CN115758193A - Distributed energy storage aggregation control method and device - Google Patents
Distributed energy storage aggregation control method and device Download PDFInfo
- Publication number
- CN115758193A CN115758193A CN202211383852.4A CN202211383852A CN115758193A CN 115758193 A CN115758193 A CN 115758193A CN 202211383852 A CN202211383852 A CN 202211383852A CN 115758193 A CN115758193 A CN 115758193A
- Authority
- CN
- China
- Prior art keywords
- cluster
- energy storage
- value
- clustering
- optimal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a method and a device for controlling distributed energy storage aggregation, relating to the technical field of energy storage control; the method comprises the steps of obtaining energy storage nodes, obtaining an upper limit value of a cluster group, calculating cluster centers and corresponding intra-class distance and inter-class distance of the cluster groups from 1 to the upper limit value of the cluster group respectively, obtaining an adjusting factor based on the value of the adjusting factor under each cluster group, substituting the adjusting factor into a balanced evaluation function, obtaining an optimal cluster group when the balanced evaluation function has the minimum value, taking the optimal cluster group as the number of initial cluster centers, taking the cluster center corresponding to the optimal cluster group as the initial cluster center, and inputting an aggregation control algorithm to realize aggregation control; the device comprises an aggregation control module; and when the equalization evaluation function has the minimum value, the initial clustering center and the number in the distributed energy storage aggregation control are correspondingly obtained.
Description
Technical Field
The invention relates to the technical field of energy storage control, in particular to a distributed energy storage polymerization control method and device.
Background
At present, with continuous consumption of traditional fossil energy, the traditional energy cannot be applied to the development of global economy, in order to solve the problems, a new energy technology is developed, the energy problem can be solved due to the appearance of the new energy technology, however, most of the new energy is limited by natural environment, the supply and demand matching capacity of a power generation party and a power utilization party is lower than the matching capacity of the traditional fossil energy, the phenomenon of wind and light abandonment can be caused, meanwhile, certain harm can be brought to the regulation and control of an electric power system, and in order to relieve the situation, the most useful method is to install a distributed energy storage module.
The name of the authorized notice number is CN111750420B, and the name is a control system and a method for a clean heating system. The system comprises a controller, a illuminometer for acquiring outdoor illuminance, a first temperature measuring instrument for acquiring outdoor air temperature, a second temperature measuring instrument for acquiring indoor air temperature, a third temperature measuring instrument for acquiring heating circulating water temperature at an outlet of a heat storage water tank and a fourth temperature measuring instrument for acquiring heating circulating water temperature at an indoor inlet; the method comprises the step of S1 heating cycle control, wherein the step of S1 heating cycle control comprises the steps of S101 data acquisition, S102 heat demand calculation, S103 running time calculation and S104 start and stop control; the high working efficiency of the clean heating system is realized through the controller, the illuminometer, the first to the fourth thermometers and the like.
The name of the authorized notice is CN111750421B, and the name is a control system based on clean energy heating. The system comprises a solar thermal collector, a heat collecting water tank, a biomass furnace, an indoor radiator, a first pump and a second pump, wherein the solar thermal collector, the heat collecting water tank and the first pump are connected in series to form a first loop, and the heat collecting water tank, the second pump, the biomass furnace and the indoor radiator are connected in series to form a second loop; the solar energy heating system realizes high energy heating efficiency through the solar heat collector, the heat collection water tank, the biomass furnace, the indoor radiator, the first pump, the second pump and the like.
Combining the above two patent documents and the prior art, the inventors know:
the distributed energy storage modules have the characteristics of poor controllability and scattered layout, and are difficult to directly receive the regulation and control management of a power system.
Based on the energy storage node aggregation control of the energy storage system architecture, the traditional Kmeans algorithm has two defects in the application. 1) The Kmeans algorithm is limited by the need to determine the initial clustering center in advance; 2) The Kmeans algorithm is also limited by the number of initial cluster centers that need to be determined in advance.
Problems with the prior art and considerations:
the technical problem of determining the initial clustering centers and the number in the distributed energy storage aggregation control is solved.
Disclosure of Invention
The invention aims to provide a method and a device for distributed energy storage aggregation control, and solves the technical problem of determining initial clustering centers and number in distributed energy storage aggregation control.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: a distributed energy storage aggregation control method comprises an aggregation control step, namely obtaining an energy storage node DESNC and obtaining a cluster group upper limit K up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the initial clustering centers into an equalization evaluation function, obtaining an optimal clustering group when the equalization evaluation function has a minimum value, taking the optimal clustering group as the number of the initial clustering centers, taking the clustering center corresponding to the optimal clustering group as the initial clustering center, and inputting the initial clustering centers into a clustering control algorithm to realize clustering control.
The further technical scheme is as follows:
equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is of the type C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
The further technical scheme is as follows: the SA-Kmeans algorithm is obtained by organically combining the simulated annealing algorithm SA and the Kmeans algorithm.
The further technical scheme is as follows: and combining the intra-class distance and the inter-class distance, introducing an adjusting factor into an equalized evaluation function, and combining an early-stopping method and an elbow method to obtain a method for reducing the size of a K value search space for obtaining an optimal clustering group.
The further technical scheme is as follows: and obtaining the optimal initial clustering center through a simulated annealing algorithm SA.
The further technical scheme is as follows: each energy storage node DESNC includes data for state of charge SOC and power.
A distributed energy storage aggregation control device comprises an aggregation control module used for obtaining an energy storage node DESNC and obtaining an upper limit K of a cluster group up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the average evaluation function into the average evaluation function, obtaining an optimal cluster group when the average evaluation function has a minimum value, taking the optimal cluster group as the number of initial cluster centers, taking the cluster center corresponding to the optimal cluster group as the initial cluster center, and inputting an aggregation control algorithm to realize aggregation control.
The further technical scheme is as follows: in the aggregation control module, the control unit is configured to control the aggregation of the plurality of cells,
equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
The device for distributed energy storage aggregation control comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the processor realizes the corresponding steps when executing the computer program.
An apparatus for distributed energy storage aggregation control includes a computer-readable storage medium storing a computer program, which when executed by a processor implements the corresponding steps described above.
Adopt the produced beneficial effect of above-mentioned technical scheme to lie in:
the first method for controlling the distributed energy storage aggregation comprises the step of aggregation control, namely obtaining an energy storage node DESNC and obtaining an upper limit K of a cluster group up Value, cluster group K from 1 to cluster group upper limit K up The values are respectively calculated to obtain a cluster center and corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the initial clustering centers into an equalization evaluation function, obtaining an optimal clustering group when the equalization evaluation function has a minimum value, taking the optimal clustering group as the number of the initial clustering centers, taking the clustering center corresponding to the optimal clustering group as the initial clustering center, and inputting the initial clustering centers into a clustering control algorithm to realize clustering control. According to the technical scheme, when the equalization evaluation function has the minimum value, the initial clustering center and the number in the distributed energy storage aggregation control are correspondingly obtained.
Secondly, the device for controlling the distributed energy storage aggregation comprises an aggregation control module, which is used for obtaining an energy storage node DESNC and obtaining an upper limit K of a cluster group up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]Substituting into the equalization evaluation function, obtaining the optimal cluster group when the equalization evaluation function has the minimum value, and obtaining the maximum valueAnd (4) taking the optimal clustering group as the number of initial clustering centers, taking the clustering center corresponding to the optimal clustering group as the initial clustering center, and inputting the initial clustering center into an aggregation control algorithm to realize aggregation control. According to the technical scheme, when the equalization evaluation function has the minimum value, the initial clustering center and the number in the distributed energy storage aggregation control are correspondingly obtained.
See detailed description of the preferred embodiments.
Drawings
FIG. 1 is a topology diagram of an energy storage system architecture;
FIG. 2 is a graph of change in SSE value;
FIG. 3 is a graph of SSE value change;
FIG. 4 is a flow chart of the SA-Kmeans algorithm of the present invention;
FIG. 5 is a graph with a plot of adjustment factor equalization function versus change in value K of the present invention;
FIG. 6 is a scatter plot of SOC and power clustering under the present invention;
fig. 7 is a graph of SOC equalization control under the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the application, its application, or uses. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application, but the present application may be practiced in other ways than those described herein, and it will be apparent to those of ordinary skill in the art that the present application is not limited to the specific embodiments disclosed below.
Example 1:
as shown in fig. 4, the present invention discloses a method for controlling distributed energy storage aggregation, which comprises the following steps:
energy storage nodes DESNC are obtained, each including data of state of charge SOC and power.
Obtaining an upper limit K of cluster group up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the average value into an equalization evaluation function formula (12), obtaining an optimal cluster group when the equalization evaluation function has a minimum value, taking the optimal cluster group as the number of initial cluster centers, taking the cluster center corresponding to the optimal cluster group as the initial cluster center, and inputting an aggregation control algorithm to realize aggregation control.
Equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
The SA-Kmeans algorithm is obtained by organically combining the simulated annealing algorithm SA and the Kmeans algorithm, the intra-class distance and the inter-class distance are combined, the adjustment factor is introduced into an equalized evaluation function, and the early-stopping method and the elbow method are combined to obtain a method for reducing the size of a K value search space for obtaining an optimal clustering group; and obtaining the optimal initial clustering center through a simulated annealing algorithm SA.
The aggregation control algorithm itself is the prior art and is not described herein again.
Example 2:
the invention discloses a distributed energy storage aggregation control device which comprises an aggregation control module which is a program module.
The aggregation control module is used for acquiring the energy storage node DESNC and the clustering group upper limit K up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the initial clustering centers into an equalization evaluation function, obtaining an optimal clustering group when the equalization evaluation function has a minimum value, taking the optimal clustering group as the number of the initial clustering centers, taking the clustering center corresponding to the optimal clustering group as the initial clustering center, and inputting the initial clustering centers into a clustering control algorithm to realize clustering control.
Equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
Example 3:
the invention discloses a distributed energy storage aggregation control device which comprises a memory, a processor and a computer program which is stored in the memory and can run on the processor, wherein the steps of embodiment 1 are realized when the processor executes the computer program.
Example 4:
the present invention discloses a computer-readable storage medium storing a computer program which, when executed by a processor, implements the steps in embodiment 1.
The research and development conception is as follows:
as shown in fig. 1, in order to solve the above problem, the present invention uses an Energy storage system architecture of an Energy Storage Cloud Controller (ESCC) and a Distributed Energy Storage Node Controller (DESNC) in a wide area as a background, and performs aggregation control on the Energy storage node controller (DESNC) in the graph based on a Kmeans algorithm (K-means clustering algorithm).
The Kmeans algorithm belongs to a classical algorithm based on a partition algorithm, is simple in structure and easy to implement, and is widely applied to various fields. However, the Kmeans algorithm determines the initial cluster centers and the number of initial cluster centers in advance, which results in a limitation in the practical application effect of the Kmeans algorithm.
The technical problem to be solved is as follows:
energy storage node aggregation control based on an energy storage system architecture solves two defects of a traditional Kmeans algorithm in the application, and 1) the Kmeans algorithm is limited by the need of determining an initial clustering center in advance; 2) The Kmeans algorithm is also limited by the number of initial cluster centers that need to be determined in advance.
The purpose of the invention is as follows:
the method mainly comprises the steps of providing the SA-Kmeans algorithm to solve the problem of initial clustering centers and providing a balanced evaluation function with an adjusting factor to solve the problem of the number of the clustering centers. The method mainly aims to improve the accuracy of the clustering model and apply the clustering model to the classification of the energy storage nodes, so that the problems of uncertainty of an initial clustering center and uncertainty of the number of the initial clustering centers of the existing Kmeans algorithm are solved.
The technical contribution is as follows:
the invention point of the invention corresponds to the following implementation steps of step one, step two and step three. The invention organically combines an SA algorithm (simulated annealing algorithm) and a Kmeans algorithm, provides an SA-Kmeans algorithm, and finds the optimal initial clustering center through the characteristic of strong global optimizing capability of the SA algorithm, thereby improving the clustering effect. The invention introduces an equalization evaluation function by combining the intra-class distance and the inter-class distance, provides an equalization evaluation function with an adjustment factor, combines an early-stopping method and an elbow method, and provides a method for reducing the size of a K value search space. And finally, integrating the proposed method to realize an energy storage node aggregation control method based on an SA-Kmeans algorithm.
The method comprises the following steps:
the SA-Kmeans algorithm comprises the following steps:
inputting: SOC and power data of n energy storage nodes (DESNC), and a clustering group number K.
And (3) outputting: k clustering groups
1) To the initialization cooling schedule [ control temperature T 0 Markov chain Length L i Setting a temperature attenuation factor alpha, determining the number n of energy storage nodes (DESNC) participating in clustering and the number K of cluster groups, setting a function of an object as y = f (x), and randomly selecting a clustering center in a feasible solution space to solve x 0 And calculates an objective function value f (x) at that time 00 )。
In the formula (1), the reaction mixture is,
is the shortest distance, x, of the data represented by each energy storage node (DESNC) to the different cluster centers under its current grouping i～ Columns are SOC data, x ～j The columns are power data.
2) For which the temperature T is equal to the next value T in the cooling table i . (first iteration T = T) 0 )。
T i ＝T 0 ×α k (2)
In the formula (2), k is the number of iterations of the temperature in the simulated annealing.
3) At the current solution
Randomly generating a new solution
Calculating an objective function value of a solution
(on the first iteration)
)。
In the formula (3), the reaction mixture is,
is a new solution, T i Is the temperature, m ij Is a set of random numbers randomly generated K columns and 2 rows, where m ij Obey N (0,1).
4) And judging whether to accept the new solution.
4.1 If)
Then accept the new solution
4.2 If)
Then calculate
And calculate
Then randomly generating an in-range [0,1]Obeying uniformly distributed random numbers r, and if r < p, accepting a new solution
5) At a temperature T i Next, 3) and 4) are repeated with L i Next, the process is carried out.
6) Determining whether a stopping criterion, i.e. temperature T, is fulfilled i And (4) if the current value is less than the set threshold value, outputting K optimal clustering centers if the current value is met, and otherwise returning to the step (3) to continue iteration.
In the formula (4), the reaction mixture is,
is a current solution, T, near the center of the best cluster i Is the temperature, m ij Is a set of random numbers randomly generated K columns and 2 rows, where m ij Compliance with N (0,1), c i_best Is the optimal cluster center, c i_best_x Is the abscissa of the optimum cluster center, c i_best_y Is the ordinate of the center of the best cluster,
is an arbitrary clustering center in the SA algorithm.
7) Starting from the first energy storage node (DESNC), calculating the distances from the energy storage node to the K cluster centers, and classifying the energy storage node into a group which is close to the nearest cluster center until the classification of the n energy storage nodes is completed.
In the formula (5), d (x) ij ,c i_best ) Represents any sample in the cluster to the cluster center c i_best A distance of (C) i Is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i Cluster center of the cluster.
8) And (4) calculating and averaging the sum of the data of each grouped energy storage node, and replacing the clustering center in the step (7) with a new clustering center shown in a formula (6).
In the formula (6), the reaction mixture is,
is the average of the abscissa and ordinate of all samples in the cluster.
9) And 7) to 8) are repeated until a convergence condition is met, namely the position change of the clustering center in the two running processes is smaller than a preset value.
10 K cluster groups and K cluster centers are output.
In the formula (7), C i Is the ith cluster, any two clusters C i And C i+1 The samples of the energy storage nodes in the cluster are different, the sum of the number of the energy storage nodes in all the clusters is equal to n, and K cluster centers are 8) the cluster center of the latest iteration.
The method comprises the following steps:
theoretically, the K value should be selected from 1 to n, but the K value obtained according to the traditional elbow method often appears in the initial stage, and as the K value increases, the error sum of squares decreases slowly, the K value at this time is far larger than the optimal K value, once the K value exceeds the optimal value, and when the K value is calculated again, only the operation time is increased in vain, so that the search range of the K value needs to be narrowed.
Therefore, the invention proposes an equalized evaluation function with an adjustment factor for the selection of the value K.
The equalized evaluation function comparison emphasizes the following two aspects:
1. the interior of each cluster should be compact, and the greater the similarity within a cluster, the more compact this cluster is.
2. While the distance between the clusters should be as far as possible and the smaller the inter-cluster similarity, the farther the distance between the clusters.
The definition of the distance within a class is to calculate the distance of each sample of a class (cluster) to the center of its cluster.
In formula (8), L intra Is an intra-class distance, C i Is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i Cluster center of cluster, x i～ Columns are SOC data, x ～j The columns are power data.
The inter-class distance is defined as the distance between the centers of the clusters belonging to different classes (clusters).
In the formula (9), L inter Is the distance between the classes,
are the different cluster centers at the value of K.
The intra-class distance represents the compactness of each cluster under the clustering, the inter-class distance represents the distance between different clusters, the overall quality of the clustering is defined as the combination of the intra-class distance and the inter-class distance, and the balanced evaluation function is proposed according to the concept.
Defining the equalization evaluation function as the square and quadratic root of the intra-class distance and the inter-class distance:
in the formula (10), L intra Is an intra-class distance, L inter Is the inter-class distance.
When the equalized evaluation function is used, the imbalance in the class and between the classes can be effectively balanced, and when the equalized function reaches the minimum, the optimal clustering number is the optimal K value.
K＝min(J k ) (11)
In the equation (11), K is an optimum K value calculated by the equalization evaluation function.
However, the traditional equalization evaluation function has a defect: as the sample space increases, the intra-class distance becomes larger and the inter-class distance becomes almost constant. The K value in this case is usually not the optimal K value. In order to overcome the phenomenon, the invention provides an equalization evaluation function with an adjusting factor alpha, and the expression of the equalization evaluation function is as follows:
equation (12) is the equalization evaluation function with adjustment factors by calculating J at different K values k-best Let J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is of the type C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
Meanwhile, a method for reducing the size of a K value search space is provided by combining an early-stop method and an elbow method. If the SSE difference value corresponding to the K value is smaller than the set threshold value theta for two times, then K is determined at the moment up The size of the value is set to the upper limit of the K-value search range, while the lower limit is still 1.
In the formula (13), SSE is the sum of squares of errors, C i Is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is C i Number of samples of a cluster, c i_best Is C i Cluster center of the cluster.
In the formula (14), SSE i And SSE i+1 Is the sum of squares of errors, K, represented by different cluster groups up Is the upper limit of the K search range.
The method comprises the following third step:
the process based on the SA-Kmeans algorithm and the equalization evaluation function algorithm with the adjustment factor comprises the following steps:
inputting: data of n energy storage nodes (DESNC)
And (3) outputting: optimum K value
1) The step SA-Kmeans algorithm in the present invention is invoked, and K is calculated using formula (13) and formula (14) up The value is obtained.
2) The value of K goes to 1 to K up The cycle of (2).
2.1 Invoke the step-SA-Kmeans algorithm in the present invention and determine i initial cluster centers.
2.2 Calculate the intra-class distance L using equation (8) and equation (9), respectively intra And the distance L between classes inter The value is obtained.
2.3 ) loop execution 2.1 and 2.2 until the loop is complete.
3) The value of the adjustment factor alpha is calculated for different values of K and the value of alpha is determined.
4) And (5) obtaining an optimal K value by using the formula (12), and ending the algorithm.
The invention concept of the application is as follows:
when a user uses a Kmeans algorithm, the user needs to set an initial clustering center, but the user does not know the initial clustering center in advance, and then a random setting mode can be adopted, but because the Kmeans objective function is not convex and may contain a plurality of local minimum values, the local minimum and the local maximum are easy to be trapped in the process of solving the objective function, and the obtained clustering effect is not ideal.
Therefore, the selection of the initial clustering center is very important, and in order to solve the problem of initializing the clustering center, the simulated annealing algorithm (SA) is adopted to solve the problem.
The SA has the advantages of strong global optimizing capability and capability of well solving the problem that the Kmeans is easy to sink into local minimum during clustering. Furthermore, the invention provides a novel SA-Kmeans hybrid algorithm to optimize the Kmeans algorithm.
The Kmeans algorithm has another problem, namely the selection of the K value, and in order to solve the problem, a plurality of methods are proposed by scholars, and the widely used method is the elbow method. However, the result graph of the elbow method may show an insignificant possibility of an optimal K value, and the elbow method considers only the intra-class distance.
According to the defects of the elbow method, the equalized evaluation function is adopted to calculate the optimal K value, but the equalized evaluation function has a defect that the intra-class distance is larger and larger along with the increase of the sample space, and the inter-class distance is almost unchanged, so that the obtained K value is not the optimal K value.
In order to overcome the phenomenon, the invention provides an equalization evaluation function with an adjusting factor. Meanwhile, the thought of the early-stopping method is introduced, the early-stopping method and the elbow method are combined, and a method for reducing the size of the K value search space is provided.
Description of the technical scheme:
aiming at two defects of the Kmeans algorithm: 1) The Kmeans algorithm is limited by the need to determine the initial clustering center in advance; 2) The Kmeans algorithm is also limited by the number of initial cluster centers that need to be determined in advance. The method mainly aims to improve the accuracy of a clustering model and apply the clustering model to the classification of energy storage nodes, so that the problems of uncertainty of the initial clustering centers and uncertainty of the number of the clustering centers of the existing Kmeans algorithm are solved.
When the Kmeans algorithm is used, a user needs to set an initial clustering center by himself, when the user does not know the initial clustering center, a random setting mode can be adopted, but because the Kmeans objective function is not convex, a plurality of local minimum values can be contained, and therefore the local minimum and the local maximum can be easily caused in the process of solving the objective function. Therefore, the selection of the initial clustering center is particularly important, and in order to solve the problem of initializing the clustering center, the simulated annealing algorithm (SA) is adopted to solve the problem.
The SA has the advantages of strong global optimizing capability and capability of well solving the problem that the Kmeans clustering is easy to fall into local minimum, but has the defect of slow searching speed. In practical application, the parameters of the SA algorithm need to be adjusted properly through an empirical method, so that the convergence rate of the algorithm can be increased, the global optimization capability of the algorithm can be improved, and if the parameters are not adjusted properly, the search of the SA becomes very slow.
The Kmeans algorithm has another problem, namely the selection of the K value, and many scholars propose a plurality of methods for solving the problem, wherein the widely used method is the elbow method, and the objective function of the elbow method is SSE, namely the sum of squares of errors.
In the formula (13), SSE is the sum of squares of errors, C i Is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is C i Number of samples of a cluster, c i_best Is C i Cluster center of the cluster.
The elbow method has the core idea: as the K value is increased, the sample division is more fine, the aggregation degree of each cluster is increased, and the square sum of errors is gradually reduced.
When K is smaller than the true clustering number, the increase of K can improve the aggregation degree of each cluster, the descending amplitude of SSE is larger at the moment, when K reaches the true clustering number and is increased, the obtained aggregation degree is reduced, the descending amplitude of SSE is slowed down at the moment, and then the descending amplitude tends to be flat along with the continuous increase of the K value.
The method is called elbow method because the SSE and K relation graph is the shape of an elbow, and the K value corresponding to the elbow is the real clustering number of the data.
However, this method has limitations and is not applicable in all situations where the elbow may be present in some particular situations.
As shown in fig. 2, the optimal K value of fig. 2 is 2.
As shown in fig. 3, however, fig. 3 cannot read out the optimal K value, and at this time, when the K value is selected, misjudgment occurs, so that the true K value cannot be obtained, and the elbow method only considers the intra-class distance and does not consider the inter-class distance.
Combining the ideas, an SA-Kmeans algorithm is provided to solve the problem of the initial clustering center, and the operation idea of the hybrid algorithm is that the SA algorithm is used to find the optimal clustering center at the initial stage of the hybrid algorithm, and when the temperature T is higher than the temperature T i Outputting K optimal clustering centers when the stopping criterion is met, switching to a Kmeans algorithm, finishing clustering grouping by using the Kmeans algorithm, and outputting K optimal clustering centers when the change of the clustering centers meets the convergence conditionClustering groups and optimal clustering centers, and exiting the SA-Kmeans algorithm.
FIG. 4 shows a flowchart of the SA-Kmeans algorithm.
Aiming at the defects of the elbow method, a balance evaluation function with an adjusting factor is provided to solve the problem of the number of clustering centers.
Equation (12) is the equalization evaluation function with adjustment factors by calculating J at different K values k-best Let J k-best Is the optimal K value, alpha is the adjustment factor,
α∈[0，1]，C i is the ith cluster, x ij Is of the type C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
In order to reduce the amount of computation, a method of reducing the size of the K value search space by combining the early-stop method and the elbow method has been proposed.
In the formula (13), SSE is the sum of squares of errors, C i Is the ith cluster, x ij Is of the type C i Sample of a cluster, n j Is C i Number of samples of a cluster, c i_best Is C i Cluster center of the cluster.
In the formula (14), SSE i And SSE i+1 Is the sum of squares of errors, K, represented by different cluster groups up Is the upper limit of the K search range.
The method has the advantages that the existing Kmeans algorithm is improved from two defects, and the method is successfully applied to the classification of the energy storage cloud platform energy storage nodes (DESNC).
1. The SA-Kmeans algorithm is provided by organically combining the SA algorithm and the Kmeans algorithm, and the optimal initial clustering center is found through the characteristic of strong global optimizing capability of the SA algorithm, so that the clustering effect is improved.
2. The invention introduces an equalization evaluation function by combining the intra-class distance and the inter-class distance, provides an equalization evaluation function with an adjusting factor, and simultaneously combines an early-stopping method and an elbow method to provide a method for reducing the size of a K value search space.
3. The two improved algorithms are not only suitable for the classification of the energy storage nodes, but also can be expanded into other fields.
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail with reference to the accompanying drawings.
As shown in FIG. 1, the experimental platform used in the invention, the clustering object realized by the invention is the energy storage node (DESNC) in the graph.
The upper energy storage cloud control layer is mainly used for receiving scheduling information, electricity price and other information of a third party, and the SOC and the power of the energy storage nodes (DESNC) of the lower layer participate in aggregation control to achieve uniform regulation and control in a wide area range.
The middle-layer energy storage node control layer is responsible for receiving scheduling information of the energy storage cloud, sending instructions to the distributed energy storage modules in real time and underground, calculating real-time SOC and power of the energy storage nodes (DESNC) according to data (SOC and power) uploaded by the Distributed Energy Storage Modules (DESM), and simultaneously participating in aggregation control of the energy storage cloud by the two data of the SOC and the power of the energy storage nodes, wherein the number of the energy storage nodes participating in aggregation control is 100.
The bottom distributed energy storage module control layer is mainly composed of Distributed Energy Storage Modules (DESM), uploads self data (SOC and power) to an energy storage node (DESNC) to which the distributed energy storage modules belong every 15min, and is applied to peak clipping and valley filling of a power grid.
As shown in fig. 5 to 7, an energy storage node aggregation control method based on an SA-Kmeans algorithm is implemented, and includes the following steps:
A. determining an optimal clustering number K value of energy storage nodes
By using the SA-Kmeans algorithm and the equalization evaluation function algorithm with the adjustment factor in the third step of the technical scheme, 100 data sets containing the SOC and the power energy storage node (DESNC) are input, and the optimal K value of 4 can be obtained by running the algorithm.
As shown in fig. 5, for convenience of observation, a graph of the evaluation function of equalization with adjustment factor is plotted. The intra-class distance is a graph processed by an adjustment factor, and the optimal K value is 4.
B. And clustering the energy storage nodes participating in aggregation control according to the optimal clustering number K value obtained from the part A.
By using the SA-Kmeans algorithm in the first step of the technical scheme, 100 data sets containing SOC and power storage nodes (DESNC) and the optimal cluster number of 4 are input, and 4 cluster groups and 4 cluster centers of the 4 cluster groups can be obtained by running the algorithm.
As shown in fig. 6, for convenience of observation, the effect graph after clustering is plotted, with the horizontal axis representing the state of charge SOC and the vertical axis representing the power.
C. Evaluation of A, B two parts
After the clustering is completed, it is important to evaluate the obtained clustering model by adopting a reasonable evaluation index. And evaluating the quality of the clustering model into an internal evaluation index and an external evaluation index. However, the external evaluation index is often used only by the result of known clustering, and has certain limitation, so the invention adopts the internal evaluation index for evaluation.
The internal evaluation index is the contour coefficient adopted by the invention, and the formula is as follows
In equation (15), a (i) represents the average distance between the sample and the other samples in the cluster in which the sample is located, and b (i) represents the average distance between the sample and the other cluster samples.
The contour coefficient S (i) takes the value of [ -1,1]. If the contour coefficient is negative, the cluster-in distance is larger than the cluster-between distance, and the clustering result is very bad.
The overall profile coefficient SC is
In equation (16), SC is the total profile coefficient, and n is the number of energy storage nodes.
The overall contour coefficient is 0.5571, and the average level of the whole is over 0.5 from the overall contour coefficient, so that the improved algorithm has strong practicability.
D. And clustering the energy storage nodes at the cloud end of the energy storage cloud platform by adopting an SA-Kmeans algorithm, and performing charging and discharging verification on the distributed energy storage modules by taking one day as a period.
From the viewpoint of protecting the service life of the Distributed Energy Storage Module (DESM), the mode of regulating and controlling the Distributed Energy Storage Module (DESM) adopted by the invention is a mode of charging and discharging once a day.
The form of charging and discharging at one time in one day is as follows: in one period of one day, the energy storage cloud terminal issues a charging instruction to the energy storage node at 0-6 hours, the energy storage cloud terminal issues a discharging instruction to the energy storage node at 10-12 hours, and the energy storage cloud terminal issues a discharging instruction to the energy storage node at 18-22 hours.
And at 0-6, taking 15min as a small period, continuously clustering the SOC and the power of the energy storage nodes (DESNC) by adopting an SA-Kmeans algorithm, and sending the power required to be consumed by the power grid to each energy storage node (DESNC) by the energy storage cloud according to the capacity of the energy storage node (DESNC) in a proportional relation according to the scheduling plan of the power grid on the day. The energy storage node (DESNC) performs SOC balance control according to the SOC and power relation of all Distributed Energy Storage Modules (DESM) of the node, wherein the power distribution relation of the SOC and the power is as follows:
in the formula (17), the compound represented by the formula (I),
the energy storage cloud is issued to each energy storage node (DESNC) i ) The power scheduling information of (a) the power scheduling information,
is DESM ij Power reference of, K scc_ij Is DESM ij The middle layer control coefficient of (1).
The invention adopts SOC balance control to prevent the overcharge/overdischarge phenomenon of the Distributed Energy Storage Module (DESM) caused by SOC unbalance control, thereby endangering the service life of the Distributed Energy Storage Module (DESM).
In a charging state, in order to ensure that the output power of a Distributed Energy Storage Module (DESM) does not exceed the allowable value in the SOC equalization process, an SOC equalization control strategy based on rated charging power and an SOC mean value is adopted, and in the charging state, a middle-layer control coefficient K scc_ij Comprises the following steps:
in formula (18), P rateij_ch For distributed energy storage modules DESM ij Rated charging power, SOC ij For distributed energy storage modules DESM ij M is the current energy storage node (DESNC) i ) Lower energy storage module (DESM) ij ) The total number.
At 10-12 and 18-22, still taking 15min as a small period to store energyNode (DESNC) i ) The SOC and the power are uninterruptedly clustered by adopting an SA-Kmeans algorithm, and the power required by the peak regulation of the power grid at the moment is transmitted to each energy storage node (DESNC) by the energy storage cloud according to the capacity of the energy storage node in a proportional relation i ) Above.
In the discharged state, each Distributed Energy Storage Module (DESM) ij ) The corresponding middle layer coefficient is changed, and the middle layer coefficient is as follows:
in the formula (19), P rateij_dis Is DESM ij Rated discharge power, SOC ij Is DESM ij The state of charge of (a). And m is the total number of the energy storage modules under the current node.
As shown in fig. 6, the clustered 100 energy storage nodes are divided into 4 groups, i.e., (1) low SOC low power, (2) low SOC high power, (3) high SOC low power, and (4) high SOC high power. In the charging state, the principle of high SOC low SOC and low SOC high charge is implemented, so the order of the groups is (1) → (2) → (3) → (4), and in the discharging state, the principle of high SOC high discharge and low SOC low discharge is implemented, and the order of the groups is (4) → (3) → (2) → (1). The Distributed Energy Storage Module (DESM) can be mobilized as much as possible, various conditions of the power grid can be flexibly met, and the service life of the Distributed Energy Storage Module (DESM) can be prolonged.
Now for convenience of observation, a storage node (DESNC) is used i ) Two lower Distributed Energy Storage Modules (DESM) ij ) Verification, two Distributed Energy Storage Modules (DESM) ij ) Has the following parameters P ratei1_ch ＝P ratei2_ch ＝2kW，P ratei1_dis ＝P ratei2_dis And =2kW, and the charging power is set to be positive and the discharging power is set to be negative. Two Distributed Energy Storage Modules (DESM) ij ) Receiving a charging command at 0-6 power consumption valley period, receiving a discharging command at 10-12 power consumption peak period, and receiving a discharging command at 18-22 point at the second power consumption peak periodAnd (4) an electric instruction.
As shown in FIG. 7, the horizontal axis is time, the left vertical axis is power, and the right vertical axis is state of charge SOC, and during these three periods, the energy storage node (DESNC) i ) Two subordinate Distributed Energy Storage Modules (DESM) ij ) The charging power and the discharging power can be flexibly adjusted according to the SOC of the user, and the rated power of the user cannot be exceeded.
After the application runs secretly for a period of time, the feedback of field technicians has the advantages that:
the invention improves two defects of the existing Kmeans algorithm and is successfully applied to the aggregation control of the energy storage cloud platform energy storage node (DESNC).
1) The SA-Kmeans algorithm is provided by organically combining the SA algorithm and the Kmeans algorithm, and the optimal initial clustering center is found through the characteristic of strong global optimizing capability of the SA algorithm, so that the clustering effect is improved.
2) The invention introduces an equalization evaluation function by combining the intra-class distance and the inter-class distance, provides an equalization evaluation function with an adjustment factor, combines an early-stopping method and an elbow method, and provides a method for reducing the size of a K value search space.
3) The two improved algorithms are not only suitable for the classification of the energy storage nodes, but also can be expanded into other fields.
At present, the technical scheme of the invention has been subjected to a pilot plant test, namely a small-scale test of the product before large-scale mass production; after the pilot plant test is finished, user use investigation is developed in a small range, and the investigation result shows that the user satisfaction is high; the preparation of products for formal production for industrialization (including intellectual property risk early warning research) has been started.
Claims (10)
1. A method for controlling distributed energy storage polymerization is characterized in that: the method comprises the steps of aggregation control, energy storage node DESNC is obtained, and clustering group upper limit K is obtained up Value, cluster group K from 1 to cluster group upper limit K up The values are respectively calculated to obtain the clustering centers and the faciesCorresponding inter-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the initial clustering centers into an equalization evaluation function, obtaining an optimal clustering group when the equalization evaluation function has a minimum value, taking the optimal clustering group as the number of the initial clustering centers, taking the clustering center corresponding to the optimal clustering group as the initial clustering center, and inputting the initial clustering centers into a clustering control algorithm to realize clustering control.
2. The method of claim 1, wherein:
equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
C i is the ith cluster, x ij Is a member of C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
3. The method of claim 1, wherein: the SA-Kmeans algorithm is obtained by organically combining the simulated annealing algorithm SA and the Kmeans algorithm.
4. The method of claim 3, wherein: and combining the intra-class distance and the inter-class distance, introducing an adjusting factor into an equalized evaluation function, and combining an early-stopping method and an elbow method to obtain a method for reducing the size of a K value search space for obtaining an optimal clustering group.
5. The method of claim 3, wherein: and obtaining the optimal initial clustering center through a simulated annealing algorithm SA.
6. The method of claim 1, wherein: each energy storage node DESNC includes data for state of charge SOC and power.
7. A distributed energy storage polymerization control device is characterized in that: the cluster system comprises an aggregation control module for obtaining an energy storage node DESNC and obtaining a cluster group upper limit K up Value, cluster group K from 1 to cluster group upper limit K up Respectively calculating values to obtain a cluster center and a corresponding intra-class distance L intra And the distance L between classes inter Obtaining an adjustment factor alpha based on the value of the adjustment factor under each cluster group, wherein the alpha belongs to [0,1 ]]And substituting the average evaluation function into the average evaluation function, obtaining an optimal cluster group when the average evaluation function has a minimum value, taking the optimal cluster group as the number of initial cluster centers, taking the cluster center corresponding to the optimal cluster group as the initial cluster center, and inputting an aggregation control algorithm to realize aggregation control.
8. The distributed energy storage aggregation control device according to claim 7, wherein: in the aggregation control module, the control unit is configured to control the aggregation of the plurality of cells,
equation (12) is the equalization evaluation function with adjustment factors by calculating J at each K value k-best ，J k-best Is the optimal K value, alpha is the adjustment factor,
C i is the ith cluster, x ij Is of the type C i Sample of a cluster, n j Is the number of samples of the cluster, c i_best Is C i The cluster center of the cluster is determined,
are the different cluster centers at the value of K.
9. An apparatus for distributed energy storage aggregation control, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: the processor when executing the computer program realizes the corresponding steps in any of claims 1 to 6.
10. An apparatus for distributed energy storage aggregation control comprising a computer readable storage medium storing a computer program, characterized in that: the computer program when executed by a processor implements the corresponding steps of any of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211383852.4A CN115758193A (en) | 2022-11-07 | 2022-11-07 | Distributed energy storage aggregation control method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211383852.4A CN115758193A (en) | 2022-11-07 | 2022-11-07 | Distributed energy storage aggregation control method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115758193A true CN115758193A (en) | 2023-03-07 |
Family
ID=85356913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211383852.4A Pending CN115758193A (en) | 2022-11-07 | 2022-11-07 | Distributed energy storage aggregation control method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115758193A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118011835A (en) * | 2024-04-08 | 2024-05-10 | 唐山阿诺达自动化有限公司 | Blast furnace ironmaking control system based on machine learning |
| CN118353072A (en) * | 2024-06-18 | 2024-07-16 | 中建安装集团有限公司 | Distributed energy storage aggregation scheduling control system and method based on artificial intelligence |
| CN118353072B (en) * | 2024-06-18 | 2024-10-15 | 中建安装集团有限公司 | Distributed energy storage aggregation scheduling control system and method based on artificial intelligence |
-
2022
- 2022-11-07 CN CN202211383852.4A patent/CN115758193A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118011835A (en) * | 2024-04-08 | 2024-05-10 | 唐山阿诺达自动化有限公司 | Blast furnace ironmaking control system based on machine learning |
| CN118353072A (en) * | 2024-06-18 | 2024-07-16 | 中建安装集团有限公司 | Distributed energy storage aggregation scheduling control system and method based on artificial intelligence |
| CN118353072B (en) * | 2024-06-18 | 2024-10-15 | 中建安装集团有限公司 | Distributed energy storage aggregation scheduling control system and method based on artificial intelligence |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109165788B (en) | Optimization method of combined cooling heating and power system | |
| WO2023274425A1 (en) | Multi-energy capacity optimization configuration method for wind-solar-water-fire storage system | |
| CN111191820B (en) | Site selection and volume fixation optimization planning method for energy storage device in comprehensive energy system | |
| CN108206543B (en) | Energy router based on energy cascade utilization and operation optimization method thereof | |
| CN109066800B (en) | Power distribution network flexible load cooperative scheduling method based on particle swarm algorithm | |
| CN111340274A (en) | Virtual power plant participation-based comprehensive energy system optimization method and system | |
| CN109799851B (en) | User side load response method based on load cluster temperature regulation and control | |
| CN109245093A (en) | A kind of supply of cooling, heating and electrical powers distributed busbar protection collaboration Optimization Scheduling | |
| CN108471143A (en) | Micro-grid multi-energy method for optimizing scheduling based on positive and negative feedback particle cluster algorithm | |
| CN115640982B (en) | Pumped storage priority regulation-based day-ahead optimal scheduling method for multi-energy complementary system | |
| CN113255198B (en) | Multi-objective optimization method for combined cooling heating and power supply micro-grid with virtual energy storage | |
| CN109861302A (en) | A kind of energy internet based on leader-followers games optimal control method a few days ago | |
| CN114529075A (en) | Comprehensive energy system distribution robustness optimization scheduling method considering wind and light prediction error | |
| CN111860950A (en) | Probability assessment method for power adjustability of virtual power plant | |
| CN110445127A (en) | A kind of var Optimization Method in Network Distribution and system towards multiple stochastic uncertainty | |
| CN115758193A (en) | Distributed energy storage aggregation control method and device | |
| CN115187018A (en) | Double-layer optimized scheduling method and device for park comprehensive energy system | |
| CN114970096A (en) | Power system flexibility quantitative evaluation method and system based on probability optimal power flow | |
| CN114861521A (en) | Method for capacity optimization configuration of combined cooling heating and power type microgrid | |
| CN112001639B (en) | Adjustable capability assessment method for comprehensive energy system energy demand and storage medium | |
| CN117375048A (en) | Power distribution method for multi-objective optimized energy storage system | |
| CN110266059B (en) | Novel energy supply system optimal configuration method based on triangular comprehensive evaluation | |
| CN115940284B (en) | Operation control strategy of new energy hydrogen production system considering time-of-use electricity price | |
| CN115912421A (en) | Power distribution network energy storage site selection constant-volume multi-objective optimization method and system | |
| CN115796533A (en) | Virtual power plant double-layer optimization scheduling method and device considering clean energy consumption |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |