Black start recovery sequence generation method and system

Abstract

The application discloses a black start recovery sequence generation method and system, and the method comprises the following steps: for a black start request, an improved A-star algorithm embedded in a graph database is used to perform A-star search with pruning on a power grid attribute graph model corresponding to a target power grid to obtain a corresponding search result; and an optimal recovery path node sequence is generated based on the search result; wherein, based on the improved A-star algorithm, the steps of determining a current expansion node and performing node expansion are as follows: a node with the minimum evaluation function value is taken out from a priority queue as the current expansion node; when the current expansion node does not coincide with the target node, corresponding candidate nodes are obtained based on the power grid attribute graph model, the candidate nodes are adjacent nodes connected with the current expansion node, and the candidate nodes satisfy preset electrical safety constraints; and each candidate node is added to the priority queue. The application can effectively improve search efficiency and reduce calculation cost.

Description

Technical Field

[0001] This invention relates to the field of power grid operation control technology, and in particular to a method and system for generating black-start recovery sequences based on graph databases and an improved A-star algorithm. Background Technology

[0002] Following a major power outage, the black start recovery process of the power system is a crucial link in ensuring the safety of social power supply. The generation of a black start recovery path requires starting from units with self-starting capabilities (black start power sources) and finding the optimal power transmission path to start units without self-starting capabilities (non-black start units) or restore important loads.

[0003] Currently, the generation of the black boot path mainly relies on three methods: First, dispatchers manually create dispatching data based on experience, which is inefficient and difficult to adapt to changes in power grid topology. Second, offline computation based on traditional graph theory algorithms (such as depth-first search (DFS) and breadth-first search (BFS)) or mathematical programming methods (such as integer programming) usually simplifies the power grid into an unweighted graph or a statically weighted graph, and performs post-process verification in conjunction with power flow checks. The third is the power grid modeling scheme based on graph databases. This type of scheme only uses graph databases for data management and executes core path search algorithms (such as Dijkstra's algorithm and integer programming) in a computing engine outside the graph database. Summary of the Invention

[0004] The purpose of this invention is to address the problems of low search efficiency, high computational overhead, difficulty in adapting to dynamic changes in power grid topology, and inability to quickly generate optimal recovery sequences that meet security constraints in existing methods. This invention proposes a black-start recovery sequence generation method based on graph databases and an improved A-star algorithm.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: Firstly, a method for generating a black boot recovery sequence is provided, including the following steps: For black start requests, an improved A-star algorithm embedded in the graph database is used to perform a pruned A-star search on the power grid attribute graph model corresponding to the target power grid to obtain the corresponding search results; and an optimal recovery path node sequence is generated based on the search results. The black start request is used to indicate the starting node and the target node to be restored; The power grid attribute graph model uses the busbars in the target power grid as nodes and the primary equipment in the target power grid as the connecting edges between nodes. Each node has a node dynamic attribute indicating its current energized state, and each edge has an edge dynamic attribute indicating its current switching state. Based on the improved A-star algorithm, the steps for determining the current expansion node and performing node expansion are as follows: The node with the smallest evaluation function value is taken from the priority queue as the current expansion node. The evaluation function value of the current node is determined based on the actual cost from the starting node to the current node and the heuristically estimated cost from the current node to the target node. When the current extended node coincides with the target node, a corresponding optimal recovery path node sequence is generated based on the priority queue; When the current expanding node does not overlap with the target node: Based on the power grid attribute graph model, corresponding candidate nodes are obtained. The candidate nodes are adjacent nodes connected to the current extended node, and the candidate nodes satisfy preset electrical safety constraints. Add each candidate node to the priority queue.

[0006] As one possible implementation method: The formula for calculating the actual cost is as follows: ; in: L ij For the cost of electrical distance, L ij =|Z ij |,Z ij For the corresponding impedance; T ij This comes at the cost of operation time. P ij For the cost of reactive charging power, P ij =Q c / Q limit Q c The charging power of the line, Q limit This represents the reactive power absorption limit of the corresponding black-start generator set. w1, w2, and w3 are all weighting coefficients, and they satisfy w1 + w2 + w3 = 1.

[0007] Existing technologies often treat electrical constraints as binary pass / fail judgments, which can only be verified after the fact. This application transforms the self-excitation risk into a continuously accumulative cost term by designing a reactive charging power cost, enabling the algorithm to quantitatively assess the accumulation of risk during the search process, thus achieving an improvement from simply whether it is safe to how safe it is.

[0008] As one possible implementation method: The adjacency relationships of each node are physically stored in the form of a linked list.

[0009] As one possible implementation, the electrical safety constraints include self-excitation verification and overvoltage verification: The self-excitation verification is as follows: when X c ≤k*X d If a node is found to have a self-excitation risk, the adjacent node will be pruned; where X c X is the line charging capacitance corresponding to the connection edge from the current extended node to the corresponding adjacent node. d The synchronous reactance corresponding to the starting node; The overvoltage verification is as follows: when K ov ≥K limit If an overvoltage risk is detected, the corresponding adjacent node is pruned; where K limit The maximum overvoltage multiple; the connection edge from the current extended node to the corresponding adjacent node is taken as the closing edge, K ov To increase the closing overvoltage multiple of the path corresponding to the adjacent node, it is estimated based on the cumulative charging power of the path from the starting node to the adjacent node and the charging power of the side to be closed.

[0010] As one possible implementation, while adding each candidate node to the priority queue, the parent pointer mapping and cost mapping are also updated based on the current actual cost of each candidate node. The parent pointer mapping is used to record the predecessor node of each node in the preferred queue; The cost mapping is used to record the optimal actual cost corresponding to each node in the preferred queue; The current actual cost is the actual cost from the starting node through the current extended node to the corresponding candidate node. When the candidate node is not recorded in the priority queue, the candidate node is added to the priority queue, and the current extended node is recorded as the predecessor node of the candidate node in the parent pointer mapping. The current actual cost corresponding to the candidate node is recorded as the optimal actual cost corresponding to the candidate node in the cost mapping. When the candidate node already exists in the priority queue; If the current actual cost corresponding to the candidate node is less than the optimal actual cost recorded in the cost mapping, then the predecessor node of the candidate node in the parent pointer mapping is updated to the current extended node, and the optimal actual cost in the cost mapping is updated to the current actual cost; otherwise, the parent pointer mapping and the cost mapping are not updated.

[0011] As one possible implementation method, the current actual cost corresponding to a candidate node is calculated as follows: Calculate the single-step cost from the current expansion node to the candidate node; Based on the single-step cost, the actual cost from the starting node to the current extended node is calculated to obtain the corresponding current actual cost; Specifically, the outgoing edge from the current expansion node to the candidate node is taken as the target edge. Based on the impedance magnitude of the target edge, the operating time and charging power of the corresponding device, and the preset generator self-excitation limit threshold, the single-step cost from the current expansion node to the candidate node is calculated.

[0012] As one possible implementation, the outgoing edge from the current expanded node n to the candidate node m is taken as the target edge, and the corresponding single-step cost cost(n,m) is calculated using the following formula: cost(n,m)=w4*|Z|+w5*T op +w6*Q c / Q limit ; Where |Z| is the impedance magnitude of the target side; T op Q represents the operation time of the device corresponding to the target edge. c Charging power of the target side; Q limit The corresponding black-start unit's reactive power absorption limit; w4, w5, and w6 are all weighting coefficients, satisfying w4+w5+w6=1.

[0013] As one possible implementation method: When calculating the reactive charging power cost of each node, the cumulative reactive charging power of each path is statistically analyzed based on the preferred queue and parent pointer mapping. When Q is detected... cum >α*Q limit The multi-constraint processing mechanism is triggered at any time, where Q cum Q represents the cumulative reactive charging power along the corresponding path. limit α represents the reactive power absorption limit of the corresponding black start unit, and α is a preset approach threshold. When the multi-constraint processing mechanism is triggered, the weighting coefficient corresponding to the reactive charging power cost in the actual cost calculation is adjusted based on the dynamic weighting strategy, as follows: ; Where β is the weighting amplification factor.

[0014] As one possible implementation, when a multi-constraint processing mechanism is triggered, a reactor is detected and activated based on a local optimization strategy to absorb reactive power and reduce the cumulative reactive power charging power of the corresponding path. Specifically: Search for parallel reactor nodes whose switches are closed around the current node; When the reactor node exists, a merged node is generated based on the reactor node and the current node, and the physical anchor point of the merged node is still the current node; The operational cost T corresponding to the reactor node r Determine the cumulative reactive charging power, actual cost, heuristic cost, and evaluation function value of the merging node: in: The cumulative reactive charging power of the current path is updated to max(0,Q). cum -Q r ), Q cum Q represents the cumulative reactive charging power for the current path. r This refers to the reactive power absorption capacity corresponding to the reactor node; The actual cumulative cost of merging node s' is g(s') = g(n) + T r Let g(n) be the actual cost corresponding to the current node n, and T be the cost of the node. r The cost of the operation corresponding to the reactor node; The heuristic cost of merging node s' is the heuristic cost corresponding to the current node n; When merging node s' and its evaluation function value are included in the preferred queue, the reactor nodes that are put into operation are recorded.

[0015] Secondly, the present invention proposes a black boot recovery sequence generation system for executing the method described in any one of the above.

[0016] This invention, by adopting the above technical solutions, has significant technical effects: This invention completely eliminates the massive data transfer overhead (I / O bottleneck) between the application layer and the database layer in the traditional storage-compute separation architecture by embedding the improved A-star algorithm into the graph database, and realizes millisecond-level real-time generation of the black start path.

[0017] This invention performs real-time pruning of electrical safety constraints simultaneously during node expansion. Only candidate nodes that meet the safety constraints are included in the search scope, compressing the traditional iterative process of searching first and then verifying into a single step, thereby reducing the computational complexity from O(n log n). (搜索空间 × 校验次数) Reduced to O (搜索空间) It effectively improves search efficiency and reduces the consumption of computing resources, and can reduce the amount of computing power by more than 90% under the scale of provincial power grids.

[0018] The algorithm of this invention directly performs real-time traversal and decision-making based on the dynamic attributes (current energized state, switching state) of nodes and edges in the power grid attribute graph model. When the power grid topology changes due to maintenance or fault, it can automatically adapt without remodeling or manual intervention. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an example diagram of a power grid property graph model; Figure 2 This is a schematic diagram of the calculation process of determining the current expansion node and performing a node expansion based on the improved A-star algorithm of the present invention. Detailed Implementation

[0021] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0022] To address the shortcomings of existing technologies, such as low search efficiency, high computational overhead, difficulty in adapting to dynamic changes in power grid topology, and inability to quickly generate optimal recovery sequences that meet security constraints, this application discloses a black-start recovery sequence generation method, the specific steps of which are as follows: For black start requests, an improved A-star algorithm embedded in the graph database is used to perform a pruned A-star search on the power grid attribute graph model corresponding to the target power grid to obtain the corresponding search results; and an optimal recovery path node sequence is generated based on the search results. in: The black start request is used to indicate the starting node and the target node to be restored; The power grid attribute graph model uses the busbars in the target power grid as nodes and the primary equipment in the target power grid as the connecting edges between nodes. Each node has a node dynamic attribute indicating its current energized state, and each edge has an edge dynamic attribute indicating its current switching state. A-star algorithm, a heuristic path search algorithm; This embodiment deeply integrates the A-star graph search algorithm with a graph database, breaking down the architectural barriers of storage and computation separation. The search algorithm is directly executed at the data storage layer, eliminating data migration overhead. Furthermore, the electrical constraint verification logic is deeply coupled with the node expansion process of the A-star algorithm, realizing an integrated computational paradigm of search and verification, effectively improving search efficiency and reducing computational overhead.

[0023] Power grid property diagram model: Reference Figure 1 The power grid attribute graph model uses the busbars in the target power grid as nodes and the primary equipment (transmission lines, transformers, etc.) in the target power grid as the connecting edges between nodes. Each node has a node dynamic attribute indicating its current energized state, and each edge has an edge dynamic attribute indicating its current switching state.

[0024] In this embodiment, node attributes include static node attributes and dynamic node attributes, specifically: Node static attributes are used to indicate the static characteristics of the corresponding bus, such as reference voltage, upper / lower allowable voltage limit, presence of black-start units, unit capacity, synchronous reactance, and whether it is an important load node. The node dynamic attribute is used to indicate the current energized state. That is, it only contains the current energized state, where 0 represents the node being de-energized and 1 represents the node being energized.

[0025] In this embodiment, edge attributes include static edge attributes and dynamic edge attributes, specifically: Side static attributes are used to indicate the static characteristics of the corresponding primary equipment, such as impedance, susceptance, rated current, transformer ratio, charging power, line charging capacitive reactance, etc. The edge dynamic attribute is used to indicate the current switch state, that is, it only contains the switch state, where 0 indicates that the device is open and 1 indicates that the device is closed.

[0026] The node parameters or edge parameters involved in the calculation are all stored as corresponding static attributes in the edges. Those skilled in the art can configure them according to actual needs, and this embodiment does not limit them in detail.

[0027] In this embodiment, the physical model of the power system of the target power grid is structured and mapped into a corresponding attribute graph, and the obtained attribute graph is persistently stored in a graph database, realizing an integrated storage of power grid topology, equipment parameters and operating status in the power grid attribute graph model.

[0028] As one possible implementation, the native storage structure of graph databases is used to physically store the adjacency relationships of each node in the form of a linked list. During subsequent algorithm traversal, the attribute information of adjacent nodes and corresponding edges can be directly obtained through pointer addressing, eliminating the performance overhead of traditional index queries.

[0029] Improved A-star algorithm: This embodiment proposes an improved A-star algorithm with a hybrid electrical-time cost function. This algorithm runs within the graph database in the form of a user-defined function (UDF) or a stored procedure, directly traversing the graph nodes.

[0030] Those skilled in the art, knowing the improved A-star algorithm, can easily convert it into a user-defined function (UDF) or stored procedure and embed it into the corresponding graph database; therefore, this specification will not describe it in detail. This invention, based on the aforementioned power grid attribute graph model, employs an improved A-star search algorithm embedded within the graph database. It directly completes node traversal, cost calculation, and constraint verification within the graph database, eliminating the need for cross-layer data transmission, and obtains the optimal recovery path node sequence and its total path cost. The specific steps are as follows: S100. Establish a global data structure; S110, OpenList: The open list is used to store priority queues and is sorted in ascending order according to the value of the evaluation function f(n) of the nodes. The open list is initialized to be empty and the starting node S is added to the open list based on the input data, where g(S)=0 and h(S) is the electrical heuristic distance from the starting node to the target point.

[0031] In this embodiment, the evaluation function value of the current node is determined based on the actual cost from the starting node to the current node and the heuristically estimated cost from the current node to the target node. The evaluation function f(n) is: f(n) = g(n) + h(n); Where g(n) is the actual cost from the starting node S (black start power bus) to the current node n, and h(n) is the heuristically estimated cost from the current node n to the target point T (bus of the unit to be started or important load bus); The specific functional expression of the actual cost function g(n) is as follows: ; Among them, L ij For the cost of electrical distance, T ij For the cost of operation time, P ij The power cost for reactive charging; w1, w2, and w3 are all weighting coefficients, and they satisfy w1+w2+w3=1. Those skilled in the art can set them according to actual needs, and they can also be adjusted in real time by the dispatcher according to the recovery stage (initial focus on safety, later focus on speed).

[0032] Electrical distance cost L ij By the static properties of the edge (impedance Z) ij The calculation is as follows: L ij =|Z ij |; In this embodiment, the electrical distance cost L ijIt is not a simple physical distance, but rather the impedance magnitude is calculated from the corresponding impedance. The smaller the impedance, the stronger the electrical coupling and the lower the cost. In this embodiment, the operation time cost T ij It is determined by the number of switching operations or the standardized operation time of the side. For example, for transformer branches, if tap adjustment is involved, the operation time has a higher weight. Those skilled in the art can set it according to actual needs. In this embodiment, the reactive charging power cost P ij The charging power Q of the line c and the corresponding reactive power absorption limit (i.e., self-excitation limit threshold) Q of the black start unit. limit The calculation is as follows: P ij =Q c / Q limit ; For long high-voltage lines, excessive no-load charging power may lead to generator self-excitation or power frequency overvoltage. This embodiment addresses this by reducing the reactive charging power cost P. ij The design prevents self-excitation and overvoltage, and reduces reactive charging power cost P. ij The larger the value, the higher the cost, which can guide the algorithm to avoid investing in too long lines at once.

[0033] The specific functional expression for the heuristic estimation cost h(n) is as follows: ; Among them, dist elec (n,T) represents the shortest electrical distance, v max The standardization propagation speed constant; electrical shortest distance dist elec (n,T) is the shortest electrical distance across the entire network, pre-calculated and cached using Dijkstra's algorithm based on the line impedance stored in the graph database, or the impedance magnitude between two nodes calculated in real time. In this embodiment, due to the shortest electrical distance dist elec (n,T) is the shortest path length based on static impedance, and the normalized propagation speed constant v. max It must be less than or equal to the remaining part of the actual dynamic cost g(n) after considering operation time and reactive power penalty. Therefore, this heuristic function satisfies the admissibility property, that is, it will never overestimate the actual cost of the objective, which guarantees that the A-star algorithm can find the global optimum.

[0034] S120, Parent Map: Records the predecessor node of each node in the preferred queue (open list) for final path backtracking.

[0035] S130, Cost Mapping (g-value table): Records the optimal actual cost g(n) of each node in the preferred queue. Based on the parent pointer mapping, it records the optimal actual cost g(n) from the starting node to the current node. Initially, only g(S)=0 is recorded.

[0036] S200. Obtain input data and improve the A-star algorithm to complete node traversal, cost calculation and constraint verification based on the input data, and obtain the corresponding search results. In this embodiment, the starting node S (black start power supply bus), the target node T (bus of the unit to be started or important load bus), and the current real-time topology snapshot of the power grid are obtained from the power grid attribute graph model based on the black start request as input data.

[0037] Based on the improved A-star algorithm, the current expansion node is determined, and node expansion is performed based on the current expansion node until the search is completed; Reference Figure 2 In this embodiment, when the open list is not empty, the following steps are performed: S210. Determine the current expansion node: Take the node n with the smallest evaluation function value f(n) from the priority queue (open list) and use it as the current expansion node.

[0038] Since the priority queue only contains the starting node during the initial calculation, the starting node is initially used as the current expansion node for calculation, and the node with the smallest evaluation function value is used as the current expansion node thereafter.

[0039] S220, Node Expansion: If the current extended node n coincides with the target node T, the priority queue is used as the search result, and path backtracking is performed based on the priority queue to generate the corresponding optimal recovery path node sequence. Since the current extended node n does not overlap with the target node T, a corresponding candidate node is obtained based on the power grid attribute graph model. The candidate node is an adjacent node connected to the current extended node, and the candidate node satisfies the preset electrical safety constraints.

[0040] When the current extended node n does not overlap with the target node T, the specific steps for obtaining the corresponding candidate node are as follows: S221, Adjacency Node Filtering: Call the graph database's native adjacency access interface to obtain all outgoing edges of the current node n (starting from the current node n and connecting to other parent nodes) and their corresponding adjacent nodes m.

[0041] S222, Filtering switch status: Read the dynamic attribute "switch status" of the above outgoing edge. If the value of this attribute is 0 (indicating that the device is off), directly remove the corresponding adjacent node m and continue to process the next outgoing edge.

[0042] S223, Dynamic Electrical Safety Constraint Pruning (Pre-verification): In this embodiment, electrical safety constraints include self-excitation verification and overvoltage verification, with the specific rules as follows: Self-excitation verification: For the scenario of closing a long line under no-load conditions, the line charging capacitive reactance X of the connection edge from the current extended node to the corresponding adjacent node is read from the edge static attributes of the graph database. c Read the synchronous reactance X of the unit being started (usually the generator at the starting node) from the static attributes of the starting node. d If X is satisfied c ≤k*X d (k is a safety factor, which can be configured as needed by those skilled in the art), then it is determined that the path has a risk of self-excitation, the current outgoing edge is prohibited from being selected, and the corresponding adjacent node m is pruned; Overvoltage verification: Based on the local network topology of the currently recovered path (composed of nodes in the open list), the graph database aggregation function is called to calculate the cumulative charging power of the path (i.e., the cumulative charging power of the path corresponding to the starting node to the adjacent node). Combined with the charging power of the edge to be closed (the connection edge from the current extended node to the corresponding adjacent node), the corresponding closing overvoltage multiple K is estimated. ov If K ov ≥K limit (K) limit If the maximum overvoltage multiple that the system can safely withstand is specified, then the path is determined to have an overvoltage risk, the current outgoing edge is prohibited from being selected, and the corresponding adjacent node m is pruned. In this embodiment, a simplified formula based on the transient network analyzer (TNA) principle is used to estimate the closing overvoltage multiple K. ov .

[0043] S230. Add each candidate node to the priority queue; In this embodiment, while adding each candidate node to the priority queue, the parent pointer mapping and cost mapping are also updated based on the current actual cost of each candidate node. When the candidate node m is not recorded in the priority queue, the candidate node m is added to the priority queue, and the current extended node n is recorded as the predecessor node of the candidate node m in the parent pointer mapping. The current actual cost g corresponding to the candidate node m is then recorded. candidate The optimal actual cost g(m) corresponding to the candidate node m is recorded in the cost mapping; When the candidate node m already exists in the priority queue, obtain the current optimal actual cost of the candidate node m from the cost mapping; For example, the current actual cost g corresponding to candidate node m candidate If the actual cost is less than the optimal actual cost g(m) recorded in the cost map, then update the predecessor node of the candidate node m in the parent pointer map to the current extended node n, and update the optimal actual cost g(m) in the cost map to the current actual cost g. candidate Otherwise, the parent pointer mapping and cost mapping will not be updated.

[0044] When determining the next current expansion node, the evaluation function value is calculated based on the optimal actual cost of each node in the cost map.

[0045] In this embodiment, the single-step cost from the current expanded node n to the candidate node m is calculated. Based on the single-step cost and the actual cost from the starting node to the current expanded node, the current actual cost g from the starting node through the current expanded node n to the candidate node m is calculated. candidate ; Current actual cost g candidate The specific calculation method is as follows: g candidate =g(n)+cost(n,m); Where g(n) is the actual cost from the starting node to the current extended node, and cost(n,m) is the cost per step.

[0046] If g candidate If the value is less than g(m) recorded in the cost map (or m has not yet been recorded), then: Update g(m) = g candidate .

[0047] The outgoing edges from the current expanded node n to the candidate node m are taken as the target edges, and the specific calculation method for the corresponding single-step cost cost(n,m) is as follows: cost(n,m)=w4*|Z|+w5*T op +w6*Q c / Q limit ; Where |Z| is the impedance magnitude of the target side; T op Q represents the operation time of the device corresponding to the target edge. c Charging power of the target side; Q limit The reactive power absorption limit of the corresponding black start unit is read from the node static attributes of the starting node; w4, w5, and w6 are all weighting coefficients, satisfying w4+w5+w6=1. Those skilled in the art can set them according to actual needs, or they can be dynamically adjusted according to the black start recovery stage.

[0048] During node expansion, those skilled in the art can set the execution order of dynamic electrical safety constraint pruning and single-step cost calculation according to actual needs. For example, three different implementation methods can be designed: Parallel execution: In a multi-core or distributed environment, cost calculation and constraint verification threads are started simultaneously. Processing terminates if any electrical safety constraint is not met. This method theoretically maximizes hardware utilization and is suitable for scenarios with extremely high response speed requirements.

[0049] Calculate the cost first, then prune: First calculate the path cost from the current node to the adjacent node, then perform constraint checks. If the check fails, discard the node.

[0050] Prune first, then calculate cost: After obtaining adjacent nodes, the corresponding electrical safety constraints are called first. If any electrical safety constraints are violated, the node is discarded directly without subsequent cost calculation. Only the nodes that pass the verification are calculated for single-step cost and the open list is updated. This method can avoid invalid calculations to the greatest extent and is especially suitable for scenarios with strict constraints (such as "veto" constraints in black start) and relatively independent verification logic.

[0051] In this embodiment, considering the primary security requirement for black-start recovery and the high efficiency of graph database kernel execution, a pruning-then-cost calculation order is adopted. Unsafe paths are eliminated in the early stages of node expansion, which not only ensures the strict satisfaction of electrical constraints but also significantly reduces the overhead of invalid calculations, realizing an integrated calculation paradigm of search and verification.

[0052] S240, Search complete; S241. Obtain the optimal recovery path node sequence; When the current extended node n is detected to coincide with the target node T, the search is determined to be complete. The priority queue at this time is used as the search result. At this time, path backtracking is performed based on the priority queue to generate the corresponding optimal recovery path node sequence. Specifically: Starting from the target node T, traverse backwards to the starting node S according to the parent pointer mapping to generate the optimal recovery path node sequence P={S,v1,v2,...,T}; As one possible implementation method, the total path cost g(T) corresponding to the optimal recovery path node sequence P is statistically analyzed, and the proportions of each component, namely electrical distance, operation time, and reactive charging power, are decomposed and the estimated risk level is marked. As one possible implementation, the optimal recovery path node sequence and the corresponding cost / risk data are stored in a graph database in the form of a subgraph, and then returned to the human-machine interface for visualization rendering.

[0053] S242, No path processing; If the open list is exhausted (i.e. there are no scalable nodes) and the target node T is not found, the algorithm terminates and returns an empty result, indicating that there is no feasible black-start recovery path that satisfies the security constraints under the current power grid topology.

[0054] As one possible implementation, a multi-constraint processing mechanism is introduced, which triggers adaptive adjustments when approaching electrical safety limits, so that paths that might otherwise be pruned become feasible. In this embodiment, the reactive charging power cost P of each node is calculated. ij At that time, the cumulative reactive charging power Q of each path (determined based on the preferred queue and parent pointer mapping) is calculated. cum When the cumulative reactive charging power Q along the path is monitored cum Approaching the reactive power absorption limit Q of the black start unit limit The multi-constraint processing mechanism is triggered at certain times; 100. Calculate the cumulative reactive charging power corresponding to the target path; Cumulative reactive charging power Q cum The specific calculation method is as follows: ; Where path is the set of edges from the starting node S to the current node; e is any outgoing edge in the path; Q e The reactive charging power of the outgoing side e (usually the line no-load charging power).

[0055] 200. Triggering judgment is made based on the cumulative reactive charging power and the preset reactive absorption limit; If Q is satisfied cum >α*Q limit If the reactive power accumulation is close to the limit, the multi-constraint processing mechanism will be triggered.

[0056] Those skilled in the art can set the proximity threshold α according to the actual situation; in this embodiment, α is set to 0.8. 300. When the multi-constraint processing mechanism is triggered, the weight coefficient corresponding to the reactive charging power cost in the actual cost calculation is adjusted based on the dynamic weighting strategy, and the reactor is detected and put into operation based on the local optimization strategy to absorb reactive power and reduce the corresponding path cumulative reactive charging power. 310. Dynamic Privilege Escalation Strategy: The weight w6' of the reactive power cost term in the actual cost function is linearly amplified, and the adjustment formula is as follows: ; Wherein, β is the weighting amplification factor (0.5 in this embodiment); The adjusted w6' is used for the single-step cost calculation of subsequent outgoing edges. The guiding algorithm prioritizes lines with lower charging power to avoid excessive reactive power accumulation.

[0057] 320. Local optimization strategy: When there are available parallel reactors around the current node n, reactive power is absorbed by connecting the reactors to reduce Q. cum To improve the feasibility of the route, the specific steps are as follows: 321. Retrieve the parallel reactor nodes r around the current node n whose switch states are closed; 322. When the parallel reactor node r exists, a merged node is generated based on the parallel reactor node r and the current node; Based on the reactive power absorption capacity Q corresponding to the parallel reactor node r , Input operation cost T r The cumulative reactive charging power, actual cost, heuristic cost, and evaluation function value of the merging nodes are determined according to the following rules: Physical topology location: The physical anchor point of the new state s' is still node n (corresponding to the power grid bus), and the topology connection relationship is not changed; Cumulative reactive power update: Update the cumulative reactive charging power Q of the current path. cum Updated to Q cum =max(0,Q) cum -Q r ); Actual cost update: The actual cumulative cost of merging node s' is g(s') = g(n) + T r ; Heuristic cost preservation: The heuristic cost h(s') = h(n,target) for merging node s' is not adjusted; List enqueueing and tracing: Merge node s' and its evaluation function value f(s')=g(s')+h(s') into the open list, and record the parent state node and the reactor node that has been put into operation.

[0058] When expanding adjacent nodes in the subsequent re-state s', the updated cumulative reactive charging power Q is used. cum Conduct constraint pruning verification; Q cum The reduction of ' can make lines that were originally prohibited due to excessive reactive power have the conditions to become feasible recovery paths after verification.

[0059] In this embodiment, when the A-star algorithm searches for a long line path with high charging power, if the cumulative reactive charging power Q of the path is detected... cum Approaching the reactive power absorption limit Q of the black start unit limit Instead of directly discarding the path, the algorithm triggers a multi-constraint processing mechanism; by adaptively adjusting the search strategy and optimizing the path using reactive power compensation equipment, it finally outputs the optimal recovery sequence that satisfies electrical safety constraints.

[0060] Another example of this application proposes an automatic black boot recovery sequence generation system for performing the method described in any of the above embodiments; As one possible implementation, the system includes: The recovery sequence generation module is used to perform a pruned A-star search on the power grid attribute graph model corresponding to the target power grid in response to a black start request, using an improved A-star algorithm embedded in the graph database to obtain the corresponding search results; and generate the optimal recovery path node sequence based on the search results. It may also include: The graph data storage and management module is built on a native graph database (such as Galaxybase or Neo4j) to store the real-time topology of the power grid, equipment parameters and switch status, and provides ACID transaction support and native graph traversal interface.

[0061] The embedded algorithm library module is a collection of algorithms deployed within the graph database in the form of UDFs / stored procedures. These include an improved A-star search algorithm, a self-excitation / overvoltage fast verification algorithm, and an electrical distance pre-calculation algorithm. These UDFs / stored procedures run within the database kernel and can directly access storage layer data without requiring data export.

[0062] Online learning and parameter correction module: Records the actual execution effect of historical recovery schemes (such as whether overvoltage occurred), corrects the weight coefficients in the A-star algorithm through backpropagation, and realizes meta-learning optimization.

[0063] Visualization and Human-Computer Interaction Module: Based on graph database front-end rendering technology, it dynamically displays the generation process of the recovery path and highlights key nodes and weak links.

[0064] In this embodiment, the system adopts a graph database centralized computing architecture: all power grid data is stored in Galaxybase or Neo4j distributed graph database. The application server does not directly perform path calculation, but sends Cypher / GQL query call requests. After receiving the request, the graph database executes the embedded A-star search UDF internally and only returns the final result path (usually only a few dozen nodes) to the application layer.

[0065] Galaxybase: An existing publicly available distributed graph database designed for storing and processing large-scale graph structure data, supporting complex graph queries and analysis. Those skilled in the art can choose other distributed graph databases according to their actual needs.

[0066] Cypher: A declarative graph query language, originally introduced by the Neo4j database, used to intuitively express graph pattern matching and traversal operations.

[0067] GQL (Graph Query Language): An internationally standardized graph database query language.

[0068] This invention employs a graph database-based centralized computing architecture to achieve integrated storage of power grid topology, equipment parameters, and operating status. An improved A-star algorithm is embedded within the graph database kernel, enabling integrated storage and computation for path searching, eliminating cross-layer data transmission overhead, and significantly improving path generation efficiency. A dynamic electrical safety constraint pruning mechanism is designed, employing a strategy of pruning before cost calculation to eliminate unsafe paths in the early stages of node expansion, avoiding invalid cost calculations, reducing algorithm overhead, and achieving integrated computation of "search as verification." A multi-constraint processing mechanism is proposed; when the cumulative reactive power of a path approaches its limit, the path is optimized through dynamic weighting and the addition of parallel reactors, improving path feasibility, avoiding the accidental deletion of effective paths, and ensuring the safety and flexibility of black-start recovery. Utilizing the "pointer addressing" and "index-free adjacency" characteristics of graph databases, the algorithm can quickly obtain the attribute information of adjacent nodes and edges, adapting to dynamic changes in power grid topology without requiring remodeling, thus improving the practicality and adaptability of the method.

[0069] As the apparatus embodiment is basically similar to the method embodiment, it is described in a relatively simple manner. For relevant details, please refer to the description of the method embodiment.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of the method, terminal device (system), and computer program product according to the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] It should be noted that: The phrase "an embodiment" or "an embodiment" used in this specification means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the invention. Therefore, the phrase "an embodiment" or "an embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment.

[0076] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0077] Furthermore, it should be noted that the shapes and names of the parts and components described in the specific embodiments described in this specification may differ. All equivalent or simple variations made to the structure, features, and principles described in this patent concept are included within the protection scope of this patent. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to replace them, as long as they do not depart from the structure of this invention or exceed the scope defined in these claims, they should all fall within the protection scope of this invention.

Claims

1. A method for generating a black boot recovery sequence, characterized in that, Includes the following steps: For black start requests, an improved A-star algorithm embedded in the graph database is used to perform a pruned A-star search on the power grid attribute graph model corresponding to the target power grid to obtain the corresponding search results; and an optimal recovery path node sequence is generated based on the search results. The black start request is used to indicate the starting node and the target node to be restored; The power grid attribute graph model uses the busbars in the target power grid as nodes and the primary equipment in the target power grid as the connecting edges between nodes. Each node has a node dynamic attribute indicating its current energized state, and each edge has an edge dynamic attribute indicating its current switching state. Based on the improved A-star algorithm, the steps for determining the current expansion node and performing node expansion are as follows: From the priority queue, the node with the smallest evaluation function value is taken as the current expansion node. The evaluation function value of the current node is determined based on the actual cost from the starting node to the current node and the heuristically estimated cost from the current node to the target node. When the current extended node coincides with the target node, a corresponding optimal recovery path node sequence is generated based on the priority queue; When the current expanding node does not overlap with the target node: Based on the power grid attribute graph model, corresponding candidate nodes are obtained. The candidate nodes are adjacent nodes connected to the current extended node, and the candidate nodes satisfy preset electrical safety constraints. Add each candidate node to the priority queue; The formula for calculating the actual cost is as follows: ； in: L ij For the cost of electrical distance, L ij =|Z ij |,Z ij For the corresponding impedance; T ij This comes at the cost of operation time. P ij For the cost of reactive charging power, P ij =Q c / Q limit Q c The charging power of the line, Q limit This represents the reactive power absorption limit of the corresponding black-start generator set. w1, w2, and w3 are all weighting coefficients, and they satisfy w1 + w2 + w3 = 1.

2. The black-start recovery sequence generation method according to claim 1, characterized in that: The adjacency relationships of each node are physically stored in the form of a linked list.

3. The black-start recovery sequence generation method according to claim 1, characterized in that, The electrical safety constraints include self-excitation verification and overvoltage verification: The self-excitation verification is as follows: when X c ≤k*X d If a node is found to have a self-excitation risk, the adjacent node will be pruned; where X c X is the line charging capacitance corresponding to the connection edge from the current extended node to the corresponding adjacent node. d The synchronous reactance corresponding to the starting node; The overvoltage verification is as follows: when K ov ≥K limit If an overvoltage risk is detected, the corresponding adjacent node is pruned; where K limit The maximum overvoltage multiple; the connection edge from the current extended node to the corresponding adjacent node is taken as the closing edge, K ov To increase the closing overvoltage multiple of the path corresponding to the adjacent node, it is estimated based on the cumulative charging power of the path from the starting node to the adjacent node and the charging power of the side to be closed.

4. The black-start recovery sequence generation method according to any one of claims 1 to 3, characterized in that, While adding each candidate node to the priority queue, the parent pointer mapping and cost mapping are also updated based on the current actual cost of each candidate node. The parent pointer mapping is used to record the predecessor node of each node in the priority queue; The cost mapping is used to record the optimal actual cost corresponding to each node in the priority queue; The current actual cost is the actual cost from the starting node through the current extended node to the corresponding candidate node. When the candidate node is not recorded in the priority queue, the candidate node is added to the priority queue, and the current extended node is recorded as the predecessor node of the candidate node in the parent pointer mapping. The current actual cost corresponding to the candidate node is recorded as the optimal actual cost corresponding to the candidate node in the cost mapping. When the candidate node already exists in the priority queue; If the current actual cost corresponding to the candidate node is less than the optimal actual cost recorded in the cost mapping, then the predecessor node of the candidate node in the parent pointer mapping is updated to the current extended node, and the optimal actual cost in the cost mapping is updated to the current actual cost; otherwise, the parent pointer mapping and the cost mapping are not updated.

5. The black-start recovery sequence generation method according to claim 4, characterized in that, The method for calculating the current actual cost corresponding to a candidate node is as follows: Calculate the single-step cost from the current expansion node to the candidate node; Based on the single-step cost, the actual cost from the starting node to the current extended node is calculated to obtain the corresponding current actual cost; Specifically, the outgoing edge from the current expansion node to the candidate node is taken as the target edge. Based on the impedance magnitude of the target edge, the operating time and charging power of the corresponding device, and the preset generator self-excitation limit threshold, the single-step cost from the current expansion node to the candidate node is calculated.

6. The black-start recovery sequence generation method according to claim 5, characterized in that, Taking the outgoing edge from the current expanded node n to the candidate node m as the target edge, the corresponding single-step cost cost(n,m) is calculated using the following formula: cost(n,m)=w4*|Z|+w5*T op +w6*Q c / Q limit ; Where |Z| is the impedance magnitude of the target side; T op Q represents the operation time of the device corresponding to the target edge. c Charging power of the target side; Q limit The corresponding black-start unit's reactive power absorption limit; w4, w5, and w6 are all weighting coefficients, satisfying w4+w5+w6=1.

7. The black boot recovery sequence generation method according to claim 1, characterized in that: When calculating the reactive charging power cost of each node, the cumulative reactive charging power of each path is statistically analyzed based on the priority queue and parent pointer mapping. When Q is detected... cum >α*Q limit The multi-constraint processing mechanism is triggered at any time, where Q cum Q represents the cumulative reactive charging power along the corresponding path. limit α represents the reactive power absorption limit of the corresponding black start unit, and α is a preset approach threshold. When the multi-constraint processing mechanism is triggered, the weighting coefficient corresponding to the reactive charging power cost in the actual cost calculation is adjusted based on the dynamic weighting strategy, as follows: ； Where β is the weighting amplification factor.

8. The black-start recovery sequence generation method according to claim 7, characterized in that, When the multi-constraint processing mechanism is triggered, a reactor is detected and activated based on a local optimization strategy to absorb reactive power and reduce the cumulative reactive power charging power of the corresponding path. Specifically: Search for parallel reactor nodes whose switches are closed around the current node; When the reactor node exists, a merged node is generated based on the reactor node and the current node, and the physical anchor point of the merged node is still the current node; The operational cost T corresponding to the reactor node r Determine the cumulative reactive charging power, actual cost, heuristic cost, and evaluation function value of the merging node: in: The cumulative reactive charging power of the current path is updated to max(0,Q). cum -Q r ), Q cum Q represents the cumulative reactive charging power for the current path. r This refers to the reactive power absorption capacity corresponding to the reactor node; The actual cumulative cost of merging node s' is g(s') = g(n) + T r Let g(n) be the actual cost corresponding to the current node n, and T be the cost of the node. r The cost of the operation corresponding to the reactor node; The heuristic cost of merging node s' is the heuristic cost corresponding to the current node n; When merging node s' and its evaluation function value are included in the priority queue, the reactor node that has been put into operation is recorded.

9. A black boot recovery sequence generation system, characterized in that, Used to perform the method according to any one of claims 1 to 8.

Images