Photovoltaic power station power collection system topology optimization method and system based on three-section optimization genetic algorithm

Abstract

This invention provides a three-stage optimization genetic algorithm-based method and system for topology optimization of photovoltaic power plant collection systems, relating to the fields of photovoltaic systems and smart grids. The method includes: acquiring basic data of the photovoltaic power plant collection system, which includes a step-up substation and multiple box-type transformers; initializing a population based on the basic data of the photovoltaic power plant collection system using a hybrid initialization strategy; repeatedly executing genetic operations, a dual repair mechanism, and dynamic penalty evaluation to update the population until preset conditions are met; and determining the optimal topology network from the current population. This method has the advantage of improving the efficiency and quality of topology optimization for photovoltaic power plant collection systems.

Description

Technical Field

[0001] This invention relates to the fields of photovoltaic systems and smart grids, and in particular to a method and system for topology optimization of photovoltaic power plant collection systems using a three-stage optimization genetic algorithm. Background Technology

[0002] Topology optimization of photovoltaic (PV) power plant collection systems involves optimizing the remaining "points" and "lines" within the power plant, assuming the connection relationship between PV module units and transformer substations is already fixed by the DC-side layout and capacity matching and does not participate in optimization. "Points" primarily refer to transformer substations and step-up substations, while "lines" are the collection lines connecting them. In smart grids, PV power plants, as typical distributed power sources, require efficient coordination between their collection systems and the main power grid. Topology optimization reduces power transmission losses and improves renewable energy absorption rates by adjusting the connection methods between transformer substations and step-up substations. For example, it reduces the risk of curtailment by shortening power supply paths or balancing line loads. PV power plant collection system topology optimization is not only a key means of reducing costs and increasing efficiency in PV power plants but also a crucial technical support for smart grids to achieve source-grid-load-storage interaction and high-proportion renewable energy integration.

[0003] Currently, there are three main methods for designing and optimizing the topology of photovoltaic power plant collection systems, but all of them have significant drawbacks and are difficult to meet the actual engineering needs.

[0004] First, the design relies entirely on human experience. Engineers manually draw up connection plans based on topographic maps, equipment locations, and their own experience. This method is highly subjective, and different engineers may produce vastly different plans due to differences in experience. It is inefficient, as manual drawing and adjustments are time-consuming and labor-intensive, making it difficult to quickly respond to changes in design requirements. Furthermore, it is difficult to conduct quantitative comparisons and lacks a unified standard for evaluating the merits of different plans. More importantly, the human brain has limited capacity to process massive combinations, and the resulting plans are often far from optimal, leading to high project investment costs and failing to achieve optimal resource allocation.

[0005] Secondly, basic computer-aided design tools or simple rule-based software are used. While these tools enable digital drawing, making the design process more convenient and intuitive, their optimization logic is extremely simplistic. The generation of solutions generally lacks comprehensive consideration of actual engineering constraints. For example, failure to fully consider the current-carrying capacity limitations of different cable specifications may lead to line overload and safety hazards; ignoring differences in unit cost makes it impossible to effectively control costs while ensuring quality; and failure to comply with requirements such as prohibiting line crossings during construction increases construction difficulty and costs. Therefore, the generated solutions have serious defects in terms of economics or feasibility and are difficult to directly apply to actual projects.

[0006] Third, while general optimization algorithms are used for automated design, directly applying them to the specific and complex engineering problem of photovoltaic power plants exposes a series of problems. In the initial stage of the algorithm, traditionally randomly generated initial connection schemes are mostly structurally chaotic and violate multiple constraints, like blindly starting on the wrong path, leading to slow convergence in the early stages and increasing optimization time and computational costs. During the evolutionary process, when invalid or non-compliant individuals are generated, the algorithm's repair mechanism becomes rigid, typically employing simple, fixed rules for repair, which is inefficient and ineffective, often causing the optimization process to fall into a stalemate of "constant destruction without effective reconstruction," making it difficult to lead to a high-quality feasible solution.

[0007] Therefore, there is a need to provide a method and system for topology optimization of photovoltaic power plant collector systems using a three-stage optimization genetic algorithm, in order to improve the efficiency and quality of topology optimization of photovoltaic power plant collector systems. Summary of the Invention

[0008] This invention provides a three-stage optimization genetic algorithm for the topology optimization of a photovoltaic power plant collection system, comprising: acquiring basic data of the photovoltaic power plant collection system, wherein the photovoltaic power plant collection system includes a step-up substation and multiple box-type transformers; initializing a population based on the basic data of the photovoltaic power plant collection system using a hybrid initialization strategy; repeatedly executing genetic operations, a dual repair mechanism, and dynamic penalty evaluation to update the population until preset conditions are met; and determining the optimal topology network from the current population.

[0009] Furthermore, a hybrid initialization strategy is employed to initialize the population based on the fundamental data of the photovoltaic power station's data collection system. This includes: generating multiple first individuals using the K-Means clustering algorithm; generating multiple second individuals using the nearest neighbor heuristic; generating multiple third individuals using the minimum spanning tree heuristic; generating multiple fourth individuals using the random generation strategy; and initializing the population based on the multiple first individuals, multiple second individuals, multiple third individuals, and multiple fourth individuals.

[0010] Furthermore, the basic data of the photovoltaic power station collection system includes at least the planar location coordinates of each box-type transformer and the step-up substation. Using the K-Means clustering algorithm, multiple first bodies are generated based on the basic data of the photovoltaic power station collection system. This includes: determining multiple box-type transformer clusters based on each box-type transformer using the K-Means clustering algorithm; for each box-type transformer cluster, generating the corresponding topology structure using a minimum spanning tree heuristic strategy and feasibility verification, where the edge weights are at least related to cable length, unit cost, power loss, and construction difficulty; and for each box-type transformer cluster, selecting the box-type transformer closest to the step-up substation as the cluster head connected to the step-up substation, thus generating multiple first bodies.

[0011] Furthermore, based on the basic data of the photovoltaic power station collection system, multiple second entities are generated using the nearest neighbor heuristic strategy. When generating each second entity, starting from a box-type transformer or step-up substation, the next node is found using the nearest neighbor heuristic strategy. When connecting to the next node, it is determined whether the cumulative power of the current branch exceeds the allowable capacity of the selected cable. If so, branch reconfiguration or cable selection adjustment is triggered.

[0012] Furthermore, the dual repair mechanism includes: performing multiple routine repairs for each individual generated through genetic manipulation; calculating the fitness improvement rate corresponding to each routine repair for each routine repair; determining whether the routine repair is invalid based on the fitness improvement rate corresponding to the routine repair; if so, incrementing the count of the routine repair failure counter by 1; otherwise, resetting the count of the routine repair failure counter to 0; and performing intelligent repair when the count of the routine repair failure counter is greater than a preset count threshold.

[0013] Furthermore, the conventional repair includes: performing a destruction operation on the individual until the individual becomes an acyclic graph; identifying all connected components in the destroyed individual using a breadth-first search or disjoint-set data structure algorithm, and determining whether there are isolated connected components that do not contain the booster station; if so, calculating the Euclidean distance between each node in the isolated connected component and the booster station or a connected component containing the booster station, determining the target node from the isolated connected components, performing a connectivity operation based on the target node to generate a connected individual; and performing a segment removal operation on the connected individual.

[0014] Furthermore, the intelligent repair includes: determining the target local subnetwork based on the counted multiple conventional repairs; for the target subnetwork, using the booster station as the root, running the nearest neighbor heuristic algorithm to generate a new local network.

[0015] Furthermore, the dynamic penalty evaluation includes: determining the current penalty coefficient based on the current iteration number and the maximum iteration number; calculating the fitness of each individual repaired by the dual repair mechanism based on the current penalty coefficient; and selecting multiple individuals repaired by the dual repair mechanism to form the next generation population based on the fitness of each individual repaired by the dual repair mechanism.

[0016] Furthermore, based on the current penalty coefficient, the fitness of the individual repaired by the double repair mechanism is calculated, including: calculating the cost of the individual repaired by the double repair mechanism; determining the constraint violation of the individual repaired by the double repair mechanism; and calculating the fitness of the individual repaired by the double repair mechanism based on the cost of the individual repaired by the double repair mechanism, the current penalty coefficient, and the constraint violation.

[0017] This invention provides a photovoltaic power plant collector system topology optimization system based on a three-stage optimization genetic algorithm. The system executes the aforementioned three-stage optimization genetic algorithm for photovoltaic power plant collector system topology optimization. It includes: a data acquisition module for acquiring basic data of the photovoltaic power plant collector system, wherein the photovoltaic power plant collector system includes a step-up substation and multiple box-type transformers; a population initialization module for initializing the population based on the basic data of the photovoltaic power plant collector system using a hybrid initialization strategy; and a topology optimization module for repeatedly executing genetic operations, a dual repair mechanism, and dynamic penalty evaluation to update the population until preset conditions are met, thereby determining the optimal topology network from the current population.

[0018] Compared with existing technologies, the photovoltaic power plant collector system topology optimization method and system based on a three-stage optimization genetic algorithm provided by this invention has at least the following beneficial effects: 1. The hybrid initialization strategy can provide high-quality and diverse initial populations. It fully utilizes the engineering characteristics of photovoltaic power plants to generate initial topologies that meet basic feasibility requirements, avoiding illegal solutions (such as isolated nodes and loops) caused by purely random initialization. At the same time, the diversified initial populations cover a wider solution space, providing a rich gene pool for subsequent genetic operations. This allows the algorithm to quickly locate potential high-quality regions in the early stages of evolution, significantly shortening the convergence time and improving optimization efficiency.

[0019] 2. The dual-repair mechanism, through the synergy of conventional and intelligent repair, ensures efficient generation of feasible solutions and effective breakthroughs in optimization deadlocks during the iteration process. Conventional repair performs local corrections for common defects such as loops and isolated nodes, ensuring the legitimacy of solutions; intelligent repair identifies frequently modified areas based on historical repair data and reconstructs local topology using domain knowledge (such as nearest neighbor heuristics), breaking the local optimum trap. This combined strategy of local correction and global reconstruction enables the algorithm to maintain stable evolution even when facing complex engineering constraints, avoiding search interruptions due to individual illegal solutions, thereby significantly improving the robustness of the optimization process.

[0020] 3. The dynamic penalty evaluation mechanism adaptively adjusts the penalty coefficient for constraint violations, simulating an intelligent search behavior of exploration followed by refinement. In the early stages of evolution, a lower penalty coefficient allows the algorithm to explore more boundary solutions, expanding the search range. As iterations progress, the penalty coefficient gradually increases, forcing solutions to converge towards the feasible region and strengthening local exploitation capabilities. This mechanism effectively balances global exploration and local exploitation, preventing the algorithm from prematurely falling into local optima or losing diversity due to excessive penalties. The final output topology solution strictly satisfies engineering constraints (such as acyclicity and full connectivity) while minimizing total investment costs, providing a practical and economically viable optimization solution for real-world engineering projects. Attached Figure Description

[0021] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a flowchart illustrating a photovoltaic power plant collector system topology optimization method using a three-stage optimization genetic algorithm, as shown in some embodiments of this specification. Figure 2 This is a flowchart illustrating the process of determining the optimal topology network according to some embodiments of this specification; Figure 3 This is a schematic diagram of an optimal topology network according to some embodiments of this specification; Figure 4 This is a schematic diagram of a photovoltaic power plant collector system topology optimization system based on a three-stage optimization genetic algorithm, as shown in some embodiments of this specification. Detailed Implementation

[0022] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0023] Figure 1 This is a flowchart illustrating a photovoltaic power plant collector system topology optimization method using a three-stage optimization genetic algorithm, as shown in some embodiments of this specification. Figure 1 As shown, the three-stage optimization genetic algorithm for photovoltaic power plant collector system topology optimization can include the following steps.

[0024] Step 110: Obtain basic data of the photovoltaic power station's collection system.

[0025] The photovoltaic power station's power collection system includes a step-up substation and multiple box-type transformers.

[0026] Specifically, the basic data of a photovoltaic power station's collection system should include at least the planar coordinates of each box-type transformer and the step-up substation.

[0027] The basic data for a photovoltaic power station's data collection system can also include a cable parameter library consisting of various cable types and their unit length costs, maximum allowable carrying capacity, and unit length construction costs, as well as engineering constraint parameters such as the maximum number of line crossings allowed by the system and the maximum number of transformer substations that can be directly connected to a single step-up substation. These data collectively constitute the known conditions and boundary constraints that must be satisfied in subsequent topology optimization algorithms.

[0028] Step 120: Initialize the population based on the basic data of the photovoltaic power station collection system using a hybrid initialization strategy.

[0029] Specifically, it includes: Using the K-Means clustering algorithm, multiple first volumes are generated based on the basic data of the photovoltaic power station's power collection system. Using a nearest neighbor heuristic strategy, multiple second entities are generated based on the fundamental data of the photovoltaic power plant's collection system. Using a minimum spanning tree heuristic strategy, multiple third entities are generated based on the fundamental data of the photovoltaic power station's collection system. Multiple fourth individuals are generated based on the basic data of the photovoltaic power station's collection system using a random generation strategy. The population is initialized based on multiple first individuals, multiple second individuals, multiple third individuals, and multiple fourth individuals.

[0030] Specifically, within the genetic algorithm framework, the encoding method of an individual directly determines its expressive power and the feasibility of genetic operations. Considering the unique topology of a photovoltaic power plant's collection system, this method uses edge list encoding as the representation of an individual. Edge list encoding refers to representing a complete tree-like network as an ordered set of edges, where each edge consists of two node identifiers, representing the start and end points of the collection line segment. The node identifiers are unique numbers for all box-type transformers or unique numbers for the step-up substation; the step-up substation is uniformly considered the root node in the encoding. Assuming there are N box-type transformers and 1 step-up substation in a photovoltaic power plant, the collection system topology should be a tree graph rooted at the step-up substation, covering all box-type transformers, containing a total of N edges. In edge list encoding, each individual is a list containing N edges, in the form of... ,in and All values ​​are node numbers, and each edge represents a physical cable connection between two nodes. This encoding method inherently possesses simplicity and structural integrity, capable of fully describing the connectivity of the entire network. For example, a power station contains four box-type transformers, numbered as follows: Given a single booster station, designated S, a possible topology can be represented by the following edge list: The meaning of this code is: The step-up substation S is connected to the box-type transformer. and Direct connection; and Connect them to form a branch; and Connecting them forms another branch. This list fully expresses all connections without redundant edges, meeting the requirements of a tree structure.

[0031] The connection relationship between the photovoltaic module unit and the box transformer is a predetermined fixed mapping, determined by the DC side layout and capacity matching, and therefore is not considered as an optimization object. The geographical location of the box transformer and the photovoltaic capacity it collects are both known data.

[0032] In some embodiments, multiple first bodies are generated based on the basic data of the photovoltaic power plant's power collection system using the K-Means clustering algorithm, including: Using the K-Means clustering algorithm, multiple transformer clusters are determined based on each transformer box. Specifically, the K-Means clustering algorithm is used to divide all transformer boxes into K clusters according to their spatial coordinates, aiming to group spatially adjacent transformer boxes into the same collector branch to shorten cable length. For each box-type transformer cluster, a topology corresponding to the box-type transformer cluster is generated through a minimum spanning tree heuristic strategy and feasibility verification. The weight of the edges is at least related to the cable length, unit cost, power loss, and construction difficulty, making the optimization objective more consistent with the actual investment cost. The core logic is: introduce the step-up substation as a fixed root node to ensure that the generated tree structure meets the radial topology requirements of the power collection system of "single power point convergence". The box-type transformer is used as a "node" and the candidate power collection lines connecting the box-type transformer are used as "edges". The Prim algorithm is used to select the set of edges with the minimum total weight to form a loop-free connected network. Thus, under the premise of meeting the electrical connection requirements, the cost is initially minimized. Multiple minimum spanning trees are used to generate the topology corresponding to multiple box-type transformer clusters. For each topology corresponding to each box-type transformer cluster, the box-type transformer closest to the step-up station in the box-type transformer cluster is taken as the cluster head connected to the step-up station, generating multiple first bodies. Specifically, the cluster head is connected to the step-up station, and combined with the topology corresponding to each box-type transformer cluster, multiple individuals are formed.

[0033] Specifically, feasibility verification is the core step in ensuring that the topology meets engineering constraints. Its core logic is to balance economy and security by dynamically identifying and correcting illegal branches.

[0034] Feasibility verification includes the following key steps: 1. Overload detection: Based on a cable parameter library (such as the maximum allowable power carrying capacity of different cable types), each branch in the topology is traversed, and the ratio of its actual transmission power (determined by the total power generation of the transformers within the cluster) to the cable current carrying capacity is calculated. If the power of a branch exceeds the current carrying capacity limit of its corresponding cable type (e.g., due to uneven distribution of transformer capacity or excessively long paths leading to current concentration), the branch is marked as "infeasible".

[0035] 2. Local Reconstruction Strategy: Make targeted adjustments to infeasible branches, for example: Cable model upgrade: Replace with a cable with a higher cross-sectional area or current carrying capacity (such as upgrading from 120mm² aluminum core to 240mm² copper core), but the edge weights need to be updated simultaneously to reflect changes in cost and loss; Path optimization: Reduce current density by adjusting branch connection relationships (such as bypassing high-load areas or shortening line length), for example, changing the original direct connection path to a segmented path via an intermediate transformer. Power redistribution: Redistributing power flow between transformer substations within a cluster (e.g., by adjusting inverter output or introducing energy storage devices for buffering) to reduce the load pressure on individual branches.

[0036] Iterative verification and convergence: After partial reconstruction, all branches need to be re-verified until they all meet the current carrying capacity constraints.

[0037] In some embodiments, a plurality of second bodies are generated based on the fundamental data of the photovoltaic power plant collection system using a nearest neighbor heuristic strategy, including: When generating each second entity, starting from a box-type transformer or step-up substation, the next node is found through the nearest neighbor heuristic strategy. When connecting to the next node, it is determined whether the cumulative power of the current branch exceeds the allowable capacity of the selected cable. If so, branch reconfiguration or cable selection adjustment is triggered.

[0038] Specifically, during algorithm initialization, a starting point can be randomly selected from either the booster station or any transformer substation. Based on a nearest neighbor heuristic, the node closest to the currently connected subtree from the remaining unconnected nodes is selected as the next access point. Each time a new node is accessed, the cumulative power of the current branch must be calculated in real-time, which is the sum of the generating power of all transformer substations along the path from the root node (booster station) to the current access node, and compared with the maximum allowable current carrying capacity of the selected cable type. If the cumulative power exceeds the cable capacity, for example, due to current concentration caused by accessing a high-power transformer or an excessively long path, one of the following two adaptive adjustment mechanisms is triggered: Branch reconfiguration: By adjusting the connection relationships of the current nodes, the power flow path is changed. For example, if the branch originally planned to directly connect node A to node B is overloaded, it can be changed to connect A to node B's parent node C, or the power can be diverted by adding an intermediate node (such as a temporary transformer), thereby reducing the load pressure on a single branch.

[0039] Backup cable selection adjustment: Select a cable with a higher cross-sectional area or current carrying capacity from the cable parameter library to replace the original model. For example, upgrade a 120mm² aluminum core cable to a 240mm² copper core cable. At the same time, the weight of this branch (such as cost and loss) needs to be dynamically updated to reflect the impact of the selection change on the overall cost.

[0040] The above verification and adjustment process needs to be executed iteratively until all nodes are connected and all branches meet the flow rate constraints.

[0041] This improved algorithm, through a closed-loop mechanism of dynamic node access, real-time power verification, and adaptive constraint adjustment, maintains the efficiency of the nearest neighbor heuristic algorithm while achieving precise adaptation to the engineering constraints of photovoltaic power collection systems, providing an effective technical path for generating economically feasible, safe, and reliable topology schemes.

[0042] Using a minimum spanning tree heuristic strategy, multiple third entities are generated based on the fundamental data of the photovoltaic power plant's collection system. These can include: During the initialization phase, the booster station is forcibly set as the starting node, and all new access nodes in subsequent expansion processes are ensured to connect to the booster station via a single path, guaranteeing a radial tree-like topology. It should be noted that forcibly setting the booster station as the starting node during initialization and ensuring all new access nodes connect via a single path in subsequent expansion processes aims to ensure the generated topology meets the engineering feasibility constraints of a radial tree structure, rather than limiting the uniqueness of the topology. Randomly selecting a second node can avoid the possibility of generating the same topology in each iteration, thus preventing the generation of multiple third entities.

[0043] The calculation of edge weights adopts composite weights, which comprehensively consider multiple factors such as cable length, unit cost, power loss and construction difficulty, so that the optimization target is closer to the actual engineering investment cost. After generating the global minimum spanning tree, a feasibility verification and local adjustment step is set up to identify and reconstruct branches that exceed the cable current carrying capacity constraint, so as to ensure the engineering feasibility of the output scheme.

[0044] After generating the global minimum spanning tree, feasibility verification and local adjustment steps are set up to identify and reconstruct branches exceeding the cable current-carrying capacity constraint, ensuring the engineering feasibility of the output scheme. Specifically, in the identification stage, a branch traversal algorithm based on power superposition is adopted. With the booster station as the root node, the topology structure is traversed through a depth-first search post-order traversal, accumulating the power transmitted by each branch from the leaf node to the root node, accurately locating the branches exceeding the limit within linear time complexity. In the reconstruction stage, a hierarchical reconstruction strategy is adopted: the first level uses the branch splitting method to reconnect the subtrees connected to the branches exceeding the limit to nodes with lighter loads or the booster station to divert power; the second level iteratively upgrades the cable models according to the current-carrying capacity in the cable parameter library; the third level re-runs the minimum spanning tree algorithm on the local subnetwork with the booster station as the root to generate a new topology that meets the current-carrying capacity constraint.

[0045] Using a random generation strategy, multiple fourth individuals are generated based on the fundamental data of the photovoltaic power plant's data collection system. These can include: The random connection method of adding nodes one by one is adopted: starting from the step-up substation, an unconnected transformer is randomly selected in turn and connected to a random node in the connected network. This process is repeated until all transformers are connected, thereby generating a connected and acyclic random tree.

[0046] It can be based on preset weights and Determine the number of the first, second, third, and fourth individuals, respectively. Weights and The value of must satisfy the constraint that the sum of the weights of the four strategies in the mixed initialization is 1, that is... In actual calculations, and Non-integer values ​​may occur. In such cases, rounding up or down is used for adjustment, and the number of units generated by other strategies is fine-tuned to ensure that the initial population size is 1. .for and The optimal range of values ​​was determined through orthogonal experiments and parameter sensitivity analysis on photovoltaic power plants of different sizes and geographical distributions. Experimental results show that when The value is in the range [0.25, 0.35]. The value is in the range [0.15, 0.25]. The value is in the range [0.10, 0.20]. When the value is in the range of [0.05, 0.15], the algorithm can obtain a population with both high quality and diversity during the initialization phase, laying a good foundation for subsequent evolution. The above preferred range should be regarded as an illustrative example rather than a limitation, and can be adaptively adjusted according to the specific scale and terrain features of the power plant in practical applications.

[0047] According to the preset weight and The individuals generated by the four strategies mentioned above are collected and merged to form the initial population. Specifically, based on the weights, approximately The first individual, approximately The second individual, approximately A third individual, approximately A fourth individual; the above four groups of individuals, after being merged, form a scale of [size missing]. The initial population is fused. This fusion mechanism ensures that the initial population contains both high-quality topologies generated by domain knowledge and retains the diversity features introduced by random exploration, thus laying a solid foundation for iterative optimization of subsequent genetic operations that balances quality and diversity.

[0048] Step 130: Repeat the genetic operations, dual repair mechanism and dynamic penalty assessment in sequence to update the population until the preset conditions are met.

[0049] Figure 2 This is a flowchart illustrating the process of determining the optimal topology network according to some embodiments of this specification, such as... Figure 2 As shown, specifically, genetic operations can include: Using tournament selection from the current population Random selection Individuals are selected and placed into the mating pool until the size of the mating pool matches the population size. Same; then, parent individuals are randomly paired from the mating pool with probability. Perform a subtree-based crossover operation: randomly select an edge in parent individual A, disconnect it, and swap one of its subtrees with a compatible subtree in parent individual B to generate two new offspring; then, for each offspring individual, apply a probability... Perform mutation operations, including randomly reconnecting an edge or swapping the endpoints of two edges.

[0050] like Figure 2 As shown, in some embodiments, the dual repair mechanism includes: For each individual generated through genetic manipulation, multiple routine repairs are performed. For each routine repair, the corresponding fitness improvement rate is calculated. Based on the fitness improvement rate, it is determined whether the routine repair is ineffective. For example, if the fitness improvement rate corresponding to the routine repair is lower than a preset threshold... If the failure rate is not met, the regular repair is considered invalid. If so, the regular repair failure counter is incremented by 1; otherwise, the regular repair failure counter is reset to 0. When the regular repair failure counter is greater than or equal to the preset counting threshold N (e.g., 10, 12), intelligent repair is performed. If the number of regular repairs is greater than the maximum number (e.g., 30) and the number of regular repair failures is less than the preset counting threshold, then dual repair is completed, and intelligent repair is not performed.

[0051] The fitness improvement rate corresponding to conventional repair can be calculated based on the following formula: in, This represents the fitness improvement rate corresponding to conventional repair. To repair the fitness value of the previous individual, These are the fitness values ​​of the repaired individuals. All values ​​are calculated using the fitness function.

[0052] In some embodiments, conventional repairs include: The algorithm performs destructive operations on individuals until the individual becomes an acyclic graph. Specifically, it traverses the network using Depth-First Search (DFS), dynamically detects all closed loops, and for each loop, the algorithm identifies and removes the edge with the highest cost (such as the longest cable, the highest unit cost, or the greatest power loss). It prioritizes removing edges that have a greater impact on economics and iterates through the removal process until no loops exist in the network. The algorithm uses breadth-first search (BFS) or disjoint-set data structure to identify all connected components in the destroyed individual and determine whether there are isolated connected components that do not contain the booster station. If so, calculate the Euclidean distance between each node in the isolated connected component and the booster station or the connected component containing the booster station, determine the target node from the isolated connected component, and perform a connectivity operation based on the target node to generate connected individuals. For example, take the node with the smallest Euclidean distance as the target node, and connect the target node to the node in the connected component containing the booster station or the booster station with the smallest Euclidean distance. That is, if the plane distance between the target node and the booster station is the smallest, then connect the target node to the booster station. If the Euclidean distance between the target node and a certain node in the connected component containing the booster station is the smallest, then connect the target node to that node. For connected individuals, perform the operation to remove intersecting segments.

[0053] Specifically, a line segment intersection detection algorithm (such as a sweep-line based geometric determination method) traverses all cable paths to identify pairs of intersecting line segments on the two-dimensional plane. For each pair of intersecting line segments, the algorithm employs two strategies to eliminate conflicts: Endpoint exchange: If the crossing segments AB and CD satisfy the exchange condition (e.g., no new crossing is formed after the exchange), then change AB to AC and CD to BD; Local reconnection: If endpoint swapping is not feasible, the intersection is eliminated by replanning intermediate paths (such as bypassing obstacles or adjusting connection angles).

[0054] In some embodiments, smart repair includes: Based on the counted multiple routine repairs, the target local subnetwork is determined; For the target subnetwork, with the booster station as the root, the nearest neighbor heuristic algorithm is run to generate a new local network.

[0055] Specifically, the modification frequency of each edge in multiple routine repairs is first counted to identify frequently adjusted local subnetworks. These areas often suffer from poor initial generation or complex loops / crossings, making it difficult for routine repairs to converge. After determining the target subnetwork, the nearest neighbor heuristic algorithm is run on all nodes in the subnetwork with the booster station as the root node. Based on a "greedy strategy," nodes are connected one by one, and each time the node is connected to the nearest node in the current part of the topology. At the same time, it is ensured that the newly added edge does not destroy the acyclicity and connectivity, thereby generating a new local topology that meets the engineering constraints. Finally, the reconstructed subnetwork is seamlessly grafted back into the main network, and the modification counter C is reset to start a new round of statistics.

[0056] The core advantage of this mechanism lies in the fact that conventional repair gradually optimizes the topology through local adjustments, but when faced with complex defects, it may get stuck in a stalemate of "repeatedly modifying the same area but failing to meet the requirements." Intelligent repair, on the other hand, proactively locates the root cause of the problem by quantifying historical modification data, namely, by frequently modifying subnetworks and performing a complete reconstruction. This preserves the optimized parts of the main network while breaking the local optimum trap by replacing the "problematic subnetwork." Through the synergy of conventional and intelligent repair, the algorithm enhances its ability to escape local optima while maintaining search efficiency, ultimately guiding the optimization process towards a high-quality solution that meets all feasibility requirements and has a lower cost.

[0057] In some embodiments, dynamic penalty assessment includes: Based on the current iteration number and maximum number of iterations Determine the current penalty coefficient. For example, when hour, Take the minimum value ,exist Within the interval, from linearly increasing to the maximum value ,when hour, Fixed as ; For each individual repaired by the double repair mechanism, calculate the fitness of the individual repaired by the double repair mechanism based on the current penalty coefficient; Based on the fitness of each individual repaired by the dual repair mechanism, select multiple individuals repaired by the dual repair mechanism to form the next generation population.

[0058] like Figure 2 As shown, in some embodiments, the fitness of an individual repaired by the dual-repair mechanism is calculated based on the current penalty coefficient, including: Calculate the cost of repairing an individual that has undergone a dual repair mechanism; Determine the amount of constraint violation by an individual who has undergone dual-repair mechanism repair; The fitness of individuals repaired through the dual-repair mechanism is calculated based on their cost, current penalty coefficient, and constraint violation amount.

[0059] Specifically, the fitness function can be: in, For fitness, This is the sum of the products of the length of each cable and its unit cost. To constrain violations, including exceeding power limits, number of line crossings, and number of directly connected transformer substations exceeding limits.

[0060] Specifically, the fitness function is: in, For fitness; This represents the total cost of the cable. For reference cost, the value is the total cost of cables when all box-type transformers are directly connected to the substation. This is a dynamic penalty coefficient. The formula for calculating the amount of violation is as follows: in, The branch power exceeds the limit (kW). The rated current carrying capacity (kW) of the selected cable; This represents the number of line crossings in the network topology. This represents the maximum number of crossovers allowed by the system. The number of box-type transformers directly connected to the substation exceeds the allowable limit. This refers to the maximum number of transformer substations that can be directly connected to the booster station.

[0061] Based on the fitness of individuals repaired through the dual repair mechanism, selected from highest to lowest fitness level. Individuals constitute the next generation population. .

[0062] like Figure 2 As shown, the preset conditions for determining whether optimization should terminate may include: the improvement rate of the fitness of the best individual being less than the minimum threshold for multiple consecutive generations, the number of iterations reaching the maximum number of iterations, etc.

[0063] Step 140: Determine the optimal topology network from the current population.

[0064] Specifically, from the current population, the individual with the highest fitness is selected as the optimal topology network. After decoding it into a specific topology connection scheme, it is output in multiple dimensions through a graphical user interface. First, all node positions are plotted in a two-dimensional coordinate system, and the optimal connection relationships between transformer substations and between transformer substations and step-up substations are clearly marked with lines, forming a visual topology connection diagram. At the same time, a detailed material and cost list is generated, listing the start and end points of each connection line, the selected cable type, the laying length, the actual power load, and the itemized costs. Finally, a summary report is provided, giving key economic indicators such as the total investment cost of cables in the power station collection system and the total length of cables, forming a complete optimization scheme that can be directly used to guide engineering design and procurement decisions.

[0065] The following section, based on experiments, explains the beneficial effects of the three-stage optimization genetic algorithm method for photovoltaic power plant collector system topology optimization.

[0066] A comparative experiment was conducted in a typical large-scale photovoltaic power plant simulation scenario comprising 120 transformer substations and one step-up substation. The method of this invention was compared with two benchmark methods: one using a reference scheme derived from traditional manual experience planning; the other using a standard genetic algorithm with completely random initialization. Optimization was performed under the same hardware platform and parameter settings, and the optimal topology network determined by this method is as follows: Figure 3 As shown in the experimental results, the proposed method significantly improves both optimization efficiency and solution quality compared to the traditional standard genetic algorithm. Specifically, under the premise of achieving the same cost convergence accuracy, the average number of iterations required by the proposed method is reduced by approximately 35%, demonstrating a faster convergence speed. From the economic analysis of the optimization results, the topology scheme obtained by the proposed method reduces the total investment cost of cables and construction by an average of approximately 11.2% compared to the manual experience scheme, and further optimizes by an average of approximately 7.5% compared to the scheme obtained by the standard genetic algorithm. In addition, in ten independent repeated experiments, the proposed method can consistently converge to a high-quality solution with similar costs, while the optimization results of the standard genetic algorithm show significant fluctuations. This proves that the proposed method effectively enhances the robustness of the algorithm through its three-stage mechanism (i.e., genetic operation, dual repair mechanism, and dynamic penalty evaluation), avoiding getting trapped in local optima, thus providing a stable, reliable, and economically efficient automated solution for the planning and design of the power collection system of large-scale photovoltaic power plants.

[0067] Figure 4 This is a schematic diagram of a photovoltaic power plant collector system topology optimization system based on a three-stage optimization genetic algorithm, as shown in some embodiments of this specification. Figure 4 As shown, the photovoltaic power plant collector system topology optimization system based on the three-stage optimization genetic algorithm can include a data acquisition module, a population initialization module, and a topology optimization module.

[0068] The data acquisition module is used to acquire basic data of the photovoltaic power station's power collection system, which includes a step-up substation and multiple box-type transformers. The population initialization module is used to initialize the population based on the basic data of the photovoltaic power station's collection system using a hybrid initialization strategy. The topology optimization module is used to repeatedly execute genetic operations, a dual repair mechanism, and dynamic penalty evaluation to update the population until preset conditions are met, and to determine the optimal topology network from the current population.

[0069] The three-stage optimization genetic algorithm-based photovoltaic power plant collector system topology optimization system can be used to execute the three-stage optimization genetic algorithm-based photovoltaic power plant collector system topology optimization method, which will not be elaborated here.

[0070] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A topology optimization method for photovoltaic power plant collector systems using a three-stage optimization genetic algorithm, characterized in that, include: Acquire basic data of the photovoltaic power station's power collection system, wherein the photovoltaic power station's power collection system includes a step-up substation and multiple box-type transformers; The population is initialized based on the fundamental data of the photovoltaic power plant's collection system using a hybrid initialization strategy. Repeatedly execute genetic operations, dual repair mechanisms, and dynamic penalty assessments to update the population until the preset conditions are met; Determine the optimal topology network from the current population.

2. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 1, characterized in that, Using a hybrid initialization strategy, the population is initialized based on the fundamental data of the photovoltaic power plant's collection system, including: Using the K-Means clustering algorithm, multiple first volumes are generated based on the basic data of the photovoltaic power station's power collection system. Using a nearest neighbor heuristic strategy, multiple second entities are generated based on the fundamental data of the photovoltaic power plant's collection system. Using a minimum spanning tree heuristic strategy, multiple third entities are generated based on the fundamental data of the photovoltaic power station's collection system. Multiple fourth individuals are generated based on the basic data of the photovoltaic power station's collection system using a random generation strategy. The population is initialized based on multiple first individuals, multiple second individuals, multiple third individuals, and multiple fourth individuals.

3. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 2, characterized in that, The basic data of the photovoltaic power station collection system includes at least the planar coordinates of each box-type transformer and step-up station; Using the K-Means clustering algorithm, based on the basic data of the photovoltaic power plant's data collection system, multiple first bodies are generated, including: Using the K-Means clustering algorithm, multiple box-type transformer clusters are determined based on each box-type transformer; For each box-type transformer cluster, the topology corresponding to the box-type transformer cluster is generated through the minimum spanning tree heuristic strategy and feasibility verification. The weight of the edge is at least related to the cable length, unit cost, power loss and construction difficulty. For each box-type transformer cluster, the box-type transformer closest to the step-up station in the cluster is taken as the cluster head connected to the step-up station, generating multiple first bodies.

4. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 3, characterized in that, Using a nearest neighbor heuristic strategy, multiple second volumes are generated based on the fundamental data of the photovoltaic power plant's collection system, including: When generating each second entity, starting from a box-type transformer or step-up substation, the next node is found through the nearest neighbor heuristic strategy. When connecting to the next node, it is determined whether the cumulative power of the current branch exceeds the allowable capacity of the selected cable. If so, branch reconfiguration or cable selection adjustment is triggered.

5. The photovoltaic power plant collector system topology optimization method based on a three-stage optimization genetic algorithm according to any one of claims 1-4, characterized in that, The dual repair mechanism includes: For each individual generated through genetic manipulation, multiple routine repairs are performed. For each routine repair, the fitness improvement rate corresponding to the routine repair is calculated. Based on the fitness improvement rate corresponding to the routine repair, it is determined whether the routine repair is invalid. If so, the count of the routine repair failure counter is incremented by 1; otherwise, the count of the routine repair failure counter is reset to 0. When the count of the routine repair failure counter is greater than the preset count threshold, intelligent repair is performed.

6. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 5, characterized in that, The routine repairs include: Perform destructive operations on individuals until the individual becomes an acyclic graph; All connected components in the damaged individual are identified by breadth-first search or disjoint-set data structure algorithm, and it is determined whether there are isolated connected components that do not contain the booster station. If so, calculate the Euclidean distance between each node in the isolated connected component and the booster station or the connected component containing the booster station, determine the target node from the isolated connected component, perform the connection operation based on the target node, and generate the connected individual. For connected individuals, perform the operation to remove intersecting segments.

7. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 5, characterized in that, The intelligent repair includes: Based on the counted multiple routine repairs, the target local subnetwork is determined; For the target subnetwork, with the booster station as the root, the nearest neighbor heuristic algorithm is run to generate a new local network.

8. The photovoltaic power plant collector system topology optimization method based on a three-stage optimization genetic algorithm according to any one of claims 1-4, characterized in that, The dynamic penalty assessment includes: The current penalty coefficient is determined based on the current iteration count and the maximum iteration count. For each individual repaired by the double repair mechanism, calculate the fitness of the individual repaired by the double repair mechanism based on the current penalty coefficient; Based on the fitness of each individual repaired by the dual repair mechanism, select multiple individuals repaired by the dual repair mechanism to form the next generation population.

9. The photovoltaic power plant collector system topology optimization method based on the three-stage optimization genetic algorithm according to claim 8, characterized in that, Based on the current penalty coefficient, calculate the fitness of individuals repaired by the dual repair mechanism, including: Calculate the cost of repairing an individual that has undergone a dual repair mechanism; Determine the amount of constraint violation by an individual who has undergone dual-repair mechanism repair; The fitness of individuals repaired through the dual-repair mechanism is calculated based on their cost, current penalty coefficient, and constraint violation amount.

10. A photovoltaic power plant collector system topology optimization system based on a three-stage optimization genetic algorithm, characterized in that, A photovoltaic power plant collector system topology optimization method for executing the three-stage optimization genetic algorithm according to any one of claims 1-9 includes: The data acquisition module is used to acquire basic data of the photovoltaic power station's power collection system, wherein the photovoltaic power station's power collection system includes a step-up substation and multiple box-type transformers; The population initialization module is used to initialize the population based on the basic data of the photovoltaic power station's collection system using a hybrid initialization strategy. The topology optimization module is used to repeatedly execute genetic operations, a dual repair mechanism, and dynamic penalty evaluation to update the population until preset conditions are met, and to determine the optimal topology network from the current population.

Images