Peak load shifting potential evaluation system and method based on two-stage clustering

Abstract

The invention discloses a peak load shifting potential evaluation system based on two-stage clustering, and the system is characterized in that a load data collection module is used for collecting historical power load data of a specified user, thereby obtaining a power load curve of the specified user during the implementation period of a peak load shifting project; the clustering analysis module is used for performing clustering analysis on the power load curve of the specified user during the implementation period of the peak load shifting project to obtain a user typical load curve set and a user power utilization regularity parameter; the power consumption mode determination mode is used for performing DTW and kmeans clustering analysis on the user typical load curve set to obtain a user power consumption mode; and the peak valley filling potential evaluation module is used for calculating a peak load shifting response factor according to the user power consumption mode parameter and the user power consumption regularity parameter, and calculating the peak load shifting potential of the user according to the peak load shifting response factor. According to the method, the peak load shifting potential evaluation of the user can be completed only according to the historical power consumption data.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid planning, and specifically to a peak shifting and valley filling potential evaluation system and method based on two-stage clustering. Background Art

[0002] With the continuous growth of global energy consumption, the dominant position of fossil fuels in the energy mix has led to a year-on-year increase in carbon emissions, further exacerbating the greenhouse effect. To address this challenge, countries are actively promoting energy system reforms and vigorously developing new energy power generation represented by solar and wind energy. However, new energy sources such as solar and wind energy are often affected by environmental factors such as weather and temperature, and their energy output exhibits significant intermittency and randomness.

[0003] Due to the characteristic that electricity cannot be stored in large quantities, the supply side and demand side of the power grid must be balanced, otherwise transmission, distribution, and power consumption equipment will be damaged. Therefore, when the power generation power is not balanced with the power load, it is necessary to use regulatory means to rebalance the power grid supply and demand. Currently, there are mainly two methods to solve the imbalance between power supply and demand in the power system: one is to start from the supply side and increase the reserve capacity of the system, but the cost of this peak shaving method is extremely high; the other is to start from the demand side and achieve supply-demand balance by shifting peaks and filling valleys to adjust the user load.

[0004] Currently, the research on peak shifting and valley filling at home and abroad is mainly divided into two types: one is to conduct potential analysis on general loads to evaluate their potential for participating in peak shifting and valley filling, and the other is to conduct potential analysis on a specific load, such as air conditioners, electric vehicles, typical buildings, etc., to obtain the accurate peak shifting and valley filling potential of the load.

[0005] In the first type, the most commonly used method is to use the historical electricity consumption data and response data of users to evaluate the potential of users to participate in peak shifting and valley filling through historical response data. The advantage of this method is that all users can be evaluated, but the disadvantage is that it can only be evaluated based on historical response data, and the evaluation of users who did not participate in the response and had a small response volume before is very inaccurate. In the second type, the commonly used method is to conduct a detailed modeling of the specific load, and then solve the potential of the load to participate in peak shifting and valley filling according to the actual parameters. The advantage of this method is that it can accurately estimate the demand response potential of specific loads, but the disadvantage is that it cannot be used to evaluate the total peak shifting and valley filling potential of a large area in a certain region.

[0006] There is no good method for evaluating the peak shifting and valley filling potential of a certain region in the potential analysis of general loads. Therefore, it is necessary to conduct research in this area. For the industrial loads in a certain region, a peak shifting and valley filling potential evaluation method based on two-stage clustering is proposed to evaluate the peak shifting and valley filling potential of industrial users in a certain region. Summary of the Invention

[0007] The object of the present invention is to provide a peak shifting and valley filling potential evaluation system and method based on two-stage clustering. Aiming at industrial power loads, the present invention obtains the electricity consumption regularity through single-user clustering analysis and obtains the electricity consumption pattern through user typical curve clustering analysis, thereby calculating the peak shifting and valley filling potential of users, and completing the evaluation of the peak shifting and valley filling potential of users only based on historical electricity consumption data.

[0008] To achieve this object, the peak shifting and valley filling potential evaluation system based on two-stage clustering designed by the present invention, as Figure 7 shown, includes a load data acquisition module, a clustering analysis module, an electricity consumption pattern determination module, and a peak shifting and valley filling potential evaluation module;

[0009] The load data acquisition module is used to collect the historical power load data of a specified user, so as to obtain the power load curve of the specified user during the implementation of the peak shifting and valley filling project;

[0010] The clustering analysis module is used to perform clustering analysis on the power load curve of the specified user during the implementation of the peak shifting and valley filling project to obtain a set of user typical load curves and user electricity consumption regularity parameters;

[0011] The electricity consumption pattern determination module is used to perform DTW and kmeans clustering analysis on the set of user typical load curves to obtain the user electricity consumption pattern, and determine the corresponding user electricity consumption pattern parameters according to the user electricity consumption pattern. The user electricity consumption pattern includes single-peak type, double-peak type, high load factor type, and peak avoidance type;

[0012] The peak shifting and valley filling potential evaluation module calculates the peak shifting and valley filling response factor according to the user electricity consumption pattern parameters and the user electricity consumption regularity parameters, and calculates the peak shifting and valley filling potential of the user according to the peak shifting and valley filling response factor.

[0013] In the above technical solution, the historical power load data is historical electricity consumption data.

[0014] In the above technical solution, the user electricity consumption regularity parameters are represented by the clustering upper limit. The K-means algorithm is used to perform clustering analysis on the power load curve of the specified user during the implementation of the peak shifting and valley filling project. Starting from K = 2, the number of clusters K is continuously increased (that is, the initial value of K is 2, and then it is gradually increased by 1 each time, that is, the classification should be at least divided into two categories) until each power load curve is clustered into a corresponding category and the clustering stops (that is, when there is only one curve in a category, the clustering stops. Clustering itself is equivalent to grouping some data. If there is a data in a separate group, it means that the set classification number has reached the upper limit). At this time, the clustering number K-1 is the clustering upper limit K of the user max , the stronger the electricity consumption regularity of the user with a lower clustering upper limit. In fact, having each curve as a category is meaningless, so the maximum clustering number is K-1.

[0015] In the above technical solution, the calculation method of the user's typical load curve set is as follows: First, calculate the silhouette index (SI) in the clustering quality evaluation index for each clustering number K. Then, select the optimal clustering number according to the silhouette index of the clustering quality evaluation index. Finally, determine the user's typical load curve according to the optimal clustering number. Use K-means clustering to divide the load curves into K categories, and each category has a clustering center. The clustering center of the category with the most load curves is the clustering center of the user's typical load curve.

[0016] In the above technical solution, the calculation method of the silhouette index of the clustering quality evaluation index is as follows:

[0017]

[0018] In the formula, p n represents the vector of the nth daily load curve; represents the mean value of all daily load curves; c k represents the kth clustering center vector; d represents the Euclidean distance between two vectors. The K with the smallest SI is the optimal clustering number K opt of the user, SI represents the silhouette index, p x represents the vector of the xth daily load curve, p y represents the vector of the yth daily load curve, d(p x , p y ) represents the distance between any xth and yth daily load curves, and N represents the number of daily load curves;

[0019] The optimal clustering number K opt The clustering center with the most daily load curves under it is the user's typical load curve P tpy The optimal clustering divides the load curves into K opt categories, and each category has a clustering center. The clustering center of the category with the most load curves is the user's typical load curve P tpy . Perform the above operations on m users to obtain the user's typical load curve set.

[0020] In the above technical solution, the specific method for performing DTW and kmeans clustering analysis on the user's typical load curve set is as follows:

[0021] From the load factor, daily peak-valley difference rate, early peak period load factor, late peak period load factor, and valley period load factor in the user's typical load curve set, form a feature data set D = {X1, X2,..., X m}; Use the K-means objective function to perform on the feature data set D = {X1, X2,..., X m} Clustering is performed, and the number of clusters K is set to 4. The user's electricity consumption pattern is determined from the clustering result graph of the user's typical load curve obtained by clustering. The user's electricity consumption patterns include single-peak type, double-peak type, high load factor type, and peak-shaving type.

[0022] The objective function of K-means is shown in formula (2). The process is to first calculate the DTW distance from each time series to the central sequence of its corresponding classification, and then accumulate and sum the distances to achieve clustering;

[0023]

[0024] where c i represents the clustering center of C i , C i represents the set of the i-th class, DTW represents the DTW distance from X to c i , K represents the number of clusters, K = 4, E represents the objective function value of the clustering process, and X is an element belonging to C i , representing the vector of user characteristic values;

[0025] The c i is calculated according to the Dynamic Time Warping Barycenter Averaging algorithm (DBA). The Dynamic Time Warping Barycenter Averaging algorithm continuously optimizes the average centroid of multiple time series through the DTW iteration method until convergence. The specific steps are as follows:

[0026] In the first step, randomly select K samples from the feature dataset D = {X1, X2,..., X m} as the initial clustering centers c i ;

[0027] In the second step, use DTW to calculate the distance from each sample to the center point c i , and divide the samples into the nearest clusters according to the distance from each sample to the center point c i ;

[0028] In the third step, use the Dynamic Time Warping Barycenter Averaging algorithm (DBA) to recalculate the new center c i of each cluster;

[0029] In the fourth step, repeat the second step and the third step until the centers of each cluster no longer change.

[0030] In the above technical solution, the specific method for calculating the DTW distance from each time series to the clustering center sequence c of its corresponding classification is:

[0031] For the time series X m =(m1, m2,..., m5) and the clustering center sequence c = (n1, n2,..., n5), the calculation process of the DTW distance is as follows:

[0032] First, construct a 5×5 matrix M using the time series X m and the time series c. The (i, j) - th term in the matrix M represents the Euclidean distance between the time - series data points m i and n j ;

[0033] Then, find a planned path through the matrix to minimize the cumulative Euclidean distance. The planned path W=(w1, w2, …, w k ). The elements in W represent the points passed from the point (m1, n1) in the matrix M to the point (m i , n i ). If w k =(m i , n j ), then w k +1=(m i +1, n j ), (m i , n j +1) or (m i +1, n j +1);

[0034] Then, find the minimum cumulative value according to Equation (3), that is, find the optimal path on the matrix M.

[0035]

[0036] where w i represents the i - th element of the planned path W, and k represents the maximum number of steps of the path, 5≤k<9;

[0037] G(m i , n i ) = d(m i , n i )+min(G(m i-1 , m i-1 ), G(m i-1 , n i ), G(m i , n i-1 ))(4)

[0038] where G(m i , n i ) represents the cumulative distance from the point (m1, n1) in the matrix M to the point (m i , n i ), d(m i , n i ) represents the Euclidean distance between the time - series data points m i and n i , min(G(mi-1 , m i-1 ), G(m i-1 , n i ), G(m i , n i-1 )) represents the cumulative distance of the smallest neighboring element that can reach the point (m i , n i ) in matrix M;

[0039] i, j = 1, 2, …, 5; G(0, 0) = 0; G(m i , 0) = G(0, n i ) = +∞. It can be seen from Equation (4) that

[0040] DTW(X m , c) = G(m5, n5) (5)

[0041] Among them, DTW(X m , c) represents the DTW distance from the time series X m to the clustering center sequence c, that is, the distance from each sample to the clustering center c, and G(m5, n5) represents the cumulative distance from the starting point (m1, n1) to the ending point (m5, n5) in matrix M.

[0042] In the above technical solution, the peak-shifting and valley-filling response factor is a value between 0 and 1, which can be used to represent the actual response of the potential for industrial power load peak-shifting and valley-filling. 0 means that users cannot participate in peak-shifting and valley-filling, and 1 means that all the potential of users can be used for peak-shifting and valley-filling. The calculation of the peak-shifting and valley-filling response factor mainly considers two aspects of parameters: the electricity consumption pattern parameter and the user's electricity consumption regularity parameter. The calculation of the peak-shifting and valley-filling potential factor is shown in Equation (6):

[0043] Factor i = αβ (6)

[0044] Among them, Factor i represents the peak-shifting and valley-filling potential factor, α represents the parameter of the user's electricity consumption pattern. The parameter of the electricity consumption pattern is set to a value between 0 and 1 for different types, as shown in Table 4 for details. β represents the user's electricity consumption regularity parameter, which is determined according to the range of Kmax, as shown in Table 4 for details.

[0045] In the above technical solution, the specific formula for calculating the user's peak-shifting and valley-filling potential based on the peak-shifting and valley-filling response factor is:

[0046]

[0047] Among them, Capacity i represents the peak-shifting and valley-filling potential of user i, Factor iDenote the peak-shifting and valley-filling potential factor as P av.e_peak,i Denote the early peak load of user i as P av.l_peak,i Denote the late peak load of user i as P av.valley,i Denote the valley load of user i

[0048] A method for evaluating the peak-shifting and valley-filling potential of the above system is as follows Figure 1 shown: It includes the following steps

[0049] Step 1: Collect the historical power load data of the specified user to obtain the power load curve of the specified user during the implementation of the peak-shifting and valley-filling project

[0050] Step 2: Conduct cluster analysis on the power load curve of the specified user during the implementation of the peak-shifting and valley-filling project to obtain the user typical load curve set and user electricity consumption regularity parameters

[0051] Step 3: Conduct DTW and kmeans cluster analysis on the user typical load curve set to obtain the user electricity consumption pattern, and determine the corresponding user electricity consumption pattern parameters according to the user electricity consumption pattern

[0052] Step 4: Calculate the peak-shifting and valley-filling response factor according to the user electricity consumption pattern parameters and user electricity consumption regularity parameters, and calculate the peak-shifting and valley-filling potential of the user according to the peak-shifting and valley-filling response factor

[0053] Advantages of the present invention

[0054] The present invention can realize the analysis of the peak-shifting and valley-filling potential of industrial loads based on historical electricity consumption data, reduce the difficulty of peak-shifting and valley-filling potential analysis, and does not require the peak-shifting and valley-filling data of users. It breaks away from the dead cycle that the peak-shifting and valley-filling project needs to conduct peak-shifting and valley-filling potential analysis, while the potential analysis requires peak-shifting and valley-filling data. The present invention realizes the general evaluation of the peak-shifting and valley-filling potential of industrial loads, can provide reference for the power grid company to carry out the peak-shifting and valley-filling project, and reduce the peak load Brief description of the drawings

[0055] Figure 1 is the flow chart for calculating the peak-shifting and valley-filling potential of the present invention

[0056] Figure 2 is the clustering result diagram of the furniture manufacturing industry users provided by the embodiment (K = 4)

[0057] Figure 3 is the typical load curve diagram of users under the optimal clustering number K = 4 provided by the embodiment

[0058] Figure 4 is the clustering result diagram of the user typical load curve provided by the embodiment

[0059] Figure 5Representative curve graphs of four clustering results provided by the embodiments of the present invention;

[0060] Figure 6 Box plot for calculating the peak shaving and valley filling potential provided by the embodiments of the present invention;

[0061] Figure 7 Structural schematic diagram of the present invention. Detailed implementation manners

[0062] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0063] In this example, a peak shaving and valley filling potential evaluation framework is used to estimate the peak shaving and valley filling potential of a certain prefecture-level city in China. The experimental data set comes from the real load data of a certain prefecture-level city in China from June 1, 2022 to August 31, 2022, consisting of 92 days of data, with a granularity of 1 h, 24 data points per day, and a total of 5,538 industrial users. The load base is large enough to show a good aggregation effect at the overall level, providing a data basis for verifying the technical solution proposed in this application. In this embodiment, the clustering analysis of a single user, the clustering analysis expression of the user's typical curve, and the peak shaving and valley filling potential algorithm are described separately.

[0064] Clustering analysis of a single user

[0065] Cluster the daily load curves of a single user for the past three months. The goal is to obtain the user's typical load curve and the regularity of the user's electricity consumption, where the regularity of the user's electricity consumption is represented by the clustering upper limit. The lower the clustering upper limit, the stronger the regularity of the user's electricity consumption.

[0066] The calculation steps are as follows:

[0067] First, select the daily load curves of the user for the past three months. Each daily load curve has one point per 1 h, consisting of a total of 24 points.

[0068] Secondly, use the K-means algorithm to cluster the daily load curves of the user. Start from K = 2 (i.e., the initial value of K is 2, and then gradually increase by 1 each time, that is, the classification should be at least divided into two categories) and continuously increase the number of clusters K until a curve is clustered into one category and then stop clustering. At this time, the number of clusters K - 1 is the clustering upper limit K of the user max .

[0069] Then, calculate the clustering quality evaluation index - the silhouette coefficient (SI) under each clustering number K. The calculation method is shown in Equation (1). Finally, select the best clustering number according to the silhouette coefficient SI. For the selected best clustering number, each category has a clustering center, and the clustering center of the category with the most load curves is the clustering center of the user's typical load curve.

[0070]

[0071] In the formula, p n represents the vector of the nth daily load curve; represents the mean value of all daily load curves; c k represents the kth cluster center vector; d represents the Euclidean distance between two vectors, and the K with the smallest SI is the optimal cluster number K for the user opt , SI represents the silhouette index, p x represents the vector of the xth daily load curve, p y represents the vector of the yth daily load curve, d(p x , p y ) represents the distance between any xth and yth daily load curves, and N represents the number of daily load curves.

[0072] In this example, taking a user in a furniture manufacturing industry as an example, the K-means algorithm is used to continuously increase the cluster number K starting from K = 2. When K = 5, a situation where one curve is clustered into one class will occur. Then, the upper limit of the cluster number K for this user max = 4. The clustering results are as shown in the appendix Figure 2 . The bold curves in the figure are the cluster centers of each class.

[0073] Table 1 gives the SI values of the clustering results of this user under different cluster numbers. It can be seen from Table 1 that the optimal cluster number for this user is 4.

[0074] Table 1 SI values under different cluster numbers

[0075]

[0076] Then, consider the number of daily load curves included in each cluster when the cluster number is 4, as shown in Table 2. The one with the largest number of daily load curves is the typical load curve of this user, and the appendix Figure 3 is obtained.

[0077] Table 2 Number of load curves in different classes under the optimal cluster number

[0078]

[0079] After clustering and analyzing 5538 users respectively, 5538 typical load curves can be obtained.

[0080] Clustering analysis of user typical curves

[0081] Use the DTW and K-means algorithms to perform clustering analysis on these 5538 typical load curves. For the time series X m=(m1, m2, …, m5) and the cluster center sequence c=(n1, n2, …, n5). The calculation process of the DTW distance is as follows:

[0082] First, construct a 5×5 matrix M using the time series X m and the time series c. The (i, j) item in the matrix M represents the Euclidean distance between the time series data points m i and n j ;

[0083] Then, find a planned path through the matrix to minimize the cumulative Euclidean distance. The planned path W=(w1, w2, …, w k ). The elements in W represent the points passed from the point (m1, n1) in the matrix M to the point (m i , n i ). If w k =(m i , n j ), then w k +1=(m i +1, n j ), (m i , n j +1) or (m i +1, n j +1);

[0084] Then, find the minimum cumulative value according to Equation (2), that is, find the optimal path on the matrix M.

[0085]

[0086] where w i represents the i-th element of the planned path W, and k represents the maximum number of steps of the path, 5 ≤ k < 9;

[0087] G(m i , n i ) = d(m i , n i ) + min(G(m i-1 , m i-1 ), G(m i-1 , n i ), G(m i , n i-1 )) (3)

[0088] where G(m i , n i ) represents the cumulative distance from the point (m1, n1) in the matrix M to the point (m i , n i ), and d(m i , ni ) represents the time series data point m i and n i 's Euclidean distance, min(G(m i-1 , m i-1 ), G(m i-1 , n i ), G(m i , n i-1 )) represents the cumulative distance of the minimum neighboring elements that can reach the point (m i , n i ) in matrix M;

[0089] i, j = 1, 2, …, 5; G(0, 0) = 0; G(m i , 0) = G(0, n i ) = +∞, as can be seen from Equation (3)

[0090] DTW(X m , c) = G(m5, n5) (4)

[0091] where DTW(X m , c) represents the DTW distance from time series X m to the cluster center sequence c, that is, the distance from each sample to the cluster center c, and G(m5, n5) represents the cumulative distance from the starting point (m1, n1) to the ending point (m5, n5) in matrix M.

[0092] Given the data set D = {X1, X2, …, X m}, a total of 5 features are selected, the load rate of the daily load curve, the daily peak - valley difference rate, the peak - period load rate, and the valley - period load rate, and their definitions and physical meanings are shown in Table 3.

[0093] Table 3 Definitions and Physical Meanings of 5 Features

[0094]

[0095] Note: P represents the load, and the subscripts av, max, min represent the mean, maximum, and minimum respectively, and the subscripts e_peak, l_peak, valley represent the early peak period, late peak period, and valley period respectively.

[0096] Let the number of clusters K = 4, C = {c1, c2, …, c K}, C is the aggregation of each cluster center. The objective function of K - means is shown in Equation (5), and the process is to first calculate the DTW distance from each time series to the center sequence of its corresponding cluster, and then sum up the distances.

[0097]

[0098] Among them, c i represents the clustering center of C i , C i represents the set of the i-th class, DTW represents the DTW distance from X to c i , K represents the number of clusters, K = 4, E represents the objective function value of the clustering process, X is an element belonging to C i , and represents the vector of user characteristic values;

[0099] The c i is calculated according to the Dynamic Time Warping Barycenter Averaging algorithm (DBA). The Dynamic Time Warping Barycenter Averaging algorithm continuously optimizes the average centroid of multiple time series through the method of DTW iteration until convergence. The specific steps are as follows:

[0100] The first step is to randomly select K samples from the feature data set D = {X1, X2, …, X m} as the initial clustering centers c i ;

[0101] The second step is to use DTW to calculate the distance from each sample to the center point c i , and divide the samples into the nearest clusters according to the distance from each sample to the center point c i ;

[0102] The third step is to use the Dynamic Time Warping Barycenter Averaging algorithm (DBA) to recalculate the new center c i of each cluster;

[0103] The fourth step is to repeat the second step and the third step until the centers of each cluster no longer change.

[0104] The obtained clustering results are as shown in the appendix Figure 4 . Among them, the users of the first class are high load rate type users, the users of the second and fourth classes are single-peak type users and double-peak type users respectively, and both belong to peak-valley regulation type users. The third class includes peak-valley avoidance type and a small number of users who do not belong to the above three classes. It should be noted that although both the second and fourth are peak-valley regulation types, the second has obvious double peaks and has better load transferability than the single peak. Therefore, it is necessary to distinguish it from the fourth during clustering.

[0105] Case analysis of the peak-valley regulation potential algorithm

[0106] The peak-valley regulation response factor is a value between 0 and 1, which can be used to represent the actual response of the potential of industrial power load peak-valley regulation. 0 means that the user cannot participate in peak-valley regulation, and 1 means that all the potential of the user can be used for peak-valley regulation. The calculation of the peak-valley regulation response factor mainly considers two aspects of parameters: the suitability parameter of the user load participating in peak-valley regulation and the reducibility parameter caused by the user industry characteristics.

[0107] Table 4 Peak Shifting and Valley Filling Response Factor Parameters

[0108]

[0109]

[0110] The calculation of the peak shifting and valley filling potential factor is shown in Equation (6):

[0111] Factor i =αβ (6)

[0112] Among them, Factor i represents the peak shifting and valley filling potential factor, α represents the parameter of the user's electricity consumption pattern, and the parameter of the electricity consumption pattern is set to a value between 0 and 1 for different types, as shown in Table 4; β represents the parameter of the user's electricity consumption regularity, which is determined according to the range of the clustering upper limit, as shown in Table 4.

[0113] In this example, the specific formula for calculating the peak shifting and valley filling potential of the user according to the peak shifting and valley filling response factor is:[[]]

[0114]

[0115] Among them, Capacity i represents the peak shifting and valley filling potential of user i, Factor i represents the peak shifting and valley filling potential factor, P av.e_peak,i represents the early peak load of user i, P av.l_peak,i represents the late peak load of user i, and P av.valley,i represents the valley load of user i.

[0116] Therefore, according to Equation (7), the peak shifting and valley filling potential corresponding to 5,538 users can be calculated, and the top 10 are shown in Table 5.

[0117] Table 5 Top 10 Peak Shifting and Valley Filling Potentials and Clustering Categories

[0118]

[0119]

[0120] For clustering 3 (peak staggering type), first do not substitute Factor = 0 into the calculation, and the box plot shown in the appendix is obtained. From the appendix Figure 6 shown. From the appendix Figure 6It is not difficult to see that Clusters 2 and 4 generally have certain potential, which is consistent with the peak shifting and valley filling potential factors set empirically in Table 4. However, the peak shifting and valley filling potential of Cluster 3 is basically negative, indicating that there is basically no peak shifting and valley filling potential, which shows the rationality of setting the potential factor to 0. For Cluster 0, it is a high-load type, and the potential is concentrated around 0, which is in line with its property of balanced peak and valley. The only high potential value appears because the user's own electricity consumption level is extremely large, resulting in huge power changes even for small fluctuations.

[0121] In summary, this application designs a peak shifting and valley filling potential evaluation method based on two-stage clustering analysis. This method uses the historical electricity consumption data of users to evaluate the electricity consumption patterns and regularities of users, and calculates the potential factor for each user. Finally, the case of using this method to evaluate the peak shifting and valley filling potential of a certain prefecture-level city in China also proves its high practical value.

[0122] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. A peak shifting and valley filling potential evaluation system based on two-stage clustering, characterized in that: It includes a load data acquisition module, a clustering analysis module, an electricity consumption pattern determination module, and a peak shaving and valley filling potential evaluation module; The load data acquisition module is used to collect the historical power load data of a specified user, so as to obtain the power load curve of the specified user during the implementation of the peak shaving and valley filling project; The clustering analysis module is used to perform clustering analysis on the power load curve of the specified user during the implementation of the peak shaving and valley filling project to obtain a set of user typical load curves and user electricity consumption regularity parameters; The electricity consumption pattern determination module is used to perform DTW and kmeans clustering analysis on the set of user typical load curves to obtain the user electricity consumption pattern, and determine the corresponding user electricity consumption pattern parameters according to the user electricity consumption pattern; The peak shaving and valley filling potential evaluation module calculates the peak shaving and valley filling response factor according to the user electricity consumption pattern parameters and user electricity consumption regularity parameters, and calculates the peak shaving and valley filling potential of the user according to the peak shaving and valley filling response factor.

2. The peak shifting and valley filling potential evaluation system based on two-stage clustering according to claim 1, wherein: The user's electricity consumption regularity parameter is represented by the clustering upper limit. The K-means algorithm is used to perform clustering analysis on the power load curves of specified users during the implementation of the peak shifting and valley filling project. Starting from K = 2, the number of clusters K is continuously increased until each power load curve is clustered into a corresponding category and then the clustering stops. At this time, the number of clusters K - 1 is the clustering upper limit K of the user max .

3. The peak-shifting and valley-filling potential assessment system based on two-stage clustering according to claim 1 is characterized in that: The calculation method of the set of user typical load curves is as follows: First, calculate the silhouette coefficient in the clustering quality evaluation index for each clustering number K, then select the optimal clustering number according to the silhouette coefficient of the clustering quality evaluation index, and finally determine the user typical load curve according to the optimal clustering number.

4. The peak shifting and valley filling potential evaluation system based on two-stage clustering according to claim 3, characterized in that: The calculation method of the silhouette coefficient of the clustering quality evaluation index is as follows: where p n represents the vector of the nth daily load curve; represents the mean of all daily load curves; c k represents the kth cluster center vector; d represents the Euclidean distance between two vectors, and the K with the smallest SI is the optimal clustering number K for the user opt , SI represents the silhouette index, p x represents the vector of the xth daily load curve, p y represents the vector of the yth daily load curve, d(p x , p y ) represents the distance between any xth and yth daily load curves, and N represents the number of daily load curves; Optimal number of clusters K opt The cluster center with the most next-day load curves is the typical load curve P of the user tpy , the optimal clustering divides the load curves into K opt classes, each class has a cluster center, and the cluster center of the class with the most load curves is the typical load curve P of the user tpy , perform the above operations on m users to obtain the set of typical load curves of the users 5. The peak shifting and valley filling potential evaluation system based on two-stage clustering according to claim 3, characterized in that: The specific method for performing DTW and kmeans clustering analysis on the set of user typical load curves is as follows: From the typical load curve of the user, the load rate, the daily peak-valley difference rate, the early peak load rate, the late peak load rate and the valley load rate are collected to form a characteristic data set D = {X1, X2, ..., X m }; Use K-means objective function to perform feature data set D = {X1, X2, ..., X m } perform clustering; The objective function of K-means is shown in formula (2). The process is to first calculate the DTW distance from each time series to the central sequence of its corresponding classification, then accumulate and sum the distances to achieve clustering, and determine the user electricity consumption pattern through clustering; Among them, c i Indicates C i The cluster center, C i represents the set of class i, DTW represents X to c i The distance DTW distance, K represents the number of clusters, E represents the objective function value of the clustering process, X is the number of clusters belonging to C i The elements of , represent the vector of user feature values; Said c i Calculated according to the dynamic time warping barycenter averaging algorithm, which continuously optimizes the average centroid of multiple time series through the DTW iteration method until convergence. The specific steps are as follows: The first step is to randomly select K samples from the feature dataset D = {X1, X2, …, X m} as the initial clustering centers c i ; Step 2: Use DTW to calculate the distance from each sample to the center point c i and divide the samples into the nearest clusters according to the distance from each sample to the center point c i ; In the third step, use the dynamic time warping barycenter averaging algorithm to recalculate the new center c of each cluster i ; Fourth step, repeat the second step and the third step until the center of each cluster no longer changes.

6. The peak shifting and valley filling potential evaluation system based on two-stage clustering according to claim 5, characterized in that: The specific method for calculating the DTW distance from each time series to the clustering center sequence c of its corresponding classification is as follows: For the time series X m =(m1, m2, …, m5) and the cluster center sequence c = (n1, n2, …, n5), the calculation process of the DTW distance is as follows: First, use the time series X m and the time series c to construct a 5×5 matrix M, where the (i, j)-th term in the matrix M represents the Euclidean distance between the time series data points m i and n j ; Then, find a planning path through the matrix to minimize the cumulative Euclidean distance, the planning path W = (w1, w2, ..., w k ), the elements in W represent the distance from the point (m1,n1) in the matrix M to the point (m i ,n i ) passes through the point, if w k =(m i ,n j ), then w k +1=(m i +1,n j )、(m i ,n j +1) or (m i +1,n j +1); Then, find the minimum cumulative according to formula (3), that is, find the optimal path on matrix M; where, w i represents the i-th element of the planned path W, and k represents the maximum number of steps of the path; G(m i ,n i ) = d(m i ,n i ) + min(G(m i-1 ,m i-1 ), G(m i-1 ,n i ), G(m i ,n i-1 ))(4) Among them, G(m i , n i ) represents the cumulative distance from the point (m1, n1) in the matrix M to the point (m i , n i ). d(m i , n i ) represents the Euclidean distance between the time series data points m i and n i . min(G(m i-1 , m i-1 ), G(m i-1 , n i ), G(m i , n i-1 )) represents the cumulative distance of the smallest neighboring element that can reach the point (m i , n i ) in the matrix M; i, j = 1, 2, …, 5; G(0, 0) = 0; G(m i , 0) = G(0, n i ) = +∞, as can be seen from Equation (4) DTW(X m , c) = G(m5, n5) (5) Among them, DTW(X m , c) represents the DTW distance from the time series X m to the cluster center sequence c, that is, the distance from each sample to the cluster center c, and G(m5, n5) represents the cumulative distance from the starting point (m1, n1) in the matrix M to the ending point (m5, n5) in the matrix M.

7. The peak-shifting and valley-filling potential assessment system based on two-stage clustering according to claim 1 is characterized in that: The peak shaving and valley filling response factor is a value between 0 and 1, which can be used to represent the actual response of the potential of industrial power load peak shaving and valley filling. 0 means that the user cannot participate in peak shaving and valley filling, and 1 means that all the potential of the user can be used for peak shaving and valley filling.

8. The peak-shifting and valley-filling potential assessment system based on two-stage clustering according to claim 7 is characterized in that: The calculation of the peak shaving and valley filling response factor mainly considers two aspects of parameters: electricity consumption pattern parameters and user electricity consumption regularity parameters. The calculation of the peak shaving and valley filling potential factor is shown in formula (6): Factor i = αβ (6) Among them, Factor i represents the potential factor for peak shifting and valley filling, α represents the parameter of the user's electricity consumption pattern, and β represents the parameter of the user's electricity consumption regularity.

9. The peak shifting and valley filling potential evaluation system based on two-stage clustering according to claim 1 or 8, characterized in that: The specific formula for calculating the peak shaving and valley filling potential of the user according to the peak shaving and valley filling response factor is: Among them, Capacity i represents the peak-shifting and valley-filling potential of user i, Factor i represents the peak-shifting and valley-filling potential factor, P av.e_peak,i represents the early peak load of user i, P av.l_peak,i represents the late peak load of user i, P av.valley,i represents the valley period load of user i.

10. A method for evaluating the peak shaving and valley filling potential of the system according to claim 1, characterized in that: It includes the following steps: Step 1: Collect the historical power load data of a specified user, so as to obtain the power load curve of the specified user during the implementation of the peak shaving and valley filling project; Step 2: Perform clustering analysis on the power load curve of the specified user during the implementation of the peak shaving and valley filling project to obtain a set of user typical load curves and user electricity consumption regularity parameters; Step 3: Perform DTW and kmeans clustering analysis on the set of user typical load curves to obtain the user electricity consumption pattern, and determine the corresponding user electricity consumption pattern parameters according to the user electricity consumption pattern; Step 4: Calculate the peak shaving and valley filling response factor according to the user electricity consumption pattern and user electricity consumption regularity parameters, and calculate the peak shaving and valley filling potential of the user according to the peak shaving and valley filling response factor.

Images