Distributed power supply cluster scheduling method based on NJW spectral clustering algorithm

Abstract

The invention discloses a distributed power supply cluster scheduling method based on an NJW spectral clustering algorithm, and belongs to the field of resource scheduling. The objective of the invention is to solve the problem that an existing clustering mode is limited and complex in application range. The method comprises the following steps: 1, constructing each distributed power supply model; 2, dividing the distributed power supply models with similar characteristics into one clustering device by adopting a clustering algorithm, and obtaining clustering parameters in each clustering device; 3, according to the power track of the virtual power plant, establishing a power consumption objective function for clustering parameters in each clustering device; and step 4, solving the power consumption objective function, and obtaining the minimum consumed power of each clustering device, so that any clustering device is scheduled according to the power scheduling requirement of the virtual power plant, and the scheduling of the distributed power supply is realized. The method is used for scheduling the distributed power supply cluster.

Description

Technical Field

[0001] The invention relates to the scheduling of distributed power sources and belongs to the field of resource scheduling. Background Art

[0002] In actual engineering applications, the dispatching structure of virtual power plants is divided into three layers: the main grid, virtual power plant operators, and distributed power generation resources. The process of using virtual power plants to manage distributed power sources is as follows: Figure 2 shown.

[0003] Faced with the recent bidding of massive distributed power sources (distributed resources) for virtual power plants, in order to ensure that the dispatching power instructions are fairly distributed among the distributed units and maximize the power boundary range after aggregation, some scholars have proposed a method of "achieving fair power distribution among energy storage resources while maintaining the premise that all distributed energy storage units respond to the dispatching power instructions." This method is only applicable to equipment with energy storage properties, such as batteries, electric vehicles and other equipment. The brief process of this method is as follows:

[0004] Assuming that there are n energy storage units in the cluster, taking the i-th energy storage resource as an example, the fairness of power allocation among energy storage resources is quantitatively described as follows.

[0005] 1) When the device is charging

[0006] Assume that at the beginning of the scheduling time t0, the SOC of a distributed energy storage unit i is Assume that it is at a constant power P i C If charging is carried out, then its SOC value at time t is for:

[0007]

[0008] Where: S i is the total capacity of device i.

[0009] The time required to charge to the maximum state Calculated by the following formula

[0010]

[0011] Where: is the maximum SOC of energy storage unit i.

[0012] 2) When the device is discharged

[0013] It is also assumed that the SOC of distributed energy storage unit i is Assuming a constant power Discharge, then its SOC value at time for:

[0014]

[0015] The time required to discharge to the minimum SOC Calculated by the following formula

[0016]

[0017] Where: is the minimum SOC of energy storage unit i.

[0018] According to the proposed principle of power allocation fairness, the proposed distributed scheduling strategy aims to "satisfy the output of each distributed energy storage unit within its power boundary and realize the or Same". For ease of handling, remove the style.

[0019]

[0020] Where: are the consensus variables during charging and discharging respectively; W i c is the difference between the maximum energy of energy storage unit i and the energy at time t0; W i d is the difference between the energy of energy storage unit i at time t0 and the minimum energy.

[0021] The dispatching goal is to achieve the optimal dispatching of energy storage resources while maintaining the response of all distributed energy storage units to dispatching power commands. or As similar as possible". The mathematical model for charging is shown below.

[0022]

[0023] Where: c is the cluster consensus variable during charging; P i min is the power lower limit of energy storage unit i; P i max is the upper power limit of energy storage unit i; P c It is the power scheduling instruction during charging, which is a positive number.

[0024] The following is the mathematical model for discharge.

[0025]

[0026] Where: d is the cluster consensus variable during charging; P d It is the power dispatch instruction during discharge, which is a negative number.

[0027] The first line of the above two equations is to realize the energy storage unit As much as possible, the second and third rows are used to correct the consensus variables of the energy storage resources with power exceeding the limit after consensus, and the last row is to ensure that the total charging and discharging power of all energy storage units can accurately respond to the dispatching instructions. The solution algorithm is mainly solved through the consistency algorithm.

[0028] The advantage of the above method is that it can accurately determine the scheduling plan of each device, and its power instruction is highly accurate. However, in order to implement the consistency algorithm, a large number of communication equipment needs to be installed, which greatly increases the investment cost of the equipment. In addition, this method is only applicable to equipment with energy storage properties, and is no longer applicable to the scheduling of distributed resources such as wind turbines and photovoltaics. Therefore, the existing method has a limited scope of application and is complex. Summary of the invention

[0029] The purpose of the present invention is to solve the problems of limited application scope and complexity of existing clustering methods, and propose a distributed power cluster scheduling method based on NJW spectral clustering algorithm.

[0030] A distributed power cluster scheduling method based on NJW spectral clustering algorithm, the method comprising the following contents:

[0031] Step 1: Build each distributed power model;

[0032] Step 2: using a clustering algorithm to divide the distributed power models with similar characteristics into a clusterer, and obtaining clustering parameters in each clusterer;

[0033] Step 3: According to the power trajectory of the virtual power plant, a power consumption objective function is established for the clustering parameters in each clusterer;

[0034] Step 4: Solve the power consumption objective function to obtain the minimum power consumption of each clusterer, so as to schedule any clusterer according to the power scheduling requirements of the virtual power plant and realize the scheduling of distributed power sources.

[0035] Preferably, in step 1, each distributed power supply model is expressed as:

[0036]

[0037] Where P L,j,t is the power demand of distributed load j; P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,t is the energy demand of distributed generation j; E L,j,min-de and E L,j,max-de are the lower and upper limits of the energy demand of distributed resource j, respectively; T start,L,jand T end,L,j are the demand start time and end time of distributed generation j respectively; ΔT is the scheduling time interval, and t represents time t.

[0038] Preferably, in step 2, a clustering algorithm is used to classify distributed power supply models with the same characteristics into a clusterer, and the specific process is:

[0039] Step 21: Perform a first normalization process on each distributed power source model to obtain all normalized distributed power source data;

[0040] Step 22: Use Euclidean distance to measure the differences between all normalized distributed power sources to obtain a difference matrix;

[0041] Step 23, using the Gaussian kernel function to convert the difference matrix into an adjacency matrix;

[0042] Step 24: perform a second normalization process on the adjacency matrix to obtain a normalized adjacency matrix. According to a given number of aggregators, split the normalized adjacency matrix into matrices of a preset number of aggregators, and divide the split matrices into corresponding aggregators.

[0043] Step 25: Use the K-means algorithm to cluster the matrix in each aggregator to obtain the clustering parameters in each clusterer.

[0044] Preferably, in step 21, the normalized distributed power supply data is expressed as:

[0045]

[0046] Where N L,i is the number of distributed resources at node i; x L,j is the original demand parameter of distributed resource j; x′ L,j is the normalized demand parameter of distributed resource j.

[0047] Preferably, in step 22, the difference matrix D(i,j) is expressed as:

[0048]

[0049] Where α1, α2, α3, α4, α5 and α6 are attribute sets, P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; P L,i,min and P L,i,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,min and E L,j,max are the lower and upper limits of the energy demand of distributed resource j respectively; E L,i,minand E L,i,max are the lower and upper limits of the energy demand of distributed resource i respectively; T start,L,j and T end,L,j are the starting time and ending time of the demand of distributed generation j, T start,L,i and T end,L,i are the start time and end time of the demand of distributed generation i respectively.

[0050] Preferably, in step 23, the adjacency matrix K(i,j) is expressed as:

[0051]

[0052] In the formula, is the Gaussian kernel function parameter.

[0053] Preferably, the clustering parameters in each clusterer are expressed as:

[0054]

[0055] Where N L,k,i is the number of distributed resources inside aggregator k at node i; and The starting demand time and the ending demand time of aggregator k at node i are respectively; and are the upper and lower power demand limits of aggregator k at node i, respectively; and are the upper and lower bounds of energy demand of aggregator k at node i, respectively.

[0056] Preferably, in step 3, the power consumption objective function is expressed as:

[0057]

[0058] In the formula, is the actual power consumption of aggregator k at node i at time t; N AGG,i is the number of aggregators at node i, and M is the coefficient.

[0059] The beneficial effects of the present invention are:

[0060] The scheduling method proposed in the present invention can effectively reduce the data dimension by constructing an adjacency matrix, splitting the normalized adjacency matrix into multiple spatial matrices, and dividing the split spatial matrices into corresponding aggregators. At the same time, the method can fully consider the characteristics of various resources themselves. Compared with methods such as kmeans that only rely on the Euclidean distance between resource operation curves, by constructing features that can characterize resource characteristics, it can effectively consider the temporal correlation between resources, greatly improve the accuracy of clustering results, and effectively improve the accuracy and rationality of aggregation. In addition, the present application has a wide range of applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a distributed power cluster aggregation algorithm based on NJW spectral clustering algorithm;

[0062] Figure 2 This is a flow chart of distributed power supply management and control. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0065] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0066] Example:

[0067] Combination Figure 1 This embodiment describes a distributed power cluster scheduling method based on the NJW spectral clustering algorithm, the method comprising the following contents:

[0068] Step 1: Build each distributed power model;

[0069] Step 2: using a clustering algorithm to divide the distributed power models with similar characteristics into a clusterer, and obtaining clustering parameters in each clusterer;

[0070] Step 3: According to the power trajectory of the virtual power plant, a power consumption objective function is established for the clustering parameters in each clusterer;

[0071] Step 4: Solve the power consumption objective function to obtain the minimum power consumption of each clusterer, so as to schedule any clusterer according to the power scheduling requirements of the virtual power plant and optimize the scheduling of distributed resources.

[0072] Through the distributed load aggregator, massive distributed resources can provide huge distribution to the virtual power plant agent without increasing excessive computing burden. This embodiment clusters massive distributed resources with similar characteristics into the same group to form an aggregator to participate in the scheduling and operation of the virtual power plant. The demand parameters of the distributed resources can be obtained based on historical statistical data or probability distribution.

[0073] In this embodiment, in step 1, each distributed power supply model is expressed as:

[0074]

[0075] Where P L,j,t is the power demand of distributed load j; P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,t is the energy demand of distributed generation j; E L,j,min-de and E L,j,max-de are the lower and upper limits of the energy demand of distributed resource j, respectively; T start,L,j and T end,L,j are the demand start time and end time of distributed generation j respectively; ΔT is the scheduling time interval, and t represents time t.

[0076] For each distributed resource, it has a series of parameters to reflect its demand information. The general model of distributed resources can be expressed as Formula 1.

[0077] In this embodiment, in step 2, a clustering algorithm is used to divide distributed power supply models with the same characteristics into a clusterer. The specific process is as follows:

[0078] Step 21: Perform a first normalization process on each distributed power source model to obtain all normalized distributed power source data;

[0079] Step 22: Use Euclidean distance to measure the differences between all normalized distributed power sources to obtain a difference matrix;

[0080] Step 23, using the Gaussian kernel function to convert the difference matrix into an adjacency matrix;

[0081] Step 24: perform a second normalization process on the adjacency matrix to obtain a normalized adjacency matrix. According to the given number of aggregators, split the normalized adjacency matrix into multiple space matrices, and divide each space matrix into a corresponding aggregator.

[0082] Step 25: Use the K-means algorithm to cluster the matrix in each aggregator to obtain the clustering parameters in each clusterer.

[0083] After the modeling is completed, each resource needs to be clustered to form a distributed resource cluster to participate in power grid dispatching. The NJW spectral clustering algorithm is used to divide distributed resources into different groups according to their characteristics.

[0084] The normalized adjacency matrix is ​​split into matrices with a preset number of aggregators. According to the design requirements, several maximum eigenvalues ​​in the normalized adjacency matrix are first taken to form a space matrix, and then several eigenvalues ​​are taken from other eigenvalues ​​in the normalized adjacency matrix to form a space matrix, and finally classification is achieved.

[0085] In this embodiment, in step 21, the normalized distributed power supply data is expressed as:

[0086]

[0087] Where N L,i is the number of distributed resources at node i; x L,j is the original demand parameter of distributed resource j; x′ L,j is the normalized demand parameter of distributed resource j.

[0088] In this embodiment, in step 22, the difference matrix D(i,j) is expressed as:

[0089]

[0090] Where α1, α2, α3, α4, α5 and α6 are attribute sets, P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; P L,i,min and P L,i,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,min and E L,j,max are the lower and upper limits of the energy demand of distributed resource j respectively; E L,i,min and E L,i,max are the lower and upper limits of the energy demand of distributed resource i respectively; T start,L,j and T end,L,j are the starting time and ending time of the demand of distributed generation j, T start,L,i and T end,L,i are the start time and end time of the demand of distributed generation i respectively.

[0091] In this embodiment, in step 23, the adjacency matrix K(i,j) is expressed as:

[0092]

[0093] In the formula, is the Gaussian kernel function parameter.

[0094] In this embodiment, the clustering parameters in each clusterer are expressed as:

[0095]

[0096] Where N L,k,i is the number of distributed resources inside aggregator k at node i; and The starting demand time and the ending demand time of aggregator k at node i are respectively; and are the upper and lower power demand limits of aggregator k at node i, respectively; and are the upper and lower bounds of energy demand of aggregator k at node i, respectively.

[0097] According to the distributed resource clustering results, for each distributed resource aggregator, the demand parameter information of the aggregator can be obtained according to all its internal distributed resource parameters.

[0098] In this embodiment, in step 3, the power consumption objective function is expressed as:

[0099]

[0100] In the formula, is the actual power consumption of aggregator k at node i at time t; N AGG,i is the number of aggregators at node i, and M is the coefficient.

[0101] Each aggregator formulates an operation plan for internal distributed resources based on the reference power trajectory obtained in the pre-optimization phase of the virtual power plant. For a specific aggregator k, there are two optimization goals: Goal 1: The overall output power of the aggregator should follow the reference power trajectory F1 obtained in the pre-optimization phase as much as possible. DIS (P L,j,t ); Goal 2: Reduce the number of starts and stops of each distributed resource and promote the stable operation of distributed resources In addition, optimization goal 1 has a higher priority than optimization goal 2.

[0102] In addition, the power consumption of each distributed resource at each moment should meet its demand model; this optimization model is a typical quadratic programming model and can be directly solved using CPLEX.

[0103] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the present invention. It should therefore be understood that many modifications may be made to the exemplary embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in a manner different from that described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be used in other described embodiments.

Claims

1. A distributed power cluster scheduling method based on NJW spectral clustering algorithm, characterized in that: The method comprises the following: Step 1: Build each distributed power model; Step 2: using a clustering algorithm to divide the distributed power models with similar characteristics into a clusterer, and obtaining clustering parameters in each clusterer; Step 3: According to the power trajectory of the virtual power plant, a power consumption objective function is established for the clustering parameters in each clusterer; Step 4: Solve the power consumption objective function to obtain the minimum power consumption of each clusterer, so as to schedule any clusterer according to the power scheduling requirements of the virtual power plant and realize the scheduling of distributed power sources.

2. According to claim 1, a distributed power cluster scheduling method based on NJW spectral clustering algorithm is characterized in that: In step 1, each distributed power model is expressed as: Where P L,j,t is the power demand of distributed load j; P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,t is the energy demand of distributed generation j; E L,j,min-de and E L,j,max-de are the lower and upper bounds of energy demand of distributed resource j, respectively; T start,L,j and T end,L,j are the demand start time and end time of distributed generation j respectively; ΔT is the scheduling time interval, and t represents time t.

3. According to the distributed power cluster scheduling method based on the NJW spectral clustering algorithm of claim 1, it is characterized in that: In step 2, a clustering algorithm is used to divide distributed power models with the same characteristics into a clusterer. The specific process is as follows: Step 21: Perform a first normalization process on each distributed power source model to obtain all normalized distributed power source data; Step 22: Use Euclidean distance to measure the differences between all normalized distributed power sources to obtain a difference matrix; Step 23, using the Gaussian kernel function to convert the difference matrix into an adjacency matrix; Step 24: perform a second normalization process on the adjacency matrix to obtain a normalized adjacency matrix. According to a given number of aggregators, split the normalized adjacency matrix into matrices of a preset number of aggregators, and divide the split matrices into corresponding aggregators. Step 25: Use the K-means algorithm to cluster the matrix in each aggregator to obtain the clustering parameters in each clusterer.

4. According to claim 3, a distributed power cluster scheduling method based on NJW spectral clustering algorithm is characterized in that: In step 21, the normalized distributed power data is expressed as: Where N L,i is the number of distributed resources at node i; x L,j is the original demand parameter of distributed resource j; x′ L,j is the normalized demand parameter of distributed resource j.

5. According to claim 4, a distributed power cluster scheduling method based on NJW spectral clustering algorithm is characterized in that: In step 22, the difference matrix D(i,j) is expressed as: Where α1, α2, α3, α4, α5 and α6 are attribute sets, P L,j,min and P L,j,max are the lower and upper limits of the power demand of distributed load j respectively; P L,i,min and P L,i,max are the lower and upper limits of the power demand of distributed load j respectively; E L,j,min and E L,j,max are the lower and upper bounds of energy demand of distributed resource j, respectively; E L,i,min and E L,i,max are the lower and upper limits of energy demand of distributed resource i, respectively; T start,L,j and T end,L,j are the starting time and ending time of the demand of distributed generation j, T start,L,i and T end,L,i are the start time and end time of the demand of distributed generation i respectively.

6. A distributed power cluster scheduling method based on NJW spectral clustering algorithm according to claim 5, characterized in that: In step 23, the adjacency matrix K(i,j) is expressed as: In the formula, is the Gaussian kernel function parameter.

7. The distributed power cluster scheduling method based on the NJW spectral clustering algorithm according to claim 6 is characterized in that: The clustering parameters within each clusterer are expressed as: Where N L,k,i is the number of distributed resources inside aggregator k at node i; and The starting demand time and the ending demand time of aggregator k at node i are respectively; and are the upper and lower power demand limits of aggregator k at node i, respectively; and are the upper and lower bounds of energy demand of aggregator k at node i, respectively.

8. The distributed power cluster scheduling algorithm based on the NJW spectral clustering algorithm according to claim 7 is characterized in that: In step 3, the power consumption objective function is expressed as: In the formula, k=1,2,...,N AGG,i , is the actual power consumption of aggregator k at node i at time t; N AGG,i is the number of aggregators at node i, and M is the coefficient.

Images