Method for splicing hollow-core and solid-core optical fibers based on multi-objective wolf pack algorithm

Abstract

The application discloses a hollow-core solid-core optical fiber fusion coupling method based on a multi-target wolf swarm algorithm and relates to the technical field of optical fiber fusion, which comprises the following steps: identifying three types of dimensional features and determining five-dimensional decision variables; adopting Latin hypercube sampling to collect multiple groups of fusion data, adopting a radial basis function to match a Gaussian kernel function, and constructing a multi-target function vector with an insertion loss, an end face deformation rate and an end face echo loss as optimization targets; performing multi-target wolf swarm algorithm iteration optimization, outputting a Pareto optimal process parameter solution set; and matching an optimal fusion parameter combination from the Pareto optimal process parameter solution set to complete arc fusion coupling of the hollow-core optical fiber and the solid-core optical fiber. The application adopts the multi-target wolf swarm algorithm, optimizes three mutually restrictive indexes of the insertion loss, the end face deformation rate and the end face echo loss, is different from artificial debugging and a single-target algorithm, and improves the debugging efficiency of the fusion parameters and the fusion stability under different working conditions.

Description

Technical Field

[0001] This invention relates to the field of optical fiber fusion splicing technology, specifically a hollow-to-solid fiber fusion splicing coupling method based on a multi-objective wolf pack algorithm. Background Technology

[0002] With the rapid construction of new power systems and ultra-high voltage transmission networks, the power backbone communication network faces stringent requirements for ultra-low latency, ultra-long-distance transmission, and high-reliability operation. Traditional solid-core optical fibers, limited by physical latency and nonlinear effects, cannot meet the needs of ultra-high voltage relay protection, flexible DC control, and other services. Hollow-core optical fibers, with their air-guiding properties, have become the core direction for upgrading power communication networks.

[0003] The large-scale application of hollow-core optical fiber requires interconnection with the existing massive solid-core optical fiber network. Hollow-core to solid-core optical fiber fusion splicing is the core process for connecting these two types of optical fibers and a key step in the fabrication of low-loss couplers. Currently, in the power communication field, fiber connection is mainly completed by manually adjusting fusion parameters in combination with conventional fusion splicing equipment. At the same time, single-objective intelligent optimization algorithms are also used to a small extent in fusion parameter adjustment scenarios.

[0004] Current fiber optic fusion splicing technology has several shortcomings: 1. Multiple performance indicators are mutually restrictive, making manual debugging difficult: During fiber optic fusion splicing, insertion loss, end-face microstructure integrity, and end-face return loss are contradictory. Reducing insertion loss can easily lead to end-face collapse, while suppressing reflection loss can increase fusion loss. Manual trial and error cannot achieve simultaneous optimization of all three performance indicators. 2. Low debugging efficiency and poor product consistency: Manual debugging of fusion splicing parameters is time-consuming, and the fusion effect fluctuates significantly depending on different batches of hollow-core optical fibers and complex power conditions in the field, making it difficult to stably achieve the power engineering performance indicator of coupler insertion loss. 3. Traditional single-objective algorithms have strong limitations and cannot balance multiple conflicting technical requirements. The output parameters are only suitable for a single scenario and cannot meet the differentiated needs of various power communication links such as UHV trunk lines and regional branch lines.

[0005] Based on this, a hollow-to-solid fiber fusion splicing coupling method based on a multi-target wolf pack algorithm is now provided, which can eliminate the drawbacks of existing technical solutions. Summary of the Invention

[0006] The purpose of this invention is to provide a hollow-to-solid fiber fusion splicing coupling method based on a multi-objective wolf pack algorithm to solve the problems in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The hollow-to-solid fiber fusion coupling method based on the multi-objective wolf pack algorithm includes the following steps: Step S1: Based on the physical coupling mechanism of hollow-core to solid-core fiber fusion splicing, identify three types of dimensional features that affect the splicing quality, and determine five-dimensional decision variables based on the three types of dimensional features; Step S2: Use Latin hypercube sampling to collect multiple sets of fusion splicing data, and use radial basis functions combined with Gaussian kernel functions to construct a multi-objective function vector with insertion loss, end face deformation rate and end face return loss as optimization objectives; Step S3: Using the multi-objective function vector as the test benchmark function, execute the multi-objective wolf pack algorithm for iterative optimization and output the Pareto optimal process parameter solution set; Step S4: Match the optimal fusion splicing parameter combination for the corresponding power communication operating condition from the Pareto optimal process parameter solution set, and use the matched optimal fusion splicing parameters to complete the arc fusion coupling of hollow fiber and solid fiber. The five-dimensional decision variables include the number of progressive discharges, the main discharge power, the discharge offset position, the fiber optic skewing angle, and the length of the self-focusing fiber. The three-dimensional features include energy input, thermal field distribution, and geometric matching.

[0008] Further, the process for determining the five-dimensional decision variables in step S1 is as follows: the number of progressive discharges and the main discharge power are extracted through the energy input dimension, which are used to control the melting depth of the quartz glass; the discharge offset position is extracted through the thermal field distribution dimension, which is used to change the center of the heat source and realize asymmetric heating of the hollow fiber; the fiber cleavage angle and the self-focusing fiber length are extracted through the geometric matching dimension, which are used to determine the end-face echo suppression capability and the mode field matching degree between heterogeneous fibers; and the search boundary for the values ​​of the five-dimensional decision variables is limited according to the hardware limits of the fusion splicing equipment.

[0009] Further, step S2 specifically includes: Several sets of fusion splicing data were selected as fusion splicing samples. Each set of fusion splicing data was recorded simultaneously with five-dimensional decision variables and measured values ​​of insertion loss, end face deformation rate, and end face return loss. Substitute the coordinates of the fusion sample, the weight coefficients of each optimization objective, and the linear drift polynomial into the Gaussian kernel radial basis function to establish the insertion loss sub-objective function, the end face deformation rate sub-objective function, and the end face return loss sub-objective function respectively. Combine them to form an iteratively computeable multi-objective function vector. The multi-objective function vector is represented as follows: , , ; in, For a multi-objective function vector, To minimize the optimization objective, For insertion loss, For end face deformation rate, For end-face return loss, superscript It is the transpose symbol. Five-dimensional decision variables, For Gaussian kernel radial basis functions, This is the serial number of the welded sample. For the first The coordinates corresponding to each welded sample It is a linear drift polynomial. For the first The first welded sample, the first The weight coefficients corresponding to each optimization objective To optimize the index number of the target, Let Gaussian kernel function be expressed as: , Distance is the independent variable. It is an exponential function. The width of the Gaussian kernel.

[0010] Furthermore, the multi-target wolf pack algorithm in step S3 includes population initialization, adaptive grouping, roaming behavior, summoning and besieging behavior, and population update operation.

[0011] Furthermore, the specific process for population initialization includes: exist A predetermined number of artificial wolves are randomly generated in a dimensional space. The position of each artificial wolf corresponds to a set of fused data. The values ​​of each dimension of each artificial wolf's position are constrained within the value search boundary. Maximum iteration count, maximum walk count, distance decision factor, step size factor, and update ratio factor are set. Among these, the th... The location of the artificial wolf is represented as , For the first Artificial wolves in the first The location of 4-dimensional space, , , For the search space dimension, This represents the total number of artificial wolves.

[0012] Furthermore, the specific process of adaptive grouping includes: Based on the function value information of the current population in the solution space, select several wolves and generate several wolf groups with these wolves as leaders. Add the remaining artificial wolves dominated by each wolf after non-dominance sorting to the corresponding wolf group. If there is a situation where a single artificial wolf is dominated by multiple wolves at the same time, then assign the artificial wolf to the nearest wolf group.

[0013] Furthermore, the specific process of the wandering behavior includes: The number of preset walking directions is set. The artificial wolf searches for prey in several walking directions. The five-dimensional position is updated with a fixed walking step size and the objective function fitness is calculated. The odor concentration in all directions is compared with the objective function fitness. After recording the prey odor concentration in each walking direction, the optimal walking direction is selected to move forward or stay still. The artificial wolf's position is updated. If the odor concentration perceived by the artificial wolf during the walking process exceeds that of the alpha wolf in the group, that is, the objective function fitness of the corresponding artificial wolf is higher than that of the alpha wolf in the group, then the artificial wolf is replaced as the new alpha wolf. The adaptive grouping is re-executed after the walking behavior ends. The search process for the artificial wolf is represented as follows: ; in, Artificial wolves To the After searching in the first wandering direction, in the second The location of 4-dimensional space, For the first time to wander Artificial wolves in the first The location of 4-dimensional space, For the direction of travel index, , To preset the number of travel directions, In the first The step length of walking in 3D space.

[0014] Furthermore, the specific process of the summoning and siege actions includes: After the wandering behavior ends, the alpha wolf summons the artificial wolves in the group. After receiving the summons, the artificial wolves approach the alpha wolf's position with a running stride. If the distance between the artificial wolf and the alpha wolf during the summoning process is less than or equal to the encirclement radius, the encirclement behavior is switched to, and the optimal process parameters are finely searched with the encirclement stride. The position update formula for the summoning and siege actions is: ; in, Indicates the first During the next iteration, the artificial wolf was at... The position in the dimensional space is the updated position. Indicates the first During the next iteration, the artificial wolf was at... The location of 4-dimensional space, and Indicates in The stride length for a raid and the stride length for an attack in 3D space. Indicates the first During the next iteration, the alpha wolf was in the... The location of 4-dimensional space, For the artificial wolf and the alpha wolf in Distance in 1-dimensional space, For the siege radius, The coefficients are random numbers in the interval [-1, 1]. This indicates the process of updating the artificial wolf's position during the summoning action. This indicates the location update process of the artificial wolf during the siege. The step size relationship between the roaming, summoning, and siege behaviors is expressed as follows: ; in, For walking stride length, For a long stride, To surround and attack the stride length.

[0015] Furthermore, the specific process of the population update operation includes: After the wolf pack finishes its wandering, summoning, and besieging behaviors, all artificial wolves are sorted according to Pareto level and crowding distance. The artificial wolf with the worst objective function fitness and the artificial wolf with the best objective function fitness in the population are selected. The artificial wolf with the worst objective function fitness is eliminated, and new artificial wolf individuals are generated based on the best artificial wolf to supplement the population. The formula for generating the new artificial wolf individual is expressed as follows: , ; in, This indicates the newly created artificial wolf individual. This represents the optimal artificial wolf. A random number between [-1, 1] This represents the number of artificial wolves eliminated.

[0016] Furthermore, the multi-objective wolf pack algorithm in step S3 also includes a judgment operation: judging whether the population has reached the maximum number of iterations. If the maximum number of iterations has been reached, the non-dominated solutions in the population are output; otherwise, the adaptive grouping operation is continued.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention employs a multi-objective wolf pack algorithm to optimize three mutually constraining indicators: insertion loss, end-face deformation rate, and end-face return loss. Unlike manual debugging and single-objective algorithms, this invention can stably control the coupler insertion loss range and solve the problem of end-face collapse in hollow fiber fusion splices. 2. This invention automatically completes parameter iterative optimization through a multi-objective wolf pack algorithm, replacing manual trial and error, thereby shortening parameter debugging time and improving the efficiency of coupler R&D and mass production. 3. The output of this invention is a standardized optimal fusion splicing parameter combination, avoiding parameter deviations caused by manual operation, so that the fusion splicing effect remains stable under different batches of optical fibers and different working conditions, making it easy to adapt to different scenarios such as UHV trunk links and ordinary power branch links, and has good versatility. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method steps of the present invention.

[0019] Figure 2 This is a schematic diagram of the Pareto front for multi-objective optimization of hollow-solid fiber fusion splicing according to the present invention.

[0020] Figure 3 This is a schematic diagram of the system architecture of the present invention.

[0021] Figure labeling notes: Variable limitation module 100, multi-objective function module 200, wolf pack algorithm iteration module 300, and process execution module 400. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0023] To address the stringent requirements for fiber optic transmission delay and reliability in new power systems and ultra-high voltage communication networks, hollow-core fiber (HCF) offers significant advantages in air-guided light transmission. However, it needs to be fused with existing solid-core fiber for interoperability. Considering that traditional single-objective optimization algorithms cannot balance multiple conflicting indicators and have limited applicability, this invention provides a hollow-core to solid-core fiber fusion coupling method based on a multi-objective wolf pack algorithm. The core of this method is to abstract the hollow-core to solid-core fiber fusion problem into a nonlinear mathematical model, transforming the complex physical process of hollow-core to solid-core fiber fusion into a high-dimensional multi-objective optimization problem. The air hole structure of hollow-core fiber (HCF) is prone to collapse under high-temperature electric arcs, while the mode field mismatch between solid-core fiber (SCF) and hollow-core fiber leads to severe insertion loss. This invention uses a multi-objective wolf pack algorithm to search for the Pareto optimal equilibrium point in the solution space, thereby obtaining the optimal combination of fusion parameters. This optimizes the three mutually constraining indicators of insertion loss, end-face deformation rate, and end-face return loss, avoiding parameter deviations caused by manual operation and achieving stable fusion splicing results for different batches and operating conditions. The specific technical solution is shown below.

[0024] Example 1

[0025] In this embodiment, as Figure 1 As shown, this invention provides a hollow-to-solid fiber fusion coupling method based on a multi-objective wolf pack algorithm, specifically including the following steps: Step S1: Based on the physical coupling mechanism of hollow-core to solid-core fiber fusion splicing, identify three types of dimensional features that affect the fusion splicing quality, and determine five-dimensional decision variables based on the three types of dimensional features. The five-dimensional decision variables include the number of progressive discharges, the main discharge power, the discharge offset position, the fiber skew angle, and the length of the self-focusing fiber. The three types of dimensional features include energy input, thermal field distribution, and geometric matching. Step S2: Multiple sets of fusion data are collected using Latin hypercube sampling. A multi-objective function vector is constructed using radial basis functions combined with Gaussian kernel functions, with insertion loss, end face deformation rate, and end face return loss as optimization objectives. The multi-objective function vector is composed of loss sub-objective functions, end face deformation rate sub-objective functions, and end face return loss sub-objective functions. Each sub-objective function is constructed by fitting the fusion samples through Gaussian kernel radial basis functions to simulate the mapping relationship between the corresponding optimization objectives and the five-dimensional decision variables. Step S3: Using the multi-objective function vector as the test benchmark function, execute the multi-objective wolf pack algorithm for iterative optimization and output the Pareto optimal process parameter solution set. The iterative optimization is based on the multi-objective function vector and is performed in the physical parameter space consisting of five-dimensional decision variables and constrained by the value search boundary. Step S4: Match the optimal fusion splicing parameter combination for the corresponding power communication operating condition from the Pareto optimal process parameter solution set. Use the matched optimal fusion splicing parameters to complete the arc fusion coupling of hollow fiber and solid fiber. Write the optimal fusion splicing parameter combination into the controller of the fusion splicing equipment through the device interface (such as Ethernet or serial port). The controller adjusts the voltage, current, discharge time and displacement platform coordinates of the discharge electrode according to the above parameters, thereby performing the fusion splicing operation on the optical fiber.

[0026] Specifically, in step S1, to ensure that the dimensions of the mathematical model can both cover the core process parameters and avoid redundant calculations, this invention implements a variable extraction process based on a physical coupling mechanism. First, the energy exchange logic of hollow-core to solid-core fiber fusion splicing is analyzed, and three dimensional characteristics affecting the splicing quality are identified: energy input, thermal field distribution, and geometric matching. Then, the number of progressive discharges is extracted through the energy input dimension. With main discharge power Number of gradual discharges With main discharge power Used to control the melting depth of quartz glass, the discharge offset position is extracted through the thermal field distribution dimension. Discharge offset position To achieve asymmetric heating of hollow optical fibers by altering the heat source center, the extraction basis lies in the difference in thermal diffusivity between heterogeneous fibers. Changing the heat source center through offset is the only physical method to achieve asymmetric heating of hollow optical fibers. The fiber cleavage angle is extracted through geometric matching dimensions. With the length of the self-focusing fiber Fiber optic slant angle With the length of the self-focusing fiber The five-dimensional decision variables are used to determine the end-face echo suppression capability and the mode field matching degree between heterogeneous fibers. The search boundary of the five-dimensional decision variables is limited by the hardware limit of the fusion splicing equipment. The extraction logic lies in the mode field overlap integral theory. These two directly determine the end-face echo suppression capability and the mode field matching degree between heterogeneous fibers. Through the aforementioned hollow-to-solid fiber fusion splicing physical coupling mechanism and physical logic, this invention ultimately converges and extracts the infinitely possible fusion splicing process parameters into a five-dimensional decision vector. , as the input benchmark for the mathematical model.

[0027] Specifically, in step S2, since the loss and deformation during the welding process have highly nonlinear and non-explicit characteristics, this invention can construct an analytical mathematical model by collecting sample points and using radial basis functions. This mathematical model transforms the abstract physical problem into a multi-objective function vector. Specifically, it includes: Several sets of fusion splicing data were selected as fusion splicing samples. Each set of fusion splicing data was recorded simultaneously with five-dimensional decision variables and measured values ​​of insertion loss, end face deformation rate, and end face return loss. Substituting the weld sample coordinates, the weight coefficients of each optimization objective, and the linear drift polynomial into the Gaussian kernel radial basis function, we establish the insertion loss sub-objective function, the end face deformation rate sub-objective function, and the end face return loss sub-objective function, respectively. These are combined to form an iteratively computeable multi-objective function vector. Through this modeling method, the originally complex physical welding process is transformed into a definite mathematical test function space, enabling the algorithm to perform high-frequency iterative calculations on this mathematical surface. The multi-objective function vector representation is as follows: , , ; in, For a multi-objective function vector, To minimize the optimization objective, For insertion loss, For end face deformation rate, For end-face return loss, superscript It is the transpose symbol. Five-dimensional decision variables, For Gaussian kernel radial basis functions, This is the serial number of the welded sample. For the first The coordinates corresponding to each welded sample It is a linear drift polynomial. For the first The first welded sample, the first The weight coefficients corresponding to each optimization objective To optimize the index number of the target, Let Gaussian kernel function be expressed as: , Distance is the independent variable. It is an exponential function. The width of the Gaussian kernel.

[0028] Specifically, in step S3, the present invention will construct the above-mentioned... As a benchmark function for testing the multi-objective wolf pack algorithm, the multi-objective wolf pack algorithm was then used for optimization. The multi-objective wolf pack algorithm includes population initialization, adaptive grouping, wandering behavior, summoning and besieging behavior, and population update operation. The population size was set to 40~100. The specific process of population initialization includes: exist Randomly generate a preset number of [number] in dimensional space Each artificial wolf has a location corresponding to a set of fused data. The values ​​of each dimension of each artificial wolf's location are constrained within the value search boundary. Maximum iteration count, maximum walk count, distance decision factor, step size factor, and update ratio factor are set. Among these, the... The location of the artificial wolf is represented as , For the first Artificial wolves in the first The location of 4-dimensional space, , , For the search space dimension, This represents the total number of artificial wolves; The specific process of adaptive grouping includes: Based on the function value information of the current population in the solution space, select several wolves and generate several wolf groups with several wolves as leaders. Add the remaining artificial wolves dominated by each wolf after non-dominance sorting to the corresponding wolf group. If there is a situation where a single artificial wolf is dominated by multiple wolves at the same time, then assign the artificial wolf to the nearest wolf group. The specific process of wandering behavior includes: Set the preset number of wandering directions, artificial wolf To several ( The artificial wolf searches for prey in each direction, updates its five-dimensional position with a fixed walking step size, and calculates the objective function fitness. It compares the odor concentration in all directions with the objective function fitness, records the prey odor concentration in each direction, selects the optimal walking direction to move forward or remain stationary, and updates the artificial wolf's position. If the artificial wolf perceives an odor concentration higher than that of the alpha wolf in the group during the walking process, i.e., the objective function fitness of the corresponding artificial wolf is higher than that of the alpha wolf in the group, then the artificial wolf is replaced as the new alpha wolf, and adaptive grouping is re-executed after the walking behavior ends. The search process for artificial wolves can be represented as follows: ; in, Artificial wolf To the After searching in the first wandering direction, in the second The location of 4-dimensional space, For the first time to wander Artificial wolves in the first The location of 4-dimensional space, For the direction of travel index, , To preset the number of travel directions, In the first The step length for walking in 3D space; The specific process of summoning and siege actions includes: After the roaming behavior ends, the alpha wolf summons the artificial wolves in its group. Upon receiving the summons, the artificial wolves approach the alpha wolf's location with a sprinting stride. If the summoning process occurs... With the alpha wolf When the distance between them is less than or equal to the siege radius, the siege behavior is switched to siege behavior, and the optimal process parameters are finely searched with the siege step size; The location update formula for summoning and siege actions is: ; in, Indicates the first During the next iteration, the artificial wolf was at... The position in the dimensional space is the updated position. Indicates the first During the next iteration, the artificial wolf was at... The location of 4-dimensional space, and Indicates in The stride length for a raid and the stride length for an attack in 3D space. Indicates the first During the next iteration, the alpha wolf was in the... The location of 4-dimensional space, For the artificial wolf and the alpha wolf in Distance in 1-dimensional space, For the siege radius, For the random coefficients in the interval [-1, 1], the upper part of the above formula... This describes the location update process of the summoned artificial wolf, part two. This indicates the location update process of the artificial wolf during the siege. Because the artificial wolf is farther from the alpha wolf during the summoning process, its stride length is longer during this phase. Conversely, during the encirclement process, the artificial wolf is closer to the alpha wolf, resulting in a shorter stride length. Therefore, the stride length relationship between the roaming, summoning, and encirclement behaviors can be expressed as follows: ; in, For walking stride length, For a long stride length, To surround and attack the stride length; The specific process of population update operations includes: After the wolf pack ceases its wandering, summoning, and attacking behaviors, all artificial wolves are ranked according to Pareto level and crowding distance, and the wolves with the worst objective function fitness in the population are selected. Artificial wolves and the optimal fitness objective function Artificial wolves are selected, and the one with the worst fitness as the objective function is eliminated. Artificial wolves were created, and the optimal artificial wolf was generated. A new artificial wolf individual was added to the population; The formula for generating the new artificial wolf individual is expressed as follows: , ; in, This indicates the newly created artificial wolf individual. This represents the optimal artificial wolf. A random number between [-1, 1] The number of artificial wolves eliminated; It also includes a judgment operation: determine whether the population has reached the maximum number of iterations. If the maximum number of iterations has been reached, output the non-dominated solutions in the population; otherwise, continue to execute the adaptive grouping operation.

[0029] To further demonstrate the application value of the hollow-solid fiber fusion coupling method based on the multi-objective wolf pack algorithm in this invention, the entire process of mathematical modeling and optimization in this invention is described in detail below, taking the fusion coupling of an anti-resonant hollow fiber and a conventional solid communication fiber as an example.

[0030] Step S1: Determine five-dimensional decision variables based on three types of features, and limit the search boundary of the five-dimensional decision variables according to the hardware limitations of the welding equipment: the number of progressive discharges. The range of values ​​is Second, main discharge power The range of values ​​is W, discharge offset position The range of values ​​is mm, fiber optic skewing angle for Self-focusing fiber length The range of values ​​is mm.

[0031] Step S2: Thirty initial experimental sample points were selected using Latin hypercube sampling. Table 1 shows the observation data of three representative experimental sample points. (See Table 1 below.) The data pairs are substituted into the radial basis function framework, and a multi-objective function is constructed by combining the radial basis function with the Gaussian kernel function, with insertion loss, end face deformation rate and end face return loss as optimization objectives. Table 1 - Schematic diagram of experimental sample observation data

[0032] With insertion loss For example, the calculation logic is as follows: using a Gaussian kernel ,set up Through matrix operations , This is the distance matrix between sample points. To determine the corresponding loss values, 30 weighting coefficients were calculated. Finally, the insertion loss is obtained. The explicit mathematical expression for it is: , ; Similarly, establish the sub-objective function of end face deformation rate. and end-face return loss sub-objective function At this point, the fiber optic splicing process, which was originally impossible to express with a formula, has been successfully transformed into three definite, continuously differentiable mathematical surface functions.

[0033] Step S3: Use the above multi-objective function vector as the test environment for the multi-objective wolf pack algorithm to perform iterations, specifically including: Initialize the population: In Randomly generate a preset number of [number] in dimensional space Each artificial wolf has a location corresponding to a set of fused data. The values ​​of each dimension of each artificial wolf's location are constrained within the value search boundary, and a maximum number of iterations is set. Maximum number of walks Distance criterion factor Step size factor and update the scaling factor ; Adaptive grouping: An adaptive grouping strategy is executed on the initialized population. Based on the function value information of the current population in the solution space, several alpha wolves are selected. The best individual in each group is defined as the alpha wolf. Several alpha wolf groups are generated with several alpha wolves as leaders. The remaining artificial wolves dominated by each alpha wolf after non-dominance sorting are added to the corresponding alpha wolf group. If there is a situation where a single artificial wolf is dominated by multiple alpha wolves at the same time, the artificial wolf is assigned to the nearest alpha wolf group. Roaming behavior: After grouping, individuals other than the alpha wolf will move towards their surroundings. The artificial wolf roams in one direction. The location during the movement is Select The best position among the possible directions is chosen as the new position for the artificial wolf. If, during its movement, the artificial wolf finds a prey with a higher scent concentration than the alpha wolf in the group, then the artificial wolf is replaced as the new alpha wolf. This process continues until the maximum number of moves is reached. After the walking behavior ends, adaptive grouping is re-executed, and the next behavior mode is entered; Summoning Behavior: After the roaming behavior ends, the alpha wolf in each group summons the other artificial wolves in the group, and the surrounding artificial wolves respond accordingly. The artificial wolf initiates a charge towards the alpha wolf. During the charge, if the artificial wolf detects a prey scent concentration superior to that of the alpha wolf in its pack, it will initiate the summons in place of the alpha wolf. During the charge, if the artificial wolf's position relative to the alpha wolf... Less than Then, a siege will begin; Encirclement behavior: After a sprint, the artificial wolf is close to its pack's alpha wolf and perceives the alpha wolf's location as the prey's position. The other artificial wolves then launch a siege attack on the prey. Encirclement behavior can be used... express; Population Update: After the wolf pack finishes its wandering, summoning, and siege behaviors, a population update based on crowding distance is performed, resulting in culling. One artificial wolf, with Join new One artificial wolf; Judgment operation: Determine whether the population has reached the maximum number of iterations. If it has reached the maximum number of iterations, output the non-dominated solutions in the population; otherwise, continue to execute the adaptive grouping operation. After optimization using a multi-objective wolf pack algorithm, the Pareto optimal set of process parameters is obtained, such as... Figure 2 As shown, Figure 2 This is a scatter plot of the three-dimensional Pareto Front, derived from multi-objective optimization simulation results of the hollow-to-solid fiber fusion splicing process. In multi-objective optimization, the various optimization objectives are often contradictory and cannot all be optimal simultaneously. The Pareto Front is the set of all non-dominated solutions. The X-axis represents insertion loss. (dB), insertion loss The insertion loss of the fiber optic splice is represented by a value that is as small as possible, indicating low optical transmission loss. The Y-axis represents the end-face deformation rate. This refers to the percentage deformation of the fiber structure after fusion splicing. A smaller value is better, as smaller deformation indicates better fusion quality. The Z-axis represents the end-face return loss. The higher the value, the better, as it indicates higher end-face return loss and less reflected light. Figure 2 Each scatter point represents a candidate combination of welding parameters, and the color represents the evaluation dimension (objective function fitness value), used to distinguish the overall quality of the solution. The redder the color, the higher the fitness value of the objective function corresponding to that solution.

[0034] Step S4: Select a set of welding parameters from the Pareto optimal process parameter solution set, expressed as: ; The above optimal solution The data was fed into the actuator of the fusion welding equipment to obtain the final measurement results. After the fusion was completed, the measured data are as follows: Insertion loss for End face deformation rate for End-face return loss for .

[0035] In summary, the multi-objective wolf pack algorithm proposed in this invention can find the optimal balance among three conflicting indicators: insertion loss, end face deformation rate, and end face return loss. The solution set is evenly and continuously distributed in space, proving that the multi-objective wolf pack algorithm has excellent global optimization capabilities and can provide diversified parameter schemes for different power communication link requirements.

[0036] Example 2 The difference from Example 1 is that, as in Example 2, ... Figure 3 As shown, the present invention also provides a hollow-to-solid fiber fusion splicing coupling system based on a multi-objective wolf pack algorithm, used to execute the hollow-to-solid fiber fusion splicing coupling method based on a multi-objective wolf pack algorithm shown in Embodiment 1 above. The system includes a variable limiting module 100, a multi-objective function module 200, a wolf pack algorithm iteration module 300, and a process execution module 400. Among them, the variable constraint module 100 is used to determine and constrain the five-dimensional decision variables that affect the fusion splicing quality of hollow-core and solid-core optical fibers and their value search boundaries; The multi-objective function module 200 is used to construct a multi-objective function vector and constraints covering insertion loss, end face deformation rate, and end face return loss as optimization objectives; The wolf pack algorithm iteration module 300 is based on the multi-objective wolf pack algorithm. It performs multi-objective search and screening on the solution set of welding parameters and outputs the Pareto optimal solution set of process parameters. The process execution module 400 selects the optimal combination of welding parameters according to the power link scenario and controls the welding equipment to complete the self-focusing beam expansion and progressive discharge welding process.

[0037] In summary, this invention focuses on the hollow-to-solid fiber fusion splicing coupling process, extracts five-dimensional decision variables, constructs a multi-objective function vector based on experimental data, and uses a multi-objective wolf pack algorithm to iterate through the entire process of population initialization, adaptive grouping, wandering behavior, summoning and besieging behavior, and population update to obtain the Pareto optimal process parameter solution set. The Pareto scatter plot clearly shows the trade-off relationship between insertion loss, end-face deformation rate, and end-face return loss. The optimal combination of fusion splicing parameters can be selected according to different power communication operating conditions. This invention effectively improves the shortcomings of existing fusion splicing parameter debugging, such as difficulty in simultaneously optimizing multiple indicators, limited adaptability to various scenarios, and poor stability in mass production, and has good application prospects.

[0038] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A hollow-to-solid fiber fusion splicing coupling method based on a multi-objective wolf pack algorithm, characterized in that, Specifically, the following steps are included: Step S1: Based on the physical coupling mechanism of hollow-core to solid-core fiber fusion splicing, identify three types of dimensional features that affect the splicing quality, and determine five-dimensional decision variables based on the three types of dimensional features; Step S2: Use Latin hypercube sampling to collect multiple sets of fusion splicing data, and use radial basis functions combined with Gaussian kernel functions to construct a multi-objective function vector with insertion loss, end face deformation rate and end face return loss as optimization objectives; Step S3: Using the multi-objective function vector as the test benchmark function, execute the multi-objective wolf pack algorithm iteratively to find the optimal solution set of Pareto optimal process parameters; Step S4: Match the optimal fusion splicing parameter combination for the corresponding power communication operating condition from the Pareto optimal process parameter solution set, and use the matched optimal fusion splicing parameters to complete the arc fusion coupling of hollow fiber and solid fiber. The five-dimensional decision variables include the number of progressive discharges, the main discharge power, the discharge offset position, the fiber optic skewing angle, and the length of the self-focusing fiber. The three-dimensional features include energy input, thermal field distribution, and geometric matching.

2. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 1, characterized in that, The process for determining the five-dimensional decision variables in step S1 is as follows: The number of progressive discharges and the main discharge power are extracted through the energy input dimension. These are used to control the melting depth of the quartz glass. The discharge offset position is extracted through the thermal field distribution dimension. This offset position is used to change the center of the heat source and achieve asymmetric heating of the hollow fiber. The fiber cleaving angle and the self-focusing fiber length are extracted through the geometric matching dimension. These are used to determine the end-face echo suppression capability and the mode field matching degree between heterogeneous fibers. The search boundary for the values ​​of the five-dimensional decision variables is limited according to the hardware limits of the fusion splicing equipment.

3. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 1, characterized in that, Step S2 specifically includes: Several sets of fusion splicing data were selected as fusion splicing samples. Each set of fusion splicing data was recorded simultaneously with five-dimensional decision variables and measured values ​​of insertion loss, end face deformation rate, and end face return loss. Substitute the coordinates of the fusion sample, the weight coefficients of each optimization objective, and the linear drift polynomial into the Gaussian kernel radial basis function to establish the insertion loss sub-objective function, the end face deformation rate sub-objective function, and the end face return loss sub-objective function respectively. Combine them to form an iteratively computeable multi-objective function vector. The multi-objective function vector is represented as follows: ， ， ； in, For a multi-objective function vector, To minimize the optimization objective, For insertion loss, For end face deformation rate, For end-face return loss, superscript It is the transpose symbol. Five-dimensional decision variables, For Gaussian kernel radial basis functions, This is the serial number of the welded sample. For the first The coordinates corresponding to each welded sample It is a linear drift polynomial. For the first The first welded sample, the first The weight coefficients corresponding to each optimization objective To optimize the index number of the target, Let Gaussian kernel function be expressed as: , Distance is the independent variable. It is an exponential function. is the width of the Gaussian kernel.

4. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 1, characterized in that, The multi-target wolf pack algorithm in step S3 includes population initialization, adaptive grouping, roaming behavior, summoning and besieging behavior, and population update operation.

5. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 4, characterized in that, The specific process for population initialization includes: exist A predetermined number of artificial wolves are randomly generated in a 3D space. The position of each artificial wolf corresponds to a set of fused data. The values ​​of each dimension of each artificial wolf's position are constrained within the value search boundary. Maximum iteration count, maximum walk count, distance decision factor, step size factor, and update ratio factor are set. Among these, the th... The location of the artificial wolf is represented as , For the first Artificial wolves in the first The location of 4-dimensional space, , , For the search space dimension, This represents the total number of artificial wolves.

6. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 5, characterized in that, The specific process of adaptive grouping includes: Based on the function value information of the current population in the solution space, select several wolves and generate several wolf groups with these wolves as leaders. Add the remaining artificial wolves dominated by each wolf after non-dominance sorting to the corresponding wolf group. If there is a situation where a single artificial wolf is dominated by multiple wolves at the same time, then assign the artificial wolf to the nearest wolf group.

7. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 6, characterized in that, The specific process of the wandering behavior includes: The number of preset walking directions is set. The artificial wolf searches for prey in several walking directions. The five-dimensional position is updated with a fixed walking step size and the objective function fitness is calculated. The odor concentration in all directions is compared with the objective function fitness. After recording the prey odor concentration in each walking direction, the optimal walking direction is selected to move forward or stay still. The artificial wolf's position is updated. If the odor concentration perceived by the artificial wolf during the walking process exceeds that of the alpha wolf in the group, that is, the objective function fitness of the corresponding artificial wolf is higher than that of the alpha wolf in the group, then the artificial wolf is replaced as the new alpha wolf. The adaptive grouping is re-executed after the walking behavior ends. The search process for the artificial wolf is represented as follows: ; in, Artificial wolves To the After searching in the first wandering direction, in the second The location of 4-dimensional space, For the first time to wander Artificial wolves in the first The location of 4-dimensional space, For the direction of travel index, , To preset the number of travel directions, In the first The step length of walking in 3D space.

8. The hollow-to-solid fiber fusion splicing coupling method based on the multi-objective wolf pack algorithm according to claim 7, characterized in that, The specific process for the summoning and siege actions includes: After the wandering behavior ends, the alpha wolf summons the artificial wolves in the group. After receiving the summons, the artificial wolves approach the alpha wolf's position with a running stride. If the distance between the artificial wolf and the alpha wolf during the summoning process is less than or equal to the encirclement radius, the encirclement behavior is switched to, and the optimal process parameters are finely searched with the encirclement stride. The position update formula for the summoning and siege actions is: ； in, Indicates the first During the next iteration, the artificial wolf was at... The position in the dimensional space is the updated position. Indicates the first During the next iteration, the artificial wolf was at... The location of 4-dimensional space, and Indicates in The stride length for a raid and the stride length for an attack in 3D space. Indicates the first During the next iteration, the alpha wolf was in the... The location of 4-dimensional space, For the artificial wolf and the alpha wolf in Distance in 1-dimensional space, For the siege radius, The coefficients are random numbers in the interval [-1, 1]. This indicates the process of updating the artificial wolf's position during the summoning action. This indicates the location update process of the artificial wolf during the siege. The step size relationship between the roaming, summoning, and siege behaviors is expressed as follows: ; in, For walking stride length, For a long stride length, To surround and attack the stride length.

9. The hollow-core to solid-core optical fiber fusion coupling method based on the multi-objective wolf pack algorithm according to claim 8, characterized in that, The specific process of the population update operation includes: After the wolf pack finishes its wandering, summoning, and besieging behaviors, all artificial wolves are sorted according to Pareto level and crowding distance. The artificial wolf with the worst objective function fitness and the artificial wolf with the best objective function fitness in the population are selected. The artificial wolf with the worst objective function fitness is eliminated, and new artificial wolf individuals are generated based on the best artificial wolf to supplement the population. The formula for generating the new artificial wolf individual is expressed as follows: , ; in, This indicates the newly created artificial wolf individual. This represents the optimal artificial wolf. A random number between [-1, 1] This represents the number of artificial wolves eliminated.

10. The hollow-to-solid fiber fusion coupling method based on the multi-objective wolf pack algorithm according to claim 4, characterized in that, The multi-objective wolf pack algorithm in step S3 also includes a judgment operation: judging whether the population has reached the maximum number of iterations. If the maximum number of iterations has been reached, the non-dominated solutions in the population are output; otherwise, the adaptive grouping operation is continued.

Images