identified according to the weighted calculation function f n for the corresponding load type:

[0157]

[0158] Among them, is the meteorological data of the i-th sampling point after weighted calculation for the n-th type of load; is the original meteorological data of the i-th sampling point of the k-th type of meteorological data of the day to be identified; Y wn ∈R 24×1 is the meteorological vector of the day to be identified after weighted calculation for the n-th type of load.

[0159] 1) Initialization:

[0160] ① Calculate the initial comprehensive difference between the load and meteorological data of each element in the historical dictionary and the load and meteorological data of the day to be identified:

[0161] First, calculate the initial comprehensive difference between the load data of each element in the comprehensive historical load data dictionary and the load data of the day to be identified, that is, calculate the inner product of the load vector Y of the day to be identified and the comprehensive historical load data dictionary A to obtain the initial load inner product matrix:

[0162]

[0163] Among them, is the initial load inner product matrix between each element in the comprehensive historical load data dictionary and the load data of the day to be identified; is the initial load inner product between the load data of the n-th type of load and the d-th day element in the comprehensive historical load data dictionary and the load data of the day to be identified; The above formula can be written as the following detailed equation:

[0164]

[0165] where y i is the load data collected at the i-th sampling point for the day to be identified, is the load data of the n-th type of load at the i-th sampling point on the d-th day.

[0166] Then, calculate the comprehensive difference between the meteorological data of each element in the comprehensive historical meteorological data dictionary and the meteorological data of the day to be identified, that is, calculate the meteorological vector Y wn of the day to be identified and the comprehensive historical meteorological data dictionary A w to obtain the initial meteorological inner product matrix:

[0167] G w = Y wn T A w

[0168]

[0169] where G w is the meteorological inner product matrix of each element in the comprehensive historical meteorological data dictionary and the meteorological data of the day to be identified; g w,n,d is the meteorological inner product of the meteorological data of the n-th type of load, the d-th day element in the comprehensive historical meteorological data dictionary and the meteorological data of the day to be identified; the above formula can be written as the following detailed equation:

[0170]

[0171] where is the meteorological data of the i-th sampling point after weighted calculation of the n-th type of load, is the weighted historical meteorological data of the n-th type of load at the d-th day and the i-th sampling point.

[0172] Finally, calculate the initial comprehensive difference between each element in the historical dictionary and the load and meteorological data of the day to be identified, that is, calculate the initial comprehensive inner product matrix G 0 :

[0173]

[0174]

[0175] where G 0 is the initial comprehensive inner product matrix, is the initial comprehensive inner product of the n-th type of load, the d-th day element in the comprehensive historical load data dictionary and the load and meteorological data of the day to be identified, and ⊙ represents the Hadamard product of matrices;

[0176] The above formula can be written as the following detailed equation:

[0177]

[0178] In the formula, is the initial load inner product of the load data of the nth type of load and the dth-day element in the comprehensive historical load data dictionary and the load data of the day to be identified, and g w,n,d is the meteorological inner product of the meteorological data of the nth type of load and the dth-day element in the comprehensive historical meteorological data dictionary and the meteorological data of the day to be identified.

[0179] ② To attempt to solve the component proportion vector, select the element with the smallest comprehensive difference from the curve to be identified, that is, select the element with the largest inner product in the initial inner product matrix G 0 to form the initial index list Γ 0 :

[0180] Γ 0 = {G 0 maximum value index}

[0181] ③ According to the index list Γ 0 select the elements in the comprehensive historical load data dictionary A to form the initial index dictionary matrix Through the least squares method, according to the above load component identification model and the initial index dictionary matrix solve the initial component proportion vector and calculate the initial residual r 0 :

[0182]

[0183] Among them, is the initial index dictionary matrix containing only the elements in the index list Γ 0 Y is the load vector of the day to be identified, is the initial component proportion vector calculated according to the index list Γ 0 r 0 is the initial residual vector.

[0184] 2) Start the loop, and the loop sequence number j = 1.

[0185] ① Regard the residual vector r j-1 as the load vector of the day to be identified, and obtain the comprehensive inner product matrix G j-1 of each element in the comprehensive historical load data dictionary A and the residual vector r j according to the following formula:

[0186]

[0187] In the formula, G j is the comprehensive inner product matrix obtained in the jth iteration, is the load inner product matrix obtained from the j-th iteration, G w is the meteorological inner product matrix, r j-1 is the residual vector obtained from the (j - 1)-th iteration, and ⊙ represents the Hadamard product of matrices.

[0188] ② Select, except for the elements marked in the index list Γ j-1 the element with the largest inner product in the comprehensive inner product matrix G j and add it to the index list to update the index list Γ j :

[0189] Γ j ={G j-1 , G j index of the maximum value}

[0190] In the formula, Γ j is the index list obtained from the j-th iteration, and Γ j-1 is the index list obtained from the (j - 1)-th iteration.

[0191] ③ Select elements in the comprehensive historical load data dictionary A according to the index list Γ j to form the index dictionary matrix Solve the component proportion vector according to the above load component identification model by the least squares method and calculate the residual vector r j :

[0192]

[0193] In the formula, is the component proportion vector obtained from the j-th iteration, Y is the load vector of the day to be identified, is the index dictionary matrix obtained from the j-th iteration, and r j is the residual vector obtained from the j-th iteration.

[0194] Let the loop sequence number j = j + 1, and repeat the above steps until the loop upper limit or the residual converges:

[0195] j ≥ j limit or ||r j - r j-1 || 2 ≤ r limit

[0196] In the formula, j limit is the preset number of iterations, ||r j - r j-1 || 2 is the 2-norm of the residual vectors obtained from the j-th iteration and the (j - 1)-th iteration, and r limit is the preset threshold of the residual.

[0197] 3) After the iteration stops, the final index list Γ is obtained. j and the final component proportion vector For the final index list Γ j the elements with the same load type in it are superimposed according to the final component proportion vector to obtain the final decomposition result:

[0198]

[0199] In the formula, is the final component proportion vector when converging to the j-th iteration; is the element proportion matrix of the n-th type of load when converging to the j-th iteration; Y n-hat is the 96-point summary curve of the identification result of the n-th type of load; X n-hat is the proportion identification result of the n-th type of load.

[0200] The identification result of the daily load to be identified in this embodiment is as Figures 2 to 4 shown.

[0201] Figure 2 shows the changes in the simulated load to be identified before and after applying noise perturbations. As Figure 2 can be seen, after applying the perturbations, the load to be identified fluctuates slightly while the overall trend remains unchanged, which better simulates the actual situation.

[0202] Figure 3 shows the 96-point load curves of the daily load vector to be identified and the four load types it contains.

[0203] Figure 4 shows the identification result using the method of the present invention, and Table 1 shows the theoretical proportion, calculated proportion and relative error of each load component. It can be seen from the charts that the overall identification effect of the load component proportion is good.

[0204] Table 1. Theoretical proportion, calculated proportion and relative error of four load types

[0205] Typical components Air conditioner Charging pile Photovoltaic Base load Theoretical proportion (%) 19.96 9.95 19.95 50.13 Calculated proportion (%) 17.31 10.40 20.53 51.75 Relative error (%) 13.27 4.56 2.90 3.22

[0206] The above are only the preferred embodiments of the present invention. All equivalent changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by the present invention.

Claims

1. A method for identifying load components in a power system based on sparse dictionary learning, characterized in that: The following steps are involved: S1) Collecting historical load data and historical meteorological data of multiple types of loads, and obtaining standardized historical load data and standardized historical meteorological data of various types of loads through preprocessing; S2) establishing a comprehensive historical load data dictionary and a comprehensive historical meteorological data dictionary based on the standardized historical load data and standardized historical meteorological data of various types of loads; S3) Obtain the load data of the day to be identified and the meteorological data of the day to be identified, and use the comprehensive historical load data dictionary and the comprehensive historical meteorological data dictionary obtained in step S2 to perform load component identification on the load data of the day to be identified using an orthogonal matching pursuit algorithm combined with meteorological factors to obtain a load component identification result, which is used for load forecasting or scheduling of the power system.

2. The method for identifying power system load components based on sparse dictionary learning according to claim 1 is characterized in that: The multiple types of loads are selected from photovoltaic, wind power, air conditioning load, small hydropower, charging piles and basic loads.

3. The method for identifying power system load components based on sparse dictionary learning according to claim 1 is characterized in that: The step S2 comprises the following steps: S2.1) Establishing historical load matrices according to the standardized historical load data of each type of load, and combining them to obtain a comprehensive historical load data dictionary; in the historical load matrix of each type of load, the column vector is the load data collected by all sampling points on the same sampling day, and the sampling point is the sampling time point; S2.2) For each type of load, a historical meteorological matrix is ​​obtained by performing correlation analysis on the standardized historical load data and the standardized historical meteorological data; the historical meteorological matrices of all loads are combined to obtain a comprehensive historical meteorological data dictionary.

4. The method for identifying power system load components based on sparse dictionary learning according to claim 3 is characterized in that: The step S2.2 is specifically as follows: S2.2.1) For each type of load, standardize the historical load data and the historical meteorological data to a uniform resolution; S2.2.2) For each type of load, the Spearman correlation coefficient between the load data and various meteorological data of each sampling day is obtained by performing correlation analysis respectively; S2.2.3) For each type of load, the weighted historical meteorological data at each sampling point on the sampling day is obtained according to the following formula based on the various meteorological data on each sampling day and the Spearman correlation coefficient of each meteorological data of the load data: In the formula, is the weighted historical meteorological data of the nth type of load at the ith sampling point on the dth day; is the kth type of meteorological data of the nth type of load at the ith sampling point on the dth day, k = 1 ~ K n , K n is the total number of meteorological types of the nth load, f n is the weighted calculation function of the nth type of load, is the average correlation coefficient between the load data of the nth type of load and the meteorological data of the kth type, ρ n,d,k is the correlation coefficient between the load data of the nth type of load on the dth day and the meteorological data of the kth type; S2.2.4) For each type of load, the weighted historical meteorological data of all sampling points on each sampling day are combined to obtain a daily unified historical meteorological vector. The daily unified historical meteorological vector is used as a column vector to construct a historical meteorological matrix for the load.

5. The method for identifying power system load components based on sparse dictionary learning according to claim 4 is characterized in that: The historical meteorological matrix for each type of load constructed in step S2.2.4 is specifically: In the formula, A wn is the historical meteorological matrix of the nth type of load, is the weighted historical meteorological data of the nth type of load at the ith sampling point on the dth day, d is the serial number of the sampling day, d = 1 ~ D n , D n is the total number of sampling days for the nth type of load, i is the serial number of the sampling point, i = 1 ~ M t , M t is the total number of sampling points on each sampling day after unifying the resolution.

6. The method for identifying power system load components based on sparse dictionary learning according to claim 4 is characterized in that: The step S3 is specifically as follows: S3.1) Taking the daily unified historical load vector in the comprehensive historical load data dictionary as an element, with the minimum load data dictionary elements and accurate representation of the total load as the solution goal, a load component identification model is established based on the comprehensive historical load data dictionary; S3.2) Obtain the load data of the day to be identified and the meteorological data of the day to be identified for each type of load, and construct them into the load vector of the day to be identified and the meteorological vector of the day to be identified respectively; use the orthogonal matching pursuit algorithm combined with meteorological factors, according to the load component identification model and the meteorological vector of the day to be identified, perform load component identification on the load vector of the day to be identified, and obtain the load component identification result.

7. The method for identifying power system load components based on sparse dictionary learning according to claim 6 is characterized in that: In step S3.1, the load component identification model is specifically: X=argmin||E||2, stY=AX+E ||X||0≤L Where, Y is the daily load vector to be identified, X is the component proportion vector, A is the comprehensive historical load data dictionary, E is the error vector, ||X||0 is the number of non-zero elements in the component proportion vector X, L is the preset upper limit of the comprehensive historical load data dictionary element selection, ||E||2 is the 2-norm of the error vector, st is the constraint condition, and X=argmin||E||2 is the value of the component proportion vector X when the 2-norm of the error vector E reaches the minimum value; Among them, the component proportion vector X is specifically: Where, X n is the element proportion matrix of the nth type of load in the comprehensive historical load data dictionary, N is the total number of load types, D n is the total number of sampling days for the nth type of load; Among them, the error vector E is specifically: E=[e 1 ... eh i ... eh M ] T ∈R M×1 In the formula, ε i is the error of the i-th sampling point, i=1~M, M is the total number of sampling points on the day to be identified.

8. The method for identifying power system load components based on sparse dictionary learning according to claim 6 is characterized in that: The step S3.2 is specifically as follows: S3.2.1) Collecting load data on the day to be identified and constructing a load vector for the day to be identified, respectively collecting meteorological data of the locations of various types of loads on the day to be identified, obtaining meteorological data on the day to be identified for various types of loads, and constructing the meteorological data on the day to be identified for the load into a meteorological vector for the day to be identified through a weighted calculation function for each type of load; S3.2.2) taking the inner product of the comprehensive historical meteorological data dictionary and the meteorological vector of the day to be identified, and obtaining a meteorological inner product matrix as the meteorological inner product matrix; at the same time, initializing the index list; S3.2.3) Take the inner product of the comprehensive historical load data dictionary and the load vector of the day to be identified to obtain a load inner product matrix, take the inner product of the load inner product matrix and the meteorological inner product matrix obtained in step S3.2.2 to obtain a comprehensive inner product matrix, and select the index corresponding to the maximum value of the elements in the comprehensive inner product matrix except the element corresponding to the index in the index list, and add it to the index list; S3.2.4) According to the index list, select elements from the comprehensive historical load data dictionary to construct an index dictionary matrix, input the index dictionary matrix as the comprehensive historical load data dictionary into the load component identification model, and obtain the component proportion vector and residual vector through least squares processing; S3.2.5) taking the residual vector as the daily load vector to be identified, and obtaining a new component proportion vector and a residual vector according to steps S3.2.3 and S3.2.4; S3.2.6) Repeat step S3.2.5 until the preset number of iterations is reached or the residual converges to obtain the final index list and component proportion vector. Superimpose the same type of elements in the index list according to the component proportion vector to obtain the load component identification result.

9. The method for identifying power system load components based on sparse dictionary learning according to claim 8, characterized in that: In step S3.2.2, the inner product of the load inner product matrix and the meteorological inner product matrix is ​​taken according to the following formula to obtain a comprehensive inner product matrix: In the formula, G j is the comprehensive inner product matrix obtained in the jth iteration, is the load inner product matrix obtained in the jth iteration, G w is the meteorological inner product matrix, and ⊙ represents the Hadamard product.

10. The method for identifying power system load components based on sparse dictionary learning according to claim 8, characterized in that: In step S3.2.1, the weighted calculation function f for each type of load is calculated according to the following formula: n The meteorological data of the load to be identified on the day is constructed as the meteorological vector of the load to be identified on the day: In the formula, is the weighted meteorological data of the ith sampling point of the nth type of load, Y is the original meteorological data of the i-th sampling point of the k-th meteorological data type on the day to be identified, wn is the daily meteorological vector to be identified for the nth type of load.

Images