Edge computing unit configuration optimization method based on partition correction
By adopting the edge computing unit configuration optimization method based on partition correction in the new distribution network, the problem of poor adaptability of single-phase grounding fault identification and positioning technology is solved, fast and accurate fault positioning is achieved, and the power supply reliability of the distribution network is improved Sex and safety.
Patent Information
- Application Number
- CN202411409242.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-10
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The single-phase grounding fault identification and positioning technology in the new distribution network has poor adaptability, resulting in slowing down the fault positioning speed and unable to meet the real-time requirements. The existing algorithms are not adaptable under different topology and fault types.
Adopting the edge computing unit configuration optimization method based on partition correction, by comprehensively considering the real-time, economic, security and balance of communication, the edge node partition planning is used and the number of partitions and nodes are optimized. Number of interactive protection and fault location.
It achieves more accurate and fast fault positioning, reduces communication delay, improves the power supply reliability and safety of the distribution network, and provides technical support for self-healing of faults under new power systems.
Smart Images

Figure CN119087137A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of methods for locating ground faults in distribution networks, and in particular to an edge computing unit configuration optimization method based on partition correction. Background Art
[0002] Active intelligent distribution network under the background of new power system is an important part of smart grid. Compared with traditional distribution network, the new distribution network has a large number of power electronic equipment with different functions such as new energy generation, energy storage, electric vehicles, etc., which makes the grid structure, operation mode and fault signal present new characteristics different from traditional distribution network. Distribution network is no longer just a one-way power provider, but is changing to a more advanced new distribution network role. Electrical signals in new distribution network present many new characteristics, which aggravates the difficulty of fault feature detection, location, isolation and recovery control of new distribution network.
[0003] The distribution network system has a wide range of installation and complex line types, which is a weak link in the power system. The current single-phase grounding fault identification and positioning technology of the distribution network has poor adaptability in the complex and changeable distribution network operating environment. The causes of single-phase grounding faults are diverse, and there are a large number of grounding faults of instantaneous high resistance, intermittent arcs, weak transient signals, strong noise backgrounds and other types. The identification and positioning of single-phase grounding faults has always been a technical difficulty. When a high-resistance or arc-light grounding fault occurs, it faces the dead zone of zero-sequence protection, resulting in multi-point grounding, phase-to-phase short circuit, cable and neutral point small resistance insulation breakdown and burning, and other extended hazards. Therefore, the effective diagnosis and positioning of grounding faults is the key to improving the reliability and safety of power supply in new distribution networks.
[0004] As the scale and structure of the distribution network become more complex, the difficulty of fault location is also increasing. On the one hand, the algorithm running time increases, the fault location speed slows down, and it cannot meet the real-time requirements of distribution network fault location; on the other hand, the number of algorithms applied to distribution network fault location is gradually increasing, but some algorithms have their own limitations. Under the conditions of different distribution network topologies and different fault types, the protection and positioning coordination mechanism changes have many problems in adaptability. Therefore, considering the actual working conditions of the distribution network, a fault isolation mechanism with adaptive correction characteristics is established to find or construct the most suitable fault location algorithm. Summary of the invention
[0005] The purpose of the present invention is to provide an edge computing unit configuration optimization method based on partition correction to solve the problems in the above-mentioned background technology.
[0006] In order to achieve the above object, the present invention adopts the following technical solutions:
[0007] A method for optimizing edge computing unit configuration based on partition correction.
[0008] The method comprises the following steps: (1) constructing an objective function for the partition planning of edge nodes by comprehensively considering the communication real-time, system economy, safety and balance factors for the topology structure of the distribution network; (2) setting the number of partitions according to different topological structures, and setting constraints on the number of partitions and communication delay; (3) using a genetic algorithm to iteratively solve the objective function of the partition planning to obtain the number of partitions and the number of nodes in each partition; on this basis, using a dimensionality reduction-clustering algorithm, the cluster center of each partition is selected as an edge node, and the IEEE33 node system is used as a model to simulate and verify the partition edge computing algorithm; (4) correcting the preliminary partitioning; and correcting the partitioning result obtained by clustering.
[0009] Further:
[0010] The step (1) is: combining the on-site working conditions of the distribution network, comprehensively considering the communication real-time, system economy, safety and balance factors to construct an objective function;
[0011] The above communication real-time performance means that in the fault section location of the distribution network, shortening the location time as much as possible can greatly improve the safety and power supply reliability of the system. The communication real-time performance increases with the increase in the number of partitions. The communication delay T is:
[0012] T=T a +T b +T c +T d (1)
[0013] In formula (1): T a Indicates the time required to send data; T b represents the delay of data propagation in the channel; T c represents the queuing delay of the packet in the router; T d It is the packet processing delay, which is small and can be ignored;
[0015]
[0016] In formula (2): P t is the capacity of the transmitted data packet; D is the data transmission distance, i.e., the channel length; v1 is the communication speed of the network; v2 is the propagation speed of data in the communication medium; R is the total number of routes; θ is the rate at which data enters the system; μ is the router transmission rate; when calculating the channel length D, the communication distance in each partition can be estimated by the following formula through equal division:
[0017]
[0018] In formula (3), n is the number of partitions; k is the number of communication nodes, i.e., the number of terminals; d is the average channel length for communication transmission between terminals in a partition; d i is the average communication distance of node i; η i is the connectivity of node i, that is, the number of lines connected to node i; d i,ε represents the length of the εth communication path connected to node i. The calculation formula of the communication delay is obtained from equations (1) to (3), as shown in equation (4).
[0019]
[0020] Further:
[0021] The economic efficiency of the partition in step (1) is:
[0022] The cost E of adding edge devices to the distribution network can be expressed as
[0023] E=E c +E t +E v (5)
[0024] Where: E c Costs required to introduce edge computing units and their supporting facilities; E t The investment cost of the remaining power distribution equipment in the system; E v is the system operation and maintenance cost; when the overall structure of the distribution network is determined, E t and E v remains basically unchanged, while E c Proportional to the number of edges, that is, E c =ns, where s is the cost of adding a single edge device; E increases with the number of partitions, and E can be expressed as
[0025] E=ns+E t +E v (6)
[0026] Further:
[0027] The security and balance in step (1) are:
[0028] Attached Figure 1 The protection diagram of the partition boundary node is shown in Figure 1, where f is the fault point and Bus is the network node. Figure 1 As shown in (a), when a fault occurs in line L4 located at the boundary of the partition, the system cannot isolate the faulty line. Figure 1(b) is a schematic diagram of cross protection of partition boundary nodes. It can be seen from the figure that the edge nodes in partitions 1 and 2 can obtain the measurement information of node Bus5. Therefore, there will be no positioning blind spot when line L4 or L5 fails, thereby improving the safety and reliability of the distribution network.
[0029] For a k-node system with n partitions, if interactive protection is to be achieved, the total number of nodes contained in each partition is K = k + (n-1);
[0030] When the distribution network is partitioned, the balance of each edge jurisdiction area is: if the scale difference between the partitions is large, the computing tasks undertaken by different edge devices will be out of balance, thereby increasing the time required for fault location; the variance δ is calculated by the following formula (7) to determine whether the partition is balanced, where W i is the number of nodes contained in edge partition i; W is the average number of total nodes divided into n partitions; the smaller the value of δ, the more balanced the partition;
[0031]
[0032] Three important factors affecting edge partitioning are identified, namely T, E, and δ. Since the three factors have different dimensions, they need to be normalized before constructing the objective function. The mathematical model is shown in the following formula (8):
[0033]
[0034] Where: x T 、x E 、x δ To assign weights to the evaluation indicators of each partition, the analytic hierarchy process can be used to obtain; T max , T min are the maximum and minimum values of the communication delay when selecting different numbers of partitions; E max 、E min , δ max , δ min Same reason.
[0035] Further:
[0036] The constraint condition of step (2) is:
[0037] First aspect: Constraints on the number of partitions:
[0038] For a system with k nodes, the upper limit of the number of partitions can be approximately √k, and the minimum number of partitions is usually 2; but considering that the population should be diverse when the genetic algorithm is used to solve the objective function, the maximum number of partitions is set to 1.5√k, and the partition number constraint is shown in formula (9):
[0039]
[0040] The second aspect: Communication delay constraints:
[0041] Maximum delay T max It should not exceed 5ms. The constraint of communication delay is shown in formula (10):
[0042] T≤T max (10)
[0043] Further:
[0044] The step (3) comprises:
[0045] First, the division of edge jurisdiction areas: According to the above method, edge partitioning is performed in the improved IEEE33 node system. The relationship between the three indicators is intuitively expressed using a 1-9 scale: x T / x δ =3,x E / x δ =2. The weight of each indicator is x T =0.5, x E =0.333, x δ =0.167;
[0046] Finally, the genetic algorithm is used to iteratively solve the zoning planning objective function to obtain the number of partitions and the number of nodes in each partition; on this basis, the clustering algorithm is used to enter the specific division of the edge jurisdiction.
[0047] Further:
[0048] The clustering algorithm is a spectral clustering algorithm. The key to the spectral clustering algorithm is to construct the Laplace matrix of the network. For the distribution network to be partitioned, its adjacency matrix J is first obtained from the topological structure. The definition of J is: if two nodes a and b in the network are directly connected, then j ab =1, otherwise j ab = 0 and j aa =0; then construct a diagonal matrix whose diagonal elements are the connection degrees of each node, called the degree matrix D. The Laplace matrix L is defined as:
[0049] L=DJ (11)
[0050] By solving the eigenvector of L and performing spectral analysis, we can obtain the distribution of network nodes in the eigenvector space.
[0051] This enables clustering.
[0052] Further:
[0053] The clustering algorithm uses spectral clustering and K-means algorithm to divide the distribution network nodes into different areas. The implementation method is as follows:
[0054] S101: Observe the topological structure of the distribution network, obtain its adjacency matrix J and degree matrix D, and obtain the Laplace matrix L; S102: Solve the eigenvalues of L and sort them. If L has multiple zero eigenvalues, the number of connected branches in the system is the number of zero eigenvalues, and the corresponding eigenvectors can be used as the basis for partitioning; if L has only one zero eigenvalue, go to step S103; S103: Take the first and second smallest non-zero eigenvalues of L, and find the corresponding eigenvectors. The clustering samples are the first and second smallest non-trivial eigenvectors; S104: After obtaining the clustering samples, use the K-means algorithm to divide the network nodes into corresponding edge partitions, and select the cluster centers of each partition as edge nodes.
[0055] Further:
[0056] The modification of the preliminary partition in step (3) is as follows:
[0057] In order to achieve interactive protection, for a k-node system with n partitions, the sum of the number of nodes in the partitions is K = k + (n-1). Following this principle, the objective function is solved. When dividing the edge jurisdiction area, the cluster sample only contains k nodes, so the partition result obtained by clustering is corrected to a certain extent. Suppose the number of nodes in each partition solved according to the objective function is W1, W2, ... W n , and the number of partition nodes obtained by clustering are G1, G2, …G n ; After completing the clustering of the k-node system, the result should be based on W1, W2, ... W n Make corrections and let W={W1,W2,…W n}, G={G1,G2,…G n}, the objective function modified based on the number of partition nodes is shown in formula (12):
[0058]
[0059] The selection rule of decision variables is shown in formula (13):
[0060]
[0061] Each partition should also meet the following conditions:
[0062]
[0063] In formula (14): Zone i is the node set of partition i; crossis a boundary intersection node.
[0064] The beneficial effects of the present invention are:
[0065] The present invention conducts research on the diagnosis, positioning and isolation technology of distributed ground faults in the new distribution network, develops an edge computing unit configuration optimization method based on partition correction, and the invention research results enrich the new distribution network distributed ground fault diagnosis and positioning method. A more accurate ground fault positioning and faster fault isolation and recovery are achieved. The invention overcomes the impact of distributed power supply access, etc., and provides strong support for the fault self-healing technology of the distribution network under the new power system; after a fault occurs in the distribution network, the fault section can be identified on the edge side and then the fault can be removed by decision, which reduces the communication delay, reduces the calculation burden of the master station, and improves the fault processing speed; the positioning model is stored in the master station and the edge node at the same time, and the two are mutually backup. Only in the extreme case that all edges and the master station are attacked, the fault positioning will be completely invalid. Therefore, the power supply reliability and safety of the distribution network are improved. The present invention provides an efficient and accurate new distribution network distributed ground fault diagnosis and positioning model, which realizes the early detection and diagnosis of the new distribution network distributed ground fault by online monitoring of the distribution network FTU operation data and using the edge computing algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0066] Attached Figure 1 A schematic diagram of protection of a partition boundary node according to a specific implementation mode of the present invention;
[0067] Attached Figure 2 An IEEE33 node system according to a specific embodiment of the present invention;
[0068] Attached Figure 3 Parameter settings for a specific implementation of the present invention;
[0069] Attached Figure 4 A distribution diagram of a first small non-trivial eigenvector of a specific embodiment of the present invention;
[0070] Attached Figure 5 A distribution diagram of first and second small non-trivial eigenvectors according to a specific embodiment of the present invention;
[0071] Attached Figure 6 The IEEE 33 node system clustering result of a specific implementation of the present invention;
[0072] Attached Figure 7 The IEEE 33-node system partitioning result of a specific implementation mode of the present invention;
[0073] Attached Figure 8This is a modified partition result of the IEEE 33-node system according to a specific implementation mode of the present invention. DETAILED DESCRIPTION
[0074] The specific implementation of the present invention is described below in conjunction with the accompanying drawings and embodiments:
[0075] The present invention is implemented by the following technical solution: A method for optimizing edge computing unit configuration based on partition correction, comprising the following steps:
[0076] 1. By comprehensively considering factors such as communication real-time, system economy, safety and balance in the distribution network topology structure, an objective function is constructed to carry out partition planning for edge nodes.
[0077] 2. Flexibly set the number of partitions according to different topological structures, and set constraints on the number of partitions and communication delay.
[0078] 3. Genetic algorithm is used to iteratively solve the objective function of partition planning to obtain the number of partitions and the number of nodes in each partition. On this basis, the dimensionality reduction-clustering algorithm is used to select the cluster center of each partition as the edge node. The IEEE33 node system is used as a model to simulate and verify the partition edge computing algorithm.
[0079] 4. Correct the preliminary partitions. When dividing the marginal jurisdiction area, the cluster samples should take into account the partition affiliation of the common nodes of the adjacent partitions, so it is necessary to make certain corrections to the partition results obtained by clustering.
[0080] Further specific embodiments are as follows:
[0081] S1: Based on the actual situation, factors such as communication real-time, system economy, security and balance are comprehensively considered to construct the objective function.
[0082] 1. Communication real-time performance: In the fault section location of the distribution network, shortening the location time as much as possible can greatly improve the safety and power supply reliability of the system. The communication real-time performance increases with the increase of the number of partitions, and the communication delay T is
[0083] T=T a +T b +T c +T d
[0084] Where: T a Indicates the time required to send data; T b represents the delay of data propagation in the channel; T c represents the queuing delay of the packet in the router; T d It is the packet processing delay, which is small and can be ignored.
[0085]
[0086] Where: P t is the capacity of the transmitted data packet; D is the data transmission distance, i.e., the channel length; v1 is the communication speed of the network; v2 is the data propagation speed in the communication medium; R is the total number of routes; θ is the rate at which data enters the system; and μ is the router transmission rate. When calculating the channel length D, the communication distance in each partition can be approximately estimated by the following formula through equal division.
[0087]
[0088] Where n is the number of partitions; k is the number of communication nodes, i.e., the number of terminals; d is the average channel length for communication transmission between terminals in a partition; d i is the average communication distance of node i; η i is the connectivity of node i, that is, the number of lines connected to node i; d i,ε represents the length of the εth communication path connected to node i. The calculation formula of the communication delay is obtained from equations (1) to (3), as shown in equation (4).
[0089]
[0090] 2. Economic efficiency of zoning: Economic efficiency is also an important indicator affecting edge zoning, which can be considered from two aspects: construction investment and operation and maintenance costs. The cost E of adding edge equipment to the distribution network can be expressed as
[0091] E=E c +E t +E v
[0092] Where: E c Costs required to introduce edge computing units and their supporting facilities; E t The investment cost of the remaining power distribution equipment in the system; E v It is the system operation and maintenance cost. When the overall structure of the distribution network is determined, E t and E v remains basically unchanged, while E c Proportional to the number of edges, that is, E c =ns, where s is the cost of adding a single edge device. It is not difficult to conclude that E increases with the number of partitions, and E can be expressed as
[0093] E=ns+E t +E v
[0094] 3. Safety and balance: Figure 1The protection diagram of the partition boundary node is shown in Figure 1, where f is the fault point and Bus is the network node. Figure 1 As shown in (a), when a fault occurs in line L4 located at the boundary of the partition, the system cannot isolate the faulty line. Figure 1 (b) is a schematic diagram of cross protection of partition boundary nodes. It can be seen from the figure that the edge nodes in partition 1 and partition 2 can obtain the measurement information of node Bus5, so there will be no positioning blind spot when line L4 or L5 fails, thereby improving the safety and reliability of the distribution network. For a k-node system with n partitions, if interactive protection is to be achieved, the total number of nodes contained in each partition is K = k + (n-1).
[0095] When zoning the distribution network, the balance of each edge jurisdiction area should also be considered. If the scale difference between the partitions is large, the computing tasks undertaken by different edge devices will be out of balance, thereby increasing the time required for fault location. The variance δ is calculated by the following formula to determine whether the partition is balanced, where W i is the number of nodes contained in edge partition i; W is the average number of total nodes divided into n partitions. The smaller the value of δ, the more balanced the partition.
[0096]
[0097] Three important factors affecting edge partitioning are identified, namely T, E, and δ. Since the three factors have different dimensions, they need to be normalized before constructing the objective function, and the mathematical model is shown in the following formula.
[0098]
[0099] Where: x T 、x E 、x δ To assign weights to the evaluation indicators of each partition, the analytic hierarchy process can be used to obtain; T max , T min are the maximum and minimum values of the communication delay when selecting different numbers of partitions; E max 、E min , δ max , δ min Same reason.
[0100] S2: Constraints
[0101] 1. Constraint on the number of partitions: For a system with k nodes, the upper limit of the number of partitions can be approximately √k, and the minimum number of partitions is usually 2. However, considering that the population should be diverse when the genetic algorithm is used to solve the objective function, the maximum number of partitions is set to 1.5√k, and the constraint on the number of partitions is shown in the formula.
[0102]
[0103] 2. Communication delay constraint: The calculation formula of communication delay is given. In order to shorten the time consumed by fault location as much as possible, it is necessary to constrain the delay of the communication network. Generally, the maximum delay T max It should not exceed 5ms. The constraint of communication delay is shown in the formula.
[0104] T≤T max
[0105] S3: Solving Algorithm
[0106] 1. Edge jurisdiction area division: According to the above method, edge partitioning is performed in the improved IEEE33 node system. Its topological structure is shown in the attached figure. Figure 2 For each parameter, see the attached Figure 3 , where c is the speed of light.
[0107] Before solving the objective function, it is necessary to select the weights of each indicator in the objective function. Assuming that security and balance are the benchmarks, communication real-time is slightly more important than them, and the importance of economy is between the above two, the relationship between the three indicators can be intuitively expressed using a 1-9 scale as follows: T / x δ =3,x E / x δ =2. The weight of each indicator is x T =0.5, x E =0.333, x δ =0.167.
[0108] Genetic algorithm is used to iteratively solve the objective function of zoning planning to obtain the number of partitions and the number of nodes in each partition. On this basis, the specific division of the edge jurisdiction can be carried out, and the clustering algorithm is needed at this time. Since spectral clustering can handle various forms of sample sets including sparse data, it has significant advantages, so this algorithm is selected to complete the zoning task. The key to the spectral clustering algorithm is to construct the Laplace matrix of the network. For the distribution network to be partitioned, its adjacency matrix J is first obtained from the topological structure. The definition of J is: if two nodes a and b in the network are directly connected, then j ab =1, otherwise j ab = 0 and j aa = 0. Then construct a diagonal matrix whose diagonal elements are the connection degrees of each node, called the degree matrix D. The Laplace matrix L is defined as
[0109] L=DJ
[0110] By solving the eigenvector of L and performing spectral analysis, we can get the distribution of network nodes in the eigenvector space, thus achieving the purpose of clustering. The spectral clustering algorithm uses the idea of dimensionality reduction. Through mapping, the clustering of high-dimensional samples can be achieved in low-dimensional space with the help of other algorithms. K-means clustering is selected.
[0111] Attach Figure 2 Taking the IEEE33 node system shown in the figure as an example, the arrangement and distribution of each element in its first small non-trivial eigenvector is shown in the attached figure. Figure 4 As shown. Figure 4 It can be seen that the IEEE 33-node system does not have a significant community structure when only the first small non-trivial eigenvector is considered. In this case, the second small non-trivial eigenvector can be introduced to improve the characteristic dimension of spectral clustering. Figure 5 is the distribution of each element in the first and second smallest non-trivial eigenvectors, where the values of the horizontal and vertical axes correspond to the first and second smallest non-trivial eigenvectors respectively. The clustering results of combining spectral clustering with K-means algorithm are shown in the attached figure. Figure 6 As shown in the figure, according to the clustering results, the system has three edge partitions. The centers of each cluster are marked in the figure, and the nodes 3, 12, and 29 adjacent to them can be selected as edge nodes. Based on the above analysis, spectral clustering and K-means algorithm can be used to divide the distribution network nodes into different areas. The implementation method is as follows.
[0112] 1) Observe the topological structure of the distribution network, obtain its adjacency matrix J and degree matrix D, and obtain the Laplace matrix L.
[0113] 2) Solve the eigenvalues of L and sort them. If L has multiple zero eigenvalues, the number of connected branches in the system is the number of zero eigenvalues, and the corresponding eigenvectors can be used as the basis for partitioning; if L has only one zero eigenvalue, proceed to step .
[0114] 3) Take the first and second smallest non-zero eigenvalues of L and find their corresponding eigenvectors. The clustering samples are the first and second smallest non-trivial eigenvectors.
[0115] 4) After obtaining the cluster samples, the K-means algorithm is used to divide the network nodes into corresponding edge partitions, and the cluster center of each partition is selected as the edge node.
[0116] 2. Partition correction: To achieve interactive protection, for a k-node system with n partitions, the sum of the number of partition nodes K = k + (n-1), and the objective function is solved according to this principle. However, when dividing the edge jurisdiction area, the cluster sample only contains k nodes, so it is necessary to make some corrections to the partition results obtained by clustering. Suppose the number of nodes in each partition solved according to the objective function is W1, W2, ... W n, and the number of partition nodes obtained by clustering are G1, G2, …G n After completing the clustering of the k-node system, the result should be based on W1, W2, …W n Make corrections and let W={W1,W2,…W n}, G={G1,G2,…G n}, the objective function modified based on the number of partition nodes is shown as follows.
[0117]
[0118] The selection rules of decision variables are shown below.
[0119]
[0120] Each partition should also meet the following conditions.
[0121]
[0122] Where: Zone i is the node set of partition i; cross is a boundary intersection node.
[0123] According to the results of the division of the edge jurisdiction area, the partition diagram of the IEEE33 node system is shown in the attached figure. Figure 7 shown.
[0124] The clustering results show that the number of nodes in partitions 1, 2, and 3 are 14, 11, and 8 respectively. By solving the objective function based on the consideration of interactive protection and balance, the number of nodes allocated to each partition is 12, 12, and 11 respectively. The partitions are modified using equations (12) and (13), and nodes 5 and 7 are determined as partition cross nodes according to equation (14). The final marginal jurisdiction area division results are shown in the attached figure. Figure 8 shown.
[0126] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the field without departing from the purpose of the present invention. These changes involve related technologies well known to those skilled in the art, which all fall within the scope of protection of the patent of the present invention.
[0127] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.
Claims
1. A method for optimizing edge computing unit configuration based on partition correction. It is characterized in that The following steps are involved: (1) By comprehensively considering the communication real-time, system economy, safety and balance factors of the distribution network topology, the objective function is constructed to plan the partition of edge nodes; (2) Setting the number of partitions according to different topological structures and setting constraints on the number of partitions and communication delay; (3) The objective function of the partition planning is iteratively solved using a genetic algorithm to obtain the number of partitions and the number of nodes in each partition. On this basis, the dimensionality reduction-clustering algorithm is used to select the cluster center of each partition as the edge node, and the IEEE33 node system is used as a model to simulate and verify the partition edge computing algorithm. (4) Correct the preliminary partitioning and correct the partitioning results obtained by clustering.
2. The edge computing unit configuration optimization method based on partition correction according to claim 1, Features: The step (1) is: Combined with the on-site working conditions of the distribution network, the objective function is constructed by comprehensively considering the communication real-time, system economy, safety and balance factors; The above communication real-time performance means that in the fault section location of the distribution network, shortening the location time as much as possible can greatly improve the safety and power supply reliability of the system. The communication real-time performance increases with the increase in the number of partitions. The communication delay T is: T=T a +T b +T c +T d (1) In formula (1): T a Indicates the time required to send data; T b represents the delay of data propagation in the channel; T c Indicates the queuing delay of packets in the router; T d It is the packet processing delay, which is small and can be ignored; In formula (2): P t is the capacity of the transmitted data packet; D is the data transmission distance, i.e., the channel length; v1 is the communication speed of the network; v2 is the propagation speed of data in the communication medium; R is the total number of routes; θ is the rate at which data enters the system; μ is the router transmission rate; when calculating the channel length D, the communication distance in each partition can be estimated by the following formula through equal division: In formula (3), n is the number of partitions; k is the number of communication nodes, i.e., the number of terminals; d is the average channel length for communication transmission between terminals in a partition; d i is the average communication distance of node i; η i is the connectivity of node i, that is, the number of lines connected to node i; d i,ε represents the length of the εth communication path connected to node i. The calculation formula of communication delay is obtained from equations (1) to (3), as shown in equation (4): .
3. The edge computing unit configuration optimization method based on partition correction as claimed in claim 2, Features: The economic efficiency of the partition in step (1) is: The cost E of adding edge devices to the distribution network can be expressed as E=E c +E t +E v (5) Where: E c Costs required to introduce edge computing units and their supporting facilities; E t The investment cost of the remaining power distribution equipment in the system; E v is the system operation and maintenance cost; when the overall structure of the distribution network is determined, E t and E v remains basically unchanged, while E c Proportional to the number of edges, that is, E c =ns, where s is the cost of adding a single edge device; E increases with the number of partitions, and E can be expressed as E=ns+E t +E v (6)。 4. The edge computing unit configuration optimization method based on partition correction as claimed in claim 3, Features: The security and balance in step (1) are: FIG1 is a schematic diagram of protection of partition boundary nodes, where f is a fault point and Bus is a network node. As shown in Figure 1(a), when line L4 at the partition boundary fails, the system cannot isolate the faulty line. Figure 1(b) is a schematic diagram of cross protection of partition boundary nodes. It can be seen from the figure that the edge nodes in partitions 1 and 2 can obtain the measurement information of node Bus5, so there will be no positioning blind spot when line L4 or L5 fails, thereby improving the safety and reliability of the distribution network. For a k-node system with n partitions, if interactive protection is to be achieved, the total number of nodes contained in each partition is K = k + (n-1); When the distribution network is partitioned, the balance of each edge jurisdiction area is: if the scale difference between the partitions is large, the computing tasks undertaken by different edge devices will be out of balance, thereby increasing the time required for fault location; the variance δ is calculated by the following formula (7) to determine whether the partition is balanced, where W i is the number of nodes contained in edge partition i; W is the average number of total nodes divided into n partitions; the smaller the value of δ, the more balanced the partition; Three important factors affecting edge partitioning are identified, namely T, E, and δ. Since the three factors have different dimensions, they need to be normalized before constructing the objective function. The mathematical model is shown in the following formula (8): Where: x T 、x E 、x δ To assign weights to the evaluation indicators of each partition, the analytic hierarchy process can be used to obtain; T max 、T min are the maximum and minimum values of the communication delay when selecting different numbers of partitions; E max 、E min ,δ max ,δ min Same reason.
5. The edge computing unit configuration optimization method based on partition correction according to claim 1, Features: The constraint condition of step (2) is: First aspect: Constraints on the number of partitions: For a system with k nodes, the upper limit of the number of partitions can be approximated as The minimum number of partitions is usually 2; however, considering that the population should be diverse when the genetic algorithm is used to solve the objective function, the maximum number of partitions is set to The partition number constraint is shown in formula (9): The second aspect: Communication delay constraints: Maximum delay T max It should not exceed 5ms. The constraint of communication delay is shown in formula (10): T≤T max (10)。 6. The edge computing unit configuration optimization method based on partition correction according to claim 1, Features: The step (3) comprises: First, the division of edge jurisdiction areas: According to the above method, edge partitioning is performed in the improved IEEE33 node system. The relationship between the three indicators is intuitively expressed using a 1-9 scale: x T / x δ =3,x E / x δ =2. The weight of each indicator is x T =0.5, x E =0.333, x δ =0.167; Finally, the genetic algorithm is used to iteratively solve the zoning planning objective function to obtain the number of partitions and the number of nodes in each partition; on this basis, the clustering algorithm is used to enter the specific division of the edge jurisdiction.
7. A partition-corrected edge computing unit configuration optimization method as claimed in claim 6, Features: The clustering algorithm is a spectral clustering algorithm. The key to the spectral clustering algorithm is to construct the Laplace matrix of the network. For the distribution network to be partitioned, its adjacency matrix J is first obtained from the topological structure. The definition of J is: if two nodes a and b in the network are directly connected, then j ab =1, otherwise j ab = 0 and j aa =0; then construct a diagonal matrix whose diagonal elements are the connection degrees of each node, called the degree matrix D. The Laplace matrix L is defined as: L=DJ (11) By solving the eigenvector of L and performing spectral analysis, we can obtain the distribution of network nodes in the eigenvector space, thereby achieving clustering.
8. A partition-corrected edge computing unit configuration optimization method as claimed in claim 6, Features: The clustering algorithm uses spectral clustering and K-means algorithm to divide the distribution network nodes into different areas. The implementation method is as follows: S101: Observe the topological structure of the distribution network, obtain its adjacency matrix J and degree matrix D, and obtain the Laplace matrix L; S102: Solve the eigenvalues of L and sort them. If L has multiple zero eigenvalues, the number of connected branches in the system is the number of zero eigenvalues, and the corresponding eigenvectors can be used as the basis for partitioning; if L has only one zero eigenvalue, proceed to step S103; S103: Take the first and second smallest non-zero eigenvalues of L and find their corresponding eigenvectors. The cluster samples are the first and second smallest non-trivial eigenvectors. S104: After obtaining the cluster samples, the K-means algorithm is used to divide the network nodes into corresponding edge partitions, and the cluster center of each partition is selected as the edge node.
9. The edge computing unit configuration optimization method based on partition correction as claimed in claim 1, Features: The modification of the preliminary partition in step (3) is as follows: In order to achieve interactive protection, for a k-node system with n partitions, the sum of the number of nodes in the partitions is K = k + (n-1). Following this principle, the objective function is solved. When dividing the edge jurisdiction area, the cluster sample only contains k nodes, so the partition result obtained by clustering is corrected to a certain extent. Suppose the number of nodes in each partition solved according to the objective function is W1, W2, ... W n , and the number of partition nodes obtained by clustering are G1, G2, …G n ; After completing the clustering of the k-node system, the result should be based on W1, W2, ... W n Make corrections and let W={W1,W2,…W n }, G={G1,G2,…G n }, the objective function modified based on the number of partition nodes is shown in formula (12): The selection rule of decision variables is shown in formula (13): Each partition should also meet the following conditions: In formula (14): Zone i is the node set of partition i; cross is a boundary intersection node.
Citation Information
Patent Citations
Optimized spectral clustering power distribution network partitioning method based on edge calculation
CN117370821A
Optimal configuration method and device for edge sensing terminal of power distribution network
CN118508419A
