Modular optimization arrangement method and system for flexible direct-current offshore platform

Abstract

The invention belongs to the field of power system engineering, and particularly relates to a flexible direct-current offshore platform modular optimization arrangement method and system. The method comprises the following steps: abstracting each functional cabin module of the offshore converter station into a three-dimensional rectangular block with an envelope size, and constructing a to-be-arranged set; then, carrying out automatic global optimization by adopting an improved genetic algorithm: generating a random layout scheme population through coding, and evaluating the compactness (total volume minimization) of the scheme and the satisfaction degree of multiple engineering constraints such as space, position, attitude, contact and the like by using a fitness function; when the algorithm executes selection, crossover and mutation operations, a random disturbance mechanism for preventing local stagnation is introduced; and finally outputting an optimal three-dimensional layout scheme meeting the termination condition. According to the method, manual design depending on experience is converted into digital and automatic optimization, a globally better compact arrangement scheme can be efficiently generated, and design efficiency, economical efficiency and engineering feasibility are improved.

Description

Technical Field

[0001] This invention belongs to the field of power system engineering, and specifically relates to a modular optimization layout method and system for flexible DC offshore platforms. Background Technology

[0002] With the large-scale development of offshore wind farms, flexible DC transmission technology, as a core supporting means for achieving efficient power transmission, relies heavily on offshore converter stations, which play crucial roles in power collection, voltage transformation, and power transmission. Offshore platforms are exposed to extreme marine environments such as high salt spray corrosion, strong winds and waves, and typhoons during construction, transportation, and installation, resulting in exceptionally high costs for structural design, material selection, and construction. Therefore, extremely stringent requirements are placed on the compactness and lightweight design of converter station platforms.

[0003] In current engineering practice, designers primarily rely on personal experience for layout planning: first, based on the main loop topology and equipment selection scheme, they manually arrange the spatial positions of each functional module, and then repeatedly check and adjust to meet basic spatial constraints. This experience-driven approach has significant drawbacks: the design process heavily depends on the individual experience accumulated by engineers, resulting in large fluctuations in the quality of designs from different personnel, making it difficult to form a standardized knowledge system; at the same time, manual adjustments are inefficient and time-consuming, especially when the converter station contains a large number of equipment units, multiple functional modules, and multiple complex constraints. Designers often struggle to comprehensively coordinate all constraints, frequently falling into a local optimization dilemma and failing to obtain a globally optimal compact layout scheme, severely restricting the economic efficiency and feasibility of platform design. Summary of the Invention

[0004] The purpose of this invention is to provide a modular optimization layout method and system for flexible DC offshore platforms, which improves the efficiency and optimization of layout design, achieves platform compactness, and reduces construction costs and risks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a modular optimization layout method for flexible DC offshore platforms is provided, including: Each functional compartment module of the offshore converter station is abstracted as a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; Set the genetic algorithm parameters and initialize the population. Each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. The fitness value of each individual in the current population is calculated. The fitness value is determined by integrating the overall platform volume optimization objective with the penalty function for the degree of violation of multiple layout constraints. Based on the fitness value, perform selection, crossover, and mutation operations on the current population to generate offspring populations; Determine if the algorithm is stuck in a local stagnation; if so, introduce random perturbation to the current best individual or population to escape the local optimal solution space; if not, proceed to the next step. Determine if the algorithm termination condition is met; if it is, output the layout scheme corresponding to the current best individual as the final optimized layout scheme; if it is not met, take the offspring population as the new current population, return to the step of calculating fitness value and continue iterative optimization.

[0006] Furthermore, the functional modules of the offshore converter station are abstracted as three-dimensional rectangular blocks with envelope dimensions, forming a set of rectangular blocks to be laid out, specifically including: Obtain the functional compartment module division information of the offshore converter station and determine the M functional compartment modules that constitute the converter station; For each functional compartment module, based on its internal equipment composition and layout, it is determined that it consists of J indivisible sub-module units, where J is an integer greater than or equal to 1; For each submodule unit, based on its maximum internal equipment size, electrical clearance requirements between equipment, and space required for equipment installation and maintenance, the maximum envelope size of the submodule unit in the three dimensions of length, width, and height is determined, thereby abstracting the submodule unit into a three-dimensional rectangular block with a defined length, width, and height. Summarize the three-dimensional rectangular blocks corresponding to all sub-module units contained in all M functional compartment modules to form a set of rectangular blocks to be laid out, where the number N of rectangular blocks in the set is equal to the sum of the number of all sub-module units.

[0007] Furthermore, the genetic algorithm parameters are set and the population is initialized. Each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space, specifically including: An individual is encoded using a sequence containing N genes, where N is the total number of three-dimensional rectangular blocks in the set of rectangular blocks; each gene corresponds to a three-dimensional rectangular block and contains the position coordinates and placement orientation information of the rectangular block in the three-dimensional space of the platform; Set the control parameters for the genetic algorithm. The parameters should include at least the population size, maximum number of iterations, selection rate for selection, crossover rate for crossover, and mutation rate for mutation. Based on the defined encoding rules and the set control parameters, an initial population is randomly generated: for each individual in the population, within the preset platform space coordinate range, the corresponding position coordinates of each gene in the sequence are randomly generated, and a placement posture is randomly selected from a preset finite number of placement postures, thereby forming an initial individual representing a random layout scheme.

[0008] Furthermore, the fitness value of each individual in the current population is calculated. The fitness value is determined by integrating the overall platform volume optimization objective with a penalty function for the degree of violation of multiple layout constraints, specifically including: Decode the current individual and determine the specific position coordinates and placement posture of each three-dimensional rectangular block in the platform space based on its gene sequence; Based on the layout of all the rectangular blocks obtained from the decoding, calculate the length L, width W and height H of the minimum outer envelope cuboid required by the platform under this layout scheme, and then calculate the total volume of the platform V=L×W×H. Based on the specific layout obtained from decoding, the degree of violation of multiple preset layout constraints is calculated. The layout constraints include at least spatial constraints, positional constraints, orientation constraints, contact constraints, and stability constraints. For any constraint i, the degree of violation is quantified as a non-negative penalty value p_i, where p_i=0 indicates that the constraint is fully satisfied. The calculated penalty values ​​p_i for each constraint are weighted and summed using preset weight coefficients w_i corresponding to the importance of each constraint, to obtain the total penalty value P=Σ(w_i×p_i) that characterizes the overall infeasibility of the layout scheme. Combining the calculated total platform volume V and total penalty value P, the fitness value of the current individual is calculated according to the fitness function F=1 / (V+α×P), where α is a positive coefficient that scales the penalty value to match the volume order. The larger the value of the fitness function F, the better the corresponding layout scheme is in terms of smaller volume and better compliance with constraints.

[0009] Furthermore, based on the fitness value, selection, crossover, and mutation operations are performed on the current population to generate offspring populations, specifically including: Based on the calculated fitness values ​​of each individual in the current population, roulette wheel selection or tournament selection is used to select individuals with high fitness values ​​from the current population as parent individuals to form a parent population for reproduction. The selected parent individuals are paired up, and for each pair of parent individuals, a crossover operation is performed according to a preset crossover probability; the crossover operation generates new offspring individuals by exchanging some genes in the gene sequences of the paired individuals. For the offspring individuals generated by the crossover operation, as well as some of the current population individuals that were not selected, a mutation operation is performed according to a preset mutation probability. The mutation operation introduces new layout features by randomly changing the position coordinates or placement posture of one or more genes in the individual's gene sequence. The new set of individuals generated by crossover and mutation operations is merged with several of the best individuals retained in the current population according to their fitness values ​​to form the offspring population for the next generation iteration.

[0010] Furthermore, determine if the algorithm is stuck in a local stagnation; if so, introduce random perturbations to the current best individual or population to escape the local optimum solution space, specifically including: Record the improvement of the global optimal fitness value during S consecutive iterations; if the global optimal fitness value does not improve after S consecutive iterations, the algorithm is determined to be stuck in a local stagnation; otherwise, the algorithm is determined not to be stuck in a local stagnation and proceeds directly to the next step; where S is a preset positive integer threshold. Select one or more target individuals from the current population and perform random perturbation operations. The target individuals include the current global best individual and / or K individuals randomly selected from the current population. For each selected target individual, one or more genes are randomly selected from its gene sequence, and the corresponding position coordinates or placement posture are randomly reset within a preset feasible range. The new individuals obtained after the perturbation operation replace the corresponding original individuals in the current population. Then, return to the step of calculating the fitness value and continue iterative optimization based on the updated population.

[0011] Furthermore, it is determined whether the algorithm termination condition is met, specifically including: Determine whether at least one preset algorithm termination condition is met; termination conditions include: the current iteration number has reached the set maximum iteration number, or the improvement of the global optimal fitness value in T consecutive iterations is less than the preset convergence threshold ε. If any termination condition is met, the individual with the highest fitness value in the current population is decoded, and the final position coordinates and placement posture information corresponding to each three-dimensional rectangular block recorded in its gene sequence are extracted to form the final optimized layout scheme and output it. If no termination condition is met, the generated offspring population is set as the new current population, and the step of calculating fitness values ​​is returned for the next iteration of optimization.

[0012] Furthermore, the degree of violation of contact constraints is calculated, and specific implementation methods include: Identify at least one pair of target rectangular blocks that have a physical connection requirement in engineering; Based on the decoded layout, calculate the minimum spatial distance between corresponding surfaces or edges that form a physical connection between a pair of target rectangular blocks; The minimum spatial distance is directly used as or proportionally converted into the penalty value p_contact corresponding to the contact constraint.

[0013] Furthermore, in the mutation operation, the gene to be mutated is selected according to the following directional strategy: Strategy 1: With the first preset probability, select the gene corresponding to the rectangular block whose current layout position causes any penalty value p_i in the total penalty value P to be greater than zero; Strategy 2: With a second preset probability, select the gene corresponding to the rectangular block whose current layout position is within a preset neighborhood range of each surface of the current platform's outer envelope cuboid.

[0014] Secondly, a modular optimized layout system for flexible DC offshore platforms is provided, including: The module abstraction module is used to abstract each functional compartment module of the offshore converter station into a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; The population initialization module is used to set the genetic algorithm parameters and initialize the population based on the set of rectangular blocks, wherein each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. The fitness calculation module is used to calculate the fitness value of each individual in the current population. The fitness value is determined by integrating the overall platform volume optimization objective with the penalty function for the degree of violation of multiple layout constraints. The genetic evolution module is used to perform selection, crossover, and mutation operations on the current population based on the fitness value to generate a progeny population; The stagnation detection and perturbation module is used to determine whether the algorithm has fallen into local stagnation, and when it is determined that it has fallen into stagnation, it introduces random perturbation to the current best individual or population to escape the local optimal solution space; The termination judgment and output module is used to determine whether the algorithm termination condition is met. If it is met, the layout scheme corresponding to the current best individual is output as the final optimized layout scheme. If it is not met, the offspring population is taken as the new current population, and the fitness calculation module is triggered to continue iterative optimization.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: By employing an improved genetic algorithm framework and combining it with a targeted random perturbation strategy, this scheme can perform efficient and systematic global search within a vast three-dimensional layout solution space. This improves search efficiency, effectively escapes local optima, and finds layout schemes with smaller overall platform volume and higher space utilization, thereby reducing the amount of construction materials, transportation, and installation costs for offshore platforms.

[0016] This solution does not oversimplify the 3D bin packing problem. Instead, it establishes a penalty function model that integrates multiple layout constraints. This transforms key engineering requirements, such as spatial constraints, positional constraints, orientation constraints, contact constraints, and stability constraints, into quantifiable metrics that the algorithm can recognize and optimize. This ensures that the automatically generated layout is not only geometrically compact but also engineering-feasible, safe, and meets operational requirements, significantly improving the practicality and reliability of the design.

[0017] The layout design process, which relies on human experience, is transformed into a repeatable, iterative, and standardized automated calculation process. This reduces over-reliance on the personal experience of senior engineers, ensures consistent and high-quality design, and significantly shortens the design cycle.

[0018] The proposed targeted mutation strategy and stagnation detection-based random perturbation mechanism do not involve blind random search, but rather guide algorithmic resources to be more concentrated on improving the weak points of the current layout. This results in faster algorithm convergence and a higher quality, more stable optimized solution. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 The largest envelope rectangle of the module Figure 2 Platform-based layout plan Detailed Implementation The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0020] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as those skilled in the art to which this application pertains. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0021] like Figure 1 As shown, a modular optimization layout method for a flexible DC offshore platform includes: Step 1: Abstract each functional compartment module of the offshore converter station into a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out. Step 2: Set the genetic algorithm parameters and initialize the population. Each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. Step 3: Calculate the fitness value of each individual in the current population. The fitness value is determined by integrating the overall platform volume optimization objective and the penalty function for the degree of violation of multiple layout constraints. Step 4: Based on the fitness value, perform selection, crossover, and mutation operations on the current population to generate offspring population; Step 5: Determine if the algorithm is stuck in a local stagnation. If so, introduce random perturbation to the current best individual or population to escape the local optimal solution space. If not, proceed to step 6. Step 6: Determine if the algorithm termination condition is met. If it is met, output the layout scheme corresponding to the current best individual as the final optimized layout scheme. If it is not met, take the offspring population as the new current population and return to step 3 to continue iterative optimization.

[0022] In one embodiment, step 1 involves abstracting each functional compartment module of the offshore converter station into a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; the process includes the following steps: Step 1.1: Obtain the functional compartment module division information of the offshore converter station and determine the M functional compartment modules that constitute the converter station; Step 1.2: For each functional compartment module, based on its internal equipment composition and layout, determine that it consists of J indivisible sub-module units, where J is an integer greater than or equal to 1. Step 1.3: For each submodule unit, based on its maximum internal equipment size, electrical clearance requirements between equipment, and space required for equipment installation and maintenance, determine the maximum envelope size of the submodule unit in the three dimensions of length, width, and height, thereby abstracting the submodule unit into a three-dimensional rectangular block with a defined length, width, and height. Step 1.4: Summarize the three-dimensional rectangular blocks corresponding to all sub-module units contained in all M functional compartment modules to form a set of rectangular blocks to be laid out, wherein the number N of rectangular blocks in the set of rectangular blocks is equal to the sum of the number of all sub-module units.

[0023] Specifically, the core of this step is to achieve a precise conversion from the actual engineering object to a model that the optimization algorithm can handle. In step 1.2, "indivisible" means that from the perspective of layout optimization, the internal equipment layout of this sub-module unit is relatively fixed and is considered a whole in this level of optimization. For example, a "valve hall module" consists of "valve tower sub-modules," "cooling equipment sub-modules," and "busbar erection sub-modules," etc. Step 1.3 is the foundation for ensuring the engineering feasibility of the optimization results. When determining the maximum envelope size, static dimensions, as well as the space margin required for dynamically stacked equipment hoisting, personnel passage, and pipeline and cable tray laying, are considered. The margin can be quantified based on relevant design specifications, operation and maintenance manuals, and safety standards to ensure that the abstracted rectangular block space can meet the needs of the entire lifecycle. This abstraction method transforms complex, irregular compartment spaces into regular geometries, simplifying the complexity of collision detection, spatial calculations, and other operations in subsequent optimization algorithms.

[0024] As an example, the M functional compartment modules may include: converter valve hall, AC switchgear room, DC switchgear room, control and protection room, auxiliary system room (including air conditioning and fire protection), transformer room (which may be considered separately if installed on the top of the platform), cable layer / pipe corridor, etc. The envelope dimension of an "AC switchgear room" sub-module unit (J=1) needs to consider the dimensions of the largest internal switchgear (e.g., length 2.5m, width 1.2m, height 2.3m), plus the front operation and maintenance passage (not less than 1.5m), the rear maintenance passage (not less than 0.8m), and the top ventilation duct space (0.5m). The final abstract dimensions of this rectangular block are: Length = switchgear depth + front and rear passage = 1.2m + 1.5m + 0.8m = 3.5m; Width = total width of parallel cabinets (assuming 4 cabinets side-by-side) = 2.5m * 4 = 10m; Height = cabinet height + top space = 2.3m + 0.5m = 2.8m.

[0025] Optionally, for modules with highly irregular shapes or containing large tilting devices, they can be abstracted as a composite of multiple smaller rectangular blocks, and in subsequent optimization, hard constraints such as "must be adjacent" or "relative position fixed" can be added to these composite rectangular blocks to approximate their real shape.

[0026] In one embodiment, step 2 involves setting the genetic algorithm parameters and initializing the population, where each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space; specifically, this includes the following steps: Step 2.1: Encode the individual using a sequence containing N genes, where N is the total number of three-dimensional rectangular blocks in the rectangular block set; each gene corresponds to a three-dimensional rectangular block and contains the position coordinates and placement posture information of the rectangular block in the three-dimensional space of the platform; Step 2.2: Set the control parameters of the genetic algorithm. The parameters should include at least the population size, maximum number of iterations, selection rate of the selection operation, crossover rate of the crossover operation, and mutation rate of the mutation operation. Step 2.3: Based on the encoding rules defined in Step 2.1 and the control parameters set in Step 2.2, an initial population is randomly generated: for each individual in the population, within the preset platform space coordinate range, the corresponding position coordinates of each gene in the sequence are randomly generated, and a placement posture is randomly selected from a preset limited number of placement postures, thereby forming an initial individual representing a random layout scheme.

[0027] Specifically, step 2.1 employs a direct encoding method, where individual gene sequences intuitively reflect the spatial state of all rectangular blocks. Position coordinates can be defined as the (x, y, z) values ​​of a specific corner point of a rectangular block (such as the geometric center or the lower left corner) in the platform's global coordinate system. Placement attitude information defines the rotation state of the rectangular block's own coordinate system relative to the platform's global coordinate system, which can be simplified to rotations of 0°, 90°, 180°, and 270° around the vertical axis (Z-axis), corresponding to different orientations of the equipment within the cabin (e.g., the cabinet facing the passageway). The random initialization in step 2.3 is the starting point for the global search. To ensure the diversity and feasibility of the initial population, the platform's spatial coordinate range can be set to a loose bounding box larger than the estimated final platform size, allowing the rectangular blocks sufficient space for random distribution. Attitude random selection is uniformly drawn from an allowed discrete set.

[0028] As an example, assume N = 50 rectangular blocks. An individual is encoded as a list of length 50, where each element is a tuple, for example: [(x1, y1, z1, orientation1), (x2, y2, z2, orientation2), ..., (x50, y50, z50, orientation50)]. The orientation value can be {0, 1, 2, 3}, corresponding to rotations of 0°, 90°, 180°, and 270° around the Z-axis, respectively. Example control parameters: population size pop_size = 100, maximum number of iterations max_gen = 500, crossover rate pc = 0.8, mutation rate pm = 0.1.

[0029] Alternatively, sequence-based indirect encoding can be used. Each individual is encoded as a priority placement sequence of rectangular blocks and a corresponding placement area or rough position label for each rectangular block. During decoding, the rectangular blocks are placed one by one into the platform space according to the sequence order, combined with a deterministic placement rule (such as a "bottom-left fill" rule). This method automatically avoids overlap, but designing the placement rules is more complex.

[0030] In one embodiment, step 3 involves calculating the fitness value of each individual in the current population. The fitness value is determined jointly by integrating the overall platform volume optimization objective and the penalty function for the degree of violation of multiple layout constraints. Specifically, this includes the following steps: Step 3.1: Decode the current individual and determine the specific position coordinates and placement posture of each three-dimensional rectangular block in the platform space based on its gene sequence; Step 3.2: Based on the layout of all rectangular blocks obtained by decoding in Step 3.1, calculate the length L, width W, and height H of the minimum outer envelope cuboid required by the platform under this layout scheme, and then calculate the total volume of the platform V = L × W × H. Step 3.3: Based on the specific layout obtained from the decoding in Step 3.1, calculate the degree of violation of the preset multiple layout constraints. The layout constraints include at least spatial constraints (ensuring that the rectangular blocks are within the platform boundary and do not overlap), position constraints, orientation constraints, contact constraints, and stability constraints. For any constraint i, the degree of violation is quantified as a non-negative penalty value p_i, where p_i=0 indicates that the constraint is fully satisfied. Step 3.4: The penalty values ​​p_i of each constraint calculated in step 3.3 are weighted and summed using the preset weight coefficient w_i corresponding to the importance of each constraint to obtain the total penalty value P=Σ(w_i×p_i) that represents the overall infeasibility of the layout scheme. Step 3.5: Combining the total platform volume V calculated in Step 3.2 and the total penalty value P calculated in Step 3.4, calculate the fitness value of the current individual according to the fitness function F=1 / (V+α×P), where α is a positive coefficient that scales the penalty value to match the volume order. The larger the value of the fitness function F, the better the corresponding layout scheme is in terms of smaller volume and better compliance with constraints.

[0031] Specifically, in step 3.2, calculating the outer envelope cuboid requires traversing the vertex coordinates of all rectangular blocks to find the maximum and minimum values ​​in the x, y, and z directions. Constraint violation metric quantization in step 3.3 is crucial. Further, Spatial constraints: Calculate the 3D interference volume between all pairs of rectangular blocks, and the volume of each rectangular block exceeding the maximum boundary of the preset platform (which can be set according to the initialization range). The penalty value p_space can be a weighted sum of the total interference volume and the total excess volume.

[0032] Location constraint: Checks whether a specific rectangular block (such as heavy equipment) is located within a specified deck level (Z coordinate range) or a planar region (XY coordinate range). The penalty value p_location can be the shortest distance from its geometric center to the specified region.

[0033] Orientation constraints: Check whether the orientation of a specific rectangular block (such as a cabinet that must be placed against a wall) meets the requirements. The penalty value p_orientation can be a Boolean value (0 if compliant, a constant if not compliant).

[0034] Contact / Adjacency Constraints: Check whether the corresponding surfaces of connected rectangular blocks (such as transformers and valve halls) are within the allowable distance range. The penalty value p_adjacency can be the absolute difference between the actual distance and the target distance.

[0035] Stability constraints: Estimate the position of the center of gravity of the entire platform and check if it is within the allowable range (e.g., projected near the center of the platform bottom). The penalty value p_stability can be the horizontal offset of the center of gravity. The weights w_i in step 3.4 need to be set according to the priority based on engineering experience. Spatial non-overlap (w_space) has the highest weight, followed by attitude and position, and then adjacency and stability. The coefficient α in step 3.5 is used to balance the order of magnitude of the objective function and the penalty term, ensuring that in the early stages of iteration, even if the layout is infeasible (P is large), the algorithm still has the motivation to improve the constraint satisfaction; in the later stages of iteration, the volume V is mainly compared among feasible solutions.

[0036] As an example, w_space = 1000.0, w_location = 100.0, w_orientation = 50.0, w_adjacency = 10.0, w_stability = 5.0. α can be estimated based on the average V and P of the initial population, for example, α = average(V) / average(P), or directly set to a large number such as α = 1000 to prioritize ensuring constraint satisfaction.

[0037] Alternatively, the fitness function can also be in the form F=1 / (V*(1+β*P)), where β is the penalty coefficient. Or, a feasibility rule can be used to handle the constraints: among all individuals, feasible solutions (P=0) are always better than infeasible solutions; V is compared among feasible solutions; P is compared among infeasible solutions. This method does not require setting weights and scaling factors, but it places higher demands on the algorithm's search capabilities.

[0038] In one embodiment, step 4 involves performing selection, crossover, and mutation operations on the current population based on its fitness value to generate a progeny population; specifically, this includes the following steps: Step 4.1: Based on the fitness values ​​of each individual in the current population calculated in Step 3, select individuals with high fitness values ​​from the current population as parent individuals using roulette wheel selection or tournament selection to form a parent population for reproduction. Step 4.2: Pair the parent individuals selected in Step 4.1 into pairs. For each pair of parent individuals, perform a crossover operation according to a preset crossover probability. The crossover operation generates new offspring individuals by exchanging some genes in the gene sequences of the paired individuals (i.e., exchanging the position and orientation information of some three-dimensional rectangular blocks). Step 4.3: Perform mutation operation on the offspring individuals generated by the crossover operation in Step 4.2, as well as the unselected individuals in the current population in Step 4.1, according to the preset mutation probability; the mutation operation introduces new layout features by randomly changing the position coordinates or placement posture of one or more genes in the individual's gene sequence. Step 4.4: The new set of individuals generated by steps 4.2 and 4.3 is merged with several of the best individuals retained in the current population according to their fitness values, and together they form the offspring population for the next generation iteration.

[0039] Specifically, in step 4.1, the tournament selection method is more robust: randomly select k individuals from the population (e.g., k=3), choose the one with the highest fitness as the parent, and repeat this process until all individuals are selected. The crossover operation in step 4.2 can use two-point crossover or uniform crossover. For direct encoding, exchange the coordinates and pose information of randomly selected segments (corresponding to a set of rectangular blocks) in the parent individual's gene sequence. It should be noted that simple exchange may produce a large number of overlapping infeasible solutions, but this will be corrected by subsequent fitness function penalties, mutations, and perturbations. The purpose of crossover is to attempt to combine excellent layout "segments" from the parent. The mutation operation in step 4.3 is crucial for maintaining population diversity and local search. A gene can be randomly selected, and its coordinates can be randomly perturbed (e.g., x=x±Δx) within a small neighborhood of its current position, or its pose can be randomly changed. The elite preservation strategy in step 4.4 is crucial; it ensures that the optimal solutions discovered during iteration are not lost, making the algorithm's convergence more stable.

[0040] As an example, the tournament size is k=3. Crossover is a two-point crossover: two random cut points are generated, and all genetic information of the two parent individuals between these two cut points is exchanged. Mutation: each gene of the offspring individual is mutated with a probability pm=0.1. If mutation occurs, there is a 50% probability of a small perturbation to the coordinates (Δx, Δy, Δz are random within ±2 meters), and a 50% probability of a random reset of the pose. The elite retention rate is set to 5%, meaning that the 5 individuals with the highest fitness in each generation directly enter the next generation.

[0041] Optionally, sequential crossover or position-based crossover can be used, which is particularly suitable for sequence-based encoding methods and can better inherit the sequential characteristics of the parent generation. The mutation operation can also be designed as adaptive mutation, increasing the mutation rate pm to enhance the exploration ability when the population diversity decreases (e.g., the fitness variance decreases).

[0042] In one embodiment, step 5 involves determining whether the algorithm has fallen into a local stagnation. If so, a random perturbation is introduced into the current optimal individual or population to escape the local optimal solution space; if not, the process proceeds to step 6. Specifically, this includes the following steps: Step 5.1: Record the improvement of the global optimal fitness value calculated in Step 3 during the S consecutive iterations; if the global optimal fitness value does not improve after S consecutive iterations, the algorithm is determined to be stuck in a local stagnation and proceeds to Step 5.2; otherwise, the algorithm is determined not to be stuck in a local stagnation and proceeds directly to Step 6; where S is a preset positive integer threshold. Step 5.2: Select one or more target individuals from the current population and perform random perturbation operations. The target individuals include the current global best individual and / or K individuals randomly selected from the current population. Step 5.3: For each target individual selected in step 5.2, randomly select one or more genes from its gene sequence, and randomly reset its corresponding position coordinates or placement posture within a preset feasible range. Step 5.4: Replace the original individuals in the current population with the new individuals obtained after the perturbation operation in Step 5.3, and then return to Step 3 to continue iterative optimization based on the updated population.

[0043] Specifically, the stagnation check in step 5.1 is the condition for triggering the "restart" or "perturbation" mechanism. The perturbation operation in steps 5.2-5.3 is stronger than regular mutation. Perturbing the current global best individual (which can be called "local search restart") aims to explore a wider range in its vicinity, hoping to find a better neighboring solution than the current best solution. Strong perturbation of K randomly selected individuals is to reinject diversity into the population, simulating a small-scale "reinitialization," helping the algorithm escape the attraction domain of the current best solution. During perturbation, the "feasible range" of random reset can be much larger than the neighborhood of mutation operation; for example, allowing coordinates to be randomly regenerated within the entire platform initialization space.

[0044] As an example, the stagnation threshold S = 20 generations. The perturbation strategy is as follows: Select the current globally optimal individual, and randomly select K = 10% * population size individuals (e.g., 10 individuals) for perturbation. For each target individual, randomly select 30% of its genes (e.g., 15 rectangular blocks), and completely and randomly reset the position coordinates of these genes within the initial platform space, as well as their poses. After perturbation, replace the original individual with a new one.

[0045] Alternatively, simulated annealing can be used, allowing for a certain probability of perturbing the optimal individual, even if its fitness temporarily deteriorates. Or, a multi-population genetic algorithm can be employed, where, when the main population stagnates, a completely randomly generated new subpopulation is introduced and merged with it for co-evolution.

[0046] In one embodiment, step 6 involves determining whether the algorithm termination condition is met. If met, the layout scheme corresponding to the current optimal individual is output as the final optimized layout scheme. If not met, the offspring population is used as the new current population, and the process returns to step 3 to continue iterative optimization. Specifically, this includes the following steps: Step 6.1, Termination Condition Judgment: Determine whether at least one preset algorithm termination condition is met; the termination conditions include: the current iteration number has reached the maximum iteration number set in step 2.2, or the improvement of the global optimal fitness value in T consecutive iterations is less than the preset convergence threshold ε. Step 6.2, Scheme Output: If any termination condition is met as determined in Step 6.1, decode the individual with the highest fitness value in the current population, extract the final position coordinates and placement posture information corresponding to each three-dimensional rectangular block recorded in its gene sequence, form the final optimized layout scheme and output it. Step 6.3, Iteration continues: If step 6.1 determines that no termination condition is met, then the offspring population generated in step 4 is set as the new current population, and step 3 is returned to perform the next iteration optimization.

[0047] Specifically, step 6.1 sets dual termination conditions to balance computational efficiency and optimization quality. The maximum number of iterations is a hard stopping condition to prevent infinite loops. The convergence judgment based on fitness improvement is a soft condition; when the optimization enters a plateau and the improvement is negligible, it is considered to have approached the optimal solution or has been sufficiently searched, and the computation can be stopped. T and ε need to be set according to the problem size and expected accuracy. The layout scheme output in step 6.2 not only includes the geometric information of each rectangular block, but should also undergo post-processing: mapping the abstract rectangular blocks back to the actual functional compartment sub-modules to generate a layout diagram or 3D model containing the exact location, size, and orientation of each compartment / equipment, and further detailed verification of the center of gravity, passageways, pipeline routes, etc.

[0048] As an example, the maximum number of iterations is max_gen=500. Convergence criteria parameters: T=50, ε=1e-6 (assuming the fitness value F is in the tens to hundreds range). The output can be a CSV table with columns including: module ID, associated compartment, center coordinates X, Y, Z, length (L), width (W), height (H), and rotation angle around the Z-axis. A 3D visualization model can also be generated for designers to review.

[0049] Optionally, termination conditions can also consider calculating the time budget or population convergence (e.g., the standard deviation of individual fitness in the population is less than a certain threshold). When outputting a solution, several suboptimal solutions (e.g., the top 3 solutions in terms of fitness) can be output simultaneously for designers to compare and make decisions based on multiple solutions.

[0050] In one embodiment, the calculation of the degree of violation of the contact constraint in step 3.3 is specifically implemented as follows: Identify at least one pair of target rectangular blocks that have a physical connection requirement in engineering; Based on the decoded layout, calculate the minimum spatial distance between corresponding surfaces or edges that form a physical connection between a pair of target rectangular blocks; The minimum spatial distance is directly used as or proportionally converted into the penalty value p_contact corresponding to the contact constraint.

[0051] Specifically, contact constraints are mainly used to optimize the layout between closely connected device modules, such as bushing connections between transformers and valve halls, and large pipe connections. First, a list of connection relationships needs to be predefined. Each item in the list specifies the IDs of two rectangular blocks and the specific faces they need to contact or approach (e.g., the "+X face" of device A needs to be close to the "-Y face" of device B). During calculation, the actual plane equations of the specified surfaces in space are calculated based on the orientation and position of the two rectangular blocks. Then, the shortest Euclidean distance d between these two parallel planes (or the edges that need to approach) is calculated. If close contact is required (distance is 0), then p_contact = |d|. If an installation gap d_target (e.g., 0.5 meters) is allowed, then p_contact = |d - d_target|. In this way, the algorithm strives to drive d towards the target value, thereby achieving precise adjacency control.

[0052] As an example, the connection relationship is as follows: Rectangular block ID_15 (transformer bushing side, +X plane) needs to be connected to rectangular block ID_23 (valve hall inlet side, -X plane), with a target installation gap d_target=0.2m. The calculated distance between the two specified planes is d=1.5m, therefore p_contact=|1.5-0.2|=1.3.

[0053] Optionally, for complex non-planar contacts, the contact constraint can be decomposed into distance constraints between multiple key connection points. The penalty p_contact can be set as the sum of squared distances between these key point pairs.

[0054] In one embodiment, the mutation operation in step 4.3 selects the gene to be mutated according to the following targeting strategy: Strategy 1: With the first preset probability, select the gene corresponding to the rectangular block whose current layout position causes any penalty value p_i in the total penalty value P to be greater than zero; Strategy 2: With a second preset probability, select the gene corresponding to the rectangular block whose current layout position is within a preset neighborhood range of each surface of the current platform's outer envelope cuboid.

[0055] Specifically, this is a heuristic mutation strategy designed to improve the efficiency and targeting of mutations, accelerating convergence. Strategy 1 (Constraint Repair Oriented): By analyzing the composition of the total penalty term P, the algorithm identifies the specific constraints violated and which rectangular blocks caused the violation. For example, if p_space is large, it identifies the rectangular blocks with the largest overlapping volume with other blocks; if p_location is not zero, it identifies rectangular blocks far from the defined region. Mutating the genes of these "problem rectangular blocks" directly attempts to repair constraint violations, effectively guiding the population towards the feasible region. Strategy 2 (Boundary Optimization Oriented): The total volume V of the platform is determined by the outer envelope cuboid. Rectangular blocks located near the outer envelope surface (e.g., less than 1 meter from the surface) are key to defining the platform boundary. Mutating the genes of these rectangular blocks (especially fine-tuning their position or rotation) may directly change a dimension among L, W, and H, thus directly affecting the optimization objective V. By prioritizing the mutation of these key blocks, the algorithm can more proactively "shape" the platform's shape to reduce its volume.

[0056] As an example, when performing mutation, for the individual to be mutated, a random number r is first generated. If r < 0.6 (first preset probability), strategy one is used, selecting 1-3 rectangular blocks that result in the maximum penalty (e.g., overlap) for mutation. If 0.6 ≤ r < 0.9 (second preset probability), strategy two is used, randomly selecting 1-2 rectangular blocks within a 0.5-meter radius of the outer envelope surface for mutation. Otherwise (r ≥ 0.9), a completely random traditional mutation is performed.

[0057] Optionally, the probabilities of these two strategies can be dynamically adjusted. In the early stages of optimization, when there are many infeasible solutions in the population, the probability of strategy one can be increased, focusing on feasibility search. In the later stages of optimization, when most solutions in the population are feasible, the probability of strategy two can be increased, focusing on fine-grained volume optimization. A third strategy can also be introduced: selecting rectangular blocks with low current fitness contribution (e.g., low space utilization, or relatively empty positions) for mutation.

[0058] The modular optimization layout method for flexible DC offshore platforms provided in this application transforms the manual layout model, which relies on personal experience and repeated trial and error, into a replicable and standardized digital automated optimization process by modularly abstracting functional compartments and using an improved genetic algorithm for automated global optimization. This improves the consistency and scientific rigor of the design. It can systematically coordinate multiple complex constraints and efficiently search for the globally optimal solution in a massive layout space, resulting in a significantly improved compactness and effectively reducing platform construction costs and material usage.

[0059] Example 2 like Figure 2 As shown, a modular optimized layout system for a flexible DC offshore platform includes: The module abstraction module is used to abstract each functional compartment module of the offshore converter station into a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; The population initialization module is used to set the genetic algorithm parameters and initialize the population based on the set of rectangular blocks, wherein each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. The fitness calculation module is used to calculate the fitness value of each individual in the current population. The fitness value is determined by integrating the overall platform volume optimization objective with the penalty function for the degree of violation of multiple layout constraints. The genetic evolution module is used to perform selection, crossover, and mutation operations on the current population based on the fitness value to generate a progeny population; The stagnation detection and perturbation module is used to determine whether the algorithm has fallen into local stagnation, and when it is determined that it has fallen into stagnation, it introduces random perturbation to the current best individual or population to escape the local optimal solution space; The termination judgment and output module is used to determine whether the algorithm termination condition is met. If it is met, the layout scheme corresponding to the current best individual is output as the final optimized layout scheme. If it is not met, the offspring population is taken as the new current population, and the fitness calculation module is triggered to continue iterative optimization.

[0060] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A modular optimized layout method for a flexible DC offshore platform, characterized in that, include: Each functional compartment module of the offshore converter station is abstracted as a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; Set the genetic algorithm parameters and initialize the population. Each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. The fitness value of each individual in the current population is calculated. The fitness value is determined by integrating the overall platform volume optimization objective with the penalty function for the degree of violation of multiple layout constraints. Based on the fitness value, perform selection, crossover, and mutation operations on the current population to generate offspring populations; Determine if the algorithm is stuck in a local stagnation; if so, introduce random perturbation to the current best individual or population to escape the local optimal solution space; if not, proceed to the next step. Determine if the algorithm termination condition is met; if it is, output the layout scheme corresponding to the current best individual as the final optimized layout scheme; if it is not met, take the offspring population as the new current population, return to the step of calculating fitness value and continue iterative optimization.

2. The modular optimized layout method for flexible DC offshore platforms according to claim 1, characterized in that, Each functional module of the offshore converter station is abstracted as a three-dimensional rectangular block with an envelope dimension, forming a set of rectangular blocks to be laid out, specifically including: Obtain the functional compartment module division information of the offshore converter station and determine the M functional compartment modules that constitute the converter station; For each functional compartment module, based on its internal equipment composition and layout, it is determined that it consists of J indivisible sub-module units, where J is an integer greater than or equal to 1; For each submodule unit, based on its maximum internal equipment size, electrical clearance requirements between equipment, and space required for equipment installation and maintenance, the maximum envelope size of the submodule unit in the three dimensions of length, width, and height is determined, thereby abstracting the submodule unit into a three-dimensional rectangular block with a defined length, width, and height. Summarize the three-dimensional rectangular blocks corresponding to all sub-module units contained in all M functional compartment modules to form a set of rectangular blocks to be laid out, where the number N of rectangular blocks in the set is equal to the sum of the number of all sub-module units.

3. The modular optimized layout method for flexible DC offshore platforms according to claim 2, characterized in that, Set the genetic algorithm parameters and initialize the population. Each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space, specifically including: An individual is encoded using a sequence containing N genes, where N is the total number of three-dimensional rectangular blocks in the set of rectangular blocks; each gene corresponds to a three-dimensional rectangular block and contains the position coordinates and placement orientation information of the rectangular block in the three-dimensional space of the platform; Set the control parameters for the genetic algorithm. The parameters should include at least the population size, maximum number of iterations, selection rate for selection, crossover rate for crossover, and mutation rate for mutation. Based on the defined encoding rules and the set control parameters, an initial population is randomly generated: for each individual in the population, within the preset platform space coordinate range, the corresponding position coordinates of each gene in the sequence are randomly generated, and a placement posture is randomly selected from a preset finite number of placement postures, thereby forming an initial individual representing a random layout scheme.

4. The modular optimized layout method for flexible DC offshore platforms according to claim 3, characterized in that, The fitness value of each individual in the current population is calculated. The fitness value is determined by combining the overall platform volume optimization objective with a penalty function for the degree of violation of multiple layout constraints, specifically including: Decode the current individual and determine the specific position coordinates and placement posture of each three-dimensional rectangular block in the platform space based on its gene sequence; Based on the layout of all the rectangular blocks obtained from the decoding, calculate the length L, width W and height H of the minimum outer envelope cuboid required by the platform under this layout scheme, and then calculate the total volume of the platform V=L×W×H. Based on the specific layout obtained from decoding, the degree of violation of multiple preset layout constraints is calculated. The layout constraints include at least spatial constraints, positional constraints, orientation constraints, contact constraints, and stability constraints. For any constraint i, the degree of violation is quantified as a non-negative penalty value p_i, where p_i=0 indicates that the constraint is fully satisfied. The calculated penalty values ​​p_i for each constraint are weighted and summed using preset weight coefficients w_i corresponding to the importance of each constraint, to obtain the total penalty value P=Σ(w_i×p_i) that characterizes the overall infeasibility of the layout scheme. Combining the calculated total platform volume V and total penalty value P, the fitness value of the current individual is calculated according to the fitness function F=1 / (V+α×P), where α is a positive coefficient that scales the penalty value to match the volume order. The larger the value of the fitness function F, the better the corresponding layout scheme is in terms of smaller volume and better compliance with constraints.

5. The modular optimized layout method for flexible DC offshore platforms according to claim 4, characterized in that, Based on the fitness value, selection, crossover, and mutation operations are performed on the current population to generate offspring populations, specifically including: Based on the calculated fitness values ​​of each individual in the current population, roulette wheel selection or tournament selection is used to select individuals with high fitness values ​​from the current population as parent individuals to form a parent population for reproduction. The selected parent individuals are paired up, and for each pair of parent individuals, a crossover operation is performed according to a preset crossover probability; the crossover operation generates new offspring individuals by exchanging some genes in the gene sequences of the paired individuals. For the offspring individuals generated by the crossover operation, as well as some of the current population individuals that were not selected, a mutation operation is performed according to a preset mutation probability. The mutation operation introduces new layout features by randomly changing the position coordinates or placement posture of one or more genes in the individual's gene sequence. The new set of individuals generated by crossover and mutation operations is merged with several of the best individuals retained in the current population according to their fitness values ​​to form the offspring population for the next generation iteration.

6. The modular optimized layout method for flexible DC offshore platforms according to claim 5, characterized in that, Determine if the algorithm is stuck in a local stagnation; if so, introduce random perturbations to the current best individual or population to escape the local optimum solution space, specifically including: Record the improvement of the global optimal fitness value during S consecutive iterations; if the global optimal fitness value does not improve after S consecutive iterations, the algorithm is determined to be stuck in a local stagnation; otherwise, the algorithm is determined not to be stuck in a local stagnation and proceeds directly to the next step; where S is a preset positive integer threshold. Select one or more target individuals from the current population and perform random perturbation operations. The target individuals include the current global best individual and / or K individuals randomly selected from the current population. For each selected target individual, one or more genes are randomly selected from its gene sequence, and the corresponding position coordinates or placement posture are randomly reset within a preset feasible range. The new individuals obtained after the perturbation operation replace the corresponding original individuals in the current population. Then, return to the step of calculating the fitness value and continue iterative optimization based on the updated population.

7. The modular optimized layout method for flexible DC offshore platforms according to claim 6, characterized in that, Determining whether the algorithm's termination condition is met includes: Determine whether at least one preset algorithm termination condition is met; termination conditions include: the current iteration number has reached the set maximum iteration number, or the improvement of the global optimal fitness value in T consecutive iterations is less than the preset convergence threshold ε. If any termination condition is met, the individual with the highest fitness value in the current population is decoded, and the final position coordinates and placement posture information corresponding to each three-dimensional rectangular block recorded in its gene sequence are extracted to form the final optimized layout scheme and output it. If no termination condition is met, the generated offspring population is set as the new current population, and the step of calculating fitness values ​​is returned for the next iteration of optimization.

8. The modular optimized layout method for flexible DC offshore platforms according to claim 4, characterized in that, Calculating the degree of violation of contact constraints can be achieved through the following methods: Identify at least one pair of target rectangular blocks that have a physical connection requirement in engineering; Based on the decoded layout, calculate the minimum spatial distance between corresponding surfaces or edges that form a physical connection between a pair of target rectangular blocks; The minimum spatial distance is directly used as or proportionally converted into the penalty value p_contact corresponding to the contact constraint.

9. The modular optimized layout method for flexible DC offshore platforms according to claim 5, characterized in that, During mutation operations, the gene to be mutated is selected according to the following targeting strategy: Strategy 1: With the first preset probability, select the gene corresponding to the rectangular block whose current layout position causes any penalty value p_i in the total penalty value P to be greater than zero; Strategy 2: With a second preset probability, select the gene corresponding to the rectangular block whose current layout position is within a preset neighborhood range of each surface of the current platform's outer envelope cuboid.

10. A modular optimized layout system for a flexible DC offshore platform, characterized in that, include: The module abstraction module is used to abstract each functional compartment module of the offshore converter station into a three-dimensional rectangular block with an envelope size, forming a set of rectangular blocks to be laid out; The population initialization module is used to set the genetic algorithm parameters and initialize the population based on the set of rectangular blocks, wherein each individual in the population represents a random layout scheme of all rectangular blocks in the set of rectangular blocks in three-dimensional space. The fitness calculation module is used to calculate the fitness value of each individual in the current population. The fitness value is determined by integrating the overall platform volume optimization objective with the penalty function for the degree of violation of multiple layout constraints. The genetic evolution module is used to perform selection, crossover, and mutation operations on the current population based on the fitness value to generate a progeny population; The stagnation detection and perturbation module is used to determine whether the algorithm has fallen into local stagnation, and when it is determined that it has fallen into stagnation, it introduces random perturbation to the current best individual or population to escape the local optimal solution space; The termination judgment and output module is used to determine whether the algorithm termination condition is met. If it is met, the layout scheme corresponding to the current best individual is output as the final optimized layout scheme. If it is not met, the offspring population is taken as the new current population, and the fitness calculation module is triggered to continue iterative optimization.

Images