Distributed resource cooperative dynamic partitioning and interval parallel recovery optimization method and system

Abstract

The application discloses a distributed resource cooperative dynamic partitioning and interval parallel recovery optimization method and system, relates to the technical field of resource cooperation optimization, and solves the problems of poor dynamic adaptability, insufficient consideration of resource constraints and lack of parallel coordination mechanism in post-disaster recovery of the prior art. The application collects real-time running data of each node after an extreme event, constructs a state vector, forms an isolated node set, further calculates the supportable power capacity of each node, establishes a partition optimization objective function integrating multiple objectives, solves the optimal partition scheme, considers the user importance level, power demand proportion and electrical distance from the power supply node in the partition, calculates a load recovery priority index, sorts the load recovery priority index to form a load recovery priority queue in the partition, and finally obtains an optimal closing sequence of each partition based on the priority queue and performs a recovery action. Through the above technical means, the application guarantees the priority recovery of key loads and the coordinated and orderly overall recovery process.

Description

Technical Field

[0001] This invention relates to the field of resource collaborative optimization technology, and in particular to a distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method and system. Background Technology

[0002] With the deepening of the construction of new power systems, the penetration rate of distributed resources on the distribution network side is increasing, and the power grid structure is showing highly decentralized and complex characteristics. Under the impact of extreme events, the distribution network is prone to large-scale failures, leading to power outages in multiple areas and forming a large number of interconnected electrical islands. How to quickly identify the status of these islands, scientifically divide the restoration areas, and efficiently execute load restoration has become a key issue in ensuring the resilience and reliability of the power system. Especially in distribution networks with a high proportion of distributed resources, each node has different supporting power capacity. If the supporting capacity of these distributed resources cannot be fully utilized for coordinated restoration, it will seriously affect the efficiency and quality of post-disaster power restoration. Therefore, there is an urgent need for an optimized method that can integrate real-time operational data, coordinate multi-source distributed resources, and achieve regional coordination and parallel restoration to improve the self-healing capability and recovery speed of the distribution network under extreme events.

[0003] Existing technologies have significant shortcomings in addressing the aforementioned scenarios. Currently, most distribution network fault recovery strategies primarily employ a centralized recovery model, where a dispatch center collects network-wide information and formulates recovery plans. However, in cases of communication outages or master station failures due to extreme events, the centralized model struggles to respond quickly due to its reliance on global information exchange. Furthermore, it fails to adequately consider the differentiated support capabilities of distributed resources within each electrical island, leading to uneven resource allocation and low recovery efficiency. In addition, existing methods often employ simple fixed-priority strategies during the load recovery phase, failing to comprehensively consider multi-dimensional factors such as user importance level, power demand ratio, and electrical distance from power supply nodes. This results in delayed recovery of critical loads and an unreasonable recovery sequence, further prolonging the overall recovery time. These shortcomings directly constrain the rapid self-healing and efficient recovery of distribution networks under extreme events.

[0004] Therefore, a distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method and system are needed. Summary of the Invention

[0005] To address the problems of low recovery efficiency caused by the inability of existing centralized recovery models to adapt to communication-constrained scenarios under extreme events and the lack of sufficient coordination of the differentiated support capabilities of distributed resources, this invention provides a method and system for distributed resource collaborative dynamic partitioning and parallel interval recovery optimization. This method enables accurate characterization of fault scenarios through real-time status perception and electrical islanding identification; scientific partitioning and collaborative scheduling of distributed resources through power capacity calculation and multi-objective partitioning optimization; and orderly parallel recovery within partitions through load recovery priority index calculation based on multi-dimensional factor fusion and optimal closing sequence generation. These technical means effectively overcome the response bottleneck of existing centralized recovery models, fully leverage the support potential of distributed resources, significantly improve the recovery speed and power supply reliability of distribution networks under extreme events, and ensure the priority recovery of critical loads and the coordinated and efficient overall recovery process. The specific technical solution is as follows: The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method includes the following steps: Real-time operating data of each node is collected after an extreme event occurs to obtain a state vector. Based on the state vector, all disconnected electrical islands are identified. All nodes contained in an electrical island are combined into an island node set. Calculate the supportable power capacity of each node in the island node set, and establish a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector. Solve the multi-objective partitioning optimization objective function to obtain the optimal partitioning scheme. Based on the optimal partitioning scheme, taking into account the user's importance level, power demand ratio, and electrical distance from the power nodes within the partition, the load restoration priority index is calculated, and the load restoration priority index of each node is sorted to form a load restoration priority queue within the partition. The optimal closing sequence for each partition is obtained based on the load restoration priority queue within the partition, and the restoration action is performed using the optimal closing sequence.

[0006] Preferably, the state vector includes at least: nodes i The measured line voltage, effective value of three-phase current, real-time predicted output value of distributed photovoltaic, current state of charge of energy storage system, current active load demand, and connection status of connected line switches.

[0007] The preferred method for calculating the supported power capacity is as follows: In the formula, Let be the power capacity that node i can support; Let be the real-time power output prediction value of the distributed photovoltaic system at node i; This represents the current state of charge of the energy storage system at node i. Let be the rated energy capacity of the energy storage system at node i; This is the expected recovery time window; This is the correction factor for energy storage availability.

[0008] The preferred objective function for multi-objective partitioning optimization is expressed as follows: In the formula, F is the comprehensive cost index of the zoning scheme; M This represents the current number of partitions. For the total load demand of the m-th partition, Let m be the set of nodes in the m-th partition; Let m be the total power capacity that the m-th partition can support. This represents the power deficit in the m-th partition; This represents the total load demand within the isolated island. For resource utilization rate of each zone Standard deviation; For weights.

[0009] Preferably, the load restoration priority index is calculated as follows: In the formula, Let be the load recovery priority index for node j; This is the load importance coefficient; Let j be the active power load requirement of node j; Total load within the zone; Let J be the electrical distance from node j to the nearest power node within the partition. For partitioning The maximum value of this type of electrical distance; These are the weighting coefficients.

[0010] Preferably, the process of obtaining the optimal closing sequence for each partition based on deep reinforcement learning is as follows: Establish a Markov decision process model, wherein the elements of the Markov decision process model include state, action, and reward; The state is defined as: In the formula, Let be the five-dimensional state input vector at time t; This represents the current average SOC of each energy storage node within the partition; This represents the predicted total photovoltaic output within the zone. This represents the average priority index of load nodes that have not yet been restored. This represents the number of nodes that have been restored. This represents the maximum node voltage deviation within the partition at the current moment. The action is defined as follows: Select one switch from the set of currently unclosed switches to perform a closing operation, and immediately perform fast power flow calculation to update the voltage distribution after closing; Define the reward function: In the formula, The immediate reward obtained after performing the closing action at time t; The priority index of the nodes restored in this decision step is used to provide positive rewards and incentives to prioritize the restoration of high-importance loads. This represents the maximum voltage deviation within the zone after the circuit is closed. The upper limit of the allowable voltage deviation, This is an indicator function that takes the value 1 when the SOC of any energy storage node is lower than a set threshold, and 0 otherwise. , , Weighting of each reward; Based on a three-layer fully connected structure consisting of an input layer, a hidden layer, and an output layer, the system takes a five-dimensional state input vector as input, outputs the Q-value of each possible action, and selects the action with the largest Q-value as the decision.

[0011] Preferably, during the recovery operation using the optimal closing sequence, the state of charge of the energy storage at node i at time t is updated, and the current mean value of the SOC of each energy storage node in the partition in the input vector is updated based on the state of charge of the energy storage at node i at time t.

[0012] Preferably, during the recovery operation using the optimal closing sequence, the boundary power value of partition m is updated and used as the boundary condition for power flow calculation in each partition, thereby updating the current average SOC of each energy storage node in the partition.

[0013] Preferably, after the optimal closing sequence is completed, the comprehensive index of the overall network recovery effect is calculated. When the comprehensive index of the overall network recovery effect is greater than the set threshold, the process terminates. Otherwise, the partition with the largest power deficit is marked as the object to be re-optimized. Based on the latest updated state of each node in the marked partition, the power capacity that can be supported is recalculated, and the optimal closing sequence is obtained again until the comprehensive index of the overall network recovery effect is greater than the set threshold.

[0014] A distributed resource collaborative dynamic partitioning and inter-area parallel recovery optimization system, applied to the method described above, includes a network status acquisition module, an electrical islanding identification module, a dynamic partitioning optimization module, a recovery path planning module, an inter-area power coordination module, and a comprehensive evaluation and feedback module, wherein: The network status acquisition module collects real-time operating data of each node after an extreme event occurs, obtains a status vector, and transmits the status vector to the electrical islanding identification module. The electrical island identification module identifies all disconnected electrical islands based on state vectors. All nodes contained in an electrical island are combined into an island node set, which is then transmitted to the dynamic partitioning optimization module. The dynamic partitioning optimization module calculates the supportable power capacity of each node in the island node set, and establishes a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector. The optimal partitioning scheme is obtained by solving the multi-objective partitioning optimization objective function, and finally the optimal partitioning scheme is transmitted to the recovery path planning module. The recovery path planning module is based on the optimal partitioning scheme. It comprehensively considers the user's importance level, power demand ratio, and electrical distance to the power nodes in the partition. It calculates the load recovery priority index and sorts the load recovery priority index of each node to form a load recovery priority queue in the partition. Then, it obtains the optimal closing sequence for each partition based on the load recovery priority queue in the partition. The inter-regional power coordination module updates the boundary power values ​​of each region in real time and communicates bidirectionally with the recovery path planning module, thereby affecting the output results of the recovery path planning module. The comprehensive evaluation and feedback module calculates the overall network recovery effect index and, based on the overall network recovery effect index, decides whether to trigger the dynamic partition optimization module and the recovery path planning module to update the optimal partition scheme and the optimal closing sequence.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention accurately identifies all disconnected electrical islands by collecting real-time operational data from each node after an extreme event and constructing a state vector, forming a set of island nodes. It then calculates the supportable power capacity of each node, establishes a multi-objective partitioning optimization objective function, and solves it to obtain the optimal partitioning scheme, achieving scientific planning and dynamic partitioning of distributed resources. Based on this, it comprehensively considers user importance level, power demand ratio, and electrical distance to power nodes within the partition, calculates a load restoration priority index, and sorts them to form a load restoration priority queue within the partition. Finally, based on the priority queue, it obtains the optimal closing sequence for each partition and executes the restoration action. Through the above technical means, this solution achieves partitioned collaborative optimization and parallel restoration of distributed resources, effectively solving the problems of slow response, uneven resource allocation, and unreasonable load restoration priorities in the centralized restoration mode of existing technologies. It significantly improves the restoration efficiency and power supply reliability of the distribution network under extreme events, ensuring the priority restoration of critical loads and the coordinated and orderly overall restoration process. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 Here is a flowchart of the real-time feedback correction process for DQN based on voltage over-limit; Figure 3 This is a flowchart of the partition adaptive optimization process based on a comprehensive index. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] Example 1 In one embodiment of the present invention, a distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method is provided, characterized in that, as Figure 1 As shown, it includes the following steps: Step 1: Collect real-time operational data of each node after the occurrence of an extreme event to obtain a state vector; Step 2: Identify all disconnected electrical islands based on state vectors. All nodes contained in an electrical island are combined into an island node set. Step 3: Calculate the power capacity that each node in the island node set can support; Step 4: Establish a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector, and solve the multi-objective partitioning optimization objective function to obtain the optimal partitioning scheme; Step 5: Based on the optimal partitioning scheme, taking into account the user's importance level, power demand ratio, and electrical distance to power nodes within the partition, calculate the load restoration priority index, and sort the load restoration priority indices of each node to form a load restoration priority queue within the partition. Step 6: Obtain the optimal closing sequence for each partition based on the load restoration priority queue within the partition, and execute the restoration action using the optimal closing sequence.

[0023] The above scheme collects real-time operational data from each node after extreme events and constructs state vectors to accurately identify all disconnected electrical islands, forming a set of island nodes. It then calculates the supportable power capacity of each node, establishes a multi-objective partitioning optimization objective function, and solves it to obtain the optimal partitioning scheme, achieving scientific planning and dynamic partitioning of distributed resources. Based on this, it comprehensively considers user importance level, power demand ratio, and electrical distance to power supply nodes within the partition, calculates the load restoration priority index, and sorts them to form a load restoration priority queue within the partition. Finally, based on the priority queue, it obtains the optimal closing sequence for each partition and executes the restoration action. Through these technical means, this scheme achieves partitioned collaborative optimization and parallel restoration of distributed resources, effectively solving the problems of slow response, uneven resource allocation, and unreasonable load restoration priorities in existing centralized restoration modes. It significantly improves the restoration efficiency and power supply reliability of the distribution network under extreme events, ensuring the priority restoration of critical loads and the coordinated and orderly overall restoration process.

[0024] Example 2 This embodiment is a further refinement of Embodiment 1, such as... Figures 1-3 As shown, the following provides a further implementation of each step in Example 1: Step 1: Collect real-time operational data of each node after an extreme event occurs to obtain a state vector.

[0025] This step involves collecting real-time operational data from each node after an extreme event occurs, using the Distribution Automation Terminal Unit (DTU), distributed photovoltaic controller, energy storage management system (BMS / PCS), and charging pile control unit. For the i-th node in the distribution network (… , N (where the total number of nodes in the distribution network is ), and its state vector is defined as: (1) In the formula, For nodes i The measured line voltage was obtained from the DTU analog signal acquisition board, with a sampling accuracy of no less than 0.2%. Let be the effective value of the three-phase current at node i; The real-time power output prediction value of the distributed photovoltaic system at node i is calculated by the photovoltaic controller based on the current irradiance and temperature using the MPPT algorithm and then sent up. The current state of charge (SOC, value range [0,1]) of the energy storage system at node i is sent up by the BMS in real time. Let i be the current active power load demand of node i. The connection status of the circuit switch connected to node i ( Indicates closure. (Indicates disconnection). Node state vector It serves as the unified data source for all subsequent computing modules, with a collection cycle set to 1 second; Each component will be extracted and used one by one in subsequent steps: The data is passed to the second step for island identification. and The data is passed to the third step for capacity calculation. The fourth and fifth steps are passed in for partition optimization and priority calculation.

[0026] Step 2: Identify all disconnected electrical islands based on state vectors. All nodes contained in an electrical island are combined into an island node set.

[0027] This step uses the connectivity status of all branches of the entire network collected in the first step { Given a node i, use a depth-first search (DFS) algorithm to traverse the entire network and identify all disconnected electrical islands. The necessary and sufficient condition for node i and node j to belong to the same electrical island is: (2) In the formula, The connection state of the switch on branch (m,n) is directly taken from the state vector of each node in the first step, which corresponds to the branch. Quantity; This is the set of all branches on the path from node i to node j in the topology graph; a product of 1 indicates that all switches on the path are closed, and nodes i and j belong to the same island; otherwise, they are different islands. After the DFS traversal is completed, module B outputs a list of island node sets. (where K is the total number of isolated islands,) (This is the set of nodes for the Kth isolated island), and this list is directly passed to the third step as the basis for the network structure of the partition optimization.

[0028] Step 3: Calculate the power capacity that each node in the island node set can support.

[0029] This step applies to each isolated island. Each node i within the node is determined according to the photovoltaic output in the node state vector equation (1). and energy storage Calculate the node's recovery time window. (For 2 hours) of available power capacity: (3) In the formula, Let be the power capacity that node i can support; and Take directly from equation (1); Let be the rated energy capacity of the energy storage system at node i, and be the known parameters on the equipment nameplate. This is the expected recovery time window; The energy storage availability correction factor is determined by comprehensively considering the minimum SOC protection threshold (retaining 20% ​​to protect battery life) and charge / discharge efficiency (typical value 0.95), according to the following rules: When hour, ;when hour, (The energy storage at this node does not participate in the support). The calculation results will be used as key constraint parameters for partition optimization in the fourth step, and will also be used to construct a spectral clustering similarity matrix to reflect the degree of resource complementarity between nodes.

[0030] Step 4: Establish a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector, and solve the multi-objective partitioning optimization objective function to obtain the optimal partitioning scheme.

[0031] This step is for isolated islands. The nodes within the system, based on the resource capacity calculated in step three, can support power capacity. The load demand in equation (1) Based on this, a multi-objective partitioning optimization objective function is established: (4) In the formula, F is the overall cost index of the partitioning scheme; M is the current number of partitions; Let m be the total load demand of the m-th partition. Let m be the set of nodes in the m-th partition; The total power capacity that the m-th partition can support (from equation (3)). This represents the power deficit in this partition. Total load demand within the island (for normalization); Resource utilization rate of each zone The standard deviation measures the degree of resource balance across intervals; weights The Analytic Hierarchy Process (AHP) is used to determine the priority based on the principle of emergency recovery; the recommended value is [value to be filled in]. , , ( The maximum value reflects the primary goal of eliminating power deficit. The solution to equation (4) employs spectral clustering, with the third step... Construct the node similarity weight matrix based on the above, and calculate its normalized spectral Laplace matrix: (5) In the formula, The normalized spectral Laplace matrix; This is a node similarity matrix. When there is a connected path between node i and node j, its elements... ( , From formula (3); (Take the mean square error of the capacity difference between all connected nodes). For a degree matrix, the diagonal elements .extract The eigenvectors corresponding to the first M smallest eigenvalues ​​form an embedding matrix. K-means clustering is then performed on these embedding vectors to obtain the node partitioning scheme. By iterating through M from 2 to the total number of power nodes in the island... Select the partitioning scheme that minimizes F in equation (4) and output it to the fifth and sixth steps.

[0032] Step 5: Based on the optimal partitioning scheme, taking into account the user's importance level, power demand ratio, and electrical distance from the power nodes within the partition, calculate the load restoration priority index, and sort the load restoration priority indices of each node to form a load restoration priority queue within the partition.

[0033] This step, after the partitioning scheme is determined, involves processing each partition. For each load node j within the zone, considering the user's importance level, power demand percentage, and electrical distance from power supply nodes within the zone, calculate the load restoration priority index: (6) In the formula, is the load recovery priority index for node j (value range [0,1], the larger the value, the higher the priority for recovery); The load importance coefficient is preset according to user category (Level 1 important users, including hospitals and emergency command centers, take 1.0; Level 2 users take 0.7; general civil loads take 0.3). The active power load demand of node j (from equation (1)) ), The total load within the zone (from equation (4)); The electrical distance from node j to the nearest power node within the partition (measured by the sum of line impedances along the path). For partitioning The maximum value of this type of electrical distance (used for normalization; the closer the distance, the larger the third term and the higher its priority). The three weighting coefficients satisfy the following conditions: The recommended value is determined using the entropy weight method based on historical typical scenario data. , , (User importance is given the highest weight). Each node After sorting, a load recovery priority queue is formed within the partition, which is directly used as the input component of the DQN state vector in step 6.

[0034] Step 6: Obtain the optimal closing sequence for each partition based on the load restoration priority queue within the partition, and execute the restoration action using the optimal closing sequence.

[0035] This step applies to each recovery partition. Module D establishes a Markov Decision Process (MDP) model and uses DQN to solve for the optimal closing sequence online. At each decision step t, the input vector observed by the DQN agent is: (7) In the formula, Let be the five-dimensional state input vector of the DQN at time t; The current average SOC of each energy storage node within the partition (calculated in real time and fed back from the SOC update equation in step 7; the initial value is taken from each node). (mean) The predicted total photovoltaic output within the zone is given by equation (1) for each node. (and the sum is updated on a rolling basis). The average priority index of load nodes that have not yet recovered (from equation (6), which is dynamically updated according to the recovery status of the nodes); This represents the number of nodes that have been restored. The maximum node voltage deviation within the partition at the current moment (from the constraint check feedback after ADMM coordination in step 8, initially set to 0). The DQN action space is defined as follows: select one switch from the set of currently unclosed switches to perform a closing operation; immediately after closing, perform fast power flow calculation to update the voltage distribution. The reward function is designed as follows: (8) In the formula, The immediate reward obtained after performing the closing action at time t; The priority index of the node restored in this decision step (from Equation (6)) is used to give a positive reward to encourage the priority restoration of high-importance loads; The maximum voltage deviation within the zone after closing the circuit (the next state is updated by the power flow calculation results and passed into equation (7), and obtained through the feedback in step eight); The upper limit of the allowable voltage deviation is set at 0.07 (i.e., ±7% of the rated voltage) according to GB / T 12325-2008. A negative penalty will be imposed if the deviation is exceeded. The indicator function is set to 1 when the SOC of any energy storage node (updated by equation (9)) is lower than the 20% protection threshold, and 0 otherwise. , , The weights for each reward were determined through hyperparameter tuning during pre-training in the RTDS simulation environment. The design principle was to determine the severity of the penalty for exceeding the limits. , ) significantly greater than positive reward ( To ensure safety, the DQN network adopts a three-layer fully connected structure: the input layer is 5-dimensional (corresponding to the five components of equation (7)), the two hidden layers each have 128 neurons, and the output layer dimension corresponds to the number of switches that can be closed. The optimal closing sequence output by DQN is passed to the seventh step for execution, and after execution, the updated state is returned to equation (7) to drive the next decision.

[0036] Example 3 This embodiment adds steps seven to nine to step nine on the basis of embodiment 1 or embodiment 2, which are used to update process parameters and evaluate recovery effect during the execution of recovery path, thereby achieving iterative optimization based on feedback.

[0037] Step 7: During the recovery operation using the optimal closing sequence, update the state of charge (SOC) of the energy storage at node i at time t, and update the current mean SOC of each energy storage node within the partition in the input vector based on the SOC of the energy storage at node i at time t. Specifically: During the recovery path execution, as the closing action gradually progresses, the energy storage systems at each node continuously discharge to support the newly added load, and their SOC is updated in real time according to the following formula: (9) In the formula, Let be the state of charge at time t for the energy stored at node i, with the initial value taken from equation (1). ; The discharge power of the energy storage node at time t is determined by the simplified power flow equation within the partition based on the current closing topology. The time step is 1 minute. Energy storage discharge efficiency (typical value 0.95); The rated energy capacity for energy storage is consistent with the definition in equation (3). After each time step update, Return to equation (7) The components form an internal closed loop of planning-execution-SOC feedback: when the average SOC continues to decline and approaches the 20% threshold, the indicator function in equation (8) triggers a penalty, and DQN automatically adjusts the subsequent closing strategy to slow down the discharge rate or prioritize switching to the topology path that can increase photovoltaic output, thereby realizing the protective scheduling of energy storage.

[0038] Step 8: During the recovery process using the optimal closing sequence, the boundary power value on the m-side of partition is updated, and this boundary power value is used as the boundary condition for power flow calculation in each partition, thereby updating the current average SOC of each energy storage node within the partition. Specifically: During parallel recovery of multiple partitions, power exchange is required between adjacent partition boundary nodes. Module E defines the partitions. Adjacent partitions The transmission power of the connecting line between them is (Positive value indicates from) Flow direction Each partition coordinates its boundary power iteratively through ADMM. The local subproblem of the next iteration is updated as follows: (10) In the formula, For the power of the m-side boundary of partition m in the first... The updated value for the next iteration; Let m be the local cost function of partition m, which is reflected in the boundary exchange power as The weighted sum of node voltage deviation and energy storage call-up cost within the partition under the given conditions is calculated by the linearized power flow equation within the partition (based on the current closing state of the DQN); For the first The mean consensus power of the partition boundary power on both sides of the next iteration is calculated as follows: (The two partitions only need to exchange this value through the boundary node); For dual variables (Lagrange multipliers), after each iteration, according to... renew; The ADMM penalty parameter is determined using a line search method to find the value that maximizes convergence. A value of [value to be filled in] is recommended. When both the original residual and the dual residual are less than Convergence is determined at a certain time. The boundary power after convergence. As boundary conditions for power flow calculation in each zone: on the one hand, update the discharge power of each node in equation (9). This drives the accurate calculation of the SOC; on the other hand, it updates the calculated voltage of each node and inputs it into equation (7). The component drives the DQN to make the next closing decision. When the voltage of any node exceeds the rated voltage ±7%, the over-limit event is recorded and counted in the constraint violation count of step nine. Meanwhile, the penalty term in equation (8) Suppressing subsequent closing operations that exceed limits creates internal constraints for power coordination and path planning.

[0039] Step Nine: After the optimal closing sequence is executed, calculate the overall network recovery effect index. The process terminates when the overall network recovery effect index is greater than the set threshold; otherwise, mark the partition with the largest power deficit as the re-optimization target. Recalculate the supportable power capacity based on the latest updated state of each node in the marked partition, and re-obtain the optimal closing sequence until the overall network recovery effect index is greater than the set threshold. Details are as follows: The parallel recovery of each partition has been completed (or the preset time limit has been reached). After (hours), module F calculates the overall network recovery effect index: (11) In the formula, The comprehensive recovery effect index (value range [0,1], the larger the value, the better the recovery effect); This is the set of load nodes for which power has been successfully restored. For the set of all load nodes; The load importance coefficient (with values ​​consistent with equation (6): 1.0 for primary users, 0.7 for secondary users, and 0.3 for general users). Let be the active power load demand of node j (from equation (1)), and the first term be the weighted load recovery rate; The actual total recovery time (in hours). The first item is the maximum allowed recovery time in hours, and the second item is the recovery timeliness score. This is the cumulative number of voltage over-limit events that occurred during the recovery process (from the record of the ADMM constraint check in step 8). The total number of closing operations is represented by the first item, and the third item is the operational safety score. , , These are three weighted coefficients, determined based on the priority principle of emergency response to extreme events (load restoration amount is the highest priority, followed by timeliness, and then safety). When When the recovery plan is deemed effective, the process is terminated; At that time, module F marks the partition with the largest power deficit as the target for re-optimization, and updates the latest state of each node in that partition after step seven (including the updated state). ,current and The data is then sent back to step three to recalculate the available power capacity. This triggers the fourth step of partition re-optimization, generating an adjusted partition scheme to continue executing subsequent steps, forming a complete outer layer optimization driven by the evaluation results of equation (11).

[0040] To further illustrate the implementation effect of this embodiment, simulation verification is performed below: This simulation example is based on a real 10kV distribution network scenario in a certain region. The distribution network includes one main line (15 km long) and three branch lines (4 km, 6 km, and 5 km long respectively), with a line impedance of 0.38 ohms per km. The network is connected to four distributed photovoltaic (total installed capacity 3.6 MW), six energy storage systems (total capacity 8.4 MWh, average SOC 0.72), one charging pile group (connected capacity 1.2 MW), and various loads (total capacity approximately 11 MW, including 1.8 MW primary load, 3.2 MW secondary load, and 6.0 MW general load). A distribution network simulation model with the above resources was built on the MATLAB / Simulink platform to simulate a scenario where a typhoon extreme event causes five main line faults and disconnections, resulting in four electrical islands within the network, with uneven distribution resource allocation within each island. The simulation platform runs three schemes for comparison: the traditional serial recovery plan method, the serial recovery method based on pure topology partitioning, and the parallel recovery optimization method proposed in this patent. The comparison indicators include partitioning scheme quality, load recovery efficiency, and constraint satisfaction.

[0041] Table 1 Comparison of Partitioning Schemes The power deficit of all five partitions generated by this patented method is zero, and the resource balance (standard deviation 0.12) is far superior to that of the pure topology partitioning method (0.38). Moreover, the online generation latency is 22 seconds, which meets the engineering requirement of no more than 30 seconds.

[0042] Table 2 Performance Comparison of Recovery Path Planning The DQN method significantly outperforms the greedy algorithm in all metrics, improving the weighted load recovery rate by 15.5 percentage points, reducing the average recovery time by 38.0%, and decreasing the number of voltage over-limits from 9 to 1, thus verifying the effectiveness of the reward function design and the state feedback mechanism.

[0043] Table 3. Overall Recovery Effect Evaluation Results Crysis 4 Initial Assessment Overall Index This triggered a re-optimization of the outer closed loop, and after re-partitioning, the energy storage scheduling strategy was adjusted, ultimately... The efficiency was improved to 0.88, verifying the effectiveness of the closed-loop feedback mechanism. Compared with the traditional serial contingency plan method, the average weighted load recovery rate of the entire network using this patented method increased from 71.2% to 90.6%, and the network recovery time was shortened from 138 minutes to 59 minutes, a reduction of 57.2%.

[0044] Example 4: In one embodiment of the present invention, a distributed resource collaborative dynamic partitioning and inter-region parallel recovery optimization system is provided, applied to the method described above. It includes a network status acquisition module, an electrical islanding identification module, a dynamic partitioning optimization module, a recovery path planning module, an inter-region power coordination module, and a comprehensive evaluation and feedback module, wherein: The network status acquisition module collects real-time operating data of each node after an extreme event occurs, obtains a status vector, and transmits the status vector to the electrical islanding identification module. The electrical island identification module identifies all disconnected electrical islands based on state vectors. All nodes contained in an electrical island are combined into an island node set, which is then transmitted to the dynamic partitioning optimization module. The dynamic partitioning optimization module calculates the supportable power capacity of each node in the island node set, and establishes a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector. The optimal partitioning scheme is obtained by solving the multi-objective partitioning optimization objective function, and finally the optimal partitioning scheme is transmitted to the recovery path planning module. The recovery path planning module is based on the optimal partitioning scheme. It comprehensively considers the user's importance level, power demand ratio, and electrical distance to the power nodes in the partition. It calculates the load recovery priority index and sorts the load recovery priority index of each node to form a load recovery priority queue in the partition. Then, it obtains the optimal closing sequence for each partition based on the load recovery priority queue in the partition. The inter-regional power coordination module updates the boundary power values ​​of each region in real time and communicates bidirectionally with the recovery path planning module, thereby affecting the output results of the recovery path planning module. The comprehensive evaluation and feedback module calculates the overall network recovery effect index and, based on the overall network recovery effect index, decides whether to trigger the dynamic partition optimization module and the recovery path planning module to update the optimal partition scheme and the optimal closing sequence.

[0045] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0046] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0047] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0048] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0049] 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 therein. Such 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, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method, characterized in that, Includes the following steps: Real-time operating data of each node is collected after an extreme event occurs to obtain a state vector. Based on the state vector, all disconnected electrical islands are identified. All nodes contained in an electrical island are combined into an island node set. Calculate the supportable power capacity of each node in the island node set, and establish a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector. Solve the multi-objective partitioning optimization objective function to obtain the optimal partitioning scheme. Based on the optimal partitioning scheme, the load restoration priority index is calculated by taking into account the user's importance level, power demand ratio, and electrical distance from the power nodes in the partition. The load restoration priority index of each node is then sorted to form a load restoration priority queue in the partition. The optimal closing sequence for each partition is obtained based on the load restoration priority queue within the partition, and the restoration action is performed using the optimal closing sequence.

2. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 1, characterized in that, The state vector includes at least: nodes i The measured line voltage, effective value of three-phase current, real-time predicted output value of distributed photovoltaic, current state of charge of energy storage system, current active load demand, and connection status of connected line switches.

3. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 2, characterized in that, The calculation method for the supported power capacity is as follows: In the formula, Let be the power capacity that node i can support; Let be the real-time power output prediction value of the distributed photovoltaic system at node i; This represents the current state of charge of the energy storage system at node i. Let be the rated energy capacity of the energy storage system at node i; This is the expected recovery time window; This is the correction factor for energy storage availability.

4. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 1, characterized in that, The objective function for multi-objective partitioning optimization is expressed as follows: In the formula, F is the comprehensive cost index of the zoning scheme; M This represents the current number of partitions. For the total load demand of the m-th partition, Let m be the set of nodes in the m-th partition; Let m be the total power capacity that the m-th partition can support. This represents the power deficit in the m-th partition; This represents the total load demand within the isolated island. For resource utilization rate of each zone Standard deviation; For weights.

5. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 1, characterized in that, The load restoration priority index is calculated as follows: In the formula, Let be the load recovery priority index for node j; This is the load importance coefficient; Let j be the active power load requirement of node j; Total load within the zone; Let J be the electrical distance from node j to the nearest power node within the partition. For partitioning The maximum value of this type of electrical distance; These are the weighting coefficients.

6. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 1, characterized in that, The process of obtaining the optimal closing sequence for each partition based on deep reinforcement learning is as follows: Establish a Markov decision process model, wherein the elements of the Markov decision process model include state, action, and reward; The state is as follows: In the formula, Let be the five-dimensional state input vector at time t; This represents the current average SOC of each energy storage node within the partition; This represents the predicted total photovoltaic output within the zone. This represents the average priority index of load nodes that have not yet been restored. This represents the number of nodes that have been restored. This represents the maximum node voltage deviation within the partition at the current moment. The action is as follows: select one switch from the set of currently unclosed switches to perform a closing operation, and immediately perform fast power flow calculation to update the voltage distribution after closing; The reward function: In the formula, The immediate reward obtained after performing the closing action at time t; The priority index of the nodes restored in this decision step is used to provide positive rewards and incentives to prioritize the restoration of high-importance loads. This represents the maximum voltage deviation within the zone after the circuit is closed. The upper limit of the allowable voltage deviation, This is an indicator function that takes the value 1 when the SOC of any energy storage node is lower than a set threshold, and 0 otherwise. , , Weighting of each reward; Based on a three-layer fully connected structure consisting of an input layer, a hidden layer, and an output layer, the system takes a five-dimensional state input vector as input, outputs the Q-value of each possible action, and selects the action with the largest Q-value as the decision.

7. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 6, characterized in that, During the recovery process using the optimal closing sequence, the state of charge of the energy storage at node i at time t is updated, and the current mean value of the SOC of each energy storage node in the partition in the input vector is updated based on the state of charge of the energy storage at node i at time t.

8. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 7, characterized in that, During the recovery process using the optimal closing sequence, the boundary power value of partition m is updated and used as the boundary condition for power flow calculation in each partition, thereby updating the current average SOC of each energy storage node in the partition.

9. The distributed resource collaborative dynamic partitioning and interval parallel recovery optimization method according to claim 8, characterized in that, After the optimal closing sequence is completed, the overall network recovery effect index is calculated. The process terminates when the overall network recovery effect index is greater than the set threshold. Otherwise, the partition with the largest power deficit is marked as the re-optimization object. The supported power capacity is recalculated based on the latest updated state of each node in the marked partition, and the optimal closing sequence is obtained again until the overall network recovery effect index is greater than the set threshold.

10. A distributed resource collaborative dynamic partitioning and interval parallel recovery optimization system, characterized in that, The method applied to any one of claims 1 to 9 includes a network status acquisition module, an electrical islanding identification module, a dynamic partitioning optimization module, a recovery path planning module, an inter-partition power coordination module, and a comprehensive evaluation feedback module, wherein: The network status acquisition module collects real-time operating data of each node after an extreme event occurs, obtains a status vector, and transmits the status vector to the electrical islanding identification module. The electrical island identification module identifies all disconnected electrical islands based on state vectors. All nodes contained in an electrical island are combined into an island node set, which is then transmitted to the dynamic partitioning optimization module. The dynamic partitioning optimization module calculates the supportable power capacity of each node in the island node set, and establishes a multi-objective partitioning optimization objective function based on the supportable power capacity and state vector. The optimal partitioning scheme is obtained by solving the multi-objective partitioning optimization objective function, and finally the optimal partitioning scheme is transmitted to the recovery path planning module. The recovery path planning module is based on the optimal partitioning scheme. It comprehensively considers the user's importance level, power demand ratio, and electrical distance to the power nodes in the partition. It calculates the load recovery priority index and sorts the load recovery priority index of each node to form a load recovery priority queue in the partition. Then, it obtains the optimal closing sequence for each partition based on the load recovery priority queue in the partition. The inter-regional power coordination module updates the boundary power values ​​of each region in real time and communicates bidirectionally with the recovery path planning module, thereby affecting the output results of the recovery path planning module. The comprehensive evaluation and feedback module calculates the overall network recovery effect index and, based on the overall network recovery effect index, decides whether to trigger the dynamic partition optimization module and the recovery path planning module to update the optimal partition scheme and the optimal closing sequence.

Images