Power grid load characteristic portraying method, system, equipment and medium

Abstract

The invention belongs to the technical field of power system load analysis, and particularly relates to a power grid load characteristic portraying method, system and device and a medium. In order to overcome the defect that an existing micro-grid load analysis method is not fine enough, the invention adopts the following technical scheme: a power grid load characteristic portraying method comprises the following steps: acquiring a micro-grid power load curve, and obtaining a clustering result by adopting LS-AP-DBI; averaging various curves in the clustering result to obtain a plurality of clustering average curves, averaging all load curves to obtain an annual average curve, comparing the clustering average curves with the annual average curve, and pasting corresponding labels on various curves; and obtaining a power grid load overall label, an integration type label and an overall label, and generating a micro-grid load power consumption portrait. The method has the beneficial effects that a load curve which is difficult to identify by clustering algorithms such as Kmeans and the like and has a short Euclidean distance and obvious characteristics can be identified, so that a finer identification portrait of the load characteristics is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system load analysis, and specifically relates to a power grid load characteristic profiling method, system, equipment and medium based on a scoring system. Background Art

[0002] With the improvement of living standards and the popularization of electric vehicles, the power consumption level of distribution-microgrids has shown a flexible and increasingly complex development trend, and the importance of load characteristic analysis of distribution-microgrids has gradually increased. Scientific and reasonable classification and analysis of load data of distribution network nodes can help distribution operators better grasp and analyze the power consumption characteristics and related characteristics of microgrids, thereby providing a basis for the formulation of refined management of microgrid load levels, and ultimately optimizing and improving load fluctuations in distribution networks in a targeted manner.

[0003] Chinese invention patent application CN116226699A discloses a method for clustering and fusing power load data of multiple types of distribution networks, including: (1) collecting historical data of busbar loads of substations, cleaning bad data and normalizing; reconstructing the cleaned data, decomposing the historical power load curve into N samples in units of 24 hours, each sample having 96-dimensional data; (2) using the Laplace eigenmap dimensionality reduction algorithm to reduce the 96-dimensional data; (3) using the classification accuracy index DBI, the variance ratio criterion CHI and the interval statistic Gap Statistic to evaluate the clustering algorithm and search for the optimal number of categories; (4) after finding the optimal number of clusters, using the Kmeans++ unsupervised learning algorithm to cluster the reduced-dimensional data, and using each power load sample sequence after clustering to visualize the original load data curve according to category.

[0004] The aforementioned invention patent application can mine the laws of power load data in a long time scale, can automatically cluster different types of loads, and then analyze the load characteristics. However, there are still some shortcomings: 1. The popular Kmeans++ algorithm is used for clustering. Kmeans++ is an improved method for Kmeans. Its main improvement is the optimization of the method of randomly initializing the cluster center of Kmeans. It is a method of classification in the time dimension. The classification results of Kmeans++ are mostly close to rectangles. Although some curves have large fluctuations, they are not classified separately because the Euclidean distance is close to the cluster center, which is not conducive to the microgrid operator's refined observation of the load curve; 2. It adopts standardization. When it comes to the comparison between different features, there is no unified standard for the benchmark values ​​of different feature indicators, and the versatility is weak; 3. It adopts direct processing of time samples; 4. It realizes the evaluation of clustering results, but does not or fails to output feature portraits. Summary of the invention

[0005] In view of the shortcomings of the existing microgrid load analysis method that is not refined enough, the present invention provides a grid load characteristic profiling method based on the LS-AP-DBI scoring system to perform a more scientific and precise analysis of the microgrid load. The present invention also provides a grid load characteristic profiling system, equipment and medium based on the LS-AP-DBI scoring system.

[0006] To achieve the above object, the present invention adopts the following technical solution: a power grid load characteristic profiling method based on the LS-AP-DBI scoring system, the power grid load characteristic profiling method based on the LS-AP-DBI scoring system comprises:

[0007] Obtain the daily power load curve of the microgrid within the year, normalize the different features of the daily power load curve within the year and score them using the Laplace score scoring method, and select the r features with the lowest scores to construct multiple cluster feature sets, 2≤r≤total number of features;

[0008] Each clustering feature set is used as a preference value and input into the neighbor propagation algorithm for clustering calculation to obtain multiple clustering results;

[0009] The Davidson-Botting index is used to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and the neighbor propagation clustering algorithm result with the best discrimination effect is taken as the optimal clustering result;

[0010] Averaging various curves in the optimal clustering result to obtain multiple cluster average curves, averaging all daily electricity load curves in a year to obtain an annual average curve, comparing the cluster average curve and the annual average curve, and attaching class labels corresponding to the characteristics to various curves in the optimal clustering result;

[0011] Obtain the overall label of the grid load, integrate the class label and the overall label, and generate the power consumption profile of the microgrid load.

[0012] As an improvement, the process of constructing the feature set using the Laplacian Score (LS) scoring method includes:

[0013] Obtain relevant features required for profiling;

[0014] Normalize the relevant features required for the portrait;

[0015] Construct the weight matrix Z, and construct an m×m adjacency matrix Z for each normalized feature. The values ​​of each element in Z are as follows:

[0016]

[0017] Where: x i 、x jare the values ​​of the i-th and j-th samples of the feature, respectively, and t is an appropriate constant;

[0018] The LS score is calculated based on the obtained adjacency matrix, and the formula is as follows:

[0019]

[0020] Where: LS r is the LS score of the rth feature, f ri represents the phasor of the i-th sample of the r-th feature (i=1,2,…,m), Var(f r ) represents the phase quantity f r The variance of r =[f r1 ,f r2 ,...,f rm ] T , T represents the transpose of the phasor.

[0021] The features are sorted in order from small to large according to the scoring results, and the importance priority of each feature decreases in turn.

[0022] As an improvement, the characteristics include at least two of maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night power consumption characteristics, peak-to-valley difference rate, load rate, fluctuation degree, and impact degree.

[0023] As an improvement, the clustering process of the Affinity Propagation (AP) algorithm includes:

[0024] Construct the similarity matrix S: S(i,j) represents the ability of point j to be the cluster center of point i, expressed as:

[0025] S(i,j)=-||x i -x j || 2 (3)

[0026] The off-diagonal elements are defined as similarity, and the diagonal elements are defined as reference degrees. The reference degree refers to the reference degree of point i as the cluster center. The larger the value, the greater the possibility of being the cluster center. The values ​​of each element are equal, which is the median of all S(i,j);

[0027] Calculate the attractiveness, the formula is:

[0028] R t+1 (i,k)=(1-λ)R′ t+1 (i,k)+λR t (i,k) (4)

[0029]

[0030] Where: R(i,k) is the attraction, which is defined as the degree to which point k is suitable as the cluster center of data point i; A(i,k) is the attribution, which is defined as the degree to which point i chooses point k as its cluster center; t is the number of iterations; λ is the damping coefficient, 0.5<λ<1;

[0031] Calculate the degree of attribution, the calculation formula is:

[0032] A t+1 (i,k)=(1-λ)A t+1 (i,k)+λA t (i,k) (6)

[0033]

[0034] The attraction R(i, k) and the attribution A(i, k) are calculated and updated iteratively until the cluster center does not change after several consecutive iterations or the maximum number of iterations is reached, and the clustering is terminated.

[0035] As an improvement, the Davies-Bouldin Index (DBI) is used to score the results after each clustering, and the feature number corresponding to the lowest DBI index is selected as the final preference value of the AP clustering algorithm. The scoring steps are as follows:

[0036] Calculate the dispersion of data points:

[0037]

[0038] Where: S i is the dispersion degree of data points in the i-th cluster; T i is the total number of data in this cluster; X j is the jth data point in the i-th cluster; A i is the central data point of the i-th category; when q=1, it represents the mean of the distance from each point to the center, and when q=2, it represents the standard deviation of the distance from each point to the center;

[0039] Calculate the distance between clusters:

[0040]

[0041] Where: M ij represents the distance between cluster i and cluster j; a ki represents the value of the kth attribute of the center point of the i-th cluster, a kj Indicates the value of the jth attribute of the center point of the i-th cluster. When p = 1, it indicates the mean of the distances from each point to the center, and when p = 2, it indicates the standard deviation of the distances from each point to the center;

[0042] Calculate the similarity between clusters:

[0043]

[0044] Where: R ij is the similarity between cluster i and cluster j;

[0045] Calculate the maximum similarity between a cluster and other clusters:

[0046] R i =max(R i1 ,R i2 ,...,R ij ,...,R iN ),i≠j (11)

[0047] Where: R i is the maximum similarity between cluster i and other clusters;

[0048] The calculation formula of DBI index is:

[0049]

[0050] Where: is the mean of the maximum similarity of each cluster, that is, the DBI index.

[0051] As an improvement, the clustering results of the AP clustering algorithm are analyzed and labeled to obtain the annual electricity consumption habits of the microgrid under the node, generate a microgrid load characteristic portrait, and visualize the data.

[0052] As an improvement, the process of labeling each type of curve with a class label corresponding to the feature includes:

[0053] Define the daily maximum load:

[0054] A 11,i =maxP i,t (13)

[0055] Where: P i,t is the power of the load curve at time t on day i, A 11,i is the maximum value of the curve on day i. When the maximum load on a single day is greater than 1.2 times the daily average load in the year, the cluster average curve is labeled as “maximum load high”;

[0056] Define the daily minimum load:

[0057] A 12,i =minP i,t (14)

[0058] Where: A 12,i is the minimum value of the i-day curve. When the maximum load of a single day is less than 0.8 times the daily average load of the year, the cluster average curve is labeled as "low minimum load";

[0059] Define the average daily load:

[0060]

[0061] Where: A 13,i is the average value of the curve on day i. When the average load on a single day is greater than 1.1 times the average daily load in the year, the cluster average curve is labeled as “low average load”;

[0062] Define the peak load period: Subtract the values ​​of the cluster average curve and the annual average curve at the same time. If the value of the daily load curve minus the daily average load curve in the previous year is positive for more than 12 periods out of 24 periods, the cluster average curve is labeled as "long peak load period";

[0063] Define the working time of day characteristics:

[0064] A 31 =P t min / P t max (16)

[0065] Where: P t min is the minimum load of the cluster average curve, P t max is the maximum load, when A 31 When the value is greater than 0.6, the label is “24-hour working system”; otherwise, the label is “non-24-hour working system”;

[0066] Define the day and night power consumption characteristics A 32 :

[0067]

[0068] Where: Q d is the daytime electricity consumption, Q n is the power consumption at night, P t is the load value on the load curve at time t;

[0069] Assume that the morning peak period is from 7:00 to 9:00 in the morning, and the evening peak period is from 5:00 to 7:00 in the afternoon. Define the characteristics of the morning and evening peaks as follows:

[0070]

[0071] Where: P mp is the average load from 7 to 9 in the morning; P ep is the average load from 5pm to 7pm; The average loads from 5am to 7am and from 9am to 11am respectively; The average loads from 3pm to 5pm and from 7pm to 9pm respectively; are the ratios of the corresponding variables in the above formula; and When and When the traffic is at a certain time, it is marked as “there is an evening peak”.

[0072] As an improvement, the overall labeling process includes:

[0073] Define monthly electricity consumption similarity:

[0074]

[0075] Where: A 41 represents the similarity of monthly electricity consumption; P i max,m is the maximum load in month i; P max,y is the maximum load in a year; when A 41 >0.85, it is marked as "high monthly electricity similarity", when 0.7≤A 41 ≤0.85, marked as "Medium Monthly Electricity Consumption Similarity", when A 41 When <0.7, it is marked as “low monthly electricity consumption similarity”;

[0076] Define quarterly electricity consumption similarity:

[0077]

[0078] Where: A 42 is the quarterly electricity consumption similarity; P i max,q is the maximum load in the i-th quarter; P max,y is the maximum load in a year. 42 >0.9, it is marked as "high quarterly electricity similarity". 42 ≤0.9, it is marked as “medium quarterly electricity consumption similarity”. 42 When <0.8, it is marked as “low similarity of quarterly electricity consumption”;

[0079] Define seasonal electricity usage patterns:

[0080]

[0081] Where: P b is the average load in spring and autumn, N sp 、N au represents the number of load curves in spring and autumn, here it is taken as 90; P i sp , P i auis the daily load in the i-th spring and autumn;

[0082] The additional load in summer and winter is expressed as follows:

[0083]

[0084] Where: P cool The additional load required for cooling in summer compared to normal times; P heat It is the additional load required for heating in winter compared to normal times. max,s , P max,w are the maximum loads in summer and winter respectively;

[0085] High and low temperature sensitivity indicators in summer and winter:

[0086]

[0087] Where: A 51 A is the summer high temperature sensitivity index; 52 is the winter low temperature sensitivity index; P t cool , P t heat It refers to the additional summer cooling load and winter heating load of the microgrid at time t; P t max,s , P t max,w It refers to the maximum load of the microgrid on day t in summer and winter;

[0088] When A 51 ≤0.1, marked as "low high temperature sensitivity", when 0.1 51 ≤0.3, marked as "high temperature sensitivity medium" 51 When A 52 ≤0.1, marked as "low low temperature sensitivity", when 0.1 52 ≤0.3, marked as "low temperature sensitivity medium", when 0.3 52 When , it is marked as “high low temperature sensitivity”;

[0089] Define the level smoothness:

[0090]

[0091] Where: B 1 , B 2 , B 3 , B 4 They represent the daily peak-to-valley difference rate, daily load rate, fluctuation degree and impact degree respectively. i max , P i ​​​​min , P i mean , P i smooth are the maximum power, minimum power, average power and the smooth curve generated by the moving average method on day i; P i,t is the power at time t on day i;

[0092] When B 1 <0.7, it is marked as “obvious peak-valley characteristics”, otherwise it is marked as “unobvious peak-valley characteristics”; when B 2 >0.85, it is marked as "load rate high", otherwise it is marked as "load rate normal". 3 >5, it is marked as “high volatility”, otherwise it is marked as “normal volatility”. 4 When it is >1.5, it is marked as “high impact”, otherwise it is marked as “normal impact”;

[0093] Define load regulation capability: According to the importance of node load, mark it as "primary load", "secondary load" or "tertiary load".

[0094] A power grid load characteristic portrait system based on the LS-AP-DBI scoring system, the power grid load characteristic portrait system based on the LS-AP-DBI scoring system comprising:

[0095] The Laplace score scoring module is used to construct a feature set for the daily power load curve of the microgrid obtained within the year by using the Laplace score scoring method, normalize and score the different features of the daily power load curve within the year, and select the r features with the lowest scores to construct multiple cluster feature sets, where 2≤r≤the total number of features;

[0096] The neighbor propagation algorithm clustering module is used to input each cluster feature set as a preference value into the neighbor propagation algorithm for clustering calculation to obtain multiple clustering results;

[0097] The Davidson-Boulding Index evaluation module is used to use the Davidson-Boulding Index to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and the neighbor propagation clustering algorithm result with the best discrimination effect is taken as the optimal clustering result;

[0098] The class label generation module is used to average various curves in the optimal clustering result to obtain multiple cluster average curves, average all daily power load curves in a year to obtain an annual average curve, compare the cluster average curve and the annual average curve, and attach class labels corresponding to the characteristics to various curves in the optimal clustering result;

[0099] An overall label generation module, used to obtain the overall label of the power grid load;

[0100] The portrait generation module is used to integrate class labels and overall labels to generate a microgrid load power consumption portrait.

[0101] A computer device includes a processor and a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by the processor, the power grid load characteristic profiling method based on the LS-AP-DBI scoring system as described above is implemented.

[0102] A computer-readable storage medium having a computer program stored thereon, which, when executed, implements the aforementioned power grid load characteristic profiling method based on the LS-AP-DBI scoring system.

[0103] The beneficial effects of the power grid load characteristic profiling method based on the LS-AP-DBI scoring system of the present invention are as follows: the core idea of ​​the Laplace score method (Laplacian Score, LS) is to use the local structural information of the data to evaluate the importance of the feature. The lower the score of the Laplace score method, the higher the discrimination degree of the feature compared with other features, and the more suitable it is as a standard for distinguishing curves. The Laplace score method prepares for the selection of preference values ​​for AP clustering; the AP algorithm (Affinity Propagation, neighbor propagation algorithm) does not need to pre-specify the number of clusters, it can automatically determine the most appropriate number of clusters for the data set, the algorithm determines the cluster center through information transmission between data points, and each data point decides whether to become a cluster center based on its own information and information received from other points, but requires two important parameters, one is the preference value, which is used to control the clustering tendency. The larger the value, the more factors the AP clustering algorithm considers, and the more the number of clusters may be in the end; the other is the damping coefficient, which is used to control the convergence effect of the AP algorithm; DBI (Davies-Bouldin The Davidson-Borgin Index takes into account both the consistency (tightness) within the cluster and the separation between clusters. The method of the present invention, through the collaboration of the LS algorithm and the AP algorithm, can identify a small number of load curves with obvious characteristics, thereby achieving a more detailed recognition portrait of the load characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 It is a flow chart of a method for profiling power grid load characteristics according to an embodiment of the present invention.

[0105] Figure 2 It is a graphical flow chart of the method for profiling the characteristics of power grid loads according to an embodiment of the present invention.

[0106] Figure 3 It is the annual load curve of the microgrid of the embodiment of the present invention (including the load curve of each day in the year).

[0107] Figure 4This is a type of load curve after clustering in the embodiment of the present invention, in which morning and evening peaks are obvious.

[0108] Figure 5 This is a type of load curve after clustering in the embodiment of the present invention in which the base load is high and the load demand in the afternoon is even higher.

[0109] Figure 6 This is a special type of electricity consumption in the clustered load curve of the embodiment of the present invention.

[0110] Figure 7 It is the load characteristic portrait of the microgrid output after clustering curve analysis.

[0111] Figure 8 It is a normalized curve of the annual load curve of the microgrid in the embodiment of the present invention.

[0112] Fig. 9 and Fig.10 The existing Kmeans clustering algorithm is used to Figure 3 Two types of curves are obtained after classifying the load curves. DETAILED DESCRIPTION

[0113] The technical solutions of the embodiments of the present invention are explained and described below, but the following embodiments are only preferred embodiments of the present invention, not all. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the protection scope of the present invention.

[0114] See also Figures 1 to 8 , a power grid load characteristic profiling method based on the LS-AP-DBI scoring system according to an embodiment of the present invention, the power grid load characteristic profiling method based on the LS-AP-DBI scoring system comprises:

[0115] Step S1, obtaining the daily power load curve of the microgrid within the year, normalizing different features of the daily power load curve within the year and scoring them using the Laplace score scoring method, and selecting the r features with the lowest scores to construct multiple clustering feature sets, 2≤r≤total number of features;

[0116] Step S2: input each cluster feature set as a preference value into the nearest neighbor propagation algorithm for clustering calculation to obtain multiple clustering results;

[0117] Step S3, using the Davidson-Botting index to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and taking the neighbor propagation clustering algorithm result with the best discrimination effect as the optimal clustering result;

[0118] Step S4, averaging various curves in the optimal clustering result to obtain multiple cluster average curves, averaging all daily power load curves in a year to obtain an annual average curve, comparing the cluster average curve and the annual average curve, and attaching class labels corresponding to the characteristics to various curves in the optimal clustering result;

[0119] Step S5: Obtain the overall label of the grid load, integrate the class label and the overall label, and generate a power consumption profile of the microgrid load.

[0120] Figure 1 The main flow chart of the method of this embodiment is shown. Figure 2 The main process of the method of this embodiment is illustrated in a graphical manner.

[0121] In this embodiment, in step S1, the process of constructing a feature set using the LS scoring method includes:

[0122] Step S11, obtaining relevant features required for portrait;

[0123] Step S12, normalizing relevant features;

[0124] Step S13: construct a weight matrix Z. For each normalized feature, an m×m adjacency matrix Z is constructed. The values ​​of the elements in Z are as follows:

[0125]

[0126] Where: x i 、x j are the values ​​of the i-th and j-th samples of the feature respectively, and t is an appropriate constant;

[0127] Step S14: Calculate the LS score based on the obtained adjacency matrix. The formula is as follows:

[0128]

[0129] Where: LS r is the LS score of the rth feature, f ri represents the phasor of the i-th sample of the r-th feature (i=1,2,…,m), Var(f r ) represents the phase quantity f r The variance of r =[f r1 ,f r2 ,...,f rm ] T , T represents the transpose of the phasor.

[0130] See also Figure 8This embodiment processes the data by normalization instead of normalization. After normalization, in each feature, at least one curve feature value is 0 and at least one curve feature value is 1. The maximum and minimum values ​​of the feature are used as the upper and lower bounds to construct the Laplace weight matrix, which increases the discrimination of the curve features.

[0131] Step S15: Sort the features in order from small to large according to the scoring results, with the importance priority of each feature decreasing in order.

[0132] In this embodiment, in step S11, the relevant features required for the portrait include maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night power consumption characteristics, peak-to-valley difference rate, load rate, fluctuation, and impact.

[0133] In this embodiment, see Figure 2 , assuming there are 10 features, after LS scoring is completed, the maximum load score is 0.2, the peak-to-valley difference score is 0.25, the minimum load score is 0.3, the load rate score is 0.6..., and the volatility score is 1.1, then the multiple cluster feature sets constructed from small to large are:

[0134] Cluster feature set 1: maximum load, peak-to-valley difference

[0135] Cluster feature set 2: maximum load, peak-to-valley difference, minimum load

[0136] Cluster feature set 3: maximum load, peak-to-valley difference, minimum load, load rate score

[0137] …

[0138] Cluster feature set 9: maximum load, peak-to-valley difference, minimum load, load rate score, ..., volatility.

[0139] Then, each cluster feature set is used as a preference value to perform AP clustering calculations in turn, and DBI scores are performed in turn, and the AP clustering result with the best DBI score is selected as the final result.

[0140] In this embodiment, in step S2, the clustering process of the AP clustering algorithm includes:

[0141] Step S21, construct a similarity matrix S: S(i,j) represents the ability of point j to be the cluster center of point i, expressed as:

[0142] S(i,j)=-||x i -x j || 2 (27)

[0143] The non-diagonal elements are defined as similarity, and the diagonal elements are defined as reference degrees. The reference degree refers to the reference degree of point i as the cluster center. The larger the value, the greater the possibility of being the cluster center. Here, the reference degree is 0 (in the actual AP clustering calculation, the diagonal elements will affect the update of the belonging degree and the attraction degree. The larger the value, the easier it is to become the center of the AP clustering algorithm. This embodiment does not preset any curve to be more inclined to be the cluster center, but hopes that the AP clustering algorithm can automatically generate a suitable cluster center, so the reference degree is 0 here); the values ​​of each element are equal, which is the median of all S(i,j); the similarity matrix is ​​used to quantify the similarity between data points. Similarity and difference, the larger S(i,j), the closer the distance between data point i and data point j is (that is, the more similar). In the AP algorithm, similarity S(i,j) is understood as the ability of data point j to serve as the clustering center of data point i. The higher the similarity, the more such curves can be clustered into one category; reference degree refers to the reference degree of point i as the clustering center. For example, the matrix S is a 10*10 matrix with 10 elements: maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night power consumption characteristics, peak-to-valley difference rate, load rate, volatility, and impact. The reference degree refers to the reference significance of a certain feature as the clustering center. The larger the reference degree, the higher the reference significance.

[0144] Step S22: Calculate the attractiveness, the calculation formula is:

[0145] R t+1 (i,k)=(1-λ)R′ t+1 (i,k)+λR t (i,k) (28)

[0146]

[0147] Where: R(i,k) is the attraction, which is defined as the degree to which point k is suitable as the cluster center of data point i; A(i,k) is the attribution, which is defined as the degree to which point i chooses point k as its cluster center; t is the number of iterations; λ is the damping coefficient, 0.5<λ<1;

[0148] Step S23: Calculate the degree of belonging, the calculation formula is:

[0149] A t+1 (i,k)=(1-λ)A t+1 (i,k)+λA t (i,k) (30)

[0150]

[0151] Step S24, iteratively update and calculate the attraction degree R(i, k) and the attribution degree A(i, k) until the cluster center does not change after several consecutive iterations or the maximum number of iterations is reached, and the clustering is terminated.

[0152] In this embodiment, in step S3, DBI is used to score the results after each clustering, and the feature number corresponding to the lowest DBI index is selected as the final preference value of the AP clustering algorithm. The scoring steps are as follows:

[0153] Step S31, calculating the dispersion degree of data points:

[0154]

[0155] Where: S i is the dispersion degree of data points in the i-th cluster; T i is the total number of data in this cluster; X j is the jth data point in the i-th cluster; A i is the central data point of the i-th category; when q=1, it represents the mean of the distance from each point to the center, and when q=2, it represents the standard deviation of the distance from each point to the center;

[0156] Step S32: Calculate the distance between clusters:

[0157]

[0158] Where: M ij represents the distance between cluster i and cluster j; a ki represents the value of the kth attribute of the center point of the i-th cluster, a kj Indicates the value of the jth attribute of the center point of the i-th cluster. When p = 1, it indicates the mean of the distances from each point to the center, and when p = 2, it indicates the standard deviation of the distances from each point to the center;

[0159] Step S33, calculating the similarity between clusters:

[0160]

[0161] Where: R ij is the similarity between cluster i and cluster j;

[0162] Step S34, calculate the maximum similarity between the cluster and other clusters:

[0163] R i =max(R i1 ,R i2 ,...,R ij ,...,R iN ),i≠j (35)

[0164] Where: R i is the maximum similarity between cluster i and other clusters;

[0165] Step S35: The calculation formula of DBI index is:

[0166]

[0167] Where: is the mean of the maximum similarity of each cluster, that is, the DBI index.

[0168] In this embodiment, in step S4, the clustering results of the AP clustering algorithm are analyzed and labeled to obtain the annual electricity consumption habits of the microgrid under the node, generate a microgrid load characteristic portrait, and visualize the data.

[0169] In this embodiment, the process of calculating and obtaining the overall label of the power grid load in step S4 includes:

[0170] Step S41, define the daily maximum load:

[0171] A 11,i =maxP i,t (37)

[0172] Where: P i,t is the power of the load curve at time t on day i, A 11,i is the maximum value of the curve on day i. When the maximum load on a single day is greater than 1.2 times the daily average load in the year, the cluster average curve is labeled as “maximum load high”;

[0173] Step S42: define the daily minimum load:

[0174] A 12,i =minP i,t (38)

[0175] Where: A 12,i is the minimum value of the i-day curve. When the maximum load of a single day is less than 0.8 times the daily average load of the year, the cluster average curve is labeled as "low minimum load";

[0176] Step S43, define the daily average load:

[0177]

[0178] Where: A 13,i is the average value of the i-day curve. When the maximum load of a single day is greater than 1.1 times the daily average load of the year, the cluster average curve is labeled as “low average load”;

[0179] Step S44, define the peak load period: subtract the values ​​of the cluster average curve and the annual average curve at the same time. If the daily load curve is subtracted from the daily average load curve in the previous year, if there are more than 12 periods with positive values ​​in 24 periods (i.e. 24 hours), the cluster average curve is labeled as "long peak load period"; A 2,i is the number of hours that day i exceeds the annual load curve,

[0180] Table 1 Definition of peak load period indicators

[0181]

[0182] Step S45, define the daily working time characteristics:

[0183] A 31 =P t min / P t max (40)

[0184] Step S45, define the daily working time characteristics:

[0185] A 31 =P t min / P t max (41)

[0186] Where: P t min is the minimum load of the cluster average curve, P t max is the maximum load, when A 31 When the value is greater than 0.6, it is labeled as “24-hour work system”; otherwise, it is labeled as “non-24-hour work system”.

[0187] Table 2 Definition of daily working time indicators

[0188]

[0189] Step S46: define day and night power consumption characteristics A 32 :

[0190]

[0191] Where: Q d is the daytime electricity consumption, Q n is the power consumption at night, P t is the load value on the load curve at time t;

[0192] Step S47: Assume that the morning peak period is from 7:00 to 9:00 in the morning, and the evening peak period is from 5:00 to 7:00 in the afternoon, and define the morning and evening peak characteristics as:

[0193]

[0194] Where: P mp is the average load from 7 to 9 in the morning; P ep is the average load from 5pm to 7pm; The average loads from 5am to 7am and from 9am to 11am respectively; The average loads from 3pm to 5pm and from 7pm to 9pm respectively; are the ratios of the corresponding variables in the above formula; and When and When

[0195] Table 3 Definition of morning and evening peak characteristic indicators

[0196]

[0197] In this embodiment, the calculation and obtaining of the overall label of the power grid load in step S5 includes:

[0198] Step S51, defining monthly electricity consumption similarity:

[0199] In the formula: represents the similarity of monthly electricity consumption; is the maximum monthly load; is the maximum load in a year; when, it is marked as "high similarity of monthly electricity consumption", when, it is marked as "medium similarity of monthly electricity consumption", when, it is marked as "low similarity of monthly electricity consumption", that is,

[0200] Table 4 Definition of electricity consumption similarity index in January

[0201]

[0202] The monthly electricity similarity can reflect the working and production arrangement rules of the microgrid in different months of the year. A large monthly electricity similarity indicates that the load level of each month is similar; otherwise, it indicates that the load level of each month is quite different.

[0203] Step S52: define quarterly electricity consumption similarity:

[0204]

[0205] Where: A 42 is the quarterly electricity consumption similarity; P i max,q is the maximum load in the i-th quarter; P max,y is the maximum load in a year. 42 >0.9, it is marked as "high quarterly electricity similarity". 42 ≤0.9, it is marked as “medium quarterly electricity consumption similarity”. 42 <0.8, it is marked as “low quarterly electricity consumption similarity”, i.e.

[0206] Table 5 Definition of quarterly electricity consumption similarity index

[0207]

[0208] The greater the similarity of quarterly electricity consumption, the more similar it is, indicating that the load levels of the microgrid in the four quarters are similar; otherwise, it indicates that the load levels of the microgrid in the four quarters are quite different;

[0209] Step S53, define seasonal electricity usage rules:

[0210]

[0211] Where: P b is the average load in spring and autumn, N sp 、N au represents the number of load curves in spring and autumn, here it is taken as 90; P i sp , P i au is the daily load in the i-th spring and autumn;

[0212] The additional load in summer and winter is expressed as follows:

[0213]

[0214] Where: P cool The additional load required for cooling in summer compared to normal times; P heat It is the additional load required for heating in winter compared to normal times. max,s , P max,w are the maximum loads in summer and winter respectively;

[0215] High and low temperature sensitivity indicators in summer and winter:

[0216]

[0217] Where: A 51 A is the summer high temperature sensitivity index; 52 is the winter low temperature sensitivity index; P t cool , P t heat It refers to the additional summer cooling load and winter heating load of the microgrid at time t; P t max,s , P t max,w It refers to the maximum load of the microgrid on day t in summer and winter;

[0218] When A 51 ≤0.1, marked as "low high temperature sensitivity", when 0.1 51 ≤0.3, marked as "high temperature sensitivity medium" 51 When A 52 ​​≤0.1, marked as "low low temperature sensitivity", when 0.1 52 ≤0.3, marked as "low temperature sensitivity medium", when 0.3 52 When the temperature is high, it is marked as "high low temperature sensitivity", that is,

[0219] Table 6 Definition of high and low temperature sensitivity index

[0220]

[0221] Due to temperature changes, the microgrid may experience fluctuations in electricity consumption in winter and summer. For example, in summer, air conditioners and other equipment will be used for cooling, and in winter, electrical appliances and other equipment will be used for heating, resulting in differences in electricity load.

[0222] Step S54, define the level compliance:

[0223]

[0224] Where: B 1 , B 2 , B 3 , B 4 They represent the daily peak-to-valley difference rate, daily load rate, fluctuation degree and impact degree respectively. i max , P i min , P i mean , P i smooth are the maximum power, minimum power, average power and the smooth curve generated by the moving average method on day i; P i,t is the power at time t on day i;

[0225] When B 1 <0.7, it is marked as “obvious peak-valley characteristics”; when B 2 >0.85, marked as "high load rate". 3 >5, marked as "high volatility", when B 4 When >1.5, it is marked as "high impact", i.e.

[0226] Table 7 Definition of smoothness index

[0227]

[0228] Step S55, define load regulation capability: according to the importance of node load, mark it as "primary load", "secondary load" or "tertiary load", that is,

[0229] Table 8 Classification of load importance levels

[0230] ​​

[0231] When the annual electricity consumption data of a microgrid in a certain area is known, the LS scoring method is used to score the characteristic indicators of the existing curve, such as maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night electricity consumption characteristics, peak-to-valley difference rate, load, fluctuation, impact, etc. As shown in the following table:

[0232] Table 9 LS microgrid load evaluation indicators

[0233]

[0234] Select the r (2≤r≤total number of scoring items) features with the lowest score as the clustering feature set, use r features as the preference value of the AP clustering algorithm, use DBI to score the results after each clustering, and select the number of features corresponding to the lowest DBI index as the final preference value of the AP clustering algorithm. Finally, analyze and label the results of the AP clustering algorithm (as shown in the table below), obtain the annual electricity consumption habits of the microgrid under the node, generate a microgrid load feature portrait, and visualize the data.

[0235] Table 10 Clustering curve evaluation criteria

[0236]

[0237] The method of this embodiment is based on the classification of load characteristics in space (see Figure 2 ), some rare power consumption curves can be additionally classified, which is more conducive to the distribution network operators to pay attention to abnormal power consumption in the microgrid. In other methods (such as CN116226699A-A multi-type distribution network power load data clustering fusion method), the classification results are mostly close to rectangles. Although some curves have large fluctuations, they are not separately classified because the Euclidean distance is close to the cluster center, which is not conducive to the microgrid operator's refined observation of the load curve.

[0238] The main process of the method of this embodiment is: collecting the microgrid power load curve, which usually includes 365 curves in a year, using the LS-AP-DBI method to perform clustering evaluation to obtain 21 types of curves, and there are more or less curves in the 21 types of curves, and the shapes of the curves in the 21 types of curves are similar; averaging the 21 types of curves, that is, calculating the average value, to obtain 21 cluster evaluation curves; averaging the 365 curves, that is, calculating the average value, to obtain an annual average curve. Compare the 21 cluster evaluation curves and the annual average curve to obtain the class labels of the 21 types of curves; calculate and obtain the overall label of the grid load, integrate the class label and the overall label, and generate the microgrid load power consumption portrait.

[0239] In this embodiment, after labeling the cluster curves, it is only necessary to multiply the label corresponding to the cluster average curve by the corresponding number of cluster curves to know the proportion of the feature in the total features. For example, if there are 35 curves in the cluster curve, and the average curve feature is multiplied by 35, only one judgment and 35 accumulation operations are required for each feature. If the label calculation is performed directly on the features of the 35 curves, 35 judgments and 35 accumulation operations are required. Therefore, the method of this embodiment can reduce the amount of calculation.

[0240] In this embodiment, the distribution network operator obtains multiple clustering graphs and a microgrid load characteristic portrait ( Figure 7 ). The microgrid load characteristic portrait reflects the overall situation of the microgrid power load. What can be observed from the load characteristic portrait is the general situation of the load characteristics of the microgrid. When more detailed characteristics are needed, they can be obtained through each clustering diagram.

[0241] In this embodiment, LS-AP-DBI is used as an evaluation system. Laplace scoring (LS) is a feature selection algorithm proposed by Xue et al. in 2005. It is mainly used to evaluate the importance of each feature in a data set. The core idea of ​​the Laplace scoring method is to use the local structural information of the data to evaluate the importance of the feature. The lower the score of the Laplace scoring method, the higher the discrimination of the feature compared with other features, and the more suitable it is as a criterion for distinguishing curves. In this embodiment, the Laplace scoring method is applied to the scoring of maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night power consumption characteristics, peak-to-valley difference rate, load rate, volatility, impact degree and other features to obtain scores for different feature values, in preparation for the selection of preference values ​​for AP clustering.

[0242] The AP algorithm is a clustering algorithm proposed by Brendan J. Frey and Delbert Dueck in 2007 for cluster analysis in data mining and machine learning. Unlike clustering algorithms such as Kmeans, AP clustering does not require the number of clusters to be specified in advance. It can automatically determine the most appropriate number of clusters for a data set. The algorithm determines the cluster center through information transmission between data points. Each data point determines whether to become a cluster center based on its own information and the information received from other points. Through iterative calculations of the attraction and belonging between data points, it gradually converges to a stable cluster center. However, in actual calculations, the AP clustering algorithm requires two important parameters, one is the preference value, which is used to control the clustering tendency. The larger the value, the more factors the AP clustering algorithm considers, and the more clusters may be generated in the end; the other is the damping coefficient, which is used to control the convergence effect of the AP algorithm. In this embodiment, the feature indicators after LS scoring are sorted from low to high as the preference value of the AP algorithm, and 2 feature indicator clustering results, 3 feature indicator clustering results, ..., 10 feature indicator clustering results are generated in sequence. Finally, the DBI index is used to score the above clustering results separately.

[0243] The DBI index is an index used to evaluate the clustering effect. The DBI index measures the similarity of clustering results, that is, the compactness within the cluster and the separation between clusters. The value of DBI ranges from 0 to infinity, and the smaller the value, the better the clustering effect. DBI is obtained by calculating the sum of the similarities between each cluster and other clusters, and then taking the average. The similarity is determined by comparing the maximum distance within the cluster (that is, the radius of the cluster) and the minimum distance between the clusters. The DBI index takes into account both the consistency (compactness) within the cluster and the separation between clusters. If the data points within the cluster are very close and far away from other clusters, the DBI value will be smaller. This embodiment uses the DBI index to evaluate the AP clustering results, evaluate the rationality of the clustering results, screen out the most reasonable results, and output them according to the 24-hour scale load curve corresponding to the characteristics.

[0244] Compared with the AP clustering algorithm, the Kmeans clustering algorithm and its improved algorithm are also relatively common clustering algorithms. If the AP clustering link is replaced with the Kmeans clustering algorithm or its improved algorithm to divide the microgrid load curve, it is easy to ignore some load curves with similar Euclidean distance but obvious characteristics, such as CN116226699A-A multi-type distribution network power load data clustering fusion method Figure 3 The lowest load curve at the 60th dimension and the high load curve at the 77th dimension in (27days) Figure 3The low load curve of the 50th dimension in (66days). This is because the Kmeans clustering algorithm needs to determine the number of virtual cluster centers in advance and classify curves with similar Euclidean distances into one category.

[0245] See also Figures 3 to 8 ,against Figure 3 The microgrid annual load curve shown in FIG. 1 is classified by the LS-AP-DBI scoring system of the embodiment of the present invention to obtain Figures 4 to 6 , the final output Figure 7 .

[0246] from Figure 3 It can be seen that the annual power load of the microgrid is complex. The power load of the microgrid is mostly concentrated in the range of 170-270kW. There are many curves and the curve characteristics are messy.

[0247] Figures 4 to 6 These are the three types of load curves obtained by this method, which illustrate the frequency of the microgrid's electricity consumption habits in 365 days and the electricity consumption characteristics under these electricity consumption habits. Figures 4 to 6 It can be seen that the classified curve can more comprehensively represent the load characteristics throughout the year, providing a basis for the refined load analysis of the microgrid.

[0248] Figure 4 This is a type of load curve with obvious morning and evening peaks in the clustered load curve of this embodiment. Figure 4 There are 35 middle curves, indicating that the microgrid has this power consumption characteristic for 35 days a year. Figure 4 From the evaluation results, we can get the characteristics of the microgrid: "normal load peak, low load valley, normal average load, normal peak load time, 24-hour working system, similar power consumption during the day and night, evening peak, obvious peak-valley characteristics, and normal load". It can be observed in the clustering curve that the morning load is mainly concentrated around 7 o'clock, but its value is not high. There will be an obvious trough at 14 o'clock, and then the load will climb again, reaching the evening peak around 19 o'clock. This type of label evaluates the power consumption of the microgrid in the morning and evening, and expresses the potential information of the curve with labels, providing a basis for the distribution network operator to formulate refined services. If the valley load of this type of power consumption feature is low, for the distribution network operator, the low valley power consumption has a large space for peak shaving and valley filling. At this time, the distribution network operator can take measures to store part of the electric energy to meet the peak power demand in the future period and ensure the stability of the microgrid.

[0249] Figure 5 It is a type of load curve after clustering in the embodiment of the present invention in which the base load is high and the load demand in the afternoon is higher. Figure 5 There are 10 middle curves, indicating that the microgrid has this power consumption characteristic for 10 days a year. Figure 5From the evaluation results, we can get the characteristics of the microgrid: "high load peak, normal load valley, high average load, long peak load time, 24-hour working system, large daily power consumption, no obvious morning and evening peaks, obvious peak-valley characteristics, and high load rate". The main characteristics of this type of curve are high electricity demand. The lowest load value in the clustered curve is around 250kW, and the highest load can reach around 425kW. This type of load has a large electricity demand, and the load demand rises at night, which is easy to impact the stability of the microgrid. If the distribution network operator can take certain measures, such as adjusting the peak and valley electricity price period, and incentivizing the behavior of reducing electricity consumption during the peak period of microgrid electricity consumption, so as to transfer the load during the peak period, the power supply reliability of the microgrid can be improved.

[0250] Figure 6 This is a special type of electricity consumption in the clustered load curve of the embodiment of the present invention. Figure 6 There are 3 curves in total, indicating that the microgrid has this power consumption characteristic for 3 days a year. Figure 6 From the evaluation results, we can get the characteristics of the microgrid: "normal load peak, low load valley, normal average load, normal peak load time, 24-hour working system, large night power consumption, no obvious morning and evening peaks, obvious peak-valley characteristics, and normal load". It can be observed from the figure that the load of this type of curve is low, and the load is within 200kW per hour. It can be seen from the figure that this type of load is relatively economical and the number is small, and the power consumption will surge around 17:00. Distribution network operators can explore the reasons for this type of power consumption in microgrids, combine the weather, holidays and other conditions of the microgrid on the day for comprehensive analysis, and formulate corresponding refined services. Kmeans tends to classify this type of load curve with other load curves, and it is not easy to observe such characteristics.

[0251] See also Fig. 9 , using the existing Kmeans method Figure 3 One of the results obtained by clustering the load curves. Fig. 9 It can be seen that the principle of Kmeans clustering is Euclidean distance. If one type of curve consumes more electricity in the morning and less electricity in the afternoon, and the other type of curve consumes less electricity in the morning and more electricity in the afternoon, then these two types of curves are easily classified into one category, and it is difficult to analyze the load characteristics in detail.

[0252] See also Fig.10 , using the existing Kmeans method Figure 3 One of the results obtained by clustering the load curves. Figure 3Some curves in the graph have similar trends during the morning peak, but the load classification is unclear at noon and evening. One type of curve has high load at noon and low load at night; another type of curve has high load at night and low load at noon. Kmeans clustering classifies them together. The load curve around 10 o'clock is messy, with one type at about 200kW and another type at 240kW, which blurs the load characteristics.

[0253] Kmeans clustering classifies curves with similar Euclidean distances within 24 hours into one category, but the clustered curves are difficult to reflect the characteristics of this type of load. Curves with different electricity consumption habits are clustered in one category. If the average value of each moment of this type of load curve is connected into a line, the curve will be approximate to a straight line, and it is not similar to a real electricity consumption curve. This is because the cluster center of Kmeans clustering is a virtual point. According to the Kmeans algorithm process, the Kmeans clustering algorithm generates 21 virtual points in the space of 365 points (the dimension of a point is 24 dimensions, that is, 24 hours of load data), and moves the positions of these 21 virtual points each iteration until the sum of the distances from the 365 actual points to their nearest virtual points is the shortest. If the cluster center is a virtual point instead of an actual point, the original characteristics of the curve will be blurred, making the characteristics of this type of curve difficult to distinguish. The AP clustering algorithm uses 21 curves out of 365 curves as cluster centers, so that the curves attract similar curves. Therefore, the method used in this embodiment is more meaningful than Kmeans clustering, and makes up for the shortcomings of the Kmeans clustering algorithm in load feature classification.

[0254] The beneficial effects of the power grid load feature profiling method based on the LS-AP-DBI scoring system of the embodiment of the present invention are as follows: the core idea of ​​the Laplace score method (Laplacian Score, LS) is to use the local structural information of the data to evaluate the importance of the feature. The lower the score of the Laplace score method, the higher the discrimination degree of the feature compared with other features, and the more suitable it is as a standard for distinguishing curves. The Laplace score method prepares for the selection of preference values ​​for AP clustering; the AP algorithm (Affinity Propagation, neighbor propagation algorithm) does not need to pre-specify the number of clusters, it can automatically determine the most appropriate number of clusters for the data set, the algorithm determines the cluster center through information transmission between data points, and each data point decides whether to become a cluster center based on its own information and information received from other points, but requires two important parameters, one is the preference value, which is used to control the clustering tendency. The larger the value, the more factors the AP clustering algorithm considers, and the more the number of clusters may be in the end; the other is the damping coefficient, which is used to control the convergence effect of the AP algorithm; DBI (Davies-Bouldin Index, Davidson Bouldin Index) also considers The consistency (closeness) within the cluster and the separation between clusters are taken into account; the method of the present invention, through the collaboration of the LS algorithm and the AP algorithm, can identify a small number of load curves with obvious characteristics, thereby realizing a more detailed recognition portrait of the load characteristics; the feature data is processed by normalization rather than per-unit, and after the eigenvalues ​​are normalized, in each feature, at least one curve eigenvalue is 0, and at least one curve eigenvalue is 1, and the maximum and minimum values ​​of the feature are used as the upper and lower bounds to construct the Laplace weight matrix, which increases the differentiation of the curve features, and each eigenvalue has at least one curve feature at 0 or 1, maximizing the use of the feature differentiation area.

[0255] The embodiment of the present invention also provides a power grid load characteristic portrait system based on the LS-AP-DBI scoring system, and the power grid load characteristic portrait system based on the LS-AP-DBI scoring system includes:

[0256] The Laplace score scoring module is used to construct a feature set for the daily power load curve of the microgrid obtained within the year by using the Laplace score scoring method, normalize and score the different features of the daily power load curve within the year, and select the r features with the lowest scores to construct multiple cluster feature sets, where 2≤r≤the total number of features;

[0257] The neighbor propagation algorithm clustering module is used to input each cluster feature set as a preference value into the neighbor propagation algorithm for clustering calculation to obtain multiple clustering results;

[0258] The Davidson-Boulding Index evaluation module is used to use the Davidson-Boulding Index to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and the neighbor propagation clustering algorithm result with the best discrimination effect is taken as the optimal clustering result;

[0259] The class label generation module is used to average various curves in the optimal clustering result to obtain multiple cluster average curves, average all daily power load curves in a year to obtain an annual average curve, compare the cluster average curve and the annual average curve, and attach class labels corresponding to the characteristics to various curves in the optimal clustering result;

[0260] An overall label generation module, used to obtain the overall label of the power grid load;

[0261] The portrait generation module is used to integrate class labels and overall labels to generate a microgrid load power consumption portrait.

[0262] An embodiment of the present invention also provides a computer device, including a processor and a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by the processor, the power grid load characteristic profiling method based on the LS-AP-DBI scoring system as described above is implemented.

[0263] The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed, the aforementioned power grid load characteristic profiling method based on the LS-AP-DBI scoring system is implemented.

[0264] The above is only a specific implementation of the invention, but the protection scope of the invention is not limited thereto. Those skilled in the art should understand that the invention includes but is not limited to the contents described in the above specific implementation. Any modification that does not deviate from the functional and structural principles of the invention will be included in the scope of the claims.

Claims

1. A method for profiling power grid load characteristics based on a scoring system, characterized by: include: Obtain the daily power load curve of the microgrid within the year, normalize the different features of the daily power load curve within the year and score them using the Laplace score scoring method, and select the r features with the lowest scores to construct multiple cluster feature sets, 2≤r≤total number of features; Each clustering feature set is used as a preference value and input into the neighbor propagation algorithm for clustering calculation to obtain multiple clustering results; The Davidson-Botting index is used to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and the neighbor propagation clustering algorithm result with the best discrimination effect is taken as the optimal clustering result; Averaging various curves in the optimal clustering result to obtain multiple cluster average curves, averaging all daily electricity load curves in a year to obtain an annual average curve, comparing the cluster average curve and the annual average curve, and attaching class labels corresponding to the characteristics to various curves in the optimal clustering result; Obtain the overall label of the grid load, integrate the class label and the overall label, and generate the power consumption profile of the microgrid load.

2. The method for profiling power grid load characteristics based on a scoring system according to claim 1, characterized in that: The process of constructing a feature set using the Laplace score scoring method includes: Obtain relevant features required for profiling; Normalize relevant features; Construct the weight matrix Z, and construct an m×m adjacency matrix Z for each normalized feature. The values ​​of each element in Z are as follows: Where: x i 、x j are the values ​​of the i-th and j-th samples of the feature, respectively, and t is an appropriate constant; The Laplace score is calculated based on the obtained adjacency matrix, and the formula is as follows: Where: LS r is the Laplace score of the rth feature, f ri represents the phasor of the i-th sample of the r-th feature (i=1,2,…,m), Var(f r ) represents the phase quantity f r The variance of r =[f r1 ,f r2 ,...,f rm ] T , T represents the transpose of the phasor; The features are sorted in order from small to large according to the scoring results, and the importance priority of each feature decreases in turn.

3. The method for profiling power grid load characteristics based on a scoring system according to claim 2, characterized in that: The relevant features required for the portrait include at least two of the maximum load, minimum load, average load, peak-to-valley difference, peak load period, day and night power consumption characteristics, peak-to-valley difference rate, load rate, fluctuation, and impact.

4. The method for profiling power grid load characteristics based on a scoring system according to claim 1, characterized in that: The clustering process of the neighbor propagation algorithm includes: Construct the similarity matrix S: S(i,j) represents the ability of point j to be the cluster center of point i, expressed as: S(i,j)=-||x i -x j || 2 (3) The off-diagonal elements are defined as similarity, and the diagonal elements are defined as reference degrees. The reference degree refers to the reference degree of point i as the cluster center. The larger the value, the greater the possibility of being the cluster center. Here, the reference degree is 0; the values ​​of each element are equal, which is the median of all S(i,j); Calculate the attractiveness, the formula is: R t+1 (i,k)=(1-λ)R' t+1 (i,k)+λR t (i,k) (4) Where: R(i,k) is the attraction, which is defined as the degree to which point k is suitable as the cluster center of data point i; A(i,k) is the attribution, which is defined as the degree to which point i chooses point k as its cluster center; t is the number of iterations; λ is the damping coefficient, 0.5<λ<1; Calculate the degree of attribution, the calculation formula is: A t+1 (i,k)=(1-λ)A t+1 (i,k)+λA t (i,k) (6) The attraction R(i, k) and the attribution A(i, k) are calculated and updated iteratively until the cluster center does not change after several consecutive iterations or the maximum number of iterations is reached, and the clustering is terminated.

5. The method for profiling power grid load characteristics based on a scoring system according to claim 1, characterized in that: The Davidson-Borg index is used to score the results after each clustering, and the feature number corresponding to the lowest Davidson-Borg index is selected as the final preference value of the neighbor propagation algorithm. The scoring steps are as follows: Calculate the dispersion of data points: Where: S i is the dispersion degree of data points in the i-th cluster; T i is the total number of data in this cluster; X j is the jth data point in the i-th cluster; A i is the central data point of the i-th category; when q=1, it represents the mean of the distance from each point to the center, and when q=2, it represents the standard deviation of the distance from each point to the center; Calculate the distance between clusters: Where: M ij represents the distance between cluster i and cluster j; a ki represents the value of the kth attribute of the center point of the i-th cluster, a kj Indicates the value of the jth attribute of the center point of the i-th cluster; when p = 1, it indicates the mean of the distances from each point to the center; when p = 2, it indicates the standard deviation of the distances from each point to the center; Calculate the similarity between clusters: Where: R ij is the similarity between cluster i and cluster j; Calculate the maximum similarity between a cluster and other clusters: R i =max(R i1 ,R i2 ,...,R ij ,...,R iN ),i≠j (11) Where: R i is the maximum similarity between cluster i and other clusters; The calculation formula of Davidson-Botting Index is: Where: is the mean of the maximum similarity of each cluster, i.e., the Davidson-Bodin index.

6. The method for profiling power grid load characteristics based on a scoring system according to claim 1, characterized in that: The process of labeling various curves with class labels corresponding to the features includes: Define the daily maximum load: A 11,i =max P i,t (13) Where: P i,t is the power of the load curve at time t on day i, A 11,i is the maximum value of the i-day curve. When the maximum load of a single day is greater than 1.2 times the daily average load of the year, the cluster average curve is labeled with a high maximum load, otherwise it is labeled with a normal maximum load. Define the daily minimum load: A 12,i =min P i,t (14) Where: A 12,i is the minimum value of the i-day curve. When the maximum load of a single day is less than 0.8 times the daily average load of the year, the cluster average curve is labeled as low minimum load, otherwise it is labeled as normal minimum load; Define the average daily load: Where: A 13,i is the average value of the i-day curve. When the daily average load is greater than 1.1 times the annual daily average load, the cluster average curve is labeled with a high average load, otherwise it is labeled with a normal average load. Define the peak load period: Subtract the values ​​of the cluster average curve and the annual average curve at the same time. If the value of the daily load curve minus the daily average load curve in the previous year is positive for more than 12 periods out of 24 periods, the cluster average curve is labeled as having a long peak load period. Otherwise, it is labeled as having a normal peak load period. Define the working time of day characteristics: A 31 =P t min / P t max (16) Where: P t min is the minimum load of the cluster average curve, P t max is the maximum load, when A 31 When the value is greater than 0.6, it is labeled as a 24-hour work system; otherwise, it is labeled as a non-24-hour work system. Define the day and night power consumption characteristics A 32 : Where: Q d is the daytime electricity consumption, Q n is the power consumption at night, P t is the load value on the load curve at time t; Assume that the morning peak period is from 7:00 to 9:00 in the morning, and the evening peak period is from 5:00 to 7:00 in the afternoon. Define the characteristics of the morning and evening peaks as follows: Where: P mp is the average load from 7 to 9 in the morning; P ep is the average load from 5pm to 7pm; The average loads from 5am to 7am and from 9am to 11am respectively; The average loads from 3pm to 5pm and from 7pm to 9pm respectively; are the ratios of the corresponding variables in the above formula; and When and It is marked as evening peak.

7. The method for profiling power grid load characteristics based on a scoring system according to claim 1, characterized in that: The process of obtaining the overall label of the power grid load includes: Define monthly electricity consumption similarity: Where: A 41 represents the similarity of monthly electricity consumption; P i max,m is the maximum load in month i; P max,y is the maximum load in a year; when A 41 >0.85, it is marked as high similarity of monthly electricity consumption. 41 ≤0.85, marked as monthly electricity similarity, when A 41 When <0.7, it is marked as low similarity of monthly electricity consumption; Define quarterly electricity consumption similarity: Where: A 42 is the quarterly electricity consumption similarity; P i max,q is the maximum load in the i-th quarter; P max,y is the maximum load in a year. 42 >0.9, it is marked as high similarity of quarterly electricity consumption. 42 ≤0.9, it is marked as the quarterly electricity consumption similarity. 42 When <0.8, it is marked as low similarity of quarterly electricity consumption; Define seasonal electricity usage patterns: Where: P b is the average load in spring and autumn, N sp 、N au represents the number of load curves in spring and autumn, here it is taken as 90; P i sp , P i au is the daily load in the i-th spring and autumn; The additional load in summer and winter is expressed as follows: Where: P cool The additional load required for cooling in summer compared to normal times; P heat The additional load required for heating in winter compared to normal times; P max,s , P max,w are the maximum loads in summer and winter respectively; High and low temperature sensitivity indicators in summer and winter: Where: A 51 A is the summer high temperature sensitivity index; 52 is the winter low temperature sensitivity index; P t cool , P t heat It refers to the additional summer cooling load and winter heating load of the microgrid at time t; P t max,s , P t max,w It refers to the maximum load of the microgrid on day t in summer and winter; When A 51 ≤0.1, it is marked as low high temperature sensitivity. 51 ≤0.3, marked as high temperature sensitivity. 51 When A 52 When ≤0.1, it is marked as low sensitivity to low temperature. 52 ≤0.3, marked as low temperature sensitivity. 52 When , it is marked as high low temperature sensitivity;​​​​ Define the level smoothness: Where: B1, B2, B3, B4 represent the daily peak-to-valley difference rate, daily load rate, fluctuation degree and impact degree respectively; P i max , P i min , P i mean , P i smooth are the maximum power, minimum power, average power and the smooth curve generated by the moving average method on day i; P i,t is the power at time t on day i; When B1<0.7, it is marked as obvious peak-valley characteristics, otherwise it is marked as inconspicuous peak-valley characteristics; when B2>0.85, it is marked as high load rate, otherwise it is marked as normal load rate; when B3>5, it is marked as high fluctuation, otherwise it is marked as normal fluctuation; when B4>1.5, it is marked as high impact, otherwise it is marked as normal impact; Define load regulation capability: According to the importance of node load, mark it as "primary load", "secondary load" or "tertiary load".

8. A power grid load characteristic profiling system based on a scoring system, characterized by: The grid load characteristic profiling system based on the scoring system includes: The Laplace score scoring module is used to construct a feature set for the daily power load curve of the microgrid obtained within the year by using the Laplace score scoring method, normalize and score the different features of the daily power load curve within the year, and select the r features with the lowest scores to construct multiple cluster feature sets, where 2≤r≤the total number of features; The neighbor propagation algorithm clustering module is used to input each cluster feature set as a preference value into the neighbor propagation algorithm for clustering calculation to obtain multiple clustering results; The Davidson-Boulding Index evaluation module is used to use the Davidson-Boulding Index to evaluate the discrimination effect of the neighbor propagation algorithm under different preference value results, and the neighbor propagation clustering algorithm result with the best discrimination effect is taken as the optimal clustering result; The class label generation module is used to average various curves in the optimal clustering result to obtain multiple cluster average curves, average all daily power load curves in a year to obtain an annual average curve, compare the cluster average curve and the annual average curve, and attach class labels corresponding to the characteristics to various curves in the optimal clustering result; An overall label generation module, used to obtain the overall label of the power grid load; The portrait generation module is used to integrate class labels and overall labels to generate a microgrid load power consumption portrait.

9. A computer device comprising a processor and a storage medium, wherein the storage medium stores a computer program, characterized in that: When the computer program is executed by a processor, the method for profiling power grid load characteristics based on a scoring system as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, a method for profiling the characteristics of power grid loads based on a scoring system as described in any one of claims 1 to 7 is implemented.

Images