Harmonic collaborative management method for double-high distribution network based on cloud-edge collaboration and dynamic partitioning

Abstract

The application discloses a double-high power distribution network harmonic collaborative management method based on cloud edge collaboration and dynamic partitioning, belongs to the field of power harmonic management, and comprises the following steps: constructing a time-varying harmonic collaborative management architecture based on cloud edge collaboration including a cloud server, an edge server and a management device; constructing a derived network link matrix; selecting an observation node by using an improved PageRank algorithm based on the derived network link matrix; according to the harmonic state of the observation node and the state information of the management device, a two-stage dynamic partitioning method is used to realize dynamic partitioning of the management area of the management device; based on the dynamic partitioning result, a multi-time scale cloud edge collaborative optimization management strategy is used to optimize the management parameters; based on the long-time scale management parameters, an improved NSGA-II multi-objective optimization algorithm is used to further dynamically correct the management parameters on the short-time scale; and based on the cloud server multi-time scale optimized management parameters, the edge server controls the management device according to the management parameters to realize time-varying harmonic collaborative management. The application can realize efficient, real-time and collaborative management of the dispersed, network-wide and time-varying waves of the double-high power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system harmonic mitigation technology, and in particular to a method for harmonic collaborative mitigation of dual-high-voltage distribution networks based on cloud-edge collaboration and dynamic partitioning. Background Technology

[0002] With the development of new power systems and the widespread application of power electronic devices in the power generation-grid-load system, harmonics in distribution networks are exhibiting new characteristics. In "high-proportion" distribution networks (high proportion of renewable energy and high proportion of power electronic equipment access), the coupling and interaction between power generation, grid, and load are significantly enhanced, and harmonic sources are becoming increasingly complex and diversified. On the power generation side, the harmonic superposition effect generated by the large-scale grid connection of distributed generation (DG) inverters leads to a significant increase in harmonic voltage distortion rate. On the grid side, the interaction and superposition of harmonics within the grid exacerbates harmonic pollution. On the load side, the access of harmonic sources such as smart home devices and commercial facilities intensifies harmonic pollution in the distribution network. Therefore, studying the emission characteristics of harmonic sources at various levels and the interaction and superposition of harmonics is crucial for developing harmonic mitigation strategies.

[0003] Harmonics generated by grid-connected distributed generation (DG) inverters are influenced by various factors, including the inverter's topology, control strategy, grid connection conditions, and interaction with the grid. DG grid-connected inverters typically employ pulse width modulation (PWM) technology. Existing technologies include: some studies investigating the harmonic interaction and superposition between the inverter and the grid, finding that within the inverter's control bandwidth, dynamic interaction between the inverter and the grid can cause harmonic resonance; others have studied the "harmonic amplification" phenomenon that grid-connected inverters exhibit under small disturbance conditions, caused by phase errors in the phase-locked loop (PLL) leading to increased harmonic content in the inverter's output current; and still others have proposed a machine learning-based time-series classification method to locate forced oscillation sources, which offers short computation time, high accuracy, and robustness. However, this machine learning-based method cannot guarantee the accuracy of the results, especially when the forced oscillation source is on the load side and exhibits a dynamic response closely related to the load. Another study proposed a sensorless control strategy for VDAPF (Vehicle-Actuated Grid-Distributed Power Filter) based on a series LC filter, effectively reducing system cost and improving system reliability. In addition, MFGCI and VDAPF have similar structures and are usually used as auxiliary equipment for harmonic control. While realizing active power transmission of DG, they use their redundant capacity to achieve harmonic compensation.

[0004] The aforementioned existing technologies primarily explore harmonic mitigation and suppression of harmonics generated by MFGCI grid connection by controlling the impedance of VDAPF or MFGCI. Although MFGCI and nonlinear loads meet the mitigation requirements, numerous small harmonics still exist at other nodes, and their superposition can lead to severe harmonic problems. Both devices can mitigate harmonics within a certain range, but their mitigation scope is limited. The cloud-edge collaborative architecture offers a new approach to these problems, but existing technologies have not yet solved core issues such as dynamic partitioning, multi-timescale optimization, and real-time parameter tuning.

[0005] Therefore, there is an urgent need for an efficient and adaptive harmonic co-control method. Summary of the Invention

[0006] To address the issues of one-sided selection of observation nodes, fixed governance areas, and insufficient real-time performance in harmonic mitigation of high-voltage and high-efficiency power distribution networks, this invention proposes a collaborative harmonic mitigation method for high-voltage and high-efficiency power distribution networks based on cloud-edge collaboration and dynamic partitioning. The aim is to achieve accurate characterization of the entire network's harmonic state: key observation nodes are dynamically selected through an improved PageRank algorithm; the governance area of ​​the mitigation equipment is dynamically adapted by employing a two-stage partitioning approach to divide the control area of ​​the mitigation equipment; and multi-scale optimization of mitigation parameters is achieved by combining cloud servers and edge servers to optimize and correct harmonic mitigation parameters for both short and long time periods in real time.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0008] A method for coordinated harmonic mitigation in high-voltage power distribution networks based on cloud-edge collaboration and dynamic zoning includes the following steps:

[0009] Step 1: Construct a time-varying harmonic collaborative governance architecture based on cloud-edge collaboration. The cloud server is responsible for selecting observation nodes, dynamic partitioning, and optimizing governance parameters, while the edge server is responsible for collecting harmonic characteristics and the status of governance equipment, and controlling the governance equipment to perform harmonic governance.

[0010] Step 2: Based on the cloud-edge collaborative time-varying harmonic governance architecture, virtual nodes are designed to establish a derived network link matrix, taking into account node efficiency and comprehensive sensitivity index. An observation node selection method based on the derived network link matrix and an improved PageRank algorithm is proposed to select a limited number of observation nodes to effectively reflect the harmonic governance effect of the entire network.

[0011] Step 3: Based on the harmonic state of the observation nodes and the state information of the treatment equipment, a two-stage dynamic zoning method is proposed to realize the dynamic zoning of the treatment equipment area, ensuring the accuracy of zoning and the balance of the number of nodes in the area.

[0012] Step 4: Based on the dynamic partitioning results, a multi-timescale cloud-edge collaborative optimization governance strategy is proposed to optimize the governance parameters;

[0013] Step 5: Based on the governance parameters, propose an improved NSGA-II multi-objective optimization algorithm to further dynamically correct the governance parameters on a short time scale;

[0014] Step 6: Based on the multi-timescale optimized governance parameters of the cloud server, the edge server controls the governance equipment to achieve time-varying harmonic collaborative governance.

[0015] A further improvement to the technical solution of this invention is as follows: In step 1, the cloud-edge collaborative time-varying harmonic governance architecture includes a cloud server deployed on a power Internet of Things cloud platform, several edge servers deployed on distribution network substations or key nodes, governance equipment, data monitoring devices installed on distribution network observation nodes, and a communication network; the governance equipment includes a voltage-sensing active power filter (VDAPF) and a multi-functional grid-connected inverter (MFGCI), which regulates harmonic voltage through virtual harmonic conductance; the data monitoring device is used to collect harmonic voltage / current data in real time and upload it to the edge server; the communication network uses the OPC UA protocol to realize three-level data transmission between the cloud server, edge server, governance equipment, and data monitoring device.

[0016] A further improvement to the technical solution of the present invention is that the task division between the cloud server and the edge server is as follows:

[0017] (1) Observation node selection: The edge server is responsible for data acquisition, harmonic feature extraction, and calculation of voltage reactive power sensitivity and harmonic sensitivity index; the cloud server is responsible for the spatiotemporal distribution of harmonics in the whole network, harmonic state prediction, and selection of observation nodes using the improved PageRank algorithm.

[0018] (2) Two-stage partitioning: The edge server is responsible for calculating the comprehensive sensitivity, modularity index, communication real-time index and workload balancing index, and uploading them to the cloud server; the cloud server is responsible for using the k-means++ algorithm and simulated annealing algorithm to perform two-stage partitioning, with a time scale of 5 minutes, and dynamically updating the partitioning results based on harmonic state information and governance equipment capacity.

[0019] (3) Harmonic mitigation: The edge server is responsible for providing the cloud server with harmonic and mitigation equipment status information, and for communication between the mitigation equipment. It controls the mitigation equipment based on the mitigation parameters. The cloud server uses the simulated annealing algorithm to optimize the mitigation parameters on a long time scale of 15 minutes based on the day-ahead harmonic and mitigation equipment remaining capacity prediction data and the partitioning results. Based on the day-ahead harmonic and mitigation equipment remaining capacity prediction data and the partitioning results, the improved NSGA-II multi-objective optimization algorithm is used to optimize and correct the mitigation parameters on a short time scale of 5 minutes. The corrected mitigation parameters are then sent to the edge server.

[0020] A further improvement to the technical solution of this invention lies in the following: Step 2 specifically includes the following: constructing a network link matrix using node efficiency indicators and comprehensive sensitivity indicators; for a distribution network with n nodes, considering the interrelationships between nodes, constructing a link matrix X. n×n As shown below:

[0021]

[0022] In the formula, x ij Let be the weight of the edge between node i and node j, that is, the overall influence of node i on node j, i∈1,2,…,n, j∈1,2,…,n; n is the number of nodes in the network;

[0023] According to the equivalent diagram of the derived network structure, virtual nodes and other nodes are not actually connected by lines; the diagram only reflects the comprehensive influence between virtual nodes and other nodes. The derived network link matrix X′ (n+1)×(n+1) As shown below:

[0024]

[0025] When i∈[1,n] and j∈[1,n]:

[0026]

[0027] When i = n + 1 and j ∈ [1, n + 1]:

[0028]

[0029] When i∈[1,n+1] and j=n+1:

[0030]

[0031] In the formula, x ij 、x′ ij These are the elements in the network link matrix and the elements in the derived network matrix, respectively;

[0032] An improvement to the PageRank algorithm, the improved Google matrix G′:

[0033] G′=αX′+(1-α)E′ (6)

[0034] In the formula, X′ is the link matrix of the (n+1)th order derivative network; α is the weight value, α∈[0,1];

[0035] In the traditional PageRank algorithm, the PR value of each webpage is evenly distributed among all the webpages it points to; the elements in the self-rebooting vector E′=e / n are uniformly distributed, where e=(1,1,1,…,1). T The self-restart vector is used to reflect the observer's level of interest in the webpage content; in power systems, when the harmonic voltage of a node changes, it generally affects other nodes located on the shortest path between nodes; therefore, the betweenness index is introduced to redefine E′:

[0036] E′=f X′ (E n (7)

[0037] In the formula, f X′ For the function corresponding to the derived link network; E n as follows:

[0038]

[0039] In the formula, b jk b is the number of shortest paths between node j and node k; jik Let i be the number of nodes i that pass through the shortest path between nodes j and k, where i = 1, 2, 3, ..., n; j = 1, 2, 3, ..., n; k = 1, 2, 3, ..., n.

[0040] A further improvement to the technical solution of the present invention is that, in step 3, during the two-stage dynamic partitioning process, the following is included:

[0041] 3.1 The first stage of partitioning uses comprehensive sensitivity as the partitioning index to establish a partitioning matrix, then determines the number of partitions, and then uses the k-means++ algorithm for partitioning;

[0042] 3.2 During the second-stage partitioning, the MFGCI governance area is equivalent to a generalized node, which is contained within the VDAPF governance area. Therefore, the generalized node area is subject to collaborative governance by VDAPF and MFGCI, and the governance devices exchange information through edge servers. A partitioning evaluation function is constructed by combining three partitioning evaluation indicators: modularity index, real-time communication index, and workload balancing index. The second-stage partitioning evaluation function is solved using the simulated annealing algorithm to finally determine the control area of ​​VDAPF.

[0043] A further improvement to the technical solution of this invention lies in the fact that, when using the k-means++ algorithm for partitioning, the number of partitions must first be determined, and the specific steps are as follows:

[0044] 3.1.1 Initial parameter setting: Set the initial number of clusters to 2, use the k-means++ algorithm for clustering, and calculate the sum of clustering errors under the current number of clusters;

[0045] 3.1.2 Error and Difference Calculation: Calculate the difference between the sum of errors of the current clustering result and the sum of errors of the previous clustering result; if the difference is less than the set threshold, the optimal number of partitions is considered to be the current number of clusters minus 1; otherwise, the number of clusters is incremented by 1, and the above steps are repeated until the condition is met.

[0046] A further improvement to the technical solution of this invention lies in that: the modularity index, the real-time communication index, and the workload balancing index are specifically as follows:

[0047] 3.2.1 Modularity Metric: The modularity metric is an evaluation indicator for measuring the quality of partitioning results. The modularity value ranges from 0 to 1, with a higher value indicating better partitioning results. The expression for the modularity Q is shown below:

[0048]

[0049] In the formula, A ij A is a weight matrix consisting of the weights between node i and node j. ij ∈[0,1]; m is the sum of the weights of all edges in the network; The sum of the weights of the lines connected to node i; c i c j These are the community numbers of nodes i and j, respectively;

[0050] Indicator D i,j S represents the electrical distance or coupling strength between nodes. i,j It characterizes the combined impact of changes in node i on the fundamental voltage stability and harmonic voltage distortion of node j; due to index D i,j and S i,j Since they are not on the same order of magnitude, the Min-Max normalization method is used to normalize the two indicator data separately, as shown in the following formula:

[0051] A ij =γD′ i ,j+(1-γ)S′ i,j (13)

[0052] In the formula, D′ i,j S′ i,j They represent the indicators D respectively.i,j and S i,j The normalization; γ is the weighting factor, with a value of 0.4-0.6;

[0053] 3.2.2 Communication Real-Time Performance Indicators: The greater the signal transmission delay, the greater the impact on real-time performance, and the less conducive it is to real-time harmonic voltage control. Therefore, it is necessary to minimize the communication real-time performance indicators. The communication real-time performance indicator T is:

[0054]

[0055] In the formula, T d T p T represents the data transmission and propagation time, respectively; r Indicates router latency; P d Indicates data size; v d v p L represents the data transmission rate of the communication network and the data transmission rate of the channel, respectively. p λ represents the channel length; g represents the number of routers; r θ represents the average data arrival rate; θ represents the router's service efficiency.

[0056] 3.2.3 Workload Balancing Indicators: The number of observation nodes within the control area of ​​each edge server varies, and the load on each observation node also differs, resulting in variations in the workload of each edge server. A more balanced workload across edge servers indicates a more reasonable control area by the governance equipment, preventing any single governance device from operating at high intensity for extended periods and extending its operating time. For any given edge server, its workload is:

[0057]

[0058] In the formula, w i For the workload of edge server i; l j ρ represents the load of observation node j; p represents the number of observation nodes within the control area of ​​edge server i.

[0059] Given k edge servers, what is the average workload of all edge servers in the network? for:

[0060]

[0061] Variance is used to measure workload balance; therefore, the smaller the workload balance index, the more balanced the workload. This results in the workload balance index L. load for:

[0062]

[0063] The partition evaluation function is constructed as follows:

[0064]

[0065] The partition is considered optimal when the partition evaluation function g reaches its minimum value. Let ε = 0.3 and φ = 0.4. With the goal of minimizing the partition evaluation function, the simulated annealing algorithm is used to obtain the optimal partitioning result. Through two-stage dynamic partitioning, the control areas of two types of governance equipment are determined respectively, realizing the dynamic partitioning of the governance area of ​​the governance equipment. This can better cope with dynamically changing harmonics and provide a basis for intraday short-timescale rolling optimization and correction.

[0066] A further improvement to the technical solution of this invention lies in the following: In step 4, specifically, it includes: making day-ahead predictions of the power grid's harmonics and the active power output of new energy sources based on historical data, optimizing the period with a 15-minute time scale, and establishing a THDv of each observation node in the entire network. i With minimizing the sum as the optimization objective, the equivalent virtual harmonic conductance of the governance equipment is optimized on a day-ahead long-time scale. On the cloud server, the simulated annealing algorithm is used to solve the day-ahead long-time scale optimization problem, and the final optimized solution set is the conductance value of each governance equipment under different harmonics. At the same time, the obtained conductance values ​​are sent to the edge server, as shown in the following formula:

[0067]

[0068] In the formula, THDv i n represents the total voltage distortion rate at observation node i; m R represents the total number of observation nodes in distribution network area m; M represents the total number of areas; R i Let be the weight value of node i, which is also the PR value of node i; where:

[0069]

[0070] In the formula, U 1,i U is the effective value of the fundamental voltage at node i; h,i Let be the effective value of the h-th harmonic voltage at node i;

[0071] The constraints are for the treatment capacity of the treatment equipment, where the treatment capacity of VDAPF must not exceed the rated capacity, and the treatment capacity of MFGCI should not exceed the remaining capacity, as follows:

[0072]

[0073]

[0074] in:

[0075]

[0076] In the formula: P j,MFGCI (t) represents the actual active power output of the MFGCI at time t; S′ j,MFGCI (t) represents the actual remaining capacity at node j of the configured MFGCI at time t; S i,VDAPF S represents the rated capacity of the VDAPF at node i; j,N For the rated capacity of MFGCI; G h,i To configure the h-th harmonic equivalent conductance of VDAPF at node i of VDAPF;

[0077] Based on the optimization results and intraday forecast data issued by the cloud server over a long period of time, an intraday short-term optimization correction function for the edge server is constructed with a 5-minute optimization correction cycle and the change in conductivity of the governance equipment within each region as the optimization variable. The optimization correction function uses the THDv of each observation node within each VDAPF partition as the optimization variable. i The objective is to minimize the sum, and the constraints are consistent with the long-term optimization objective function.

[0078] The intraday short-timescale optimization correction objective function for region c at time t is:

[0079]

[0080] In the formula, THDv i,t n is the total harmonic voltage distortion rate of observation node i within region c at time t; c U represents the number of observation nodes contained in region c; 1,i,t U represents the effective value of the fundamental voltage at node i within region c at time t; h,i,t Let be the effective value of the h-th harmonic voltage of node i in region c at time t;

[0081] For partition c, the harmonic current prediction bias of the observed nodes is ΔI. h,i The harmonic power flow equations are as follows:

[0082]

[0083] In the formula, ΔI h,i The prediction bias value injected into the h-th harmonic current at observation node i; ΔG h,i,t Y represents the h-th equivalent virtual harmonic conductance increment of node i connected to the governance device at time t; i,i Let be the self-admittance of node i.

[0084] A further improvement to the technical solution of this invention lies in step 5, where the improved NSGA-II multi-objective optimization algorithm incorporates the beluga optimization algorithm, including an exploration phase and a development phase; the mathematical models for position updates in the two phases are shown below:

[0085] Exploration phase:

[0086]

[0087] In the formula, It is the new position of the i-th beluga whale in the j-th dimension; p j It is randomly selected from d dimensions, where j = 1, 2, ..., d; It is the position of the i-th beluga whale in dimension j; This indicates the current position of the i-th beluga whale; This represents the current location of the r-th beluga whale, where r represents a randomly selected beluga whale; r1 and r2 are random numbers between 0 and 1, used to enhance the random operators in the exploration phase;

[0088] Development phase:

[0089]

[0090] In the formula, It is the current position of the i-th beluga whale; It is the current location of a random beluga whale;

[0091] It is the optimal position for the i-th beluga whale; This is the optimal location among the whales; r3 and r4 are random numbers in the range (0,1), used to enhance the random operators during the exploration phase; C1 is used to measure the random jump intensity of the Levy flight intensity, C1 = 2r4(1-t / T) max );L F It is the Levy flight function. Where u and v are normally distributed random numbers, β = 1.5.

[0092] The technological advancements achieved by this invention due to the adoption of the above technical solutions are as follows:

[0093] This invention acquires voltage and current data of all nodes through power grid monitoring devices, dynamically selects key observation nodes using an improved PageRank algorithm that integrates node efficiency and comprehensive sensitivity indicators, and achieves adaptive adjustment of the governance area based on a two-stage dynamic partitioning mechanism (k-means++ partitioning of MFGCI region and simulated annealing optimization of VDAPF region). In the cloud, the virtual harmonic conductance parameters are globally optimized at a 15-minute cycle using the simulated annealing algorithm, and the governance parameters are rolled over at a 5-minute cycle by integrating the White Whale optimization mechanism to improve the NSGA-II algorithm on the edge server. Attached Figure Description

[0094] Figure 1 This is a flowchart of a dual-high-voltage power distribution network harmonic collaborative governance method based on cloud-edge collaboration and dynamic partitioning provided in an embodiment of the present invention;

[0095] Figure 2 This is the architecture of the time-varying harmonic collaborative governance scheme based on cloud-edge collaboration provided in the embodiments of the present invention;

[0096] Figure 3 This is an equivalent diagram of the derived network structure provided in the embodiments of the present invention;

[0097] Figure 4 This is a flowchart of the improved NSGA-II algorithm provided in the embodiments of the present invention;

[0098] Figure 5 This is a schematic diagram of the configuration scheme under the harmonic scenario provided in the embodiment of the present invention;

[0099] Figure 6 This invention provides three governance strategies for each node's THDv under two governance device configuration schemes in a harmonic scenario. i Line graph;

[0100] Figure 7 This is the THDv of each observation node before and after mitigation in a harmonic scenario provided by the embodiments of the present invention. i Line graph. Detailed Implementation

[0101] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0102] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments:

[0103] like Figure 1 As shown, a method for coordinated harmonic mitigation in a high-voltage power distribution network based on cloud-edge collaboration and dynamic zoning includes the following steps:

[0104] Step 1: Construct a time-varying harmonic collaborative governance architecture based on cloud-edge collaboration. The cloud server is responsible for selecting observation nodes, dynamic partitioning, and optimizing governance parameters, while the edge server is responsible for collecting harmonic characteristics and the status of governance equipment, and controlling the governance equipment to perform harmonic governance.

[0105] like Figure 2 As shown, the cloud-edge collaborative governance architecture for time-varying harmonics includes cloud servers deployed on the power Internet of Things cloud platform, several edge servers deployed on distribution network substations or key nodes, governance equipment, data monitoring devices installed on distribution network observation nodes, and communication networks.

[0106] The governance equipment includes a Voltage Detection Active Power Filter (VDAPF) and a Multi-functional grid-connected inverter (MFGCI), which regulates harmonic voltage through virtual harmonic conductance; a data monitoring device: collects harmonic voltage / current data in real time and uploads it to the edge server; and a communication network, which uses the OPC UA protocol to realize three-level data transmission between the cloud server, the edge server, the governance equipment, and the data monitoring device.

[0107] The task division between cloud servers and edge servers is as follows:

[0108] (1) Observation node selection: The edge server is responsible for data acquisition, harmonic feature extraction, and calculation of indicators such as voltage reactive power sensitivity and harmonic sensitivity; the cloud server is responsible for the spatiotemporal distribution of harmonics across the entire network, harmonic state prediction, and selection of observation nodes using the improved PageRank algorithm.

[0109] (2) Two-stage partitioning: The edge server is responsible for calculating the comprehensive sensitivity, modularity index, communication real-time index and workload balancing index, and uploading them to the cloud server; the cloud server is responsible for using the k-means++ algorithm and simulated annealing algorithm to perform two-stage partitioning, with a time scale of 5 minutes, and dynamically updating the partitioning results based on harmonic state information and governance equipment capacity.

[0110] (3) Harmonic mitigation: The edge server is responsible for providing the cloud server with harmonic and mitigation equipment status information, and for communication between the mitigation equipment. It controls the mitigation equipment based on the mitigation parameters. The cloud server uses the simulated annealing algorithm to optimize the mitigation parameters (conductance) on a long time scale of 15 minutes based on the day-ahead harmonic and mitigation equipment remaining capacity prediction data and the partitioning results. Based on the day-ahead harmonic and mitigation equipment remaining capacity prediction data and the partitioning results, the improved NSGA-II multi-objective optimization algorithm is used to optimize and correct the mitigation parameters on a short time scale of 5 minutes. The corrected mitigation parameters are then sent to the edge server.

[0111] Step 2: Based on the cloud-edge collaborative time-varying harmonic governance architecture, virtual nodes are designed to establish a derived network link matrix, taking into account both node efficiency and comprehensive sensitivity index. An observation node selection method based on the derived network link matrix and an improved PageRank algorithm is proposed to select a limited number of observation nodes to effectively reflect the harmonic governance effect of the entire network.

[0112] The observation node selection method can effectively identify observation nodes that represent the harmonic level of the entire network. The selection of observation nodes takes into account both mitigation time and effectiveness, and the selected observation nodes can effectively reduce optimization time and computational complexity, laying the foundation for rapid and effective harmonic mitigation of the distribution network.

[0113] Specifically, this includes: constructing a network link matrix using node efficiency and comprehensive sensitivity indices. Node efficiency indices reflect the transmission relationships between nodes, thus indicating their locational relationships. The comprehensive sensitivity index encompasses fundamental voltage reactive power sensitivity and harmonic sensitivity, simultaneously reflecting the impact of harmonic voltage changes at one node on other nodes and voltage distortion. For a distribution network with n nodes, a link matrix X is constructed considering the interrelationships between nodes. n×n As shown in Equation 1:

[0114]

[0115] Where, x ij Let be the weight of the edge between node i and node j, that is, the overall influence of node i on node j, i∈1,2,…,n, j∈1,2,…,n; n is the number of nodes in the network;

[0116] According to the equivalent diagram of the derived network structure, virtual nodes and other nodes are not actually connected by lines; the diagram only reflects the comprehensive influence between virtual nodes and other nodes. The derived network link matrix X′ (n+1)×(n+1) As shown below:

[0117]

[0118] When i∈[1,n] and j∈[1,n]:

[0119]

[0120] When i = n + 1 and j ∈ [1, n + 1]:

[0121]

[0122] When i∈[1,n+1] and j=n+1:

[0123]

[0124] In the formula, x ij 、x′ ij These are the elements in the network link matrix and the elements in the derived network matrix, respectively;

[0125] like Figure 3 As shown, the PageRank algorithm has been improved, and the improved Google matrix G′ is:

[0126] G′=αX′+(1-α)E′ (6)

[0127] In the formula, X′ is the link matrix of the (n+1)th order derivative network; α is the weight value, α∈[0,1];

[0128] In the traditional PageRank algorithm, the PR value of each webpage is evenly distributed among all the webpages it points to. Therefore, the elements in the self-rebooting vector E′ = e / n are uniformly distributed, where e = (1,1,1,…,1). T The self-restart vector is used to reflect the observer's level of interest in webpage content. In power systems, when the harmonic voltage of a node changes, it generally has a significant impact on other nodes located on the shortest path between nodes. Therefore, the betweenness index is introduced to redefine E′:

[0129] E′=f X′ (E n (7)

[0130] In the formula, f X′ For the function corresponding to the derived link network; E n as follows:

[0131]

[0132] In the formula, b jk b is the number of shortest paths between node j and node k; jikLet i be the number of nodes i that pass through the shortest path between nodes j and k, where i = 1, 2, 3, ..., n; j = 1, 2, 3, ..., n; k = 1, 2, 3, ..., n.

[0133] Step 3: Based on the harmonic state of the observation nodes and the state information of the treatment equipment, a two-stage dynamic zoning method is proposed to realize the dynamic zoning of the treatment equipment area, ensuring the accuracy of the zoning and the balance of the number of nodes in the area.

[0134] The two-stage dynamic partitioning process includes:

[0135] 3.1 The first stage of partitioning uses comprehensive sensitivity as the partitioning index to establish a partition matrix, determine the number of partitions, and then uses the k-means++ algorithm for partitioning. When using the k-means++ algorithm for partitioning, the number of partitions must first be determined. The specific steps are as follows:

[0136] 3.1.1 Initial parameter setting: Set the initial number of clusters to 2, use the k-means++ algorithm for clustering, and calculate the sum of clustering errors under the current number of clusters;

[0137] 3.1.2 Error and Difference Calculation: Calculate the difference between the sum of errors of the current clustering result and the sum of errors of the previous clustering result. If the difference is less than the set threshold, the optimal number of partitions is considered to be the current number of clusters minus 1; otherwise, the number of clusters is incremented by 1, and the above steps are repeated until the condition is met.

[0138] 3.2 During the second-stage partitioning, the MFGCI governance area is equivalent to a generalized node, which is included within the VDAPF governance area. Therefore, this generalized node area is subject to collaborative governance by VDAPF and MFGCI, and the governance devices exchange information through edge servers. This invention proposes to construct a partitioning evaluation function by combining three partitioning evaluation indicators: modularity index, real-time communication index, and workload balancing index. The partitioning evaluation function is solved using a simulated annealing algorithm to ultimately determine the VDAPF control area. Specifically, this includes:

[0139] 3.2.1 Modularity Metric: The modularity metric is an evaluation indicator for measuring the quality of partitioning results. The modularity value ranges from 0 to 1, with a higher value indicating better partitioning results. The expression for the modularity Q is shown below:

[0140]

[0141] In the formula, A ij A is a weight matrix consisting of the weights between node i and node j. ij ∈[0,1]; m is the sum of the weights of all edges in the network; The sum of the weights of the lines connected to node i; c ic j These are the community numbers of nodes i and j, respectively.

[0142] Indicator D i,j S represents the electrical distance or coupling strength between nodes. i,j This represents the combined impact of changes in node i (primarily considering the equivalent harmonic conductance changes of the harmonic mitigation equipment connected to it) on the fundamental voltage stability and harmonic voltage distortion of node j. Index D i,j and S i,j Since they are not of the same magnitude, the Min-Max normalization method is used to normalize the two indicator data separately, as shown in the following formula:

[0143] A ij =γD′ i,j +(1-γ)S′ i,j (13)

[0144] In the formula, D′ i,j S′ i,j They represent the indicators D respectively. i,j and S i,j The normalization; γ is the weighting factor, which is generally taken as 0.4-0.6;

[0145] 3.2.2 Communication Real-Time Performance Indicators: The greater the signal transmission delay, the greater the impact on real-time performance, and the less conducive it is to real-time harmonic voltage control. Therefore, it is necessary to minimize the communication real-time performance indicators. The communication real-time performance indicator T is:

[0146]

[0147] In the formula, T d T p T represents the data transmission and propagation time, respectively; r Indicates router latency; P d Indicates data size; v d v p L represents the data transmission rate of the communication network and the data transmission rate of the channel, respectively. p λ represents the channel length; g represents the number of routers; r θ represents the average data arrival rate; θ represents the router's service efficiency.

[0148] 3.2.3 Workload Balancing Indicators: The number of observation nodes within each edge server's control area varies, and the load on each observation node also differs, resulting in variations in the workload of each edge server. A more balanced workload across edge servers indicates a more reasonable control area for the governance equipment, preventing any single governance device from operating at high intensity for extended periods and extending its operating time. For any given edge server, its workload is:

[0149]

[0150] In the formula, w i For the workload of edge server i; l j ρ is the load of observation node j; p is the number of observation nodes within the control area of ​​edge server i.

[0151] Given k edge servers, what is the average workload of all edge servers in the network? for:

[0152]

[0153] Variance is used to measure workload balance; therefore, the smaller the workload balance index, the more balanced the workload. This yields the workload balance index L. load for:

[0154]

[0155] The partition evaluation function is constructed as follows:

[0156]

[0157] The partition is considered optimal when the partition evaluation function g reaches its minimum value. Let ε = 0.3 and φ = 0.4. With the goal of minimizing the partition evaluation function, the simulated annealing algorithm is used to obtain the optimal partitioning result. Through two-stage dynamic partitioning, the control areas of two types of governance equipment are determined respectively, realizing dynamic partitioning of the governance area of ​​the governance equipment. This can better cope with dynamically changing harmonics and provide a basis for intraday short-timescale rolling optimization and correction.

[0158] Step 4: Based on the dynamic partitioning results, a multi-timescale cloud-edge collaborative optimization governance strategy is proposed to optimize the governance parameters;

[0159] Specifically, based on the results of the two dynamic partitioning methods, day-ahead long-term scale collaborative optimization is completed on the cloud server. The cloud server, relying on its powerful computing and storage capabilities, primarily undertakes non-real-time, global tasks such as harmonic characteristic analysis, data storage and analysis, and long-term harmonic operating parameter optimization. Therefore, day-ahead long-term scale optimization and governance are performed on the cloud server. Based on historical data, day-ahead predictions of grid harmonics and active power output from new energy sources are made, with a 15-minute long-term optimization cycle, establishing a THDv (Total Harmonic Discharge) scale for all observation nodes across the entire network. iWith minimizing the harmonic conductance as the optimization objective, a day-ahead long-timescale optimization is performed on the equivalent harmonic conductance of the governance equipment. On the cloud server, a simulated annealing algorithm is used to solve the day-ahead long-timescale optimization problem, ultimately obtaining the optimized solution set consisting of the conductance values ​​of each governance device under different harmonics. Simultaneously, the obtained conductance values ​​are distributed to the edge server, as shown below:

[0160]

[0161] In the formula, THDv i n represents the total voltage distortion rate at observation node i; m R represents the total number of observation nodes in distribution network area m; M represents the total number of areas; R i Let be the weight value of node i, which is also the PR value of node i; where:

[0162]

[0163] In the formula, U 1,i U is the effective value of the fundamental voltage at node i; h,i Let be the effective value of the h-th harmonic voltage at node i.

[0164] The constraints are for the treatment capacity of the treatment equipment, where the treatment capacity of VDAPF must not exceed the rated capacity, and the treatment capacity of MFGCI should not exceed the remaining capacity, as follows:

[0165]

[0166] in:

[0167]

[0168] In the formula: P j,MFGCI (t) represents the actual active power output of the MFGCI at time t; S′ j,MFGCI (t) represents the actual remaining capacity at node j of the configured MFGCI at time t; S i,VDAPF S represents the rated capacity of the VDAPF at node i; j,N For the rated capacity of MFGCI; G h,i To configure the h-th harmonic equivalent conductance of the VDAPF at node i.

[0169] Based on the optimization results and intraday forecast data distributed by the cloud server over a long period of time, an intraday short-term optimization correction function for the edge server is constructed with a 5-minute optimization correction period and the change in conductivity of the governance equipment within each region as the optimization variable. This optimization function uses the THDv of each observation node within each VDAPF partition as the optimization variable. i The objective is to minimize the sum, and the constraints are consistent with the long-term optimization objective function.

[0170] The intraday short-timescale optimization correction objective function for region c at time t is:

[0171]

[0172] In the formula, THDv i,t n is the total harmonic voltage distortion rate of observation node i within region c at time t; c U represents the number of observation nodes contained in region c; 1,i,t U represents the effective value of the fundamental voltage at node i within region c at time t; h,i,t Let be the effective value of the h-th harmonic voltage of node i in region c at time t.

[0173] For partition c, the harmonic current prediction deviation of the observed node is given by the following equation:

[0174]

[0175] In the formula, ΔI h,i The prediction bias value injected into the h-th harmonic current at observation node i; ΔG h,i,t Y represents the h-th equivalent harmonic conductance increment of node i connected to the governance device at time t; i,i Let be the self-admittance of node i.

[0176] The objective function for the intraday short-term rolling optimization correction of the constructed edge server is a multi-objective function, and the number of objective functions is the same as the number of governance areas of the governance device.

[0177] Step 5: To address the short-term optimization performance issue of edge servers, an improved NSGA-II multi-objective optimization algorithm is proposed to enhance governance effectiveness.

[0178] Specifically, to obtain more effective governance parameters, an improved NSGA-II multi-objective optimization algorithm is used to solve the intraday short-timescale optimization correction objective function. The final optimized solution set represents the change in conductivity of each governance device under different harmonics. For example... Figure 4 As shown, the specific steps of the improved NSGA-II multi-objective optimization algorithm are as follows:

[0179] S1. Set initial parameters and initialize the population.

[0180] S2. Use the non-dominated sorting algorithm to sort the individuals in the population, determine the non-dominated level and crowding distance of the individuals, and obtain the Pareto front.

[0181] S3. The iteration begins, and a binary competition selection method is used to select an individual from the current population as the parent.

[0182] S4. Offspring generation. The process of generating offspring consists of six steps:

[0183] 4.1 The offspring generated in step three using the binary competition selection method are used as the initial population;

[0184] 4.2 Calculate the fitness value of each individual, sort the individuals according to the fitness value, and find the position of the optimal individual;

[0185] 4.3 Calculate the balance factor and the whale's fall probability. The balance factor determines whether the beluga whale optimization algorithm shifts from exploration to development. The balance factor can be obtained from the following formula:

[0186] B f =B0(1-t / 2T) max (27)

[0187] In the formula, B0 represents a random number between (0,1) in each iteration; t represents the current iteration number; T max This indicates the maximum number of iterations.

[0188] Balance factor B f When the value is greater than 0.5, it is considered the exploration stage, with a balance factor of B. f When B is ≤0.5, it is considered the development stage. As the number of iterations increases, B... f The range decreases from (0,1) to (0,0.5), indicating that the probability of entering the development and exploration phase changes, and the probability of entering the development phase increases with the number of iterations.

[0189] Won's probability of falling f We obtain it from the following formula:

[0190] W f =0.1-0.5t / T max (28)

[0191] 4.4 Determine if the balance factor is greater than 0.5. The exploration phase is established by considering the swimming behavior of beluga whales. When B f >0.5, enter the exploration phase, updating individual positions based on random numbers and roulette wheel selection. The development phase is inspired by the predation behavior of beluga whales, which can cooperate in hunting and change positions based on the locations of their companions. When B... f If the value is ≤0.5, the development phase begins, where the individual's position is updated based on random numbers and roulette wheel selection. The mathematical model for position update in both phases is shown below:

[0192] Exploration phase:

[0193]

[0194] In the formula, It is the new position of the i-th beluga whale in the j-th dimension; p jIt is randomly selected from d dimensions, where j = 1, 2, ..., d; It is the position of the i-th beluga whale in dimension j; This indicates the current position of the i-th beluga whale; This represents the current location of the r-th beluga whale, where r represents a randomly selected beluga whale; r1 and r2 are random numbers between 0 and 1, used to enhance the random operators in the exploration phase;

[0195] Based on the difference between odd and even dimensions, the updated position reflects the synchronous or mirror behavior of beluga whales when swimming or diving.

[0196] Development phase:

[0197]

[0198] In the formula, It is the current position of the i-th beluga whale; It is the current location of a random beluga whale; It is the optimal position for the i-th beluga whale; This is the optimal location among the whales; r3 and r4 are random numbers in the range (0,1), used to enhance the random operators during the exploration phase; C1 is used to measure the random jump intensity of the Levy flight intensity, C1 = 2r4(1-t / T) max );L F It is the Levy flight function. Where u and v are normally distributed random numbers,

[0199] 4.5 When B f ≤W f Then, the individual enters the whale fall behavior phase and updates its location. The mathematical model for location update is as follows:

[0200]

[0201] Where r5, r6, and r7 are random numbers between (0, 1); u b l b C1 and C2 are the upper and lower bounds of the variables, respectively; C2 is the step size factor related to the whale descent probability and population size, C2 = 2W. f ×n, where n is the population size.

[0202] 4.6 Determine if the maximum number of iterations has been reached. If not, repeat steps 4.3-4.6. If the maximum number of iterations has been reached, the updated individual position is the generated descendant.

[0203] S5. Merge the parent and offspring generations to form a new population.

[0204] S6. Use the non-dominated sorting algorithm again to sort the new population and calculate the crowding.

[0205] S7. Select a new population based on the non-dominated ordination results.

[0206] S8. Determine if the number of iterations has reached the predetermined value. If not, repeat steps S3 to S7. If it has, the latest population obtained is the optimal solution.

[0207] Step 6: Based on the governance parameters optimized by the cloud server over multiple time scales, the edge server controls the governance equipment to achieve time-varying harmonic collaborative governance.

[0208] To verify the advantages of the improved NSGA-II multi-objective optimization algorithm proposed in this invention, two metrics, hypervoume (HV) and inverted generational distance (IGD), are used to measure the quality of the obtained Pareto approximate optimal solution set. HV measures the volume of the objective space; a larger HV indicates a larger hypervoume occupied by the solution set, better convergence, and a more uniform distribution. Its expression is as follows:

[0209]

[0210] In the formula, ν(x,p) represents the hypervolume of the space formed between solution x and reference point p in the approximate optimal solution set X obtained by the algorithm.

[0211] IGD is used to measure the distance between the solution generated by an algorithm and the true Pareto. The smaller the IGD, the better the algorithm's diversity and convergence. Its expression is as follows:

[0212]

[0213] In the formula, d i Let x be the Euclidean distance between the reference point x on the Pareto approximation front X and the nearest solution p; p is a point in the optimal solution set P.

[0214] The implementation process of this solution is illustrated below with a specific example.

[0215] Figure 5 This invention provides a configuration scheme for a harmonic scenario. After two-stage dynamic partitioning on the cloud server, the final partitioning result of the IEEE 33 nodes under two governance device configuration schemes for a harmonic scenario is as follows: Figure 4As shown, different governance areas are represented by different colors, with the red circle indicating the MFGCI governance area. The red nodes are observation nodes; during partitioning, it's necessary to ensure a balanced number of observation nodes within each VDAPF governance area to guarantee a balanced workload for each VDAPF during harmonic governance. When harmonic fluctuations occur that significantly alter the partitioning results, a two-stage dynamic partitioning approach is used to recalculate the partitioning and dynamically update the results. Configuration scheme one involves configuring 3 VDAPFs, and configuration scheme two involves configuring 4 VDAPFs.

[0216] To verify the advantages of the proposed long-term harmonic co-management using VDAPF and MFGCI cloud servers, this invention selects three harmonic optimization and management strategies for comparison, as shown below:

[0217] Strategy 1: The present invention proposes to use VDAPF and MFGCI for long-term harmonic synergistic management.

[0218] Strategy 2: Use VDAPF alone for long-term scale governance.

[0219] Strategy 3: Point-to-point decentralized governance, using CDAPF alone for harmonic control.

[0220] Taking the harmonics in Scenario 1 as the target of governance, Strategies 1 and 2 use the THDv of all observed nodes across the network. i The optimization objective is to minimize the sum. The simulated annealing algorithm is used for optimization. The installation location and capacity of each treatment device in the two strategies are shown in Table 1. The two configuration schemes are used for verification.

[0221] Table 1 CDAPF Configuration Parameters

[0222]

[0223] See Figure 6 In (a) and (b), the THDv of each node under three governance strategies with two governance equipment configuration schemes in harmonic scenario 1 i .

[0224] Comparing the three mitigation strategies under two different mitigation device configurations in a harmonic scenario, using the CDAPF mitigation strategy alone on CDAPF-installed nodes is effective in mitigating harmonics, but THDv still exists on nodes without CDAPF installed. i The exceeding of limits indicates that traditional decentralized harmonic mitigation methods are no longer suitable for the distributed time-varying harmonics of "high-voltage" distribution networks. However, the harmonic mitigation strategy using VDAPF alone can reduce the THDv of all nodes in the distribution network. iWhile significantly reduced, some nodes still fail to meet governance requirements. This invention proposes a collaborative governance strategy using VDAPF and MFGCI, which enables all network-wide observation nodes to achieve THDv under both configuration schemes. i All values ​​were below the limit of 4%, indicating that long-term collaborative optimization can effectively improve the harmonic distortion of the distribution network, and the effect is better than using VDAPF alone.

[0225] Figure 7 THDv of each observation node before and after mitigation in a harmonic scenario provided as an example of the present invention i When harmonics fluctuate within the long-term collaborative optimization governance cycle of the cloud server, and the virtual conductance of the long-term collaborative optimization governance (strategy one) using VDAPF and MFGCI is still used for governance, it will cause the node THDv to increase. i If the limit is exceeded, it is necessary to perform rolling optimization correction on the harmonic virtual conductance based on the partitioning results using short-term rolling optimization on the edge server. To verify the necessity and effectiveness of two-stage dynamic partitioning, three governance strategies under two governance device configuration schemes are compared in a harmonic scenario.

[0226] Strategy 4: Perform short-time-scale rolling optimization and governance based on the two-stage dynamic partitioning results of this invention;

[0227] Strategy 5: Perform short-timescale rolling optimization and governance based on the first-stage partitioning results (MFGCI region);

[0228] Strategy 6: Perform short-timescale rolling optimization governance based on the second-stage partitioning results (VDAPF region).

[0229] Depend on Figure 7 It can be seen that when harmonics fluctuate during the long-term optimization and governance cycle of the cloud server, the THDv of certain observation nodes will increase. i It may exceed the limit. In this case, consider using the governance equipment to perform rolling optimization and correction governance in the governance area to reduce the node THDv. i The result of less than 4% validates the necessity of partitioning. Comparing different strategies, it can be seen that the governance region considering both VDAPF and MFGCI, after rolling optimization correction, performs better than the governance region considering only VDAPF or MFGCI.

[0230] To verify the advantages of the improved NSGA-II multi-objective optimization algorithm proposed in this invention compared to NSGA-II, this invention uses two metrics, HV and IGD, to measure the quality of the obtained Pareto approximate optimal solution set, and selects commonly used two-objective test functions ZDT1, ZDT2, ZDT3, ZDT4, and ZDT6 as test functions. The Pareto optimal solution sets for each test function are obtained using the two algorithms.

[0231] The HV and IGD of each test function obtained using the two algorithms are shown in the table.

[0232] Table 2 HV and IGD for each test function

[0233]

[0234] A larger HV indicates a larger hypervolume for the solution set, better convergence, and a more uniform distribution. A smaller IGD indicates better algorithm diversity and convergence. As shown in the table, the Pareto optimal solution sets obtained by the algorithm proposed in this invention for each test function are better than those of NSGA-II, proving that the method proposed in this invention performs better than NSGA-II.

[0235] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for coordinated harmonic mitigation in a dual-high-voltage power distribution network based on cloud-edge collaboration and dynamic partitioning, characterized in that, Includes the following steps: Step 1: Construct a time-varying harmonic collaborative governance architecture based on cloud-edge collaboration. The cloud server is responsible for selecting observation nodes, dynamic partitioning, and optimizing governance parameters, while the edge server is responsible for collecting harmonic characteristics and the status of governance equipment, and controlling the governance equipment to perform harmonic governance. Step 2: Based on the cloud-edge collaborative time-varying harmonic governance architecture, virtual nodes are designed to establish a derived network link matrix, taking into account node efficiency and comprehensive sensitivity index. An observation node selection method based on the derived network link matrix and an improved PageRank algorithm is proposed to select a limited number of observation nodes to effectively reflect the harmonic governance effect of the entire network. The cloud-edge collaborative time-varying harmonic governance architecture includes cloud servers deployed on the power Internet of Things cloud platform, several edge servers deployed on distribution network substations or key nodes, governance equipment, data monitoring devices installed on distribution network observation nodes, and communication networks. The governance equipment includes a voltage-sensing active filter (VDAPF) and a multi-functional grid-connected inverter (MFGCI), which regulates harmonic voltage through virtual harmonic conductance. The data monitoring device is used to collect harmonic voltage / current data in real time and upload it to the edge server; The communication network uses the OPC UA protocol to achieve three-level data transmission between the cloud server, edge server, governance equipment, and data monitoring device. Step 3: Based on the harmonic state of the observation nodes and the state information of the treatment equipment, a two-stage dynamic zoning method is proposed to realize the dynamic zoning of the treatment equipment area, ensuring the accuracy of zoning and the balance of the number of nodes in the area. The two-stage dynamic partitioning process includes: 3.1 The first stage of partitioning uses comprehensive sensitivity as the partitioning index to establish a partitioning matrix, then determines the number of partitions, and then uses the k-means++ algorithm for partitioning; 3.2 During the second-stage partitioning, the MFGCI governance area is equivalent to a generalized node, which is contained within the VDAPF governance area. Therefore, the generalized node area is subject to collaborative governance by VDAPF and MFGCI, and the governance devices communicate with each other through edge servers. A partitioning evaluation function is constructed by combining three partitioning evaluation indicators: modularity, real-time communication, and workload balancing. The second-stage partitioning evaluation function is solved using the simulated annealing algorithm to finally determine the control area of ​​VDAPF. Step 4: Based on the dynamic partitioning results, a multi-timescale cloud-edge collaborative optimization governance strategy is proposed to optimize the governance parameters; Step 5: Based on the governance parameters, propose an improved NSGA-II multi-objective optimization algorithm to further dynamically correct the governance parameters on a short time scale; Step 6: Based on the multi-timescale optimized governance parameters of the cloud server, the edge server controls the governance equipment to achieve time-varying harmonic collaborative governance.

2. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, The task division between the cloud server and the edge server is as follows: (1) Observation node selection: The edge server is responsible for data acquisition, harmonic feature extraction, and calculation of voltage reactive power sensitivity and harmonic sensitivity index; the cloud server is responsible for the spatiotemporal distribution of harmonics in the whole network, harmonic state prediction, and selection of observation nodes using the improved PageRank algorithm. (2) Two-stage partitioning: The edge server is responsible for calculating the comprehensive sensitivity, modularity index, communication real-time index and workload balancing index, and uploading them to the cloud server; The cloud server is responsible for performing two-stage partitioning using the k-means++ algorithm and simulated annealing algorithm, with a time scale of 5 minutes, and dynamically updating the partitioning results based on harmonic state information and the capacity of the governance equipment. (3) Harmonic mitigation: The edge server is responsible for providing the cloud server with harmonic and mitigation equipment status information, for communication between mitigation equipment, and for controlling the mitigation equipment based on the mitigation parameters; Based on the day-ahead harmonic and remaining capacity prediction data of the governance equipment, and the partitioning results, the cloud server uses the simulated annealing algorithm to optimize the governance parameters on a day-ahead long-term time scale with a 15-minute time scale. Based on intraday harmonic and governance equipment remaining capacity prediction data and zoning results, an improved NSGA-I multi-objective optimization algorithm is used to optimize and correct the governance parameters on a 5-minute time scale intraday, and the corrected governance parameters are sent to the edge server.

3. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, Step 2 specifically includes the following: A network link matrix is ​​constructed using node efficiency and comprehensive sensitivity indices. For a distribution network with n nodes, the link matrix is ​​constructed considering the relationships between the nodes. As shown below: (1) In the formula, Let be the weight of the edge between node i and node j, that is, the overall influence of node i on node j. , n is the number of nodes in the network; According to the equivalent diagram of the derived network structure, virtual nodes and other nodes are not actually connected by lines; the diagram only reflects the comprehensive influence between virtual nodes and other nodes. The derived network link matrix... As shown below: (2) When satisfied ,and hour: (3) When satisfied ,and hour: (4) When satisfied ,and hour: (5) In the formula, , These are the elements in the network link matrix and the elements in the derived network matrix, respectively; An improvement to the PageRank algorithm, the resulting Google matrix. : (6) In the formula, Let be the link matrix of the (n+1)th order derivative network; For the weight value, ; In the traditional PageRank algorithm, the PR value of each webpage is evenly distributed among all the webpages it points to; self-restarting vector The element values ​​in the array are uniformly distributed, where The self-restart vector is used to reflect the observer's level of interest in webpage content; in power systems, when the harmonic voltage of a node changes, it generally affects other nodes located on the shortest path between nodes; therefore, the betweenness index is introduced to... Redefining: (7) In the formula, For functions corresponding to the derived link network; as follows: (8) (9) (10) In the formula, The number of shortest paths between node j and node k; Let i be the number of nodes that pass through node i in the shortest path between nodes j and k. ; ; .

4. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, When using the k-means++ algorithm for partitioning, the number of partitions must first be determined. The specific steps are as follows: 3.1.1 Initial parameter settings: Set the initial number of clusters to 2, use the k-means++ algorithm to perform clustering, and calculate the sum of clustering errors under the current number of clusters; 3.1.2 Error and difference calculation: Calculate the difference between the sum of errors of the current clustering result and the sum of errors of the previous clustering result; if the difference is less than the set threshold, the optimal number of partitions is considered to be the current number of clusters minus 1. Otherwise, increment the cluster count by 1 and repeat the above steps until the condition is met.

5. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, The modularity metric, the real-time communication metric, and the workload balancing metric are specifically as follows: 3.2.1 Modularity Index Modularity is an evaluation metric for measuring the quality of partitioning results. Modularity values ​​range from 0 to 1; a higher value indicates better partitioning results. The expression is as follows: (11) (12) In the formula, It is a weight matrix composed of the weights between node i and node j. m is the sum of the weights of all edges in the network; The sum of the weights of the lines connected to node i; , These are the community numbers of nodes i and j, respectively; index Represents the electrical distance or coupling strength between nodes. It characterizes the degree of combined impact of changes at node i on the fundamental voltage stability and harmonic voltage distortion at node j; due to the index and Since they are not on the same order of magnitude, the Min-Max normalization method is used to normalize the two indicator data separately, as shown in the following formula: (13) In the formula, , Representing indicators and Normalization; This is a weighting factor, with a value ranging from 0.4 to 0.6; 3.2.2 Communication Real-Time Performance Indicators The greater the signal transmission delay, the greater the impact on real-time performance, and the less conducive it is to real-time harmonic voltage control. Therefore, it is necessary to minimize the communication real-time performance indicators. for: (14) In the formula, , These represent the data sending and propagation times, respectively. Indicates router latency; Indicates the size of the data; , These represent the data transmission rate of the communication network and the data transmission rate of the channel, respectively. Indicates the channel length; Indicates the number of routers; Indicates the average data arrival rate; Indicates the router's service efficiency; 3.2.3 Workload Balancing Indicators The number of observation nodes within the control area of ​​each edge server varies, and the load of each observation node also varies, resulting in differences in the workload of each edge server. A more balanced workload among the edge servers indicates a more reasonable control area by the governance device, preventing any single governance device from operating at high intensity for extended periods and extending its operating time. For any given edge server, its workload is: (15) In the formula, For the workload of edge server i; ρ represents the load of observation node j; p represents the number of observation nodes within the control area of ​​edge server i. Given k edge servers, what is the average workload of all edge servers in the network? for: (16) Workload balancing is measured using variance; therefore, a smaller workload balancing metric indicates better load balancing. for: (17) The partition evaluation function is constructed as follows: (18) The partition is considered optimal when the partition evaluation function g reaches its minimum value. ,Pick , , ; With the goal of minimizing the partition evaluation function, the simulated annealing algorithm is used to obtain the optimal partitioning result. Through two-stage dynamic partitioning, the control areas of two types of governance equipment are determined respectively, realizing the dynamic partitioning of the governance area of ​​the governance equipment. This can better cope with dynamically changing harmonics and provide a basis for intraday short-timescale rolling optimization and correction.

6. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, Step 4 specifically includes: Based on historical data, day-ahead forecasts are made for power grid harmonics and active power output from new energy sources. An optimization period is established using a 15-minute timescale, and a network of observation nodes is established. With minimizing the sum as the optimization objective, the equivalent virtual harmonic conductance of the governance equipment is optimized on a day-ahead long-time scale. On the cloud server, the simulated annealing algorithm is used to solve the day-ahead long-time scale optimization problem, and the final optimized solution set is the conductance value of each governance equipment under different harmonics. At the same time, the obtained conductance values ​​are sent to the edge server, as shown in the following formula: (19) In the formula, The total voltage distortion at observation node i; is the total number of observation nodes in distribution network area m; M is the total number of areas; Let be the weight value of node i, which is also the PR value of node i; where: (20) In the formula, Let be the effective value of the fundamental voltage at node i; Let be the effective value of the h-th harmonic voltage at node i; The constraints are for the treatment capacity of the treatment equipment, where the treatment capacity of VDAPF must not exceed the rated capacity, and the treatment capacity of MFGCI should not exceed the remaining capacity, as follows: (21) (22) in: (23) In the formula: Let t be the actual active power output of MFGCI at time t; The actual remaining capacity at node j of the configured MFGCI at time t; Let VDAPF be the rated capacity at node i; The rated capacity of MFGCI; To configure the h-th harmonic equivalent conductance of VDAPF at node i of VDAPF; Based on the optimization results and intraday forecast data distributed by the cloud server over a long period of time, an intraday short-term optimization correction function for the edge server is constructed with a 5-minute optimization correction cycle and the change in conductivity of the governance equipment within each region as the optimization variable. This optimization correction function uses the observation nodes within each VDAPF partition as the basis for the optimization correction function. The objective is to minimize the sum, and the constraints are consistent with the long-term optimization objective function. The intraday short-timescale optimization correction objective function for region c at time t is: (24) (25) In the formula, Let be the total harmonic voltage distortion rate of the observed node i within region c at time t; The number of observation nodes contained in region c; Let be the effective value of the fundamental voltage of node i in region c at time t; Let be the effective value of the h-th harmonic voltage of node i in region c at time t; For partition c, the harmonic current prediction bias of the observed nodes is: The harmonic power flow equations are as follows: (26) In the formula, The predicted deviation value injected into the h-th harmonic current of observation node i; The h-th equivalent virtual harmonic conductance increment of node i connected to the governance device at time t; Let be the self-admittance of node i.

7. The method for harmonic collaborative management of high-voltage power distribution networks based on cloud-edge collaboration and dynamic partitioning as described in claim 1, characterized in that, In step 5, the improved NSGA-II multi-objective optimization algorithm incorporates the Beluga optimization algorithm, including exploration and development phases; the mathematical models for position updates in the two phases are shown below: Exploration phase: (29) In the formula, It is the first A beluga whale in the A new position on the dimension; It is randomly selected from d dimensions, where ; It is the first A beluga whale in Position in dimensions; Indicates the first The current location of the beluga whale; Indicates the first The current location of the beluga whale. The beluga whale is represented by a randomly selected number; r1 and r2 are random numbers between 0 and 1, used to enhance the random operators in the exploration phase. Development phase: (30) In the formula, It is the first The current location of the beluga whale; It is the current location of a random beluga whale; It is the first The best spot for a beluga whale; It is the optimal location among the whales; r3 and r4 are random numbers in the range (0,1) used to enhance the random operators in the exploration phase; Random jump intensity used to measure Levy flight intensity ; It is the Levy flight function. Where u and v are normally distributed random numbers, , .

Images