Disclosure of Invention
Aiming at the problem that the existing method has limitation in numerical value extraction, the invention provides an adjustable potential aggregation method for adjustable resources of a power distribution network, which combines an automatic clustering center recognition technology with the basic ideas of PDC and k-means, improves a k-means clustering algorithm, and avoids the problems that the traditional density peak value clustering algorithm cannot automatically recognize the clustering center and the traditional k-means initial clustering algorithm center and the clustering number cannot be determined.
The technical scheme adopted by the invention is as follows: a method for aggregating the adjustable potential of the adjustable resources of a power distribution network comprises the following steps:
step 1, selecting adjustable potential data of N power distribution network power users;
step 2, classifying the adjustable potential data obtained in the step 1 according to an ADPC-kmeans clustering algorithm:
step 2.1, normalizing the adjustable potential data according to time scale, calculating the distance between any two data points, and performing ascending sequence arrangement on the distance;
step 2.2, calculating the local density of each data point, and arranging all the local densities in a descending order;
step 2.3, calculating the minimum distance of each data point;
step 2.4, a joint probability density function is established according to the density and the minimum distance of each data point, the expectation and the variance of the minimum distance of each data point under the corresponding local density are calculated according to the joint probability density function,
step 2.5, establishing a threshold function, calculating a cluster center discrimination threshold of each data point and determining a cluster center;
step 2.6, clustering the determined clustering center and N sample points to be clustered by adopting a K-means core idea to obtain a clustering result;
step 3, grading the time-interval adjustable potentials of the resident users of the various types of power distribution networks according to an adjustable potential grading method based on the clustering result to form an adjustable potential resource pool of the resident users of the power distribution networks;
and 4, positioning resources corresponding to the resource pool according to the types of the services participating in the power distribution network, reasonably aggregating various power distribution network power users through the adjustable potential superposition of the various power distribution network power users, and then calling and adjusting several types of power distribution network power users successively according to the grading result.
The invention analyzes and excavates the adjustable potential of the multi-element adjustable resources of the power distribution network based on the massive historical data of the power distribution network, then, an ADPC-Kmeans (automatic Density Peak Cluster Kmeans Algorithm) automatic Density Peak value confirmation K mean value clustering algorithm is adopted, the automatic clustering center identification technology is combined with the basic ideas of PDC and K-means, extracting and classifying the adjustable potential commonalities of the adjustable resources of the power distribution network to obtain the adjustable potential characteristics of the adjustable resources of each type of power distribution network represented by a clustering center, meanwhile, the adjustable potential grades of each type of power distribution network adjustable resources are divided into time intervals to form an adjustable potential resource pool which is convenient to call, therefore, the priority selection of the integral regulation after the adjustable resources of each type of power distribution network are aggregated can be realized, and the problems that the clustering center of the traditional density peak value clustering algorithm cannot be automatically identified and the center and the clustering number of the traditional k-means clustering initial clustering algorithm cannot be determined can be solved.
Further, the specific operation process in step 2.1 is as follows: normalizing the adjustable potential data according to time scale to form a data set X to be clustered separated by time scale, wherein the data set X is { X }
i}
NThe corresponding index set I
xN, the index set has a corresponding set of features as x
iWherein x is
i=(x
i,1,x
i,2,...,x
i,t,...,x
i,24),x
iHas a data length of 24, and represents the maximum adjustable potential power value of the user in 24 time periods in a day; calculate any two data points x
i、x
jDistance d of
ijThe data set X to be clustered has a total of D distance values, where D ═ N (N-1)/2,
and D distance values are arranged in ascending order to form a corresponding distance value sequence of the serial numbers 1-D.
Further, the specific operation process of step 2.2 is as follows: calculate each data point x
iLocal density of (p)
iForm a
Wherein the content of the first and second substances,
d
cfor truncating the distance, i.e. the distance value d corresponding to the mth sequence number in the distance sequence
m(ii) a And all data points x
iCorresponding local density values are arranged in descending order and fixed text
For local density data sets
The subscript data sets in descending order, wherein,
the purpose of the ascending order arrangement and the descending order arrangement is to facilitate subsequent calculation and quickly find the required data value.
Further, in step 2.3, the specific operation process of calculating the minimum distance of each data point is as follows:
step 2.3.1, calculate q
iMinimum distance index under data points
Step 2.3.2, mixing
Corresponding to before no sorting
Under the boundary number, so as to quickly find the required data value as each data point x
iMinimum distance value delta
i。
Further, the specific operation process in step 2.4 is as follows: from each data point x
iDensity of (p)
iAnd a minimum distance delta
iEstablishing a joint probability density function
And calculating each data point x according to the joint probability density function
iMinimum distance delta at its corresponding local density
iIs expected to
Sum variance
Wherein:
alpha and beta are relation coefficients, and alpha and beta belong to 0-1.
Further, the specific operation process in step 2.5 is as follows: establishing a threshold function TH (i), calculating a cluster center discrimination threshold of each data point xi, and determining a cluster center: data point x
iIs a threshold value TH (i) of (d) and a minimum distance delta thereof
iBy contrast, if δ
iIf more than TH (i), the point is the candidate clustering center M
kThe number is marked as Y, wherein,
further, step 2.6, clustering the determined clustering center and the N sample points to be clustered by adopting the K-means core idea, wherein the specific operation process for obtaining the clustering result is as follows:
step 2.6.1, calculate each data point xiWith each cluster center MkDistance ofFrom dikAnd according to the distance minimum principle, the data point x isiIs divided to form a partition containing NkCluster of data points Ck,k=1,2,...,Y;NkRepresenting the number of data points of each class, N being used because each class contains different data pointskRepresents; class cluster refers to the cluster with MkIs a cluster center, contains NkCluster of data points CkI.e. into the Y class;
step 2.6.2, calculate each class CkMean value of all data points and using the mean value as updated cluster center Mk′:
And step 2.6.3, returning to step 2.6.1 until the clustering center is not changed any more and clustering is finished.
Further, the operation process of acquiring the adjustable potential data in step 1 includes:
step 1.1, acquiring reference daily electricity load data of N power distribution network electricity users respectively;
step 1.2, respectively acquiring peak load daily power load data of N power distribution network power users;
and step 1.3, subtracting the reference daily power load data obtained by the user corresponding to the step 1.1 from the peak load daily power load data obtained in the step 1.2 to obtain adjustable potential data of N power distribution network power users.
Further, the specific operation process of acquiring the reference daily electricity load data of the N power distribution network electricity users in step 1.1 is as follows: determining time periods which are close to summer, have basically unchanged load, have basically no air conditioner load to be started and have cool weather, selecting 24-point load data of each user in the time period, and averaging the selected load data to obtain a reference load curve.
Further, in step 1.2, the peak load daily electricity load data is 24 load data per user maximum load day.
The invention has the following beneficial effects: the invention analyzes and excavates the adjustable potential of the multi-element adjustable resources of the power distribution network based on the massive historical data of the power distribution network, then, an automatic density peak value confirmation K-means clustering algorithm is adopted, an automatic clustering center identification technology is combined with the basic ideas of PDC and K-means, extracting and classifying the adjustable potential commonalities of the adjustable resources of the power distribution network to obtain the adjustable potential characteristics of the adjustable resources of each type of power distribution network represented by a clustering center, meanwhile, the adjustable potential grades of each type of power distribution network adjustable resources are divided into time intervals to form an adjustable potential resource pool which is convenient to call, therefore, the priority selection of the integral regulation after the adjustable resources of each type of power distribution network are aggregated can be realized, and the problems that the clustering center of the traditional density peak value clustering algorithm cannot be automatically identified and the center and the clustering number of the traditional k-means clustering initial clustering algorithm cannot be determined can be solved.
Detailed Description
The technical solutions of the embodiments of the present invention are explained and explained below with reference to the drawings of the present invention, but the following embodiments are only preferred embodiments of the present invention, and not all embodiments. Based on the embodiments in the implementation, other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present invention.
The adjustable potential aggregation method for the adjustable resources of the power distribution network, as shown in fig. 1, includes the following steps:
step 1, selecting adjustable potential data of 50 power distribution network power utilization users; as shown in fig. 2;
step 1.1, acquiring reference daily electricity load data of 50 power distribution network electricity users respectively; determining time periods which are close to summer, have basically unchanged load, have basically no air conditioner load to be started and have cool weather, selecting 24-point load data of each user in the time period, and averaging the selected load data to obtain a reference load curve;
step 1.2, respectively acquiring peak load daily power load data of 50 power distribution network power users; the peak load daily electricity load data is 24-point load data of each user maximum load day;
step 1.3, subtracting the standard daily power load data obtained by the user corresponding to the step 1.1 from the peak load daily power load data obtained in the step 1.2 to obtain adjustable potential data of 50 power distribution network power users;
and 2, classifying the adjustable potential data obtained in the
step 1 according to an ADPC-kmeans clustering algorithm: will be provided withNormalizing the potential modulation data according to time scale to form a data set X ═ X to be clustered separated by time scale
i}
50The corresponding index set I
x1, 2, 50, the index set has a corresponding set of features as x
iWherein x is
i=(x
i,1,x
i,2,...,x
i,t,...,x
i,24),x
iHas a data length of 24, and represents the maximum adjustable potential power value of the user in 24 time periods in a day; calculate any two data points x
i、x
jDistance d of
ijThere are D distance values between 50 sample points,
d distance values are arranged in ascending order to form a corresponding distance value sequence of serial numbers 1-1225;
step 2.2, calculate each data point x
iLocal density of (p)
iForm a
Wherein the content of the first and second substances,
d
cfor truncating the distance, i.e. the distance value d corresponding to the mth sequence number in the distance sequence
mIn this example, let m be [ 3% D ═ D]When the distance is 37, the truncation distance is 0.3729, which is the distance value corresponding to the sorted serial number 37; and all data points x
iCorresponding local density value descending order, defining
For local density data sets
The subscript data sets in descending order, wherein,
the purpose of the ascending order arrangement and the descending order arrangement is to facilitate subsequent calculation and quickly find the required data value.
Step 2.3, calculate the minimum distance for each data point:
step 2.3.1, calculate q
iMinimum distance index under data points
Step 2.3.2, mixing
Corresponding to before no sorting
Under the index to quickly find the desired data value as each data point x
iMinimum distance value delta
i;
Step 2.4, according to each data point x
iDensity of (p)
iAnd a minimum distance delta
iEstablishing a joint probability density function
And calculating each data point x according to the joint probability density function
iMinimum distance delta at its corresponding local density
iIs expected to
Sum variance
Wherein:
α=β=0.5;
step 2.5, establishing a threshold function TH (i), and calculating each data point x
iAnd (3) judging a threshold value and determining a clustering center: data point x
iIs a threshold value TH (i) of (d) and a minimum distance delta thereof
iBy contrast, if δ
iIf more than TH (i), the point is the candidate clustering center M
kThe number is marked as Y, wherein,
step 2.6, clustering centers determined by adopting K-means core thought clustering and 50 sample points to be clustered, wherein the specific operation process for obtaining a clustering result is as follows:
step 2.6.1, calculate each data point xiWith each cluster center MkDistance d ofikAnd according to the distance minimum principle, the data point x isiIs divided to form a partition containing NkCluster of data points Ck,k=1,2,...,Y;NkRepresenting the number of data points of each class, N being used because each class contains different data pointskRepresents; class cluster refers to the cluster with MkIs a cluster center, contains NkCluster of data points CkI.e. into the Y class;
step 2.6.2, calculate each class CkMean value of all data points and using the mean value as updated cluster center Mk′:
Step 2.6.3, returning to step 2.6.1 until the clustering center is not changed any more, and finishing clustering;
according to the steps, 6 types of adjustable potential types of the power utilization users of the power distribution network are obtained, and an adjustable potential curve of each type of the included sample points is shown in fig. 3.
And performing clustering comparison on the potential curves adjustable by 50 power distribution network resident users by adopting the traditional K-means and DPC clustering algorithms, and taking three indexes of CHI (Calinski-Harabasz Index), DBI (Davies-Bouldin Index) and VCVI (variance based clustering variance Index) as standards for judging the clustering effectiveness. Because the initial clustering centers and the clustering numbers of the traditional k-means clustering cannot be determined, the clustering results have great relation with the determination of the initial clustering centers, the fast search density peak value clustering algorithm is used for distribution, and the clustering centers cannot be automatically identified, each evaluation index is shown in table 1.
TABLE 1
|
ADPC-kmeans
|
DPC
|
k-means
|
CHI
|
38.04153
|
18.7283
|
25.97063
|
DBI
|
0.92299
|
0.9581
|
1.141662
|
VCVI
|
0.412388
|
0.6018
|
0.522962 |
The larger the CHI index is, the smaller the DBI index and the VCVI index are, and the clustering effect of the algorithm is better. As can be seen from the table, the indexes CHI, DBI and VCVI of the improved ADPC-kmeans algorithm are higher than the clustering effectiveness of DPC and kmeans.
As can be seen from comparison of fig. 4 to fig. 9, the cluster center adjustable potential curve of each class is substantially consistent with the fluctuation and shape of the adjustable potential curve after aggregation of each class. Therefore, the clustering center is used as a representative of the adjustable potential curve of each power distribution network electricity user type, and the characteristics of the adjustable potential curve of each type are analyzed.
1. Stationary analysis of tunable potential curves of various types
Calculating the stationarity index of the adjustable potential curve of each cluster center according to the peak-valley load difference and the average load of each cluster center,
wherein P is
p-v,iPeak to valley load difference, P, for the ith cluster center
a,iIs the average load of the ith cluster center. The stability index is only used for stability analysis of the adjustable potential curve, and is used for analyzing results obtained after a clustering method is adopted. When the stationarity index approaches 1, the more stable the adjustable potential curve. According to the clustering result, the stationarity indexes of each class are 1.536, 1.1605, 2.0836, 1.2836, 1.1379 and 1.6080 respectively, so that the adjustable potential stationarity of 6 classes of power distribution network resident users in one day is V class > II class > IV class > I class > VI class > III class. Taking class V as an example, the class usesThe adjustable potential of the household at 0 point to 7 points is obviously lower than that of the household at 7 points to 24 points, and the adjustable potential of the household at 7 points to 24 points is more than 4 kW. Due to the load characteristics and the electricity consumption habits of the users, the electricity consumption loads in the period have obvious adjustable potential peaks at 9 and 18 points, so that the inferred electricity consumption loads can be used as adjustable loads such as interruptible loads, transferable loads and reducible loads to participate in demand response projects.
2. Tunable total potential analysis for each type
The adjustable potential curves of the adjustable potential total amount of each type of distribution network resident users obtained by superposing the adjustable potentials of each type of distribution network resident users are shown in fig. 4(2), fig. 5(2), fig. 6(2), fig. 7(2) and fig. 8(2), fig. 9 (2). According to the clustering result and the adjustable potential curve of each type of adjustable potential total amount, the adjustable potential total amount of 24 points of each type is 922.17kW, 940.98kW, 319.16kW, 927.7069kW, 1120kW and 412.83kW, so that the adjustable potential total amount in one day is V type, II type, IV type, I type, VI type and III type.
And 3, grading the time-interval adjustable potential of each type of power distribution network resident users by using an electricity consumption peak time interval adjustable potential grading method, an electricity consumption valley time interval adjustable potential grading method and an electricity consumption time interval adjustable potential grading method based on the clustering result to form an adjustable potential resource pool of the power distribution network resident users. The specific grading method comprises the following steps:
A. adjustable potential division according to peak hours of electricity utilization
Taking the Zhejiang power saving demand response time as an example, the peak time of power consumption is divided into early peak: 10 to 11 pm, peak at noon: from point 13 to point 17. And for the early peak and the mid-day peak, calculating the average value of the adjustable potential of each clustering center in the time period, and performing adjustable grade division on each class of users according to the principle that the larger the average value is, the higher the grade is, so as to form adjustable potential division grades I, II, III, IV, V and VI in the early peak and mid-day peak. According to the clustering result, the average value of the adjustable potential in the early peak period of each type is 5.21kW, 4.31kW, 3.81kW, 3.13kW, 4.88kW and 2.64kW respectively, and the corresponding adjustable potential grades are I, III, IV, V, II and VI respectively. The average value of the adjustable potential in the peak noon period of each type is 5.02kW, 4.55kW, 3.52kW, 3.58kW, 5.35kW and 4.03kW respectively, and the corresponding adjustable potential grades are II, III, VI, V, I and IV respectively.
B. Potential division adjustable according to power consumption valley time period
Taking the Zhejiang power saving demand response period as an example, the low valley period is divided into the early morning low valley: 0 to 6 o' clock, noon valley: from point 11 to point 12. For the early morning low valley period and the midday low valley period, the average value of the adjustable potential of each clustering center in the period is calculated, and each class of users is subjected to adjustable grade division according to the principle that the larger the average value is, the higher the grade is, so that grades I, II, III, IV, V and VI of the adjustable potential in the early morning low valley period and the midday low valley period are formed. According to the clustering result, the average value of the adjustable potential of each type in the early morning valley period is respectively 2.91kW, 3.33kW, 1.81kW, 2.99kW, 2.95kW and 2.48kW, and the corresponding adjustable potential grades are respectively IV, I, VI, II, III and V. The average value of the adjustable potential of each type of the midday valley period is 5.02kW, 3.54kW, 4.86kW, 2.81kW, 5.463kW and 3.77kW respectively, and the corresponding adjustable potential grades are II, V, III, VI, I and IV respectively.
C. Segment-tunable potential partitioning by usage level
Taking the Zhejiang power saving demand response time period as an example, except the early peak: 10 to 11 pm, peak at noon: 13 o 'clock to 17 o' clock, early morning low trough: 0 to 6 o' clock, noon valley: the period other than 11 to 12 points is regarded as the usage level period, i.e., the morning flat period: 6 o 'clock to 10 o' clock, evening period: point 17 to point 0. And for the morning leveling period and the evening leveling period, calculating the average value of the adjustable potential of each clustering center in the time period, and performing adjustable grade division on each class of users according to the principle that the larger the average value is, the higher the grade is, so as to form adjustable potential division grades I, II, III, IV, V and VI in the morning leveling period and the evening leveling period. According to the clustering result, the average values of the adjustable potential in the morning leveling period of each type are respectively 4.16kW, 4.58kW, 3.59kW, 3.06kW, 5.44kW and 3.22kW, and the corresponding adjustable potential grades are respectively III, II, IV, VI, I and V. The average value of the adjustable potential of each type in the late leveling period is 1.86kW, 1.44kW, 1.77kW, 1.4kW, 1.56kW and 1.45kW respectively, and the corresponding adjustable potential grades are I, V, II, VI, III and IV respectively.
And 4, according to the types of the services participating in the power distribution network, if a peak-eliminating demand response is carried out, preferentially positioning and calling a classification result according to the grade of the adjustable potential of the power distribution network at the peak time period, calling the residents of the power distribution networks of the first type, the fifth type, the second type, the third type, the fourth type and the sixth type in sequence for the peak-eliminating demand response at the early peak time period, calling the residents of the power distribution networks of the fifth type, the first type, the second type, the sixth type, the fourth type and the third type in sequence for the peak-eliminating demand response at the peak time period, superposing the adjustable potentials of the residents of the power distribution networks, effectively aggregating the power users of the power distribution networks, and then realizing the reduction calling and the adjustment of the power users of several types of the power distribution networks according to the classification result.
If valley filling demand response is carried out, locking the result of potential grade division adjustable according to the peak period of electricity utilization and the result of potential grade division adjustable according to the middle period of electricity utilization in the resource pool, combining load types used by resident users of various power distribution networks, and realizing sequential calling of resident users of the power distribution networks with adjustable potential grades from the peak period of electricity utilization to the middle period of electricity utilization, thereby achieving the purpose of effective aggregation after the adjustable potentials of various resident users of the power distribution networks are superposed.
If real-time demand response is carried out, the adjustable potential grade division results of the corresponding time periods are positioned from the adjustable potential resource pool of the power distribution network resident users, and the power distribution network resident users with the adjustable potential grades are sequentially called by combining the load types used by the various power distribution network resident users, so that the purpose of effective aggregation after the adjustable potentials of the various power distribution network resident users are superposed is achieved.
While the invention has been described with reference to specific embodiments, the scope of the invention is not limited thereto, and those skilled in the art will appreciate that the invention includes, but is not limited to, the accompanying drawings and the description of the embodiments above. Any modification which does not depart from the functional and structural principles of the present invention is intended to be included within the scope of the claims.